Original Research

Approximately a century ago, Francis Peabody taught that “the secret of the care of the patient is in caring for the patient.”¹ His advice remains true today. Despite the advent of novel diagnostic tests, technologically sophisticated interventional procedures, and life-saving medications, perhaps the most important skill a bedside clinician can use is the ability to connect with patients.

The literature on patient-physician interaction is vast^2-11 and generally indicates that exemplary bedside clinicians are able to interact well with patients by being competent, trustworthy, personable, empathetic, and effective communicators. “Etiquette-based medicine,” first proposed by Kahn,¹² emphasizes the importance of certain behaviors from physicians, such as introducing yourself and explaining your role, shaking hands, sitting down when speaking to patients, and asking open-ended questions.

Yet, improving patient-physician interactions remains necessary. A recent systematic review reported that almost half of the reviewed studies on the patient-physician relationship published between 2000 and 2014 conveyed the idea that the patient-physician relationship is deteriorating.¹³

As part of a broader study to understand the behaviors and approaches of exemplary inpatient attending physicians,^14-16 we examined how 12 carefully selected physicians interacted with their patients during inpatient teaching rounds.

METHODS

Overview

We conducted a multisite study using an exploratory, qualitative approach to inquiry, which has been described previously.^14-16 Our primary purpose was to study the attributes and behaviors of outstanding general medicine attendings in the setting of inpatient rounds. The focus of this article is on the attendings’ interactions with patients.

We used a modified snowball sampling approach¹⁷ to identify 12 exemplary physicians. First, we contacted individuals throughout the United States who were known to the principal investigator (S.S.) and asked for suggestions of excellent clinician educators (also referred to as attendings) for potential inclusion in the study. In addition to these personal contacts, other individuals unknown to the investigative team were contacted and asked to provide suggestions for attendings to include in the study. Specifically, the US News & World Report 2015 Top Medical Schools: Research Rankings,¹⁸ which are widely used to represent the best U.S. hospitals, were reviewed in an effort to identify attendings from a broad range of medical schools. Using this list, we identified other medical schools that were in the top 25 and were not already represented. We contacted the division chiefs of general internal (or hospital) medicine, chairs and chiefs of departments of internal medicine, and internal medicine residency program directors from these medical schools and asked for recommendations of attendings from both within and outside their institutions whom they considered to be great inpatient teachers.

This sampling method resulted in 59 potential participants. An internet search was conducted on each potential participant to obtain further information about the individuals and their institutions. Both personal characteristics (medical education, training, and educational awards) and organizational characteristics (geographic location, hospital size and affiliation, and patient population) were considered so that a variety of organizations and backgrounds were represented. Through this process, the list was narrowed to 16 attendings who were contacted to participate in the study, of which 12 agreed. The number of attendings examined was appropriate because saturation of metathemes can occur in as little as 6 interviews, and data saturation occurs at 12 interviews.¹⁹ The participants were asked to provide a list of their current learners (ie, residents and medical students) and 6 to 10 former learners to contact for interviews and focus groups.

Data Collection

Observations

Two researchers conducted the one-day site visits. One was a physician (S.S.) and the other a medical anthropologist (M.H.), and both have extensive experience in qualitative methods. The only exception was the site visit at the principal investigator’s own institution, which was conducted by the medical anthropologist and a nonpracticing physician who was unknown to the participants. The team structure varied slightly among different institutions but in general was composed of 1 attending, 1 senior medical resident, 1 to 2 interns, and approximately 2 medical students. Each site visit began with observing the attendings (n = 12) and current learners (n = 57) on morning rounds, which included their interactions with patients. These observations lasted approximately 2 to 3 hours. The observers took handwritten field notes, paying particular attention to group interactions, teaching approaches, and patient interactions. The observers stood outside the medical team circle and remained silent during rounds so as to be unobtrusive to the teams’ discussions. The observers discussed and compared their notes after each site visit.

 

Interviews and Focus Groups

The research team also conducted individual, semistructured interviews with the attendings (n = 12), focus groups with their current teams (n = 46), and interviews or focus groups with their former learners (n = 26). Current learners were asked open-ended questions about their roles on the teams, their opinions of the attendings, and the care the attendings provide to their patients. Because they were observed during rounds, the researchers asked for clarification about specific interactions observed during the teaching rounds. Depending on availability and location, former learners either participated in in-person focus groups or interviews on the day of the site visit, or in a later telephone interview. All interviews and focus groups were audio recorded and transcribed.

This study was deemed to be exempt from regulation by the University of Michigan Institutional Review Board. All participants were informed that their participation was completely voluntary and that they could refuse to answer any question.

Data Analysis

Data were analyzed using a thematic analysis approach,²⁰ which involves reading through the data to identify patterns (and create codes) that relate to behaviors, experiences, meanings, and activities. The patterns are then grouped into themes to help further explain the findings.²¹ The research team members (S.S. and M.H.) met after the first site visit and developed initial ideas about meanings and possible patterns. One team member (M.H.) read all the transcripts from the site visit and, based on the data, developed a codebook to be used for this study. This process was repeated after every site visit, and the coding definitions were refined as necessary. All transcripts were reviewed to apply any new codes when they developed. NVivo^® 10 software (QSR International, Melbourne, Australia) was used to assist with the qualitative data analysis.

To ensure consistency and identify relationships between codes, code reports listing all the data linked to a specific code were generated after all the field notes and transcripts were coded. Once verified, codes were grouped based on similarities and relationships into prominent themes related to physician-patient interactions by 2 team members (S.S. and M.H.), though all members reviewed them and concurred.

RESULTS

A total of 12 attending physicians participated (Table 1). The participants were from hospitals located throughout the U.S. and included both university-affiliated hospitals and Veterans Affairs medical centers. We observed the attending physicians interact with more than 100 patients, with 3 major patient interaction themes emerging. Table 2 lists key approaches for effective patient-physician interactions based on the study findings.

Care for the Patient’s Well-Being

The attendings we observed appeared to openly care for their patients’ well-being and were focused on the patients’ wants and needs. We noted that attendings were generally very attentive to the patients’ comfort. For example, we observed one attending sending the senior resident to find the patient’s nurse in order to obtain additional pain medications. The attending said to the patient several times, “I’m sorry you’re in so much pain.” When the team was leaving, she asked the intern to stay with the patient until the medications had been administered.

Learners noticed when an attending physician was especially skilled at demonstrating empathy and patient-centered care. While education on rounds was emphasized, patient connection was the priority. One learner described the following: “… he really is just so passionate about patient care and has so much empathy, really. And I will tell you, of all my favorite things about him, that is one of them...”

The attendings we observed could also be considered patient advocates, ensuring that patients received superb care. As one learner said about an attending who was attempting to have his patient listed for a liver transplant, “He is the biggest advocate for the patient that I have ever seen.” Regarding the balance between learning biomedical concepts and advocacy, another learner noted the following: “… there is always a teaching aspect, but he always makes sure that everything is taken care of for the patient…”

Building rapport creates and sustains bonds between people. Even though most of the attendings we observed primarily cared for hospitalized patients and had little long-term continuity with them, the attendings tended to take special care to talk with their patients about topics other than medicine to form a bond. This bonding between attending and patient was appreciated by learners. “Probably the most important thing I learned about patient care would be taking the time and really developing that relationship with patients,” said one of the former learners we interviewed. “There’s a question that he asks to a lot of our patients,” one learner told us, “especially our elderly patients, that [is], ‘What’s the most memorable moment in your life?’ So, he asks that question, and patient[s] open up and will share.”

The attendings often used touch to further solidify their relationships with their patients. We observed one attending who would touch her patients’ arms or knees when she was talking with them. Another attending would always shake the patient’s hand when leaving. Another attending would often lay his hand on the patient’s shoulder and help the patient sit up during the physical examination. Such humanistic behavior was noticed by learners. “She does a lot of comforting touch, particularly at the end of an exam,” said a current learner.

 

Consideration of the “Big Picture”

Our exemplary attendings kept the “big picture” (that is, the patient’s overall medical and social needs) in clear focus. They behaved in a way to ensure that the patients understood the key points of their care and explained so the patients and families could understand. A current learner said, “[The attending] really makes sure that the patient understands what’s going on. And she always asks them, ‘What do you understand, what do you know, how can we fill in any blanks?’ And that makes the patient really involved in their own care, which I think is important.” This reflection was supported by direct observations. Attendings posed the following questions at the conclusion of patient interactions: “Tell me what you know.” “Tell me what our plan is.” “What did the lung doctors tell you yesterday?” These questions, which have been termed “teach-back” and are crucial for health literacy, were not meant to quiz the patient but rather to ensure the patient and family understood the plan.

We noticed that the attendings effectively explained clinical details and the plan of care to the patient while avoiding medical jargon. The following is an example of one interaction with a patient: “You threw up and created a tear in the food tube. Air got from that into the middle of the chest, not into the lungs. Air isn’t normally there. If it is just air, the body will reabsorb [it]... But we worry about bacteria getting in with the air. We need to figure out if it is an infection. We’re still trying to figure it out. Hang in there with us.” One learner commented, “… since we do bedside presentations, he has a great way of translating our gibberish, basically, to real language the patient understands.”

Finally, the attendings anticipated what patients would need in the outpatient setting. We observed that attendings stressed what the next steps would be during transitions of care. As one learner put it, “But he also thinks ahead; what do they need as an outpatient?” Another current learner commented on how another attending always asked about the social situations of his patients stating, “And then there is the social part of it. So, he is very much interested [in] where do they live? What is their support system? So, I think it has been a very holistic approach to patient care.”

Respect for the Patient

The attendings we observed were steadfastly respectful toward patients. As one attending told us, “The patient’s room is sacred space, and it’s a privilege for us to be there. And if we don’t earn that privilege, then we don’t get to go there.” We observed that the attendings generally referred to the patient as Mr. or Ms. (last name) rather than the patient’s first name unless the patient insisted. We also noticed that many of the attendings would introduce the team members to the patients or ask each member to introduce himself or herself. They also tended to leave the room and patient the way they were found, for example, by pushing the patient’s bedside table so that it was back within his or her reach or placing socks back onto the patient’s feet.

We noted that many of our attendings used appropriate humor with patients and families. As one learner explained, “I think Dr. [attending] makes most of our patients laugh during rounds. I don’t know if you noticed, but he really puts a smile on their face[s] whenever he walks in. … Maybe it would catch them off guard the first day, but after that, they are so happy to see him.”

Finally, we noticed that several of our attendings made sure to meet the patient at eye level during discussions by either kneeling or sitting on a chair. One of the attendings put it this way: “That’s a horrible power dynamic when you’re an inpatient and you’re sick and someone’s standing over you telling you things, and I like to be able to make eye contact with people, and often times that requires me to kneel down or to sit on a stool or to sit on the bed. … I feel like you’re able to connect with the people in a much better way…” Learners viewed this behavior favorably. As one told us, “[The attending] gets down to their level and makes sure that all of their questions are answered. So that is one thing that other attendings don’t necessarily do.”

DISCUSSION

In our national, qualitative study of 12 exemplary attending physicians, we found that these clinicians generally exhibited the following behaviors with patients. First, they were personable and caring and made significant attempts to connect with their patients. This occasionally took the form of using touch to comfort patients. Second, they tended to seek the “big picture” and tried to understand what patients would need upon hospital discharge. They communicated plans clearly to patients and families and inquired if those plans were understood. Finally, they showed respect toward their patients without fail. Such respect took many forms but included leaving the patient and room exactly as they were found and speaking with patients at eye level.

 

Our findings are largely consistent with other key studies in this field. Not surprisingly, the attendings we observed adhered to the major suggestions that Branch and colleagues² put forth more than 15 years ago to improve the teaching of the humanistic dimension of the patient-physician relationship. Examples include greeting the patient, introducing team members and explaining each person’s role, asking open-ended questions, providing patient education, placing oneself at the same level as the patient, using appropriate touch, and being respectful. Weissmann et al.²² also found similar themes in their study of teaching physicians at 4 universities from 2003 to 2004. In that study, role-modeling was the primary method used by physician educators to teach the humanistic aspects of medical care, including nonverbal communication (eg, touch and eye contact), demonstration of respect, and building a personal connection with the patients.²²In a focus group-based study performed at a teaching hospital in Boston, Ramani and Orlander²³ concluded that both participating teachers and learners considered the patient’s bedside as a valuable venue to learn humanistic skills. Unfortunately, they also noted that there has been a decline in bedside teaching related to various factors, including documentation requirements and electronic medical records.²³ Our attendings all demonstrated the value of teaching at a patient’s bedside. Not only could physical examination skills be demonstrated but role-modeling of interpersonal skills could be observed by learners.

Block and colleagues²⁴ observed 29 interns in 732 patient encounters in 2 Baltimore training programs using Kahn’s “etiquette-based medicine” behaviors as a guide.¹² They found that interns introduced themselves 40% of the time, explained their role 37% of the time, touched patients on 65% of visits (including as part of the physical examination), asked open-ended questions 75% of the time, and sat down with patients during only 9% of visits.²⁴ Tackett et al.⁷ observed 24 hospitalists who collectively cared for 226 unique patients in 3 Baltimore-area hospitals. They found that each of the following behaviors was performed less than 30% of the time: explains role in care, shakes hand, and sits down.⁷ However, our attendings appeared to adhere to these behaviors to a much higher extent, though we did not quantify the interactions. This lends support to the notion that effective patient-physician interactions are the foundation of great teaching.

The attendings we observed (most of whom are inpatient based) tended to the contextual issues of the patients, such as their home environments and social support. Our exemplary physicians did what they could to ensure that patients received the appropriate follow-up care upon discharge.

Our study has important limitations. First, it was conducted in a limited number of US hospitals. The institutions represented were generally large, research-intensive, academic medical centers. Therefore, our findings may not apply to settings that are different from the hospitals studied. Second, our study included only 12 attendings and their learners, which may also limit the study’s generalizability. Third, we focused exclusively on teaching within general medicine rounds. Thus, our findings may not be generalizable to other subspecialties. Fourth, attendings were selected through a nonexhaustive method, increasing the potential for selection bias. However, the multisite design, the modified snowball sampling, and the inclusion of several types of institutions in the final participant pool introduced diversity to the final list. Former-learner responses were subject to recall bias. Finally, the study design is susceptible to observer bias. Attempts to reduce this included the diversity of the observers (ie, both a clinician and a nonclinician, the latter of whom was unfamiliar with medical education) and review of the data and coding by multiple research team members to ensure validity. Although we cannot discount the potential role of a Hawthorne effect on our data collection, the research team attempted to mitigate this by standing apart from the care teams and remaining unobtrusive during observations.

Limitations notwithstanding, we believe that our multisite study is important given the longstanding imperative to improve patient-physician interactions. We found empirical support for behaviors proposed by Branch and colleagues² and Kahn¹² in order to enhance these relationships. While others have studied attendings and their current learners,²² we add to the literature by also examining former learners’ perspectives on how the attendings’ teaching and role-modeling have created and sustained a lasting impact. The key findings of our national, qualitative study (care for the patient’s well-being, consideration of the “big picture,” and respect for the patient) can be readily adopted and honed by physicians to improve their interactions with hospitalized patients.

Acknowledgments

The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the US Department of Veterans Affairs.

 

Funding

Dr. Saint provided funding for this study using a University of Michigan endowment.

Disclosure

The authors declare no conflicts of interest.

Approximately a century ago, Francis Peabody taught that “the secret of the care of the patient is in caring for the patient.”¹ His advice remains true today. Despite the advent of novel diagnostic tests, technologically sophisticated interventional procedures, and life-saving medications, perhaps the most important skill a bedside clinician can use is the ability to connect with patients.

The literature on patient-physician interaction is vast^2-11 and generally indicates that exemplary bedside clinicians are able to interact well with patients by being competent, trustworthy, personable, empathetic, and effective communicators. “Etiquette-based medicine,” first proposed by Kahn,¹² emphasizes the importance of certain behaviors from physicians, such as introducing yourself and explaining your role, shaking hands, sitting down when speaking to patients, and asking open-ended questions.

Yet, improving patient-physician interactions remains necessary. A recent systematic review reported that almost half of the reviewed studies on the patient-physician relationship published between 2000 and 2014 conveyed the idea that the patient-physician relationship is deteriorating.¹³

As part of a broader study to understand the behaviors and approaches of exemplary inpatient attending physicians,^14-16 we examined how 12 carefully selected physicians interacted with their patients during inpatient teaching rounds.

METHODS

Overview

We conducted a multisite study using an exploratory, qualitative approach to inquiry, which has been described previously.^14-16 Our primary purpose was to study the attributes and behaviors of outstanding general medicine attendings in the setting of inpatient rounds. The focus of this article is on the attendings’ interactions with patients.

We used a modified snowball sampling approach¹⁷ to identify 12 exemplary physicians. First, we contacted individuals throughout the United States who were known to the principal investigator (S.S.) and asked for suggestions of excellent clinician educators (also referred to as attendings) for potential inclusion in the study. In addition to these personal contacts, other individuals unknown to the investigative team were contacted and asked to provide suggestions for attendings to include in the study. Specifically, the US News & World Report 2015 Top Medical Schools: Research Rankings,¹⁸ which are widely used to represent the best U.S. hospitals, were reviewed in an effort to identify attendings from a broad range of medical schools. Using this list, we identified other medical schools that were in the top 25 and were not already represented. We contacted the division chiefs of general internal (or hospital) medicine, chairs and chiefs of departments of internal medicine, and internal medicine residency program directors from these medical schools and asked for recommendations of attendings from both within and outside their institutions whom they considered to be great inpatient teachers.

This sampling method resulted in 59 potential participants. An internet search was conducted on each potential participant to obtain further information about the individuals and their institutions. Both personal characteristics (medical education, training, and educational awards) and organizational characteristics (geographic location, hospital size and affiliation, and patient population) were considered so that a variety of organizations and backgrounds were represented. Through this process, the list was narrowed to 16 attendings who were contacted to participate in the study, of which 12 agreed. The number of attendings examined was appropriate because saturation of metathemes can occur in as little as 6 interviews, and data saturation occurs at 12 interviews.¹⁹ The participants were asked to provide a list of their current learners (ie, residents and medical students) and 6 to 10 former learners to contact for interviews and focus groups.

Data Collection

Observations

Two researchers conducted the one-day site visits. One was a physician (S.S.) and the other a medical anthropologist (M.H.), and both have extensive experience in qualitative methods. The only exception was the site visit at the principal investigator’s own institution, which was conducted by the medical anthropologist and a nonpracticing physician who was unknown to the participants. The team structure varied slightly among different institutions but in general was composed of 1 attending, 1 senior medical resident, 1 to 2 interns, and approximately 2 medical students. Each site visit began with observing the attendings (n = 12) and current learners (n = 57) on morning rounds, which included their interactions with patients. These observations lasted approximately 2 to 3 hours. The observers took handwritten field notes, paying particular attention to group interactions, teaching approaches, and patient interactions. The observers stood outside the medical team circle and remained silent during rounds so as to be unobtrusive to the teams’ discussions. The observers discussed and compared their notes after each site visit.

 

Interviews and Focus Groups

The research team also conducted individual, semistructured interviews with the attendings (n = 12), focus groups with their current teams (n = 46), and interviews or focus groups with their former learners (n = 26). Current learners were asked open-ended questions about their roles on the teams, their opinions of the attendings, and the care the attendings provide to their patients. Because they were observed during rounds, the researchers asked for clarification about specific interactions observed during the teaching rounds. Depending on availability and location, former learners either participated in in-person focus groups or interviews on the day of the site visit, or in a later telephone interview. All interviews and focus groups were audio recorded and transcribed.

This study was deemed to be exempt from regulation by the University of Michigan Institutional Review Board. All participants were informed that their participation was completely voluntary and that they could refuse to answer any question.

Data Analysis

Data were analyzed using a thematic analysis approach,²⁰ which involves reading through the data to identify patterns (and create codes) that relate to behaviors, experiences, meanings, and activities. The patterns are then grouped into themes to help further explain the findings.²¹ The research team members (S.S. and M.H.) met after the first site visit and developed initial ideas about meanings and possible patterns. One team member (M.H.) read all the transcripts from the site visit and, based on the data, developed a codebook to be used for this study. This process was repeated after every site visit, and the coding definitions were refined as necessary. All transcripts were reviewed to apply any new codes when they developed. NVivo^® 10 software (QSR International, Melbourne, Australia) was used to assist with the qualitative data analysis.

To ensure consistency and identify relationships between codes, code reports listing all the data linked to a specific code were generated after all the field notes and transcripts were coded. Once verified, codes were grouped based on similarities and relationships into prominent themes related to physician-patient interactions by 2 team members (S.S. and M.H.), though all members reviewed them and concurred.

RESULTS

A total of 12 attending physicians participated (Table 1). The participants were from hospitals located throughout the U.S. and included both university-affiliated hospitals and Veterans Affairs medical centers. We observed the attending physicians interact with more than 100 patients, with 3 major patient interaction themes emerging. Table 2 lists key approaches for effective patient-physician interactions based on the study findings.

Care for the Patient’s Well-Being

The attendings we observed appeared to openly care for their patients’ well-being and were focused on the patients’ wants and needs. We noted that attendings were generally very attentive to the patients’ comfort. For example, we observed one attending sending the senior resident to find the patient’s nurse in order to obtain additional pain medications. The attending said to the patient several times, “I’m sorry you’re in so much pain.” When the team was leaving, she asked the intern to stay with the patient until the medications had been administered.

Learners noticed when an attending physician was especially skilled at demonstrating empathy and patient-centered care. While education on rounds was emphasized, patient connection was the priority. One learner described the following: “… he really is just so passionate about patient care and has so much empathy, really. And I will tell you, of all my favorite things about him, that is one of them...”

The attendings we observed could also be considered patient advocates, ensuring that patients received superb care. As one learner said about an attending who was attempting to have his patient listed for a liver transplant, “He is the biggest advocate for the patient that I have ever seen.” Regarding the balance between learning biomedical concepts and advocacy, another learner noted the following: “… there is always a teaching aspect, but he always makes sure that everything is taken care of for the patient…”

Building rapport creates and sustains bonds between people. Even though most of the attendings we observed primarily cared for hospitalized patients and had little long-term continuity with them, the attendings tended to take special care to talk with their patients about topics other than medicine to form a bond. This bonding between attending and patient was appreciated by learners. “Probably the most important thing I learned about patient care would be taking the time and really developing that relationship with patients,” said one of the former learners we interviewed. “There’s a question that he asks to a lot of our patients,” one learner told us, “especially our elderly patients, that [is], ‘What’s the most memorable moment in your life?’ So, he asks that question, and patient[s] open up and will share.”

The attendings often used touch to further solidify their relationships with their patients. We observed one attending who would touch her patients’ arms or knees when she was talking with them. Another attending would always shake the patient’s hand when leaving. Another attending would often lay his hand on the patient’s shoulder and help the patient sit up during the physical examination. Such humanistic behavior was noticed by learners. “She does a lot of comforting touch, particularly at the end of an exam,” said a current learner.

 

Consideration of the “Big Picture”

Our exemplary attendings kept the “big picture” (that is, the patient’s overall medical and social needs) in clear focus. They behaved in a way to ensure that the patients understood the key points of their care and explained so the patients and families could understand. A current learner said, “[The attending] really makes sure that the patient understands what’s going on. And she always asks them, ‘What do you understand, what do you know, how can we fill in any blanks?’ And that makes the patient really involved in their own care, which I think is important.” This reflection was supported by direct observations. Attendings posed the following questions at the conclusion of patient interactions: “Tell me what you know.” “Tell me what our plan is.” “What did the lung doctors tell you yesterday?” These questions, which have been termed “teach-back” and are crucial for health literacy, were not meant to quiz the patient but rather to ensure the patient and family understood the plan.

We noticed that the attendings effectively explained clinical details and the plan of care to the patient while avoiding medical jargon. The following is an example of one interaction with a patient: “You threw up and created a tear in the food tube. Air got from that into the middle of the chest, not into the lungs. Air isn’t normally there. If it is just air, the body will reabsorb [it]... But we worry about bacteria getting in with the air. We need to figure out if it is an infection. We’re still trying to figure it out. Hang in there with us.” One learner commented, “… since we do bedside presentations, he has a great way of translating our gibberish, basically, to real language the patient understands.”

Finally, the attendings anticipated what patients would need in the outpatient setting. We observed that attendings stressed what the next steps would be during transitions of care. As one learner put it, “But he also thinks ahead; what do they need as an outpatient?” Another current learner commented on how another attending always asked about the social situations of his patients stating, “And then there is the social part of it. So, he is very much interested [in] where do they live? What is their support system? So, I think it has been a very holistic approach to patient care.”

Respect for the Patient

The attendings we observed were steadfastly respectful toward patients. As one attending told us, “The patient’s room is sacred space, and it’s a privilege for us to be there. And if we don’t earn that privilege, then we don’t get to go there.” We observed that the attendings generally referred to the patient as Mr. or Ms. (last name) rather than the patient’s first name unless the patient insisted. We also noticed that many of the attendings would introduce the team members to the patients or ask each member to introduce himself or herself. They also tended to leave the room and patient the way they were found, for example, by pushing the patient’s bedside table so that it was back within his or her reach or placing socks back onto the patient’s feet.

We noted that many of our attendings used appropriate humor with patients and families. As one learner explained, “I think Dr. [attending] makes most of our patients laugh during rounds. I don’t know if you noticed, but he really puts a smile on their face[s] whenever he walks in. … Maybe it would catch them off guard the first day, but after that, they are so happy to see him.”

Finally, we noticed that several of our attendings made sure to meet the patient at eye level during discussions by either kneeling or sitting on a chair. One of the attendings put it this way: “That’s a horrible power dynamic when you’re an inpatient and you’re sick and someone’s standing over you telling you things, and I like to be able to make eye contact with people, and often times that requires me to kneel down or to sit on a stool or to sit on the bed. … I feel like you’re able to connect with the people in a much better way…” Learners viewed this behavior favorably. As one told us, “[The attending] gets down to their level and makes sure that all of their questions are answered. So that is one thing that other attendings don’t necessarily do.”

DISCUSSION

In our national, qualitative study of 12 exemplary attending physicians, we found that these clinicians generally exhibited the following behaviors with patients. First, they were personable and caring and made significant attempts to connect with their patients. This occasionally took the form of using touch to comfort patients. Second, they tended to seek the “big picture” and tried to understand what patients would need upon hospital discharge. They communicated plans clearly to patients and families and inquired if those plans were understood. Finally, they showed respect toward their patients without fail. Such respect took many forms but included leaving the patient and room exactly as they were found and speaking with patients at eye level.

 

Our findings are largely consistent with other key studies in this field. Not surprisingly, the attendings we observed adhered to the major suggestions that Branch and colleagues² put forth more than 15 years ago to improve the teaching of the humanistic dimension of the patient-physician relationship. Examples include greeting the patient, introducing team members and explaining each person’s role, asking open-ended questions, providing patient education, placing oneself at the same level as the patient, using appropriate touch, and being respectful. Weissmann et al.²² also found similar themes in their study of teaching physicians at 4 universities from 2003 to 2004. In that study, role-modeling was the primary method used by physician educators to teach the humanistic aspects of medical care, including nonverbal communication (eg, touch and eye contact), demonstration of respect, and building a personal connection with the patients.²²In a focus group-based study performed at a teaching hospital in Boston, Ramani and Orlander²³ concluded that both participating teachers and learners considered the patient’s bedside as a valuable venue to learn humanistic skills. Unfortunately, they also noted that there has been a decline in bedside teaching related to various factors, including documentation requirements and electronic medical records.²³ Our attendings all demonstrated the value of teaching at a patient’s bedside. Not only could physical examination skills be demonstrated but role-modeling of interpersonal skills could be observed by learners.

Block and colleagues²⁴ observed 29 interns in 732 patient encounters in 2 Baltimore training programs using Kahn’s “etiquette-based medicine” behaviors as a guide.¹² They found that interns introduced themselves 40% of the time, explained their role 37% of the time, touched patients on 65% of visits (including as part of the physical examination), asked open-ended questions 75% of the time, and sat down with patients during only 9% of visits.²⁴ Tackett et al.⁷ observed 24 hospitalists who collectively cared for 226 unique patients in 3 Baltimore-area hospitals. They found that each of the following behaviors was performed less than 30% of the time: explains role in care, shakes hand, and sits down.⁷ However, our attendings appeared to adhere to these behaviors to a much higher extent, though we did not quantify the interactions. This lends support to the notion that effective patient-physician interactions are the foundation of great teaching.

The attendings we observed (most of whom are inpatient based) tended to the contextual issues of the patients, such as their home environments and social support. Our exemplary physicians did what they could to ensure that patients received the appropriate follow-up care upon discharge.

Our study has important limitations. First, it was conducted in a limited number of US hospitals. The institutions represented were generally large, research-intensive, academic medical centers. Therefore, our findings may not apply to settings that are different from the hospitals studied. Second, our study included only 12 attendings and their learners, which may also limit the study’s generalizability. Third, we focused exclusively on teaching within general medicine rounds. Thus, our findings may not be generalizable to other subspecialties. Fourth, attendings were selected through a nonexhaustive method, increasing the potential for selection bias. However, the multisite design, the modified snowball sampling, and the inclusion of several types of institutions in the final participant pool introduced diversity to the final list. Former-learner responses were subject to recall bias. Finally, the study design is susceptible to observer bias. Attempts to reduce this included the diversity of the observers (ie, both a clinician and a nonclinician, the latter of whom was unfamiliar with medical education) and review of the data and coding by multiple research team members to ensure validity. Although we cannot discount the potential role of a Hawthorne effect on our data collection, the research team attempted to mitigate this by standing apart from the care teams and remaining unobtrusive during observations.

Limitations notwithstanding, we believe that our multisite study is important given the longstanding imperative to improve patient-physician interactions. We found empirical support for behaviors proposed by Branch and colleagues² and Kahn¹² in order to enhance these relationships. While others have studied attendings and their current learners,²² we add to the literature by also examining former learners’ perspectives on how the attendings’ teaching and role-modeling have created and sustained a lasting impact. The key findings of our national, qualitative study (care for the patient’s well-being, consideration of the “big picture,” and respect for the patient) can be readily adopted and honed by physicians to improve their interactions with hospitalized patients.

Acknowledgments

The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the US Department of Veterans Affairs.

 

Funding

Dr. Saint provided funding for this study using a University of Michigan endowment.

Disclosure

The authors declare no conflicts of interest.

Approximately a century ago, Francis Peabody taught that “the secret of the care of the patient is in caring for the patient.”¹ His advice remains true today. Despite the advent of novel diagnostic tests, technologically sophisticated interventional procedures, and life-saving medications, perhaps the most important skill a bedside clinician can use is the ability to connect with patients.

The literature on patient-physician interaction is vast^2-11 and generally indicates that exemplary bedside clinicians are able to interact well with patients by being competent, trustworthy, personable, empathetic, and effective communicators. “Etiquette-based medicine,” first proposed by Kahn,¹² emphasizes the importance of certain behaviors from physicians, such as introducing yourself and explaining your role, shaking hands, sitting down when speaking to patients, and asking open-ended questions.

Yet, improving patient-physician interactions remains necessary. A recent systematic review reported that almost half of the reviewed studies on the patient-physician relationship published between 2000 and 2014 conveyed the idea that the patient-physician relationship is deteriorating.¹³

As part of a broader study to understand the behaviors and approaches of exemplary inpatient attending physicians,^14-16 we examined how 12 carefully selected physicians interacted with their patients during inpatient teaching rounds.

METHODS

Overview

We conducted a multisite study using an exploratory, qualitative approach to inquiry, which has been described previously.^14-16 Our primary purpose was to study the attributes and behaviors of outstanding general medicine attendings in the setting of inpatient rounds. The focus of this article is on the attendings’ interactions with patients.

We used a modified snowball sampling approach¹⁷ to identify 12 exemplary physicians. First, we contacted individuals throughout the United States who were known to the principal investigator (S.S.) and asked for suggestions of excellent clinician educators (also referred to as attendings) for potential inclusion in the study. In addition to these personal contacts, other individuals unknown to the investigative team were contacted and asked to provide suggestions for attendings to include in the study. Specifically, the US News & World Report 2015 Top Medical Schools: Research Rankings,¹⁸ which are widely used to represent the best U.S. hospitals, were reviewed in an effort to identify attendings from a broad range of medical schools. Using this list, we identified other medical schools that were in the top 25 and were not already represented. We contacted the division chiefs of general internal (or hospital) medicine, chairs and chiefs of departments of internal medicine, and internal medicine residency program directors from these medical schools and asked for recommendations of attendings from both within and outside their institutions whom they considered to be great inpatient teachers.

This sampling method resulted in 59 potential participants. An internet search was conducted on each potential participant to obtain further information about the individuals and their institutions. Both personal characteristics (medical education, training, and educational awards) and organizational characteristics (geographic location, hospital size and affiliation, and patient population) were considered so that a variety of organizations and backgrounds were represented. Through this process, the list was narrowed to 16 attendings who were contacted to participate in the study, of which 12 agreed. The number of attendings examined was appropriate because saturation of metathemes can occur in as little as 6 interviews, and data saturation occurs at 12 interviews.¹⁹ The participants were asked to provide a list of their current learners (ie, residents and medical students) and 6 to 10 former learners to contact for interviews and focus groups.

Data Collection

Observations

Two researchers conducted the one-day site visits. One was a physician (S.S.) and the other a medical anthropologist (M.H.), and both have extensive experience in qualitative methods. The only exception was the site visit at the principal investigator’s own institution, which was conducted by the medical anthropologist and a nonpracticing physician who was unknown to the participants. The team structure varied slightly among different institutions but in general was composed of 1 attending, 1 senior medical resident, 1 to 2 interns, and approximately 2 medical students. Each site visit began with observing the attendings (n = 12) and current learners (n = 57) on morning rounds, which included their interactions with patients. These observations lasted approximately 2 to 3 hours. The observers took handwritten field notes, paying particular attention to group interactions, teaching approaches, and patient interactions. The observers stood outside the medical team circle and remained silent during rounds so as to be unobtrusive to the teams’ discussions. The observers discussed and compared their notes after each site visit.

 

Interviews and Focus Groups

The research team also conducted individual, semistructured interviews with the attendings (n = 12), focus groups with their current teams (n = 46), and interviews or focus groups with their former learners (n = 26). Current learners were asked open-ended questions about their roles on the teams, their opinions of the attendings, and the care the attendings provide to their patients. Because they were observed during rounds, the researchers asked for clarification about specific interactions observed during the teaching rounds. Depending on availability and location, former learners either participated in in-person focus groups or interviews on the day of the site visit, or in a later telephone interview. All interviews and focus groups were audio recorded and transcribed.

This study was deemed to be exempt from regulation by the University of Michigan Institutional Review Board. All participants were informed that their participation was completely voluntary and that they could refuse to answer any question.

Data Analysis

Data were analyzed using a thematic analysis approach,²⁰ which involves reading through the data to identify patterns (and create codes) that relate to behaviors, experiences, meanings, and activities. The patterns are then grouped into themes to help further explain the findings.²¹ The research team members (S.S. and M.H.) met after the first site visit and developed initial ideas about meanings and possible patterns. One team member (M.H.) read all the transcripts from the site visit and, based on the data, developed a codebook to be used for this study. This process was repeated after every site visit, and the coding definitions were refined as necessary. All transcripts were reviewed to apply any new codes when they developed. NVivo^® 10 software (QSR International, Melbourne, Australia) was used to assist with the qualitative data analysis.

To ensure consistency and identify relationships between codes, code reports listing all the data linked to a specific code were generated after all the field notes and transcripts were coded. Once verified, codes were grouped based on similarities and relationships into prominent themes related to physician-patient interactions by 2 team members (S.S. and M.H.), though all members reviewed them and concurred.

RESULTS

A total of 12 attending physicians participated (Table 1). The participants were from hospitals located throughout the U.S. and included both university-affiliated hospitals and Veterans Affairs medical centers. We observed the attending physicians interact with more than 100 patients, with 3 major patient interaction themes emerging. Table 2 lists key approaches for effective patient-physician interactions based on the study findings.

Care for the Patient’s Well-Being

The attendings we observed appeared to openly care for their patients’ well-being and were focused on the patients’ wants and needs. We noted that attendings were generally very attentive to the patients’ comfort. For example, we observed one attending sending the senior resident to find the patient’s nurse in order to obtain additional pain medications. The attending said to the patient several times, “I’m sorry you’re in so much pain.” When the team was leaving, she asked the intern to stay with the patient until the medications had been administered.

Learners noticed when an attending physician was especially skilled at demonstrating empathy and patient-centered care. While education on rounds was emphasized, patient connection was the priority. One learner described the following: “… he really is just so passionate about patient care and has so much empathy, really. And I will tell you, of all my favorite things about him, that is one of them...”

The attendings we observed could also be considered patient advocates, ensuring that patients received superb care. As one learner said about an attending who was attempting to have his patient listed for a liver transplant, “He is the biggest advocate for the patient that I have ever seen.” Regarding the balance between learning biomedical concepts and advocacy, another learner noted the following: “… there is always a teaching aspect, but he always makes sure that everything is taken care of for the patient…”

Building rapport creates and sustains bonds between people. Even though most of the attendings we observed primarily cared for hospitalized patients and had little long-term continuity with them, the attendings tended to take special care to talk with their patients about topics other than medicine to form a bond. This bonding between attending and patient was appreciated by learners. “Probably the most important thing I learned about patient care would be taking the time and really developing that relationship with patients,” said one of the former learners we interviewed. “There’s a question that he asks to a lot of our patients,” one learner told us, “especially our elderly patients, that [is], ‘What’s the most memorable moment in your life?’ So, he asks that question, and patient[s] open up and will share.”

The attendings often used touch to further solidify their relationships with their patients. We observed one attending who would touch her patients’ arms or knees when she was talking with them. Another attending would always shake the patient’s hand when leaving. Another attending would often lay his hand on the patient’s shoulder and help the patient sit up during the physical examination. Such humanistic behavior was noticed by learners. “She does a lot of comforting touch, particularly at the end of an exam,” said a current learner.

 

Consideration of the “Big Picture”

Our exemplary attendings kept the “big picture” (that is, the patient’s overall medical and social needs) in clear focus. They behaved in a way to ensure that the patients understood the key points of their care and explained so the patients and families could understand. A current learner said, “[The attending] really makes sure that the patient understands what’s going on. And she always asks them, ‘What do you understand, what do you know, how can we fill in any blanks?’ And that makes the patient really involved in their own care, which I think is important.” This reflection was supported by direct observations. Attendings posed the following questions at the conclusion of patient interactions: “Tell me what you know.” “Tell me what our plan is.” “What did the lung doctors tell you yesterday?” These questions, which have been termed “teach-back” and are crucial for health literacy, were not meant to quiz the patient but rather to ensure the patient and family understood the plan.

We noticed that the attendings effectively explained clinical details and the plan of care to the patient while avoiding medical jargon. The following is an example of one interaction with a patient: “You threw up and created a tear in the food tube. Air got from that into the middle of the chest, not into the lungs. Air isn’t normally there. If it is just air, the body will reabsorb [it]... But we worry about bacteria getting in with the air. We need to figure out if it is an infection. We’re still trying to figure it out. Hang in there with us.” One learner commented, “… since we do bedside presentations, he has a great way of translating our gibberish, basically, to real language the patient understands.”

Finally, the attendings anticipated what patients would need in the outpatient setting. We observed that attendings stressed what the next steps would be during transitions of care. As one learner put it, “But he also thinks ahead; what do they need as an outpatient?” Another current learner commented on how another attending always asked about the social situations of his patients stating, “And then there is the social part of it. So, he is very much interested [in] where do they live? What is their support system? So, I think it has been a very holistic approach to patient care.”

Respect for the Patient

The attendings we observed were steadfastly respectful toward patients. As one attending told us, “The patient’s room is sacred space, and it’s a privilege for us to be there. And if we don’t earn that privilege, then we don’t get to go there.” We observed that the attendings generally referred to the patient as Mr. or Ms. (last name) rather than the patient’s first name unless the patient insisted. We also noticed that many of the attendings would introduce the team members to the patients or ask each member to introduce himself or herself. They also tended to leave the room and patient the way they were found, for example, by pushing the patient’s bedside table so that it was back within his or her reach or placing socks back onto the patient’s feet.

We noted that many of our attendings used appropriate humor with patients and families. As one learner explained, “I think Dr. [attending] makes most of our patients laugh during rounds. I don’t know if you noticed, but he really puts a smile on their face[s] whenever he walks in. … Maybe it would catch them off guard the first day, but after that, they are so happy to see him.”

Finally, we noticed that several of our attendings made sure to meet the patient at eye level during discussions by either kneeling or sitting on a chair. One of the attendings put it this way: “That’s a horrible power dynamic when you’re an inpatient and you’re sick and someone’s standing over you telling you things, and I like to be able to make eye contact with people, and often times that requires me to kneel down or to sit on a stool or to sit on the bed. … I feel like you’re able to connect with the people in a much better way…” Learners viewed this behavior favorably. As one told us, “[The attending] gets down to their level and makes sure that all of their questions are answered. So that is one thing that other attendings don’t necessarily do.”

DISCUSSION

In our national, qualitative study of 12 exemplary attending physicians, we found that these clinicians generally exhibited the following behaviors with patients. First, they were personable and caring and made significant attempts to connect with their patients. This occasionally took the form of using touch to comfort patients. Second, they tended to seek the “big picture” and tried to understand what patients would need upon hospital discharge. They communicated plans clearly to patients and families and inquired if those plans were understood. Finally, they showed respect toward their patients without fail. Such respect took many forms but included leaving the patient and room exactly as they were found and speaking with patients at eye level.

 

Our findings are largely consistent with other key studies in this field. Not surprisingly, the attendings we observed adhered to the major suggestions that Branch and colleagues² put forth more than 15 years ago to improve the teaching of the humanistic dimension of the patient-physician relationship. Examples include greeting the patient, introducing team members and explaining each person’s role, asking open-ended questions, providing patient education, placing oneself at the same level as the patient, using appropriate touch, and being respectful. Weissmann et al.²² also found similar themes in their study of teaching physicians at 4 universities from 2003 to 2004. In that study, role-modeling was the primary method used by physician educators to teach the humanistic aspects of medical care, including nonverbal communication (eg, touch and eye contact), demonstration of respect, and building a personal connection with the patients.²²In a focus group-based study performed at a teaching hospital in Boston, Ramani and Orlander²³ concluded that both participating teachers and learners considered the patient’s bedside as a valuable venue to learn humanistic skills. Unfortunately, they also noted that there has been a decline in bedside teaching related to various factors, including documentation requirements and electronic medical records.²³ Our attendings all demonstrated the value of teaching at a patient’s bedside. Not only could physical examination skills be demonstrated but role-modeling of interpersonal skills could be observed by learners.

Block and colleagues²⁴ observed 29 interns in 732 patient encounters in 2 Baltimore training programs using Kahn’s “etiquette-based medicine” behaviors as a guide.¹² They found that interns introduced themselves 40% of the time, explained their role 37% of the time, touched patients on 65% of visits (including as part of the physical examination), asked open-ended questions 75% of the time, and sat down with patients during only 9% of visits.²⁴ Tackett et al.⁷ observed 24 hospitalists who collectively cared for 226 unique patients in 3 Baltimore-area hospitals. They found that each of the following behaviors was performed less than 30% of the time: explains role in care, shakes hand, and sits down.⁷ However, our attendings appeared to adhere to these behaviors to a much higher extent, though we did not quantify the interactions. This lends support to the notion that effective patient-physician interactions are the foundation of great teaching.

The attendings we observed (most of whom are inpatient based) tended to the contextual issues of the patients, such as their home environments and social support. Our exemplary physicians did what they could to ensure that patients received the appropriate follow-up care upon discharge.

Our study has important limitations. First, it was conducted in a limited number of US hospitals. The institutions represented were generally large, research-intensive, academic medical centers. Therefore, our findings may not apply to settings that are different from the hospitals studied. Second, our study included only 12 attendings and their learners, which may also limit the study’s generalizability. Third, we focused exclusively on teaching within general medicine rounds. Thus, our findings may not be generalizable to other subspecialties. Fourth, attendings were selected through a nonexhaustive method, increasing the potential for selection bias. However, the multisite design, the modified snowball sampling, and the inclusion of several types of institutions in the final participant pool introduced diversity to the final list. Former-learner responses were subject to recall bias. Finally, the study design is susceptible to observer bias. Attempts to reduce this included the diversity of the observers (ie, both a clinician and a nonclinician, the latter of whom was unfamiliar with medical education) and review of the data and coding by multiple research team members to ensure validity. Although we cannot discount the potential role of a Hawthorne effect on our data collection, the research team attempted to mitigate this by standing apart from the care teams and remaining unobtrusive during observations.

Limitations notwithstanding, we believe that our multisite study is important given the longstanding imperative to improve patient-physician interactions. We found empirical support for behaviors proposed by Branch and colleagues² and Kahn¹² in order to enhance these relationships. While others have studied attendings and their current learners,²² we add to the literature by also examining former learners’ perspectives on how the attendings’ teaching and role-modeling have created and sustained a lasting impact. The key findings of our national, qualitative study (care for the patient’s well-being, consideration of the “big picture,” and respect for the patient) can be readily adopted and honed by physicians to improve their interactions with hospitalized patients.

Acknowledgments

The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the US Department of Veterans Affairs.

 

Funding

Dr. Saint provided funding for this study using a University of Michigan endowment.

Disclosure

The authors declare no conflicts of interest.

Sepsis affects over 1 million Americans annually, resulting in significant morbidity, mortality, and costs for hospitalized patients.^1-4 There is an increasing interest in policy-oriented approaches to improving sepsis care at both the state and national levels.^5,6 The most prominent policy is the Centers for Medicare and Medicaid Services (CMS) Sepsis CMS Core (SEP-1) program, which was formally implemented in October 2015; the program mandates that hospitals report their compliance with a variety of sepsis treatment processes (Table 1). Academic quality experts generally applaud the increased attention to sepsis but are concerned that the measure’s design and specifications advance beyond the existing evidence base.^7,8 However, remarkably little is known about how front-line hospital quality officials perceive the program and how they are responding or not responding, to the new requirements. This knowledge gap is a critical barrier to evaluating the program’s practical impact on sepsis treatment and outcomes.

We therefore sought to understand hospital stakeholders’ perceptions of the SEP-1 program in general as well as their characterization of their local hospitals’ responses to the program. We were specifically interested in obtaining a focused perspective on the policy and hospitals’ responses to the policy rather than individual physicians’ attitudes regarding sepsis care protocols, which are complex and may be independent from the policy itself.⁹ We used a qualitative research approach designed to generate both a deep and broad understanding of how hospitals are responding to SEP-1 requirements, including the resources required to implement their responses.

METHODS

Study Design, Setting, and Subjects

We conducted a qualitative study by using semistructured telephone interviews with hospital quality officers in the United States. We targeted hospital quality officers because they are in a position to provide overarching insights into hospitals’ perceptions of and responses to the SEP-1 program. We enrolled quality officers at general, short-stay, nonfederal acute care hospitals because those are the hospitals to which the SEP-1 program applies. We generated a stratified random sample of hospitals by using 2013 data from Medicare’s Healthcare Cost and Reporting Information System (HCRIS) database.¹⁰ We stratified by size (greater than or less than 200 total beds), teaching status (presence or absence of any resident physician trainees), and ownership (for-profit vs nonprofit), creating 8 mutually exclusive strata. This sampling frame was designed to ensure representativeness from a broad range of hospital types, not to enable comparisons across hospital types, which is outside the scope of qualitative research.

Within strata, we contacted hospitals in a random order by phone using the primary number listed in the HCRIS database. We asked the hospital operator to connect us to the chief quality officer or an appropriate alternative hospital administrator with knowledge of hospital quality-improvement activities. We limited participation to 1 respondent per hospital. We did not offer any specific incentives for participation.

The study was approved by the University of Pittsburgh Institutional Review Board with a waiver of signed informed consent.

Data Collection

Interviews were conducted by a trained research coordinator between February 2016 and October 2016. Interviews were conducted concurrently with data analysis by using a constant comparison approach.¹¹ The constant comparison approach involves the iterative refinement of themes by comparing the existing themes to new data as they emerge during successive interviews. We chose a constant comparison approach because we wanted to systematically describe hospital responses to SEP-1 rather than specifically test individual hypotheses.¹¹ As is typical in qualitative research, we did not set the sample size a priori but instead continued the interviews until we achieved thematic saturation.^12,13

The interview script included a mix of directed and open-ended questions about respondents’ perspectives of and hospital responses to the SEP-1 program. The questions covered the following 4 domains: hospitals’ sepsis quality-improvement initiatives before and after the Medicare reporting program, reception of the hospital responses, the approach to data abstraction and reporting, and the overall impressions of the program and its impact.^6-8,14 We allowed for updates and revisions of the interview guide as necessary to explore any new content and emergent themes. We piloted the interview guide on 2 hospital quality officers at our institution and then revised its structure again after interviews with the initial 6 hospitals. The complete final interview guide is available in the supplemental digital content.

 

Analysis

Interviews were audio recorded, transcribed, and loaded onto a secure server. We used NVivo 11 (QSR International, Cambridge, Massachusetts) for coding and analysis. We iteratively reviewed and thematically analyzed the transcripts for structural content and emergent themes, consistent with established qualitative methods.¹⁵ Three investigators reviewed the initial 20 transcripts and developed the codebook through iterative discussion and consensus. The codes were then organized into themes and subthemes. Subsequently, 1 investigator coded the remaining transcripts. The results are presented as a series of key themes supported by direct quotes from the interviews.

RESULTS

Sample Description

We performed 29 interviews prior to achieving thematic saturation. Each of the 8 strata from the sampling frame was represented by at least 3 hospitals. Hospitals in the final sample were diverse in total bed size, intensive care unit bed capacity, teaching status, and ownership (Table 2). The median interview length was 25 minutes (interquartile range, 20-32 minutes). Respondents included 6 quality coordinators, 6 quality managers, and 11 quality directors, with the remainder holding a variety of other quality-related titles. Most respondents worked in hospital quality departments, although 4 were affiliated with individual clinical departments (eg, emergency medicine and/or critical care services). Of the 9 respondents who reported their professional training, 8 were registered nurses. Eleven respondents reported participating in measure abstraction.

Perspectives on SEP-1

Respondents’ general perspectives on the SEP-1 program are outlined in Table 3, with several key themes emerging. Foremost was the sheer complexity of the measure compounded by its reliance on time-stamped clinical documentation, and in particular, the physical reassessment in individual medical notes. Respondents expressed frustration with the “all-or-none” approach to declaring sepsis treatment a “success,” which they noted was unfair and difficult to justify to their local clinicians. In part because of the time and effort required to comply with the measure and report results to CMS, several respondents noted that the measure is a uniquely burdensome addition to an already-crowded landscape of hospital quality programs. Despite the resources required to adhere to the measures’ standards and report results to CMS, respondents expressed a belief that the increased attention to sepsis is driving positive changes in hospital care and leading to improved patient outcomes.

Responses to SEP-1

Respondents identified several specific ways in which their hospitals responded to the SEP-1 mandate (Table 4), including investments in measurement, planning and coordinating sepsis-specific quality-improvement activities, improving the early identification of patients with sepsis, improving sepsis treatment and measure compliance, and addressing negative attitudes towards the implementation of the SEP-1 program.

Efforts to Collect Data for SEP-1 Reporting

Respondents reported challenges in reliably and validly measuring and reporting data for the SEP-1 program. First, patient identification and the measurement of treatment processes depends largely on manual medical record review, which is subject to variation across coders. This presents a particular challenge because the clinical definition of sepsis itself is in evolution,¹ creating the possibility that treating physicians could identify a given patient as having sepsis or septic shock based on the most up-to-date definitions but not based on the measure’s specifications or vice versa. Second, each case requires up to an hour of manual medical record review and patients who develop sepsis during prolonged hospitalizations can require several hours or more, which is an unprecedented length of time to spend abstracting data for a single measure.

In addressing these measurement challenges, investment in human resources is the rule. No respondent reported automating abstraction of all the SEP-1 data elements, underscoring concerns regarding the measurement burden of the SEP-1 program.^7,8,14 Rather, hospitals with sufficient financial resources frequently employ full-time data abstractors and individuals responsible for ongoing performance feedback, which facilitates the iterative revision of sepsis quality-improvement initiatives. In contrast, hospitals with fewer resources often rely on contracts with third-party vendors, which delays reporting and complicates efforts to use the data for individualized performance improvement.

Efforts to Coordinate Hospital Responses Across Care Teams

Complying with the measure involves the longitudinal coordination of multiple care teams across different units, so planning and executing local hospital responses required interdepartmental and multidisciplinary stakeholder involvement. Respondents were uncertain about the ideal strategy to coordinate these quality-improvement efforts, yielding iterative changes to electronic health records (EHRs), education programs, and data collection methods. This “learning by doing” is necessary because no prior CMS quality measure is as complex as SEP-1 or as varied in the sources of data required to measure and report the results. By requiring hospitals to improve coordination of care throughout the hospital, SEP-1 presents a quality-improvement and measurement challenge that may ultimately drive innovation and better patient care.

 

Efforts to Improve Sepsis Diagnosis

Several hospitals are implementing sepsis screening and alerts to speed sepsis recognition and meet the measure’s time-sensitive treatment requirements. An example of a less-intensive alert is one hospital’s lowering of the threshold for lactate values that are viewed as “critical” (and thus requiring notification of the bedside clinician). Examples of more resource-intensive alerts included electronic screening for vital sign abnormalities that trigger bedside assessment for infection as well as nurse-driven manual sepsis screening tools.

Frequently, these more intensive efforts faced barriers to successful implementation related to the broader issues of performance measurement rather than the specifics of SEP-1. EHRs generally lacked built-in electronic screening capacity, and few hospitals had the resources required for customized EHR modification. Manual screening required nurses to spend time away from direct patient care. For both electronic and manual screening, respondents expressed concern about how these new alerts would fit into a care landscape already inundated with alerts, alarms, and care notifications.^16,17

Efforts to Improve Sepsis Treatment

Many hospitals are implementing sepsis-specific treatment protocols and order sets designed to help meet SEP-1 treatment specifications. In hospitals and health systems with preexisting sepsis quality-improvement efforts, SEP-1 stimulated adaptation and acceleration of their efforts; in hospitals without preexisting sepsis-specific quality improvement, SEP-1 inspired de novo program development and implementation. These programs were wide ranging. Several hospitals implemented a process by which an initially elevated lactate value automates an order for a repeat lactate level, facilitating an assessment of the clinical response to treatment. Other examples include triggers for sepsis-specific treatment protocols and checklists that bedside nurses can begin without initial physician oversight. In 1 hospital, sepsis alerts triggered by emergency medical first responders initiate responses prior to hospital arrival in a manner analogous to prehospital alerts for myocardial infarction and stroke.^18,19

Efforts to implement these protocols encountered several common challenges. Physicians were often resistant to adopting inflexible treatment rules that did not allow them to tailor therapies to individual patients. Furthermore, even protocols and order sets that worked in 1 setting did not necessarily generalize throughout the hospital or health system, reflecting the difficulty in implementing a highly specified measure across diverse treatment environments.

Efforts to Manage Clinician Attitudes Toward SEP-1 Implementation

In addition to addressing clinicians’ behaviors, hospitals sought to address stakeholders’ attitudes when those attitudes created barriers to SEP-1 implementation. First, hospitals frequently faced a lack of buy-in from clinicians who were resistant to the idea of protocolized care in general and who were specifically skeptical that initiatives designed to increase clinical documentation would drive improvements in patient-centered outcomes. Second, respondents had to confront a hierarchical hospital culture, which manifests not only in clinical care, but also in the quality-improvement infrastructure. Many respondents reported that physicians were more receptive to performance feedback from fellow physicians rather than nonphysician quality administrators.

Respondents described a range of approaches to counteract these attitudes. First, hospitals deployed department- and profession-specific “champions” to provide peer-to-peer performance feedback supported by data demonstrating a link between process improvements and patient outcomes. Second, many respondents noted that the addition of new clinical staff, who were often younger and more receptive to new initiatives, could alter a hospital’s quality culture; in smaller hospitals, just a few individuals could significantly alter the dynamic. Finally, when other efforts failed, some respondents indicated that top-down administrative support could persuade resistant individuals to change their approach. However, this solution worked best with employed physicians and was less effective with independent physician groups without direct financial ties to hospital performance. These efforts to overcome negative attitudes toward SEP-1 implementation required individuals’ time and energy, leading to frustration at times and adding to the resources required to comply with the program.

Planning for the Future of SEP-1

Respondents anticipate that performance of the SEP-1 measure will eventually become publicly reported and incorporated into value-based purchasing calculations. Hospitals are therefore seeking greater interaction with CMS as it makes iterative revisions to the measure because respondents expect that their hospitals’ level of performance, rather than just the act of participating, will affect hospital finances. Respondents expressed a desire for more live, interactive educational sessions with CMS moving forward, rather than limiting the opportunities for clarification to online comment forums or statements elsewhere in the public record. In addition, respondents hope that public reporting and pay-for-performance could be delayed to allow more time to work out the “kinks” in measurement and reporting.

DISCUSSION

We conducted semistructured telephone interviews with quality officers in U.S. hospitals in order to understand hospitals’ perceptions of and responses to Medicare’s SEP-1 sepsis quality-reporting program. Hospitals are struggling with the program’s complexity and investing considerable resources in order to iteratively revise their responses to the program. However, they generally believe that the program is bringing much-needed attention to sepsis diagnosis and treatment. These findings have several implications for the SEP-1 measure in particular and for hospital-based quality measurement and pay-for-performance policies in general.

 

First, we demonstrate that SEP-1 consistently requires a substantial investment of resources from hospitals already struggling under the weight of numerous local, state, and national quality-reporting and improvement programs.^14,20,21 In aggregate, these programs can stretch hospitals’ resources to their limit. Respondents universally reported that the SEP-1 program is requiring dedicated staff to meet the data abstraction and reporting requirements as well as multicomponent quality-improvement initiatives. In the absence of well-established roadmaps for improving sepsis care, these sepsis quality-improvement efforts require experimentation and iterative revision, which can contribute to fatigue and frustration among quality officers and clinical staff. This process of innovation inherently involves successes, failures, and the risk of harm and opportunity costs that strain hospital resources.

Second, our study indicates how SEP-1 could exacerbate existing inequalities in our health system. Sepsis incidence and mortality are already higher in medically underserved regions.²² Given the resources required to respond to the SEP-1 program, optimal performance may be beyond the reach of smaller hospitals, or even larger hospitals, whose resources are already stretched to their limits. Public reporting and pay-for-performance can be adisadvantage to hospitals caring for underserved populations.^23,24 To the extent that responding to sepsis-oriented public policy requires resources that certain hospitals cannot access, these policies could exacerbate existing health disparities.

Third, our findings highlight some specific ways that CMS could revise the SEP-1 program to better meet the needs of hospitals and improve outcomes for patients with sepsis. Primarily, although the program’s current specifications take an “all-or-none” approach to treatment success, a more flexible approach, such as a weighted score or composite measure that combines processes and outcomes,^25,26 could allow hospitals to focus their efforts on those components of the bundle with the strongest evidence for improved patient outcomes.²⁷ Second, policy makers need to reconcile the 2 existing clinical definitions for sepsis.^1,28 CMS has already stated its plans to retain the preexisting sepsis definition,²⁹ but this does not change the reality that frontline providers and quality officials face different, and at times conflicting, clinical definitions while caring for patients. Finally, current implementation challenges may support a delay in moving the measure toward public reporting and pay-for-performance. Hospitals are already responding to the measure in a substantial way, providing an opportunity for early quantitative evaluations of the program’s impact that could inform evidence-based revisions to the measure.

Our study has several limitations. First, by interviewing only individual quality officers within each hospital, it is possible that our findings were not representative of the perspectives of other individuals within their hospitals or the hospital as a whole; indeed, to the extent that quality officers “buy in” to quality measurement and reporting, their perspectives on SEP-1 may skew more positive than other hospital staff. Our respondents represented individuals from a range of positions within the quality infrastructure, whereas “hospital quality leaders” are often chief executive officers, chief medical officers, or vice presidents for quality.³⁰ However, by virtue of our purposive sampling approach, we included respondents from a broad range of hospitals and found similar themes across these respondents, supporting the internal validity of our findings. Second, as is inherent in interview-based research, we cannot verify that respondents’ reports of hospital responses to SEP-1 match the actual changes implemented “on the ground.” We are reassured, however, by the fact that many of the perspectives and quality-improvement changes that respondents described align with the opinions and suggestions of academic quality experts, which are informed by clinical experience.^6-8 Third, while respondents believe that hospital responses to SEP-1 are contributing to improvements in treatment and outcomes, we do not yet have robust objective data to support this opinion or to evaluate the association between quality officers’ perspectives and hospital performance. A quantitative evaluation of the clinical impact of SEP-1, as well as the relationship between hospital performance and quality officers’ perspectives on the measure, are important areas for future research.

CONCLUSIONS

In a qualitative study of hospital responses to Medicare’s SEP-1 program, we found that hospitals are implementing changes across a variety of domains and in ways that consistently require dedicated resources. Giving hospitals the flexibility to focus on treatment processes with the most direct impact on patient-centered outcomes might enhance the program’s effectiveness. Future work should quantify the program’s impact and develop novel approaches to data abstraction and quality improvement.

Disclosure

Aside from federal funding, the authors have no conflicts of interest to disclose. The authors received funding from the National Institutes of Health (IJB, F32HL132461) (JMK, K24HL133444). This work was submitted as an abstract to the 2017 American Thoracic Society International Conference, May 2017.

 

Sepsis affects over 1 million Americans annually, resulting in significant morbidity, mortality, and costs for hospitalized patients.^1-4 There is an increasing interest in policy-oriented approaches to improving sepsis care at both the state and national levels.^5,6 The most prominent policy is the Centers for Medicare and Medicaid Services (CMS) Sepsis CMS Core (SEP-1) program, which was formally implemented in October 2015; the program mandates that hospitals report their compliance with a variety of sepsis treatment processes (Table 1). Academic quality experts generally applaud the increased attention to sepsis but are concerned that the measure’s design and specifications advance beyond the existing evidence base.^7,8 However, remarkably little is known about how front-line hospital quality officials perceive the program and how they are responding or not responding, to the new requirements. This knowledge gap is a critical barrier to evaluating the program’s practical impact on sepsis treatment and outcomes.

We therefore sought to understand hospital stakeholders’ perceptions of the SEP-1 program in general as well as their characterization of their local hospitals’ responses to the program. We were specifically interested in obtaining a focused perspective on the policy and hospitals’ responses to the policy rather than individual physicians’ attitudes regarding sepsis care protocols, which are complex and may be independent from the policy itself.⁹ We used a qualitative research approach designed to generate both a deep and broad understanding of how hospitals are responding to SEP-1 requirements, including the resources required to implement their responses.

METHODS

Study Design, Setting, and Subjects

We conducted a qualitative study by using semistructured telephone interviews with hospital quality officers in the United States. We targeted hospital quality officers because they are in a position to provide overarching insights into hospitals’ perceptions of and responses to the SEP-1 program. We enrolled quality officers at general, short-stay, nonfederal acute care hospitals because those are the hospitals to which the SEP-1 program applies. We generated a stratified random sample of hospitals by using 2013 data from Medicare’s Healthcare Cost and Reporting Information System (HCRIS) database.¹⁰ We stratified by size (greater than or less than 200 total beds), teaching status (presence or absence of any resident physician trainees), and ownership (for-profit vs nonprofit), creating 8 mutually exclusive strata. This sampling frame was designed to ensure representativeness from a broad range of hospital types, not to enable comparisons across hospital types, which is outside the scope of qualitative research.

Within strata, we contacted hospitals in a random order by phone using the primary number listed in the HCRIS database. We asked the hospital operator to connect us to the chief quality officer or an appropriate alternative hospital administrator with knowledge of hospital quality-improvement activities. We limited participation to 1 respondent per hospital. We did not offer any specific incentives for participation.

The study was approved by the University of Pittsburgh Institutional Review Board with a waiver of signed informed consent.

Data Collection

Interviews were conducted by a trained research coordinator between February 2016 and October 2016. Interviews were conducted concurrently with data analysis by using a constant comparison approach.¹¹ The constant comparison approach involves the iterative refinement of themes by comparing the existing themes to new data as they emerge during successive interviews. We chose a constant comparison approach because we wanted to systematically describe hospital responses to SEP-1 rather than specifically test individual hypotheses.¹¹ As is typical in qualitative research, we did not set the sample size a priori but instead continued the interviews until we achieved thematic saturation.^12,13

The interview script included a mix of directed and open-ended questions about respondents’ perspectives of and hospital responses to the SEP-1 program. The questions covered the following 4 domains: hospitals’ sepsis quality-improvement initiatives before and after the Medicare reporting program, reception of the hospital responses, the approach to data abstraction and reporting, and the overall impressions of the program and its impact.^6-8,14 We allowed for updates and revisions of the interview guide as necessary to explore any new content and emergent themes. We piloted the interview guide on 2 hospital quality officers at our institution and then revised its structure again after interviews with the initial 6 hospitals. The complete final interview guide is available in the supplemental digital content.

 

Analysis

Interviews were audio recorded, transcribed, and loaded onto a secure server. We used NVivo 11 (QSR International, Cambridge, Massachusetts) for coding and analysis. We iteratively reviewed and thematically analyzed the transcripts for structural content and emergent themes, consistent with established qualitative methods.¹⁵ Three investigators reviewed the initial 20 transcripts and developed the codebook through iterative discussion and consensus. The codes were then organized into themes and subthemes. Subsequently, 1 investigator coded the remaining transcripts. The results are presented as a series of key themes supported by direct quotes from the interviews.

RESULTS

Sample Description

We performed 29 interviews prior to achieving thematic saturation. Each of the 8 strata from the sampling frame was represented by at least 3 hospitals. Hospitals in the final sample were diverse in total bed size, intensive care unit bed capacity, teaching status, and ownership (Table 2). The median interview length was 25 minutes (interquartile range, 20-32 minutes). Respondents included 6 quality coordinators, 6 quality managers, and 11 quality directors, with the remainder holding a variety of other quality-related titles. Most respondents worked in hospital quality departments, although 4 were affiliated with individual clinical departments (eg, emergency medicine and/or critical care services). Of the 9 respondents who reported their professional training, 8 were registered nurses. Eleven respondents reported participating in measure abstraction.

Perspectives on SEP-1

Respondents’ general perspectives on the SEP-1 program are outlined in Table 3, with several key themes emerging. Foremost was the sheer complexity of the measure compounded by its reliance on time-stamped clinical documentation, and in particular, the physical reassessment in individual medical notes. Respondents expressed frustration with the “all-or-none” approach to declaring sepsis treatment a “success,” which they noted was unfair and difficult to justify to their local clinicians. In part because of the time and effort required to comply with the measure and report results to CMS, several respondents noted that the measure is a uniquely burdensome addition to an already-crowded landscape of hospital quality programs. Despite the resources required to adhere to the measures’ standards and report results to CMS, respondents expressed a belief that the increased attention to sepsis is driving positive changes in hospital care and leading to improved patient outcomes.

Responses to SEP-1

Respondents identified several specific ways in which their hospitals responded to the SEP-1 mandate (Table 4), including investments in measurement, planning and coordinating sepsis-specific quality-improvement activities, improving the early identification of patients with sepsis, improving sepsis treatment and measure compliance, and addressing negative attitudes towards the implementation of the SEP-1 program.

Efforts to Collect Data for SEP-1 Reporting

Respondents reported challenges in reliably and validly measuring and reporting data for the SEP-1 program. First, patient identification and the measurement of treatment processes depends largely on manual medical record review, which is subject to variation across coders. This presents a particular challenge because the clinical definition of sepsis itself is in evolution,¹ creating the possibility that treating physicians could identify a given patient as having sepsis or septic shock based on the most up-to-date definitions but not based on the measure’s specifications or vice versa. Second, each case requires up to an hour of manual medical record review and patients who develop sepsis during prolonged hospitalizations can require several hours or more, which is an unprecedented length of time to spend abstracting data for a single measure.

In addressing these measurement challenges, investment in human resources is the rule. No respondent reported automating abstraction of all the SEP-1 data elements, underscoring concerns regarding the measurement burden of the SEP-1 program.^7,8,14 Rather, hospitals with sufficient financial resources frequently employ full-time data abstractors and individuals responsible for ongoing performance feedback, which facilitates the iterative revision of sepsis quality-improvement initiatives. In contrast, hospitals with fewer resources often rely on contracts with third-party vendors, which delays reporting and complicates efforts to use the data for individualized performance improvement.

Efforts to Coordinate Hospital Responses Across Care Teams

Complying with the measure involves the longitudinal coordination of multiple care teams across different units, so planning and executing local hospital responses required interdepartmental and multidisciplinary stakeholder involvement. Respondents were uncertain about the ideal strategy to coordinate these quality-improvement efforts, yielding iterative changes to electronic health records (EHRs), education programs, and data collection methods. This “learning by doing” is necessary because no prior CMS quality measure is as complex as SEP-1 or as varied in the sources of data required to measure and report the results. By requiring hospitals to improve coordination of care throughout the hospital, SEP-1 presents a quality-improvement and measurement challenge that may ultimately drive innovation and better patient care.

 

Efforts to Improve Sepsis Diagnosis

Several hospitals are implementing sepsis screening and alerts to speed sepsis recognition and meet the measure’s time-sensitive treatment requirements. An example of a less-intensive alert is one hospital’s lowering of the threshold for lactate values that are viewed as “critical” (and thus requiring notification of the bedside clinician). Examples of more resource-intensive alerts included electronic screening for vital sign abnormalities that trigger bedside assessment for infection as well as nurse-driven manual sepsis screening tools.

Frequently, these more intensive efforts faced barriers to successful implementation related to the broader issues of performance measurement rather than the specifics of SEP-1. EHRs generally lacked built-in electronic screening capacity, and few hospitals had the resources required for customized EHR modification. Manual screening required nurses to spend time away from direct patient care. For both electronic and manual screening, respondents expressed concern about how these new alerts would fit into a care landscape already inundated with alerts, alarms, and care notifications.^16,17

Efforts to Improve Sepsis Treatment

Many hospitals are implementing sepsis-specific treatment protocols and order sets designed to help meet SEP-1 treatment specifications. In hospitals and health systems with preexisting sepsis quality-improvement efforts, SEP-1 stimulated adaptation and acceleration of their efforts; in hospitals without preexisting sepsis-specific quality improvement, SEP-1 inspired de novo program development and implementation. These programs were wide ranging. Several hospitals implemented a process by which an initially elevated lactate value automates an order for a repeat lactate level, facilitating an assessment of the clinical response to treatment. Other examples include triggers for sepsis-specific treatment protocols and checklists that bedside nurses can begin without initial physician oversight. In 1 hospital, sepsis alerts triggered by emergency medical first responders initiate responses prior to hospital arrival in a manner analogous to prehospital alerts for myocardial infarction and stroke.^18,19

Efforts to implement these protocols encountered several common challenges. Physicians were often resistant to adopting inflexible treatment rules that did not allow them to tailor therapies to individual patients. Furthermore, even protocols and order sets that worked in 1 setting did not necessarily generalize throughout the hospital or health system, reflecting the difficulty in implementing a highly specified measure across diverse treatment environments.

Efforts to Manage Clinician Attitudes Toward SEP-1 Implementation

In addition to addressing clinicians’ behaviors, hospitals sought to address stakeholders’ attitudes when those attitudes created barriers to SEP-1 implementation. First, hospitals frequently faced a lack of buy-in from clinicians who were resistant to the idea of protocolized care in general and who were specifically skeptical that initiatives designed to increase clinical documentation would drive improvements in patient-centered outcomes. Second, respondents had to confront a hierarchical hospital culture, which manifests not only in clinical care, but also in the quality-improvement infrastructure. Many respondents reported that physicians were more receptive to performance feedback from fellow physicians rather than nonphysician quality administrators.

Respondents described a range of approaches to counteract these attitudes. First, hospitals deployed department- and profession-specific “champions” to provide peer-to-peer performance feedback supported by data demonstrating a link between process improvements and patient outcomes. Second, many respondents noted that the addition of new clinical staff, who were often younger and more receptive to new initiatives, could alter a hospital’s quality culture; in smaller hospitals, just a few individuals could significantly alter the dynamic. Finally, when other efforts failed, some respondents indicated that top-down administrative support could persuade resistant individuals to change their approach. However, this solution worked best with employed physicians and was less effective with independent physician groups without direct financial ties to hospital performance. These efforts to overcome negative attitudes toward SEP-1 implementation required individuals’ time and energy, leading to frustration at times and adding to the resources required to comply with the program.

Planning for the Future of SEP-1

Respondents anticipate that performance of the SEP-1 measure will eventually become publicly reported and incorporated into value-based purchasing calculations. Hospitals are therefore seeking greater interaction with CMS as it makes iterative revisions to the measure because respondents expect that their hospitals’ level of performance, rather than just the act of participating, will affect hospital finances. Respondents expressed a desire for more live, interactive educational sessions with CMS moving forward, rather than limiting the opportunities for clarification to online comment forums or statements elsewhere in the public record. In addition, respondents hope that public reporting and pay-for-performance could be delayed to allow more time to work out the “kinks” in measurement and reporting.

DISCUSSION

We conducted semistructured telephone interviews with quality officers in U.S. hospitals in order to understand hospitals’ perceptions of and responses to Medicare’s SEP-1 sepsis quality-reporting program. Hospitals are struggling with the program’s complexity and investing considerable resources in order to iteratively revise their responses to the program. However, they generally believe that the program is bringing much-needed attention to sepsis diagnosis and treatment. These findings have several implications for the SEP-1 measure in particular and for hospital-based quality measurement and pay-for-performance policies in general.

 

First, we demonstrate that SEP-1 consistently requires a substantial investment of resources from hospitals already struggling under the weight of numerous local, state, and national quality-reporting and improvement programs.^14,20,21 In aggregate, these programs can stretch hospitals’ resources to their limit. Respondents universally reported that the SEP-1 program is requiring dedicated staff to meet the data abstraction and reporting requirements as well as multicomponent quality-improvement initiatives. In the absence of well-established roadmaps for improving sepsis care, these sepsis quality-improvement efforts require experimentation and iterative revision, which can contribute to fatigue and frustration among quality officers and clinical staff. This process of innovation inherently involves successes, failures, and the risk of harm and opportunity costs that strain hospital resources.

Second, our study indicates how SEP-1 could exacerbate existing inequalities in our health system. Sepsis incidence and mortality are already higher in medically underserved regions.²² Given the resources required to respond to the SEP-1 program, optimal performance may be beyond the reach of smaller hospitals, or even larger hospitals, whose resources are already stretched to their limits. Public reporting and pay-for-performance can be adisadvantage to hospitals caring for underserved populations.^23,24 To the extent that responding to sepsis-oriented public policy requires resources that certain hospitals cannot access, these policies could exacerbate existing health disparities.

Third, our findings highlight some specific ways that CMS could revise the SEP-1 program to better meet the needs of hospitals and improve outcomes for patients with sepsis. Primarily, although the program’s current specifications take an “all-or-none” approach to treatment success, a more flexible approach, such as a weighted score or composite measure that combines processes and outcomes,^25,26 could allow hospitals to focus their efforts on those components of the bundle with the strongest evidence for improved patient outcomes.²⁷ Second, policy makers need to reconcile the 2 existing clinical definitions for sepsis.^1,28 CMS has already stated its plans to retain the preexisting sepsis definition,²⁹ but this does not change the reality that frontline providers and quality officials face different, and at times conflicting, clinical definitions while caring for patients. Finally, current implementation challenges may support a delay in moving the measure toward public reporting and pay-for-performance. Hospitals are already responding to the measure in a substantial way, providing an opportunity for early quantitative evaluations of the program’s impact that could inform evidence-based revisions to the measure.

Our study has several limitations. First, by interviewing only individual quality officers within each hospital, it is possible that our findings were not representative of the perspectives of other individuals within their hospitals or the hospital as a whole; indeed, to the extent that quality officers “buy in” to quality measurement and reporting, their perspectives on SEP-1 may skew more positive than other hospital staff. Our respondents represented individuals from a range of positions within the quality infrastructure, whereas “hospital quality leaders” are often chief executive officers, chief medical officers, or vice presidents for quality.³⁰ However, by virtue of our purposive sampling approach, we included respondents from a broad range of hospitals and found similar themes across these respondents, supporting the internal validity of our findings. Second, as is inherent in interview-based research, we cannot verify that respondents’ reports of hospital responses to SEP-1 match the actual changes implemented “on the ground.” We are reassured, however, by the fact that many of the perspectives and quality-improvement changes that respondents described align with the opinions and suggestions of academic quality experts, which are informed by clinical experience.^6-8 Third, while respondents believe that hospital responses to SEP-1 are contributing to improvements in treatment and outcomes, we do not yet have robust objective data to support this opinion or to evaluate the association between quality officers’ perspectives and hospital performance. A quantitative evaluation of the clinical impact of SEP-1, as well as the relationship between hospital performance and quality officers’ perspectives on the measure, are important areas for future research.

CONCLUSIONS

In a qualitative study of hospital responses to Medicare’s SEP-1 program, we found that hospitals are implementing changes across a variety of domains and in ways that consistently require dedicated resources. Giving hospitals the flexibility to focus on treatment processes with the most direct impact on patient-centered outcomes might enhance the program’s effectiveness. Future work should quantify the program’s impact and develop novel approaches to data abstraction and quality improvement.

Disclosure

Aside from federal funding, the authors have no conflicts of interest to disclose. The authors received funding from the National Institutes of Health (IJB, F32HL132461) (JMK, K24HL133444). This work was submitted as an abstract to the 2017 American Thoracic Society International Conference, May 2017.

 

Sepsis affects over 1 million Americans annually, resulting in significant morbidity, mortality, and costs for hospitalized patients.^1-4 There is an increasing interest in policy-oriented approaches to improving sepsis care at both the state and national levels.^5,6 The most prominent policy is the Centers for Medicare and Medicaid Services (CMS) Sepsis CMS Core (SEP-1) program, which was formally implemented in October 2015; the program mandates that hospitals report their compliance with a variety of sepsis treatment processes (Table 1). Academic quality experts generally applaud the increased attention to sepsis but are concerned that the measure’s design and specifications advance beyond the existing evidence base.^7,8 However, remarkably little is known about how front-line hospital quality officials perceive the program and how they are responding or not responding, to the new requirements. This knowledge gap is a critical barrier to evaluating the program’s practical impact on sepsis treatment and outcomes.

We therefore sought to understand hospital stakeholders’ perceptions of the SEP-1 program in general as well as their characterization of their local hospitals’ responses to the program. We were specifically interested in obtaining a focused perspective on the policy and hospitals’ responses to the policy rather than individual physicians’ attitudes regarding sepsis care protocols, which are complex and may be independent from the policy itself.⁹ We used a qualitative research approach designed to generate both a deep and broad understanding of how hospitals are responding to SEP-1 requirements, including the resources required to implement their responses.

METHODS

Study Design, Setting, and Subjects

We conducted a qualitative study by using semistructured telephone interviews with hospital quality officers in the United States. We targeted hospital quality officers because they are in a position to provide overarching insights into hospitals’ perceptions of and responses to the SEP-1 program. We enrolled quality officers at general, short-stay, nonfederal acute care hospitals because those are the hospitals to which the SEP-1 program applies. We generated a stratified random sample of hospitals by using 2013 data from Medicare’s Healthcare Cost and Reporting Information System (HCRIS) database.¹⁰ We stratified by size (greater than or less than 200 total beds), teaching status (presence or absence of any resident physician trainees), and ownership (for-profit vs nonprofit), creating 8 mutually exclusive strata. This sampling frame was designed to ensure representativeness from a broad range of hospital types, not to enable comparisons across hospital types, which is outside the scope of qualitative research.

Within strata, we contacted hospitals in a random order by phone using the primary number listed in the HCRIS database. We asked the hospital operator to connect us to the chief quality officer or an appropriate alternative hospital administrator with knowledge of hospital quality-improvement activities. We limited participation to 1 respondent per hospital. We did not offer any specific incentives for participation.

The study was approved by the University of Pittsburgh Institutional Review Board with a waiver of signed informed consent.

Data Collection

Interviews were conducted by a trained research coordinator between February 2016 and October 2016. Interviews were conducted concurrently with data analysis by using a constant comparison approach.¹¹ The constant comparison approach involves the iterative refinement of themes by comparing the existing themes to new data as they emerge during successive interviews. We chose a constant comparison approach because we wanted to systematically describe hospital responses to SEP-1 rather than specifically test individual hypotheses.¹¹ As is typical in qualitative research, we did not set the sample size a priori but instead continued the interviews until we achieved thematic saturation.^12,13

The interview script included a mix of directed and open-ended questions about respondents’ perspectives of and hospital responses to the SEP-1 program. The questions covered the following 4 domains: hospitals’ sepsis quality-improvement initiatives before and after the Medicare reporting program, reception of the hospital responses, the approach to data abstraction and reporting, and the overall impressions of the program and its impact.^6-8,14 We allowed for updates and revisions of the interview guide as necessary to explore any new content and emergent themes. We piloted the interview guide on 2 hospital quality officers at our institution and then revised its structure again after interviews with the initial 6 hospitals. The complete final interview guide is available in the supplemental digital content.

 

Analysis

Interviews were audio recorded, transcribed, and loaded onto a secure server. We used NVivo 11 (QSR International, Cambridge, Massachusetts) for coding and analysis. We iteratively reviewed and thematically analyzed the transcripts for structural content and emergent themes, consistent with established qualitative methods.¹⁵ Three investigators reviewed the initial 20 transcripts and developed the codebook through iterative discussion and consensus. The codes were then organized into themes and subthemes. Subsequently, 1 investigator coded the remaining transcripts. The results are presented as a series of key themes supported by direct quotes from the interviews.

RESULTS

Sample Description

We performed 29 interviews prior to achieving thematic saturation. Each of the 8 strata from the sampling frame was represented by at least 3 hospitals. Hospitals in the final sample were diverse in total bed size, intensive care unit bed capacity, teaching status, and ownership (Table 2). The median interview length was 25 minutes (interquartile range, 20-32 minutes). Respondents included 6 quality coordinators, 6 quality managers, and 11 quality directors, with the remainder holding a variety of other quality-related titles. Most respondents worked in hospital quality departments, although 4 were affiliated with individual clinical departments (eg, emergency medicine and/or critical care services). Of the 9 respondents who reported their professional training, 8 were registered nurses. Eleven respondents reported participating in measure abstraction.

Perspectives on SEP-1

Respondents’ general perspectives on the SEP-1 program are outlined in Table 3, with several key themes emerging. Foremost was the sheer complexity of the measure compounded by its reliance on time-stamped clinical documentation, and in particular, the physical reassessment in individual medical notes. Respondents expressed frustration with the “all-or-none” approach to declaring sepsis treatment a “success,” which they noted was unfair and difficult to justify to their local clinicians. In part because of the time and effort required to comply with the measure and report results to CMS, several respondents noted that the measure is a uniquely burdensome addition to an already-crowded landscape of hospital quality programs. Despite the resources required to adhere to the measures’ standards and report results to CMS, respondents expressed a belief that the increased attention to sepsis is driving positive changes in hospital care and leading to improved patient outcomes.

Responses to SEP-1

Respondents identified several specific ways in which their hospitals responded to the SEP-1 mandate (Table 4), including investments in measurement, planning and coordinating sepsis-specific quality-improvement activities, improving the early identification of patients with sepsis, improving sepsis treatment and measure compliance, and addressing negative attitudes towards the implementation of the SEP-1 program.

Efforts to Collect Data for SEP-1 Reporting

Respondents reported challenges in reliably and validly measuring and reporting data for the SEP-1 program. First, patient identification and the measurement of treatment processes depends largely on manual medical record review, which is subject to variation across coders. This presents a particular challenge because the clinical definition of sepsis itself is in evolution,¹ creating the possibility that treating physicians could identify a given patient as having sepsis or septic shock based on the most up-to-date definitions but not based on the measure’s specifications or vice versa. Second, each case requires up to an hour of manual medical record review and patients who develop sepsis during prolonged hospitalizations can require several hours or more, which is an unprecedented length of time to spend abstracting data for a single measure.

In addressing these measurement challenges, investment in human resources is the rule. No respondent reported automating abstraction of all the SEP-1 data elements, underscoring concerns regarding the measurement burden of the SEP-1 program.^7,8,14 Rather, hospitals with sufficient financial resources frequently employ full-time data abstractors and individuals responsible for ongoing performance feedback, which facilitates the iterative revision of sepsis quality-improvement initiatives. In contrast, hospitals with fewer resources often rely on contracts with third-party vendors, which delays reporting and complicates efforts to use the data for individualized performance improvement.

Efforts to Coordinate Hospital Responses Across Care Teams

Complying with the measure involves the longitudinal coordination of multiple care teams across different units, so planning and executing local hospital responses required interdepartmental and multidisciplinary stakeholder involvement. Respondents were uncertain about the ideal strategy to coordinate these quality-improvement efforts, yielding iterative changes to electronic health records (EHRs), education programs, and data collection methods. This “learning by doing” is necessary because no prior CMS quality measure is as complex as SEP-1 or as varied in the sources of data required to measure and report the results. By requiring hospitals to improve coordination of care throughout the hospital, SEP-1 presents a quality-improvement and measurement challenge that may ultimately drive innovation and better patient care.

 

Efforts to Improve Sepsis Diagnosis

Several hospitals are implementing sepsis screening and alerts to speed sepsis recognition and meet the measure’s time-sensitive treatment requirements. An example of a less-intensive alert is one hospital’s lowering of the threshold for lactate values that are viewed as “critical” (and thus requiring notification of the bedside clinician). Examples of more resource-intensive alerts included electronic screening for vital sign abnormalities that trigger bedside assessment for infection as well as nurse-driven manual sepsis screening tools.

Frequently, these more intensive efforts faced barriers to successful implementation related to the broader issues of performance measurement rather than the specifics of SEP-1. EHRs generally lacked built-in electronic screening capacity, and few hospitals had the resources required for customized EHR modification. Manual screening required nurses to spend time away from direct patient care. For both electronic and manual screening, respondents expressed concern about how these new alerts would fit into a care landscape already inundated with alerts, alarms, and care notifications.^16,17

Efforts to Improve Sepsis Treatment

Many hospitals are implementing sepsis-specific treatment protocols and order sets designed to help meet SEP-1 treatment specifications. In hospitals and health systems with preexisting sepsis quality-improvement efforts, SEP-1 stimulated adaptation and acceleration of their efforts; in hospitals without preexisting sepsis-specific quality improvement, SEP-1 inspired de novo program development and implementation. These programs were wide ranging. Several hospitals implemented a process by which an initially elevated lactate value automates an order for a repeat lactate level, facilitating an assessment of the clinical response to treatment. Other examples include triggers for sepsis-specific treatment protocols and checklists that bedside nurses can begin without initial physician oversight. In 1 hospital, sepsis alerts triggered by emergency medical first responders initiate responses prior to hospital arrival in a manner analogous to prehospital alerts for myocardial infarction and stroke.^18,19

Efforts to implement these protocols encountered several common challenges. Physicians were often resistant to adopting inflexible treatment rules that did not allow them to tailor therapies to individual patients. Furthermore, even protocols and order sets that worked in 1 setting did not necessarily generalize throughout the hospital or health system, reflecting the difficulty in implementing a highly specified measure across diverse treatment environments.

Efforts to Manage Clinician Attitudes Toward SEP-1 Implementation

In addition to addressing clinicians’ behaviors, hospitals sought to address stakeholders’ attitudes when those attitudes created barriers to SEP-1 implementation. First, hospitals frequently faced a lack of buy-in from clinicians who were resistant to the idea of protocolized care in general and who were specifically skeptical that initiatives designed to increase clinical documentation would drive improvements in patient-centered outcomes. Second, respondents had to confront a hierarchical hospital culture, which manifests not only in clinical care, but also in the quality-improvement infrastructure. Many respondents reported that physicians were more receptive to performance feedback from fellow physicians rather than nonphysician quality administrators.

Respondents described a range of approaches to counteract these attitudes. First, hospitals deployed department- and profession-specific “champions” to provide peer-to-peer performance feedback supported by data demonstrating a link between process improvements and patient outcomes. Second, many respondents noted that the addition of new clinical staff, who were often younger and more receptive to new initiatives, could alter a hospital’s quality culture; in smaller hospitals, just a few individuals could significantly alter the dynamic. Finally, when other efforts failed, some respondents indicated that top-down administrative support could persuade resistant individuals to change their approach. However, this solution worked best with employed physicians and was less effective with independent physician groups without direct financial ties to hospital performance. These efforts to overcome negative attitudes toward SEP-1 implementation required individuals’ time and energy, leading to frustration at times and adding to the resources required to comply with the program.

Planning for the Future of SEP-1

Respondents anticipate that performance of the SEP-1 measure will eventually become publicly reported and incorporated into value-based purchasing calculations. Hospitals are therefore seeking greater interaction with CMS as it makes iterative revisions to the measure because respondents expect that their hospitals’ level of performance, rather than just the act of participating, will affect hospital finances. Respondents expressed a desire for more live, interactive educational sessions with CMS moving forward, rather than limiting the opportunities for clarification to online comment forums or statements elsewhere in the public record. In addition, respondents hope that public reporting and pay-for-performance could be delayed to allow more time to work out the “kinks” in measurement and reporting.

DISCUSSION

We conducted semistructured telephone interviews with quality officers in U.S. hospitals in order to understand hospitals’ perceptions of and responses to Medicare’s SEP-1 sepsis quality-reporting program. Hospitals are struggling with the program’s complexity and investing considerable resources in order to iteratively revise their responses to the program. However, they generally believe that the program is bringing much-needed attention to sepsis diagnosis and treatment. These findings have several implications for the SEP-1 measure in particular and for hospital-based quality measurement and pay-for-performance policies in general.

 

First, we demonstrate that SEP-1 consistently requires a substantial investment of resources from hospitals already struggling under the weight of numerous local, state, and national quality-reporting and improvement programs.^14,20,21 In aggregate, these programs can stretch hospitals’ resources to their limit. Respondents universally reported that the SEP-1 program is requiring dedicated staff to meet the data abstraction and reporting requirements as well as multicomponent quality-improvement initiatives. In the absence of well-established roadmaps for improving sepsis care, these sepsis quality-improvement efforts require experimentation and iterative revision, which can contribute to fatigue and frustration among quality officers and clinical staff. This process of innovation inherently involves successes, failures, and the risk of harm and opportunity costs that strain hospital resources.

Second, our study indicates how SEP-1 could exacerbate existing inequalities in our health system. Sepsis incidence and mortality are already higher in medically underserved regions.²² Given the resources required to respond to the SEP-1 program, optimal performance may be beyond the reach of smaller hospitals, or even larger hospitals, whose resources are already stretched to their limits. Public reporting and pay-for-performance can be adisadvantage to hospitals caring for underserved populations.^23,24 To the extent that responding to sepsis-oriented public policy requires resources that certain hospitals cannot access, these policies could exacerbate existing health disparities.

Third, our findings highlight some specific ways that CMS could revise the SEP-1 program to better meet the needs of hospitals and improve outcomes for patients with sepsis. Primarily, although the program’s current specifications take an “all-or-none” approach to treatment success, a more flexible approach, such as a weighted score or composite measure that combines processes and outcomes,^25,26 could allow hospitals to focus their efforts on those components of the bundle with the strongest evidence for improved patient outcomes.²⁷ Second, policy makers need to reconcile the 2 existing clinical definitions for sepsis.^1,28 CMS has already stated its plans to retain the preexisting sepsis definition,²⁹ but this does not change the reality that frontline providers and quality officials face different, and at times conflicting, clinical definitions while caring for patients. Finally, current implementation challenges may support a delay in moving the measure toward public reporting and pay-for-performance. Hospitals are already responding to the measure in a substantial way, providing an opportunity for early quantitative evaluations of the program’s impact that could inform evidence-based revisions to the measure.

Our study has several limitations. First, by interviewing only individual quality officers within each hospital, it is possible that our findings were not representative of the perspectives of other individuals within their hospitals or the hospital as a whole; indeed, to the extent that quality officers “buy in” to quality measurement and reporting, their perspectives on SEP-1 may skew more positive than other hospital staff. Our respondents represented individuals from a range of positions within the quality infrastructure, whereas “hospital quality leaders” are often chief executive officers, chief medical officers, or vice presidents for quality.³⁰ However, by virtue of our purposive sampling approach, we included respondents from a broad range of hospitals and found similar themes across these respondents, supporting the internal validity of our findings. Second, as is inherent in interview-based research, we cannot verify that respondents’ reports of hospital responses to SEP-1 match the actual changes implemented “on the ground.” We are reassured, however, by the fact that many of the perspectives and quality-improvement changes that respondents described align with the opinions and suggestions of academic quality experts, which are informed by clinical experience.^6-8 Third, while respondents believe that hospital responses to SEP-1 are contributing to improvements in treatment and outcomes, we do not yet have robust objective data to support this opinion or to evaluate the association between quality officers’ perspectives and hospital performance. A quantitative evaluation of the clinical impact of SEP-1, as well as the relationship between hospital performance and quality officers’ perspectives on the measure, are important areas for future research.

CONCLUSIONS

In a qualitative study of hospital responses to Medicare’s SEP-1 program, we found that hospitals are implementing changes across a variety of domains and in ways that consistently require dedicated resources. Giving hospitals the flexibility to focus on treatment processes with the most direct impact on patient-centered outcomes might enhance the program’s effectiveness. Future work should quantify the program’s impact and develop novel approaches to data abstraction and quality improvement.

Disclosure

Aside from federal funding, the authors have no conflicts of interest to disclose. The authors received funding from the National Institutes of Health (IJB, F32HL132461) (JMK, K24HL133444). This work was submitted as an abstract to the 2017 American Thoracic Society International Conference, May 2017.

 

Health literacy (HL), defined as patients’ ability to understand health information and make health decisions,¹ is a prevalent problem in the outpatient and inpatient settings.^2,3 In both settings, low HL has adverse implications for self-care including interpreting health labels⁴ and taking medications correctly.⁵ Among outpatient cohorts, HL has been associated with worse outcomes and acute care utilization.⁶ Associations with low HL include increased hospitalizations,⁷ rehospitalizations,^8,9 emergency department visits,¹⁰ and decreased preventative care use.¹¹ Among the elderly, low HL is associated with increased mortality¹² and decreased self-perception of health.¹³

A systematic review revealed that most high-quality HL outcome studies were conducted in the outpatient setting.⁶ There have been very few studies assessing effects of low HL in an acute-care setting.^7,14 These studies have evaluated postdischarge outcomes, including admissions or readmissions,^7-9 and medication knowledge.¹⁴ To the best of our knowledge, there are no studies evaluating associations between HL and hospital length of stay (LOS).

LOS has received much attention as providers and payers focus more on resource utilization and eliminating adverse effects of prolonged hospitalization.¹⁵ LOS is multifactorial, depending on clinical characteristics like disease severity, as well as on sociocultural, demographic, and geographic factors.¹⁶ Despite evidence that LOS reductions translate into improved resource allocation and potentially fewer complications, there remains a tension between the appropriate LOS and one that is too short for a given condition.¹⁷

Because low HL is associated with inefficient resource utilization, we hypothesized that low HL would be associated with increased LOS after controlling for illness severity. Our objectives were to evaluate the association between low HL and LOS and whether such an association was modified by illness severity and sociodemographics.

METHODS

Study Design, Setting, Participants

An in-hospital, cohort study design of patients who were admitted or transferred to the general medicine service at the University of Chicago between October 2012 and November 2015 and screened for inclusion as part of a large, ongoing study of inpatient care quality was conducted.¹⁸ Exclusion criteria included observation status, age under 18 years, non-English speaking, and repeat participants. Those who died during hospitalization or whose discharge status was missing were excluded because the primary goal was to examine the association of HL and time to discharge, which could not be evaluated among those who died. We excluded participants with LOS >30 days to limit overly influential effects of extreme outliers (1% of the population).

Variables

HL was screened using the Brief Health Literacy Screen (BHLS), a validated, 3-question verbal survey not requiring adequate visual acuity to assess HL.^19,20 The 3 questions are as follows: (1) “How confident are you filling out medical forms by yourself?”, (2) “How often do you have someone help you read hospital materials?”, and (3) “How often do you have problems learning about your medical condition because of difficulty understanding written information?” Responses to the questions were scored on a 5-point Likert scale in which higher scores corresponded to higher HL.^21,22 The scores for each of the 3 questions were summed to yield a range between 3 and 15. On the individual questions, prior work has demonstrated improved test performance with a cutoff of ≤3, which corresponds to a response of “some of the time” or “somewhat”; therefore, when the 3 questions were summed together, scores of ≤9 were considered indicative of low HL.^21,23

For severity of illness adjustment, we used relative weights derived from the 3M (3M, Maplewood, MN) All Patient Refined Diagnosis Related Groups (APR-DRG) classification system, which uses administrative data to classify the severity. The APR-DRG system assigns each admission to a DRG based on principal diagnosis; for each DRG, patients are then subdivided into 4 severity classes based on age, comorbidity, and interactions between these variables and the admitting diagnosis.²⁴ Using the base DRG and severity score, the system assigns relative weights that reflect differences in expected hospital resource utilization.

LOS was derived from hospital administrative data and counted from the date of admission to the hospital. Participants who were discharged on the day of admission were counted as having an LOS of 1. Insurance status (Medicare, Medicaid, no payer, private) also was obtained from administrative data. Age, sex (male or female), education (junior high or less, some high school, high school graduate, some college, college graduate, postgraduate), and race (black/African American, white, Asian or Pacific Islander [including Asian Indian, Chinese, Filipino, Japanese, Korean, Vietnamese, other Asian, Native Hawaiian, Guam/Chamorro, Samoan, other Pacific], American Indian or Alaskan Native, multiple race) were obtained from administrative data based on information provided by the patient. Participants with missing data on any of the sociodemographic variables or on the APR-DRG score were excluded from the analysis.

 

Statistical Analysis

χ² and 2-tailed t tests were used to compare categorical and continuous variables, respectively. Multivariate linear regressions were employed to measure associations between the independent variables (HL, illness severity, race, gender, education, and insurance status) and the dependent variable, LOS. Independent variables were chosen for clinical significance and retained in the model regardless of statistical significance. The adjusted R² values of models with and without the HL variable included were reported to provide information on the contribution of HL to the overall model.

Because LOS was observed to be right skewed and residuals of the untransformed regression were observed to be non-normally distributed, the decision was made to natural log transform LOS, which is consistent with previous hospital LOS studies.¹⁶ Regression coefficients and confidence intervals were then transformed into percentage estimates using the following equation: 100(e^β–1). Adjusted R² was reported for the transformed regression.

The APR-DRG relative weight was treated as a continuous variable. Sociodemographic variables were dichotomized as follows: female vs male; high school graduates vs not; African American vs not; Medicaid/no payer vs Medicare/private payer. Age was not included in the multivariate model because it has been incorporated into the weighted APR-DRG illness severity scores.

Each of the sociodemographic variables and the APR-DRG score were examined for effect modification via the same multivariate linear equation described above, with the addition of an interaction term. A separate regression was performed with an interaction term between age (dichotomized at ≥65) and HL to investigate whether age modified the association between HL and LOS. Finally, we explored whether effects were isolated to long vs short LOS by dividing the sample based on the mean LOS (≥6 days) and performing separate multivariate comparisons.

Sensitivity analyses were performed to exclude those with LOS greater than the 90th percentile and those with APR-DRG score greater than the 90th percentile; age was added to the model as a continuous variable to evaluate whether the illness severity score fully adjusted for the effects of age on LOS. Furthermore, we compared the participants with missing data to those with complete data across both dependent and independent variables. Alpha was set at 0.05; analyses were performed using Stata Version 14 (Stata, College Station, TX).

RESULTS

A total of 5983 participants met inclusion criteria and completed the HL assessment; of these participants, 75 (1%) died during hospitalization, 9 (0.2%) had missing discharge status, and 79 (1%) had LOS >30 days. Two hundred eighty (5%) were missing data on sociodemographic variables or APR-DRG score. Of the remaining (n = 5540), the mean age was 57 years (standard deviation [SD] = 19 years), over half of participants were female (57%), and the majority were African American (73%) and had graduated from high school (81%). The sample was divided into those with private insurance (25%), those with Medicare (46%), and those with Medicaid (26%); 2% had no payer. The mean APR-DRG score was 1.3 (SD = 1.2), and the scores ranged from 0.3 to 15.8.

On the BHLS screen for HL, 20% (1104/5540) had inadequate HL. Participants with low HL had higher weighted illness severity scores (average 1.4 vs 1.3; P = 0.003). Participants with low HL were also more likely to be 65 or older (55% vs 33%; P < 0.001), non-high school graduates (35% vs 15%; P < 0.001), and African American (78% vs 72%; P < 0.001), and to have Medicare or private insurance (75% vs 71%; P = 0.02). There was no significant difference with respect to gender (54% male vs 57% female; P = 0.1)

The mean and median LOS were 6 ± 5 days and 4 days (interquartile range 2-7 days), respectively. Those with low HL had a longer average LOS (6.0 vs 5.4 days; P < 0.001). In multivariate analysis controlling for APR-DRG score, gender, education, race, and insurance status, low HL was associated with an 11.1% longer LOS (95% CI, 6.1-16.1; P < 0.001; Table 1). The adjusted R² value for the regression was 25.0% including HL and 24.7% with HL excluded. Additionally, being African American (P < 0.001) and having Medicaid or no insurance (P < 0.001) were associated with a shorter LOS in multivariate analysis (Table 1). The association of HL and LOS in multivariate modeling remained significant among participants with LOS <6 days (10.2%; 95% CI, 5.6%-14.9%; P < 0.001), but not among participants with LOS ≥6 days (0.4%; 95% CI, −3.6% to 4.4%; P = 0.8).

Neither age ≥65 (P = 0.4) nor illness severity score (P = 0.5) significantly modified the effect of HL on LOS. However, the effect of HL on hospital LOS was significantly modified by gender (P = 0.02). Among men, low HL was associated with a 17.8% longer LOS (95% CI, 10.0%-25.7%; P < 0.001), but among women, low HL was associated with only a 7.7% longer LOS (95% CI, 1.9%-13.5%; P = 0.009). Among the remaining demographics, high school graduation status (P = 0.4), being African American (P = 0.6), and insurance status (P = 0.2) did not significantly modify the effect of HL on LOS. In sensitivity analysis, excluding participants with LOS above the 90th percentile of 12 days and excluding participants with illness severity scores above the 90th percentile, low HL was still associated with longer LOS (P < 0.001 and P = 0.001, respectively; Table 2). In the final sensitivity analysis, although age remained a significant predictor of longer LOS after controlling for illness severity (0.2% increase per year, 95% CI, 0.1%-0.3%; P < 0.001), low HL nevertheless remained significantly associated with longer LOS (P < 0.001; Table 2).

Finally, we compared the group with missing data (n = 280) to the group with complete data (n = 5540). The participants with missing data were more likely to have low HL (31% [86/280] vs 20%; P < 0.001) and to have Medicare or private insurance (82% [177/217] vs 72%; P = 0.002); however, they were not more likely to be 65 or older (40% [112/280] vs 37%; P = 0.3), high school graduates (88% [113/129] vs 81%; P = 0.06), African American (69% [177/256] vs 73%; P = 0.1), or female (57% [158/279] vs 57%; P = 1), nor were they more likely to have longer LOS (5.7 [n = 280] vs 5.5 days; P = 0.6) or higher illness severity scores (1.3 [n = 231] vs 1.3; P = 0.7).

 

DISCUSSION

To our knowledge, this study is the first to evaluate the association between low HL and an important in-hospital outcome measure, hospital LOS. We found that low HL was associated with a longer hospital LOS, a result which remained significant when controlling for severity of illness and sociodemographic variables and when testing the model for sensitivity to the highest values of LOS and illness severity. Additionally, the association of HL with LOS appeared concentrated among participants with shorter LOS. Relative to other predictors, the contribution of HL to the overall LOS model was small, as evidenced by the change in adjusted R² values with HL excluded.

Among the covariates, only gender modified the association between HL and LOS; the findings suggested that men were more susceptible to the effect of low HL on increased LOS. Illness severity and other sociodemographics, including age ≥65, did not appear to modify the association. We also found that being African American and having Medicaid or no insurance were associated with a significantly shorter LOS in multivariate analysis.

Previous work suggested that the adverse health effects of low HL may be mediated through several pathways, including health knowledge, self-efficacy, health skills, and illness stigma.^25-27 The finding of a small but significant relationship between HL and LOS was not surprising given these known associations; nevertheless, there may be an additional patient-dependent effect of low HL on LOS not discovered here. For instance, patients with poor health knowledge and self-efficacy might stay in the hospital longer if they or their providers do not feel comfortable with their self-care ability.

This finding may be useful in developing hospital-based interventions. HL-specific interventions, several of which have been tested in the inpatient setting,^14,28,29 have shown promise toward improving health knowledge,³⁰ disease severity,³¹ and health resource utilization.³²

Those with low HL may lack the self-efficacy to participate in discharge planning; in fact, previous work has related low HL to posthospital readmissions.^8,9 Conversely, patients with low HL might struggle to engage in the inpatient milieu, advocating for shorter LOS if they feel alienated by the inpatient experience.

These possibilities show that LOS is a complex measure shown to depend on patient-level characteristics and on provider-based, geographical, and sociocultural factors.^16,33 With these forces at play, additional effects of lower levels of HL may be lost without phenotyping patients by both level of HL and related characteristics, such as self-efficacy, health skills, and stigma. By gathering these additional data, future work should explore whether subpopulations of patients with low HL may be at risk for too-short vs too-long hospital admissions.

For instance, in this study, both race and Medicaid insurance were associated with shorter LOS. Being African American was associated with shorter LOS in our study but has been found to be associated with longer LOS in another study specifically focused on diabetes.³⁴ Prior findings found uninsured patients have shorter LOS.³⁵ Therefore, these findings in our study are difficult to explain without further work to understand whether there are health disparities in the way patients are cared for during hospitalization that may shorten or lengthen their LOS because of factors outside of their clinical need.

The finding that gender modified the effect of low HL on LOS was unexpected. There were similar proportions of men and women with low HL. There is evidence to support that women make the majority of health decisions for themselves and their familes³⁶; therefore, there may be unmeasured aspects of HL that provide an advantage for female vs male inpatients. Furthermore, omitted confounders, such as social support, may not fully capture potential gender-related differences. Future work is needed to understand the role of gender in relationship to HL and LOS.

Limitations of this study include its observational, single-centered design with information derived from administrative data; positive and negative confounding cannot be ruled out. For instance, we did not control for complex aspects affecting LOS, such as discharge disposition and goals of care (eg, aggressive care after discharge vs hospice). To address this limitation, multivariate analyses were performed, which were adjusted for illness severity scores and took into account both comorbidity and severity of the current illness. Additionally, although it is important to study such populations, our largely urban, minority sample is not representative of the U.S. population, and within our large sample, there were participants with missing data who had lower HL on average, although this group represented only 5% of the sample. Finally, different HL tools have noncomplete concordance, which has been seen when comparing the BHLS with more objective tools.^20,37 Furthermore, certain in-hospital clinical scenarios (eg, recent stroke or prolonged intensive care unit stay) may present unique challenges in establishing a baseline HL level. However, the BHLS was used in this study because of its greater feasibility.

In conclusion, this study is the first to evaluate the relationship between low HL and LOS. The findings suggest that HL may play a role in shaping outcomes in the inpatient setting and that targeting interventions toward screened patients may be a pathway toward mitigating adverse effects. Our findings need to be replicated in larger, more representative samples, and further work understanding subpopulations within the low HL population is needed. Future work should measure this association in diverse inpatient settings (eg, psychiatric, surgical, and specialty), in addition to assessing associations between HL and other important in-hospital outcome measures, including mortality and discharge disposition.

 

Acknowledgments

The authors thank the Hospitalist Project team for their assistance with data collection. The authors especially thank Chuanhong Liao and Ashley Snyder for assistance with statistical analyses; Andrea Flores, Ainoa Coltri, and Tom Best for their assistance with data management. The authors would also like to thank Nicole Twu for her help with preparing and editing the manuscript.

Disclosures

Dr. Jaffee was supported by a Calvin Fentress Research Fellowship and NIH R25MH094612. Dr. Press was supported by a career development award (NHLBI K23HL118151). This work was also supported by a seed grant from the Center for Health Administration Studies. All other authors declare no conflicts of interest.

Health literacy (HL), defined as patients’ ability to understand health information and make health decisions,¹ is a prevalent problem in the outpatient and inpatient settings.^2,3 In both settings, low HL has adverse implications for self-care including interpreting health labels⁴ and taking medications correctly.⁵ Among outpatient cohorts, HL has been associated with worse outcomes and acute care utilization.⁶ Associations with low HL include increased hospitalizations,⁷ rehospitalizations,^8,9 emergency department visits,¹⁰ and decreased preventative care use.¹¹ Among the elderly, low HL is associated with increased mortality¹² and decreased self-perception of health.¹³

A systematic review revealed that most high-quality HL outcome studies were conducted in the outpatient setting.⁶ There have been very few studies assessing effects of low HL in an acute-care setting.^7,14 These studies have evaluated postdischarge outcomes, including admissions or readmissions,^7-9 and medication knowledge.¹⁴ To the best of our knowledge, there are no studies evaluating associations between HL and hospital length of stay (LOS).

LOS has received much attention as providers and payers focus more on resource utilization and eliminating adverse effects of prolonged hospitalization.¹⁵ LOS is multifactorial, depending on clinical characteristics like disease severity, as well as on sociocultural, demographic, and geographic factors.¹⁶ Despite evidence that LOS reductions translate into improved resource allocation and potentially fewer complications, there remains a tension between the appropriate LOS and one that is too short for a given condition.¹⁷

Because low HL is associated with inefficient resource utilization, we hypothesized that low HL would be associated with increased LOS after controlling for illness severity. Our objectives were to evaluate the association between low HL and LOS and whether such an association was modified by illness severity and sociodemographics.

METHODS

Study Design, Setting, Participants

An in-hospital, cohort study design of patients who were admitted or transferred to the general medicine service at the University of Chicago between October 2012 and November 2015 and screened for inclusion as part of a large, ongoing study of inpatient care quality was conducted.¹⁸ Exclusion criteria included observation status, age under 18 years, non-English speaking, and repeat participants. Those who died during hospitalization or whose discharge status was missing were excluded because the primary goal was to examine the association of HL and time to discharge, which could not be evaluated among those who died. We excluded participants with LOS >30 days to limit overly influential effects of extreme outliers (1% of the population).

Variables

HL was screened using the Brief Health Literacy Screen (BHLS), a validated, 3-question verbal survey not requiring adequate visual acuity to assess HL.^19,20 The 3 questions are as follows: (1) “How confident are you filling out medical forms by yourself?”, (2) “How often do you have someone help you read hospital materials?”, and (3) “How often do you have problems learning about your medical condition because of difficulty understanding written information?” Responses to the questions were scored on a 5-point Likert scale in which higher scores corresponded to higher HL.^21,22 The scores for each of the 3 questions were summed to yield a range between 3 and 15. On the individual questions, prior work has demonstrated improved test performance with a cutoff of ≤3, which corresponds to a response of “some of the time” or “somewhat”; therefore, when the 3 questions were summed together, scores of ≤9 were considered indicative of low HL.^21,23

For severity of illness adjustment, we used relative weights derived from the 3M (3M, Maplewood, MN) All Patient Refined Diagnosis Related Groups (APR-DRG) classification system, which uses administrative data to classify the severity. The APR-DRG system assigns each admission to a DRG based on principal diagnosis; for each DRG, patients are then subdivided into 4 severity classes based on age, comorbidity, and interactions between these variables and the admitting diagnosis.²⁴ Using the base DRG and severity score, the system assigns relative weights that reflect differences in expected hospital resource utilization.

LOS was derived from hospital administrative data and counted from the date of admission to the hospital. Participants who were discharged on the day of admission were counted as having an LOS of 1. Insurance status (Medicare, Medicaid, no payer, private) also was obtained from administrative data. Age, sex (male or female), education (junior high or less, some high school, high school graduate, some college, college graduate, postgraduate), and race (black/African American, white, Asian or Pacific Islander [including Asian Indian, Chinese, Filipino, Japanese, Korean, Vietnamese, other Asian, Native Hawaiian, Guam/Chamorro, Samoan, other Pacific], American Indian or Alaskan Native, multiple race) were obtained from administrative data based on information provided by the patient. Participants with missing data on any of the sociodemographic variables or on the APR-DRG score were excluded from the analysis.

 

Statistical Analysis

χ² and 2-tailed t tests were used to compare categorical and continuous variables, respectively. Multivariate linear regressions were employed to measure associations between the independent variables (HL, illness severity, race, gender, education, and insurance status) and the dependent variable, LOS. Independent variables were chosen for clinical significance and retained in the model regardless of statistical significance. The adjusted R² values of models with and without the HL variable included were reported to provide information on the contribution of HL to the overall model.

Because LOS was observed to be right skewed and residuals of the untransformed regression were observed to be non-normally distributed, the decision was made to natural log transform LOS, which is consistent with previous hospital LOS studies.¹⁶ Regression coefficients and confidence intervals were then transformed into percentage estimates using the following equation: 100(e^β–1). Adjusted R² was reported for the transformed regression.

The APR-DRG relative weight was treated as a continuous variable. Sociodemographic variables were dichotomized as follows: female vs male; high school graduates vs not; African American vs not; Medicaid/no payer vs Medicare/private payer. Age was not included in the multivariate model because it has been incorporated into the weighted APR-DRG illness severity scores.

Each of the sociodemographic variables and the APR-DRG score were examined for effect modification via the same multivariate linear equation described above, with the addition of an interaction term. A separate regression was performed with an interaction term between age (dichotomized at ≥65) and HL to investigate whether age modified the association between HL and LOS. Finally, we explored whether effects were isolated to long vs short LOS by dividing the sample based on the mean LOS (≥6 days) and performing separate multivariate comparisons.

Sensitivity analyses were performed to exclude those with LOS greater than the 90th percentile and those with APR-DRG score greater than the 90th percentile; age was added to the model as a continuous variable to evaluate whether the illness severity score fully adjusted for the effects of age on LOS. Furthermore, we compared the participants with missing data to those with complete data across both dependent and independent variables. Alpha was set at 0.05; analyses were performed using Stata Version 14 (Stata, College Station, TX).

RESULTS

A total of 5983 participants met inclusion criteria and completed the HL assessment; of these participants, 75 (1%) died during hospitalization, 9 (0.2%) had missing discharge status, and 79 (1%) had LOS >30 days. Two hundred eighty (5%) were missing data on sociodemographic variables or APR-DRG score. Of the remaining (n = 5540), the mean age was 57 years (standard deviation [SD] = 19 years), over half of participants were female (57%), and the majority were African American (73%) and had graduated from high school (81%). The sample was divided into those with private insurance (25%), those with Medicare (46%), and those with Medicaid (26%); 2% had no payer. The mean APR-DRG score was 1.3 (SD = 1.2), and the scores ranged from 0.3 to 15.8.

On the BHLS screen for HL, 20% (1104/5540) had inadequate HL. Participants with low HL had higher weighted illness severity scores (average 1.4 vs 1.3; P = 0.003). Participants with low HL were also more likely to be 65 or older (55% vs 33%; P < 0.001), non-high school graduates (35% vs 15%; P < 0.001), and African American (78% vs 72%; P < 0.001), and to have Medicare or private insurance (75% vs 71%; P = 0.02). There was no significant difference with respect to gender (54% male vs 57% female; P = 0.1)

The mean and median LOS were 6 ± 5 days and 4 days (interquartile range 2-7 days), respectively. Those with low HL had a longer average LOS (6.0 vs 5.4 days; P < 0.001). In multivariate analysis controlling for APR-DRG score, gender, education, race, and insurance status, low HL was associated with an 11.1% longer LOS (95% CI, 6.1-16.1; P < 0.001; Table 1). The adjusted R² value for the regression was 25.0% including HL and 24.7% with HL excluded. Additionally, being African American (P < 0.001) and having Medicaid or no insurance (P < 0.001) were associated with a shorter LOS in multivariate analysis (Table 1). The association of HL and LOS in multivariate modeling remained significant among participants with LOS <6 days (10.2%; 95% CI, 5.6%-14.9%; P < 0.001), but not among participants with LOS ≥6 days (0.4%; 95% CI, −3.6% to 4.4%; P = 0.8).

Neither age ≥65 (P = 0.4) nor illness severity score (P = 0.5) significantly modified the effect of HL on LOS. However, the effect of HL on hospital LOS was significantly modified by gender (P = 0.02). Among men, low HL was associated with a 17.8% longer LOS (95% CI, 10.0%-25.7%; P < 0.001), but among women, low HL was associated with only a 7.7% longer LOS (95% CI, 1.9%-13.5%; P = 0.009). Among the remaining demographics, high school graduation status (P = 0.4), being African American (P = 0.6), and insurance status (P = 0.2) did not significantly modify the effect of HL on LOS. In sensitivity analysis, excluding participants with LOS above the 90th percentile of 12 days and excluding participants with illness severity scores above the 90th percentile, low HL was still associated with longer LOS (P < 0.001 and P = 0.001, respectively; Table 2). In the final sensitivity analysis, although age remained a significant predictor of longer LOS after controlling for illness severity (0.2% increase per year, 95% CI, 0.1%-0.3%; P < 0.001), low HL nevertheless remained significantly associated with longer LOS (P < 0.001; Table 2).

Finally, we compared the group with missing data (n = 280) to the group with complete data (n = 5540). The participants with missing data were more likely to have low HL (31% [86/280] vs 20%; P < 0.001) and to have Medicare or private insurance (82% [177/217] vs 72%; P = 0.002); however, they were not more likely to be 65 or older (40% [112/280] vs 37%; P = 0.3), high school graduates (88% [113/129] vs 81%; P = 0.06), African American (69% [177/256] vs 73%; P = 0.1), or female (57% [158/279] vs 57%; P = 1), nor were they more likely to have longer LOS (5.7 [n = 280] vs 5.5 days; P = 0.6) or higher illness severity scores (1.3 [n = 231] vs 1.3; P = 0.7).

 

DISCUSSION

To our knowledge, this study is the first to evaluate the association between low HL and an important in-hospital outcome measure, hospital LOS. We found that low HL was associated with a longer hospital LOS, a result which remained significant when controlling for severity of illness and sociodemographic variables and when testing the model for sensitivity to the highest values of LOS and illness severity. Additionally, the association of HL with LOS appeared concentrated among participants with shorter LOS. Relative to other predictors, the contribution of HL to the overall LOS model was small, as evidenced by the change in adjusted R² values with HL excluded.

Among the covariates, only gender modified the association between HL and LOS; the findings suggested that men were more susceptible to the effect of low HL on increased LOS. Illness severity and other sociodemographics, including age ≥65, did not appear to modify the association. We also found that being African American and having Medicaid or no insurance were associated with a significantly shorter LOS in multivariate analysis.

Previous work suggested that the adverse health effects of low HL may be mediated through several pathways, including health knowledge, self-efficacy, health skills, and illness stigma.^25-27 The finding of a small but significant relationship between HL and LOS was not surprising given these known associations; nevertheless, there may be an additional patient-dependent effect of low HL on LOS not discovered here. For instance, patients with poor health knowledge and self-efficacy might stay in the hospital longer if they or their providers do not feel comfortable with their self-care ability.

This finding may be useful in developing hospital-based interventions. HL-specific interventions, several of which have been tested in the inpatient setting,^14,28,29 have shown promise toward improving health knowledge,³⁰ disease severity,³¹ and health resource utilization.³²

Those with low HL may lack the self-efficacy to participate in discharge planning; in fact, previous work has related low HL to posthospital readmissions.^8,9 Conversely, patients with low HL might struggle to engage in the inpatient milieu, advocating for shorter LOS if they feel alienated by the inpatient experience.

These possibilities show that LOS is a complex measure shown to depend on patient-level characteristics and on provider-based, geographical, and sociocultural factors.^16,33 With these forces at play, additional effects of lower levels of HL may be lost without phenotyping patients by both level of HL and related characteristics, such as self-efficacy, health skills, and stigma. By gathering these additional data, future work should explore whether subpopulations of patients with low HL may be at risk for too-short vs too-long hospital admissions.

For instance, in this study, both race and Medicaid insurance were associated with shorter LOS. Being African American was associated with shorter LOS in our study but has been found to be associated with longer LOS in another study specifically focused on diabetes.³⁴ Prior findings found uninsured patients have shorter LOS.³⁵ Therefore, these findings in our study are difficult to explain without further work to understand whether there are health disparities in the way patients are cared for during hospitalization that may shorten or lengthen their LOS because of factors outside of their clinical need.

The finding that gender modified the effect of low HL on LOS was unexpected. There were similar proportions of men and women with low HL. There is evidence to support that women make the majority of health decisions for themselves and their familes³⁶; therefore, there may be unmeasured aspects of HL that provide an advantage for female vs male inpatients. Furthermore, omitted confounders, such as social support, may not fully capture potential gender-related differences. Future work is needed to understand the role of gender in relationship to HL and LOS.

Limitations of this study include its observational, single-centered design with information derived from administrative data; positive and negative confounding cannot be ruled out. For instance, we did not control for complex aspects affecting LOS, such as discharge disposition and goals of care (eg, aggressive care after discharge vs hospice). To address this limitation, multivariate analyses were performed, which were adjusted for illness severity scores and took into account both comorbidity and severity of the current illness. Additionally, although it is important to study such populations, our largely urban, minority sample is not representative of the U.S. population, and within our large sample, there were participants with missing data who had lower HL on average, although this group represented only 5% of the sample. Finally, different HL tools have noncomplete concordance, which has been seen when comparing the BHLS with more objective tools.^20,37 Furthermore, certain in-hospital clinical scenarios (eg, recent stroke or prolonged intensive care unit stay) may present unique challenges in establishing a baseline HL level. However, the BHLS was used in this study because of its greater feasibility.

In conclusion, this study is the first to evaluate the relationship between low HL and LOS. The findings suggest that HL may play a role in shaping outcomes in the inpatient setting and that targeting interventions toward screened patients may be a pathway toward mitigating adverse effects. Our findings need to be replicated in larger, more representative samples, and further work understanding subpopulations within the low HL population is needed. Future work should measure this association in diverse inpatient settings (eg, psychiatric, surgical, and specialty), in addition to assessing associations between HL and other important in-hospital outcome measures, including mortality and discharge disposition.

 

Acknowledgments

The authors thank the Hospitalist Project team for their assistance with data collection. The authors especially thank Chuanhong Liao and Ashley Snyder for assistance with statistical analyses; Andrea Flores, Ainoa Coltri, and Tom Best for their assistance with data management. The authors would also like to thank Nicole Twu for her help with preparing and editing the manuscript.

Disclosures

Dr. Jaffee was supported by a Calvin Fentress Research Fellowship and NIH R25MH094612. Dr. Press was supported by a career development award (NHLBI K23HL118151). This work was also supported by a seed grant from the Center for Health Administration Studies. All other authors declare no conflicts of interest.

Health literacy (HL), defined as patients’ ability to understand health information and make health decisions,¹ is a prevalent problem in the outpatient and inpatient settings.^2,3 In both settings, low HL has adverse implications for self-care including interpreting health labels⁴ and taking medications correctly.⁵ Among outpatient cohorts, HL has been associated with worse outcomes and acute care utilization.⁶ Associations with low HL include increased hospitalizations,⁷ rehospitalizations,^8,9 emergency department visits,¹⁰ and decreased preventative care use.¹¹ Among the elderly, low HL is associated with increased mortality¹² and decreased self-perception of health.¹³

A systematic review revealed that most high-quality HL outcome studies were conducted in the outpatient setting.⁶ There have been very few studies assessing effects of low HL in an acute-care setting.^7,14 These studies have evaluated postdischarge outcomes, including admissions or readmissions,^7-9 and medication knowledge.¹⁴ To the best of our knowledge, there are no studies evaluating associations between HL and hospital length of stay (LOS).

LOS has received much attention as providers and payers focus more on resource utilization and eliminating adverse effects of prolonged hospitalization.¹⁵ LOS is multifactorial, depending on clinical characteristics like disease severity, as well as on sociocultural, demographic, and geographic factors.¹⁶ Despite evidence that LOS reductions translate into improved resource allocation and potentially fewer complications, there remains a tension between the appropriate LOS and one that is too short for a given condition.¹⁷

Because low HL is associated with inefficient resource utilization, we hypothesized that low HL would be associated with increased LOS after controlling for illness severity. Our objectives were to evaluate the association between low HL and LOS and whether such an association was modified by illness severity and sociodemographics.

METHODS

Study Design, Setting, Participants

An in-hospital, cohort study design of patients who were admitted or transferred to the general medicine service at the University of Chicago between October 2012 and November 2015 and screened for inclusion as part of a large, ongoing study of inpatient care quality was conducted.¹⁸ Exclusion criteria included observation status, age under 18 years, non-English speaking, and repeat participants. Those who died during hospitalization or whose discharge status was missing were excluded because the primary goal was to examine the association of HL and time to discharge, which could not be evaluated among those who died. We excluded participants with LOS >30 days to limit overly influential effects of extreme outliers (1% of the population).

Variables

HL was screened using the Brief Health Literacy Screen (BHLS), a validated, 3-question verbal survey not requiring adequate visual acuity to assess HL.^19,20 The 3 questions are as follows: (1) “How confident are you filling out medical forms by yourself?”, (2) “How often do you have someone help you read hospital materials?”, and (3) “How often do you have problems learning about your medical condition because of difficulty understanding written information?” Responses to the questions were scored on a 5-point Likert scale in which higher scores corresponded to higher HL.^21,22 The scores for each of the 3 questions were summed to yield a range between 3 and 15. On the individual questions, prior work has demonstrated improved test performance with a cutoff of ≤3, which corresponds to a response of “some of the time” or “somewhat”; therefore, when the 3 questions were summed together, scores of ≤9 were considered indicative of low HL.^21,23

For severity of illness adjustment, we used relative weights derived from the 3M (3M, Maplewood, MN) All Patient Refined Diagnosis Related Groups (APR-DRG) classification system, which uses administrative data to classify the severity. The APR-DRG system assigns each admission to a DRG based on principal diagnosis; for each DRG, patients are then subdivided into 4 severity classes based on age, comorbidity, and interactions between these variables and the admitting diagnosis.²⁴ Using the base DRG and severity score, the system assigns relative weights that reflect differences in expected hospital resource utilization.

LOS was derived from hospital administrative data and counted from the date of admission to the hospital. Participants who were discharged on the day of admission were counted as having an LOS of 1. Insurance status (Medicare, Medicaid, no payer, private) also was obtained from administrative data. Age, sex (male or female), education (junior high or less, some high school, high school graduate, some college, college graduate, postgraduate), and race (black/African American, white, Asian or Pacific Islander [including Asian Indian, Chinese, Filipino, Japanese, Korean, Vietnamese, other Asian, Native Hawaiian, Guam/Chamorro, Samoan, other Pacific], American Indian or Alaskan Native, multiple race) were obtained from administrative data based on information provided by the patient. Participants with missing data on any of the sociodemographic variables or on the APR-DRG score were excluded from the analysis.

 

Statistical Analysis

χ² and 2-tailed t tests were used to compare categorical and continuous variables, respectively. Multivariate linear regressions were employed to measure associations between the independent variables (HL, illness severity, race, gender, education, and insurance status) and the dependent variable, LOS. Independent variables were chosen for clinical significance and retained in the model regardless of statistical significance. The adjusted R² values of models with and without the HL variable included were reported to provide information on the contribution of HL to the overall model.

Because LOS was observed to be right skewed and residuals of the untransformed regression were observed to be non-normally distributed, the decision was made to natural log transform LOS, which is consistent with previous hospital LOS studies.¹⁶ Regression coefficients and confidence intervals were then transformed into percentage estimates using the following equation: 100(e^β–1). Adjusted R² was reported for the transformed regression.

The APR-DRG relative weight was treated as a continuous variable. Sociodemographic variables were dichotomized as follows: female vs male; high school graduates vs not; African American vs not; Medicaid/no payer vs Medicare/private payer. Age was not included in the multivariate model because it has been incorporated into the weighted APR-DRG illness severity scores.

Each of the sociodemographic variables and the APR-DRG score were examined for effect modification via the same multivariate linear equation described above, with the addition of an interaction term. A separate regression was performed with an interaction term between age (dichotomized at ≥65) and HL to investigate whether age modified the association between HL and LOS. Finally, we explored whether effects were isolated to long vs short LOS by dividing the sample based on the mean LOS (≥6 days) and performing separate multivariate comparisons.

Sensitivity analyses were performed to exclude those with LOS greater than the 90th percentile and those with APR-DRG score greater than the 90th percentile; age was added to the model as a continuous variable to evaluate whether the illness severity score fully adjusted for the effects of age on LOS. Furthermore, we compared the participants with missing data to those with complete data across both dependent and independent variables. Alpha was set at 0.05; analyses were performed using Stata Version 14 (Stata, College Station, TX).

RESULTS

A total of 5983 participants met inclusion criteria and completed the HL assessment; of these participants, 75 (1%) died during hospitalization, 9 (0.2%) had missing discharge status, and 79 (1%) had LOS >30 days. Two hundred eighty (5%) were missing data on sociodemographic variables or APR-DRG score. Of the remaining (n = 5540), the mean age was 57 years (standard deviation [SD] = 19 years), over half of participants were female (57%), and the majority were African American (73%) and had graduated from high school (81%). The sample was divided into those with private insurance (25%), those with Medicare (46%), and those with Medicaid (26%); 2% had no payer. The mean APR-DRG score was 1.3 (SD = 1.2), and the scores ranged from 0.3 to 15.8.

On the BHLS screen for HL, 20% (1104/5540) had inadequate HL. Participants with low HL had higher weighted illness severity scores (average 1.4 vs 1.3; P = 0.003). Participants with low HL were also more likely to be 65 or older (55% vs 33%; P < 0.001), non-high school graduates (35% vs 15%; P < 0.001), and African American (78% vs 72%; P < 0.001), and to have Medicare or private insurance (75% vs 71%; P = 0.02). There was no significant difference with respect to gender (54% male vs 57% female; P = 0.1)

The mean and median LOS were 6 ± 5 days and 4 days (interquartile range 2-7 days), respectively. Those with low HL had a longer average LOS (6.0 vs 5.4 days; P < 0.001). In multivariate analysis controlling for APR-DRG score, gender, education, race, and insurance status, low HL was associated with an 11.1% longer LOS (95% CI, 6.1-16.1; P < 0.001; Table 1). The adjusted R² value for the regression was 25.0% including HL and 24.7% with HL excluded. Additionally, being African American (P < 0.001) and having Medicaid or no insurance (P < 0.001) were associated with a shorter LOS in multivariate analysis (Table 1). The association of HL and LOS in multivariate modeling remained significant among participants with LOS <6 days (10.2%; 95% CI, 5.6%-14.9%; P < 0.001), but not among participants with LOS ≥6 days (0.4%; 95% CI, −3.6% to 4.4%; P = 0.8).

Neither age ≥65 (P = 0.4) nor illness severity score (P = 0.5) significantly modified the effect of HL on LOS. However, the effect of HL on hospital LOS was significantly modified by gender (P = 0.02). Among men, low HL was associated with a 17.8% longer LOS (95% CI, 10.0%-25.7%; P < 0.001), but among women, low HL was associated with only a 7.7% longer LOS (95% CI, 1.9%-13.5%; P = 0.009). Among the remaining demographics, high school graduation status (P = 0.4), being African American (P = 0.6), and insurance status (P = 0.2) did not significantly modify the effect of HL on LOS. In sensitivity analysis, excluding participants with LOS above the 90th percentile of 12 days and excluding participants with illness severity scores above the 90th percentile, low HL was still associated with longer LOS (P < 0.001 and P = 0.001, respectively; Table 2). In the final sensitivity analysis, although age remained a significant predictor of longer LOS after controlling for illness severity (0.2% increase per year, 95% CI, 0.1%-0.3%; P < 0.001), low HL nevertheless remained significantly associated with longer LOS (P < 0.001; Table 2).

Finally, we compared the group with missing data (n = 280) to the group with complete data (n = 5540). The participants with missing data were more likely to have low HL (31% [86/280] vs 20%; P < 0.001) and to have Medicare or private insurance (82% [177/217] vs 72%; P = 0.002); however, they were not more likely to be 65 or older (40% [112/280] vs 37%; P = 0.3), high school graduates (88% [113/129] vs 81%; P = 0.06), African American (69% [177/256] vs 73%; P = 0.1), or female (57% [158/279] vs 57%; P = 1), nor were they more likely to have longer LOS (5.7 [n = 280] vs 5.5 days; P = 0.6) or higher illness severity scores (1.3 [n = 231] vs 1.3; P = 0.7).

 

DISCUSSION

To our knowledge, this study is the first to evaluate the association between low HL and an important in-hospital outcome measure, hospital LOS. We found that low HL was associated with a longer hospital LOS, a result which remained significant when controlling for severity of illness and sociodemographic variables and when testing the model for sensitivity to the highest values of LOS and illness severity. Additionally, the association of HL with LOS appeared concentrated among participants with shorter LOS. Relative to other predictors, the contribution of HL to the overall LOS model was small, as evidenced by the change in adjusted R² values with HL excluded.

Among the covariates, only gender modified the association between HL and LOS; the findings suggested that men were more susceptible to the effect of low HL on increased LOS. Illness severity and other sociodemographics, including age ≥65, did not appear to modify the association. We also found that being African American and having Medicaid or no insurance were associated with a significantly shorter LOS in multivariate analysis.

Previous work suggested that the adverse health effects of low HL may be mediated through several pathways, including health knowledge, self-efficacy, health skills, and illness stigma.^25-27 The finding of a small but significant relationship between HL and LOS was not surprising given these known associations; nevertheless, there may be an additional patient-dependent effect of low HL on LOS not discovered here. For instance, patients with poor health knowledge and self-efficacy might stay in the hospital longer if they or their providers do not feel comfortable with their self-care ability.

This finding may be useful in developing hospital-based interventions. HL-specific interventions, several of which have been tested in the inpatient setting,^14,28,29 have shown promise toward improving health knowledge,³⁰ disease severity,³¹ and health resource utilization.³²

Those with low HL may lack the self-efficacy to participate in discharge planning; in fact, previous work has related low HL to posthospital readmissions.^8,9 Conversely, patients with low HL might struggle to engage in the inpatient milieu, advocating for shorter LOS if they feel alienated by the inpatient experience.

These possibilities show that LOS is a complex measure shown to depend on patient-level characteristics and on provider-based, geographical, and sociocultural factors.^16,33 With these forces at play, additional effects of lower levels of HL may be lost without phenotyping patients by both level of HL and related characteristics, such as self-efficacy, health skills, and stigma. By gathering these additional data, future work should explore whether subpopulations of patients with low HL may be at risk for too-short vs too-long hospital admissions.

For instance, in this study, both race and Medicaid insurance were associated with shorter LOS. Being African American was associated with shorter LOS in our study but has been found to be associated with longer LOS in another study specifically focused on diabetes.³⁴ Prior findings found uninsured patients have shorter LOS.³⁵ Therefore, these findings in our study are difficult to explain without further work to understand whether there are health disparities in the way patients are cared for during hospitalization that may shorten or lengthen their LOS because of factors outside of their clinical need.

The finding that gender modified the effect of low HL on LOS was unexpected. There were similar proportions of men and women with low HL. There is evidence to support that women make the majority of health decisions for themselves and their familes³⁶; therefore, there may be unmeasured aspects of HL that provide an advantage for female vs male inpatients. Furthermore, omitted confounders, such as social support, may not fully capture potential gender-related differences. Future work is needed to understand the role of gender in relationship to HL and LOS.

Limitations of this study include its observational, single-centered design with information derived from administrative data; positive and negative confounding cannot be ruled out. For instance, we did not control for complex aspects affecting LOS, such as discharge disposition and goals of care (eg, aggressive care after discharge vs hospice). To address this limitation, multivariate analyses were performed, which were adjusted for illness severity scores and took into account both comorbidity and severity of the current illness. Additionally, although it is important to study such populations, our largely urban, minority sample is not representative of the U.S. population, and within our large sample, there were participants with missing data who had lower HL on average, although this group represented only 5% of the sample. Finally, different HL tools have noncomplete concordance, which has been seen when comparing the BHLS with more objective tools.^20,37 Furthermore, certain in-hospital clinical scenarios (eg, recent stroke or prolonged intensive care unit stay) may present unique challenges in establishing a baseline HL level. However, the BHLS was used in this study because of its greater feasibility.

In conclusion, this study is the first to evaluate the relationship between low HL and LOS. The findings suggest that HL may play a role in shaping outcomes in the inpatient setting and that targeting interventions toward screened patients may be a pathway toward mitigating adverse effects. Our findings need to be replicated in larger, more representative samples, and further work understanding subpopulations within the low HL population is needed. Future work should measure this association in diverse inpatient settings (eg, psychiatric, surgical, and specialty), in addition to assessing associations between HL and other important in-hospital outcome measures, including mortality and discharge disposition.

 

Acknowledgments

The authors thank the Hospitalist Project team for their assistance with data collection. The authors especially thank Chuanhong Liao and Ashley Snyder for assistance with statistical analyses; Andrea Flores, Ainoa Coltri, and Tom Best for their assistance with data management. The authors would also like to thank Nicole Twu for her help with preparing and editing the manuscript.

Disclosures

Dr. Jaffee was supported by a Calvin Fentress Research Fellowship and NIH R25MH094612. Dr. Press was supported by a career development award (NHLBI K23HL118151). This work was also supported by a seed grant from the Center for Health Administration Studies. All other authors declare no conflicts of interest.

Evidence suggests that troponin-only testing is the superior strategy to diagnose acute myocardial infarction (AMI).¹ Because of this, in February 2015, the Choosing Wisely^® campaign issued a recommendation to use troponin I or T to diagnose AMI, and not to test for myoglobin or creatine kinase-MB (CK-MB).² This recommendation was in line with guidelines from the American Heart Association and the American College of Cardiology, which recommended that myoglobin and CK-MB are not useful and offer no benefit for the diagnosis of acute coronary syndrome.³ Some institutions have developed interventions to promote troponin-only testing, reporting substantial cost savings and no negative consequences.^4,5

Despite these successes, it is likely that institutions vary with respect to the adoption of the Choosing Wisely^® troponin-only testing recommendation.⁶ Implementing this recommendation requires both promoting clinician behavior change and a strong institutional culture of high-value care.⁷ Understanding the variation across institutions of troponin-only testing could inform how to promote high-value care recommendations nationwide. We aimed to describe patterns of troponin, myoglobin, and CK-MB testing in a sample of academic teaching hospitals before and after the Choosing Wisely^® recommendation.

METHODS

Troponin, myoglobin, and CK-MB ordering data were extracted from Vizient’s (formerly University HealthSystem Consortium, Chicago, IL) Clinical Database/Resource Manager (CDB/RM^®) for all patients with a principal discharge diagnosis of AMI at all hospitals reporting all 36 months from the fourth quarter of 2013 through the third quarter of 2016. This period includes time both before and after the Choosing Wisely^® recommendation, which was released in the first quarter of 2015. Vizient’s CDB/RM contains ordering data for 300 academic medical centers and their affiliated hospitals and includes the discharge diagnoses for patients cared for by these institutions. Only patients with a principal discharge diagnosis of AMI were included because the Choosing Wisely^® recommendation is specific with regard to troponin-only testing for the diagnosis of AMI. Patients with a principal diagnosis code for subcategories of myocardial ischemia (eg, stable angina, unstable angina) were not included because of the large number of diagnosis codes for these subcategories (more than 100 in the International Classification of Diseases, Ninth Revision and the International Classification of Diseases, Tenth Revision) and because the variation in their use across institutions within the dataset limited the utility of using these codes to consistently and accurately identify patients with myocardial ischemia. Moreover, the diagnosis of AMI encompasses the subcategories of myocardial ischemia.⁸

Hospital rates of ordering cardiac biomarkers (troponin-only or troponin and myoglobin/CK-MB) were determined overall for the entire study period and for each quarter of the study period based on the total patients with a discharge diagnosis of AMI. For each quarter of the 12 study quarters, all the hospitals were divided into tertiles based on their rate of troponin-only testing per discharge diagnosis of AMI. Hospitals were then classified into 3 groups based on their tertile ranking over the full 12 study quarters. The first group included hospitals whose rate of troponin-only testing placed them in the top tertile for each and all quarters throughout the study period. The second group included hospitals whose troponin-only testing rate placed them in the bottom tertile for each and all quarters throughout the study period. The third group included hospitals whose troponin-only testing rate each quarter led to either an increase or decrease in their tertile ranking throughout the study period. χ² tests were used to test for bivariate associations among hospitals based on their rate of troponin-only testing and hospital size (number of beds), their regional geographic location, the volume of AMI patients seen at the hospital, whether the primary physician during the hospitalization was a cardiologist or other provider, and the hospitals’ quality ratings. Quality rating was based on an internal Vizient rating and the “Best Hospitals for Cardiology and Heart Surgery Rankings” as published in the US News & World Report.⁹ The Vizient quality rating is based on a composite score that combines scores from the domains of quality (hospital quality incentive scores), safety (patient safety indicators), patient-centeredness (Hospital Consumer Assessment of Healthcare Providers and Systems Hospital Survey), and equity (distribution of care by race/ethnicity, gender, and age). Simple slopes were calculated to determine the rate of change in troponin-only testing for each study quarter, and Student t tests were used to compare the rates of change of these simple slopes across study quarters.

 

RESULTS

Of the 300 hospitals in Vizient’s CDB/RM, 91 (30%, 91/300) had full reporting of data throughout the study period. These hospitals had a total of 106,954 inpatient discharges with a principal diagnosis of AMI during the study period. The overall rates of troponin-only testing for AMI discharges by hospital varied from 0% to 87.4% (Figure 1). The mean rate of troponin-only testing across all patients with a discharge diagnosis of AMI was 29.2% at the start of the study (fourth quarter of 2013) and 53.5% at the end of the study (third quarter 2016; Supplemental Figure). Nineteen hospitals (21%, 19/91; 27,973 discharges) had high rates of troponin-only testing for AMI and were in the top tertile of all hospitals throughout the study period. Thirty-four hospitals (37%, 34/91; 35,080 discharges) ordered both troponin and myoglobin/CK-MB tests to diagnose AMI, and they were in the bottom tertile of all hospitals throughout the study period. In the 38 hospitals (42%, 38/91; 43,090 discharges) that were not in the top or bottom tertile for all study quarters, the rate of troponin-only testing for AMI increased at each hospital during each quarter of the study period (Table).

Pattern of Troponin-Only Testing by Hospital Size

Of the hospitals in the top tertile of troponin-only testing throughout the study period, the majority had ≥500 beds (13/19), but the highest rate of troponin-only testing was in hospitals that had <250 beds (n = 4, troponin-only testing rate of 82/100 patients). Additionally, in hospitals that improved their troponin-only testing during the study period, hospitals that had <500 beds had higher rates of troponin-only testing than did hospitals with ≥500 beds. The differences in the rates of troponin-only testing across the 3 groups of hospitals and hospital size were statistically significant (P < 0.0001; Table).

Pattern of Troponin-Only Testing by Geographic Region

The rate of troponin-only testing also varied and was statistically significantly different when comparing the 3 groups of hospitals across geographic regions of the country (P < 0.0001). Of the hospitals in the top tertile of troponin-only testing throughout the study period, the majority were in the Midwest (n = 6) and Mid-Atlantic (n = 5) regions. However, the rate of troponin-only testing for AMI in this group was highest in hospitals in the West (86/100 patients) and/or Southeast (75/100 patients) regions, although this rate was based on a small number of hospitals in these geographic areas (n = 1 in the West, n = 2 in the Southeast). Of hospitals in the bottom tertile of troponin-only testing throughout the study period, the majority were in the Mid-Atlantic region (n = 10). Hospitals that increased their troponin-only testing during the study period were predominantly in the Midwest (n = 12) and Mid-Atlantic regions (n = 11; Table), with the hospitals in the Midwest having the highest rate of troponin-only testing in this group.

Pattern of Troponin-Only Testing by Volume of AMI Patients

Of the hospitals in the top tertile of troponin-only testing during the study period, the majority cared for ≥1500 AMI patients (n = 9), but interestingly, among these hospitals, those caring for a smaller volume of AMI patients all had higher rates of troponin-only testing per 100 patients (P < 0.0001; Table). There was no other obvious pattern of troponin-only testing based on the volume of AMI patients cared for in hospitals in either the bottom tertile of troponin-only testing or hospitals that improved troponin-only testing during the study period.

Pattern of Troponin-Only Testing by Physician Type

Of the hospitals in the top tertile of troponin-only testing throughout the study period, those where a cardiologist cared for patients with AMI had higher rates of troponin-only testing (71/100 patients) than did hospitals where patients were cared for by a noncardiologist (60/100 patients). However, of the hospitals that improved their troponin-only testing during the study period, higher rates of troponin-only testing were seen in hospitals where patients were cared for by a noncardiologist (48/100 patients) compared with patients cared for by a cardiologist (34/100 patients; Table). These differences in hospital rates of troponin-only testing during the study period based on physician type were statistically significant (P < 0.0001; Table).

Pattern of Troponin-Only Testing by Quality Rating

Hospitals that were in the top tertile of troponin-only testing and were rated highly by Vizient’s quality rating or recognized as a top hospital by the US News & World Report had higher rates of troponin-only testing per 100 patients than did hospitals in the top tertile that were not ranked highly by Vizient’s quality rating or recognized as a top hospital by the US News & World Report. However, the majority of hospitals in the top tertile of troponin-only testing were not rated highly by Vizient (n = 15) or recognized as a top hospital by the US News & World Report (n = 16). The large majority of hospitals in the bottom tertile of troponin-only testing were not recognized as high-quality hospitals by Vizient (n = 32) or the US News & World Report (n = 31). Of the hospitals that improved their troponin-only testing during the study period, the majority were not recognized as high-quality hospitals by Vizient (n = 33) or the US News & World Report (n = 36), but among this group, those hospitals recognized by Vizient as high quality (n = 5) had the highest rate of troponin-only testing (57/100 patients). The differences in the rate of troponin-only testing across the different groups of hospitals and quality ratings were statistically significant (P < 0.0001; Table).

 

The Effect of Choosing Wisely^® on Troponin-Only Testing

While in many institutions the rates of troponin-only testing were increasing before the Choosing Wisely^® recommendation was released in 2015, the release of the recommendation was associated with a significant increase in the rate of troponin-only testing in the institutions that were in the bottom tertile of troponin-only testing prior to the release of the recommendation but moved to the top tertile after the release of the recommendation (n = 5). The slope percentage of the rate of change of the 5 hospitals that went from the bottom tertile to the top tertile after the release of the Choosing Wisely^® recommendation was 5.7%. Additionally, the Choosing Wisely^® recommendation was associated with an accelerated rate of troponin-only testing in hospitals moving from the bottom tertile before the release of the recommendation to the middle tertile after the recommendation (n = 15; slope = 3.2%) and in hospitals moving from the middle tertile before the release of the recommendation to the top tertile after (n = 6; slope = 2.4%) (Figure 2). For all of these hospitals (n = 26), the increased rate of troponin-only testing in the study quarter after the Choosing Wisely^® recommendation was statistically significantly higher and different from the rate of troponin-only testing in all other study quarters, except for the period between 2014 quarter 3 and quarter 4 (P = 0.08), the period between 2015 quarter 2 and quarter 3 (P = 0.18), and 2015 quarter 3 and quarter 4 (P = 0.06), where the effect did not quite reach statistical significance (Figure 3).

DISCUSSION

In a broad sample of academic teaching hospitals, there was an overall increase in the rate of troponin-only testing starting from the fourth quarter of 2013 through the third quarter of 2016. However, there was wide variation in the adoption of troponin-only testing for AMI across institutions. Our study identified several high-performing hospitals where the rate of troponin-only testing was high prior to and after the Choosing Wisely^® troponin-only recommendation. Additionally, we identified several poor-performing hospitals, which even after the release of the Choosing Wisely® recommendation continue to order both troponin and myoglobin/CK-MB tests for the diagnosis of AMI. Lastly, we identified several hospitals in which the release of the Choosing Wisely® recommendation was associated with a significant increase in the rate of troponin-only testing for the diagnosis of AMI. 
The high-performing hospitals in our sample that were in the top tertile of troponin-only testing throughout the study period are “early adopters,” having already instituted troponin-only testing before the release of the Choosing Wisely^® troponin-only recommendation. These hospitals vary in size, geographic region of the country, volume of AMI patients cared for, whether AMI patients are cared for by a cardiologist or other provider, and quality rating. Interestingly, in these hospitals, AMI patients admitted under the care of a cardiologist had higher rates of troponin-only testing than when admitted under another physician type. This is perhaps not surprising given that cardiologists would be the most likely to be aware of the data supporting troponin-only testing prior to the Choosing Wisely^® recommendation and the most likely to institute interventions to promote troponin-only testing and disseminate this knowledge across their institution. These institutions and their practice of troponin-only testing before the Choosing Wisely^® recommendation represent the idea of positive deviance,¹⁰ whereby they had identified troponin-only testing as a superior strategy and instituted successful initiatives to reduce the use of unnecessary myoglobin and CK-MB testing before their peer hospitals and the release of the Choosing Wisely^® recommendation. Further efforts to explore and understand the additional factors that define the hospitals that had high rates of troponin-only testing prior to the Choosing Wisely^® recommendation may be helpful to understanding the necessary culture and institutional factors that can promote high-value care.

In the hospitals that demonstrated increasing adoption of troponin-only testing, there are several interesting patterns. First, among these hospitals, smaller hospitals tended to have higher overall rates of troponin-only testing per 100 patients than larger hospitals. Additionally, the hospitals with the highest rates were located in the Midwest region. These hospitals may be learning from and following the high-performing institutions observed in our data that are also located in the Midwest. Additionally, among the hospitals that significantly increased their rate of troponin-only testing, we see that the Choosing Wisely^® recommendation appeared to facilitate accelerated adoption of troponin-only testing. In these institutions, it is likely that the impact of Choosing Wisely^® was significant because there was attention to high-value care and already an existing movement underway to institute such high-value practices. For example, natural champions, leadership, infrastructure, and a supportive culture may all be prerequisites for Choosing Wisely^® recommendations to become institutionally adopted.

Lastly, in the hospitals that have continued to order myoglobin and CK-MB, future work is needed to understand and overcome barriers to adopting high-value care practices.

There are several limitations to this study. First, because this was an observational study, we cannot prove a causal relationship between the Choosing Wisely^® recommendation and the increased rates of troponin-only testing. Additionally, the Vizient CDB/RM contains reporting data for a limited number of academic medical centers only, and therefore, these results may not represent practices at nonacademic or even other academic medical centers. Our study only included patients with a principal discharge diagnosis of AMI because the Choosing Wisely^® recommendation to order troponin-only is specific for diagnosing patients with AMI. However, it is possible that the Choosing Wisely^® recommendation also has affected provider ordering in patients with diagnoses such as chest pain or angina, and these affects would not be captured in our study. Lastly, because instituting high-value care practices take time, our follow-up time may not have been long enough to capture improvement in troponin-only testing at institutions responding to and attempting to adhere to the Choosing Wisely^® recommendation to order troponin-only testing for patients with AMI.

 

Disclosure 

No other individuals besides the authors contributed to this work. This project was not funded or supported by any external grant or agency. Dr. Prochaska’s institute received funding from the Agency for Research Healthcare and Quality for a K12 Career Development Grant (AHRQ K12 HS023007) outside the submitted work. Dr. Hohmann and Dr Modes have nothing to disclose. Dr. Arora receives financial compensation as a member of the Board of Directors for the American Board of Internal Medicine and has received grant funding from the ABIM Foundation. She also receives royalties from McGraw Hill.

Evidence suggests that troponin-only testing is the superior strategy to diagnose acute myocardial infarction (AMI).¹ Because of this, in February 2015, the Choosing Wisely^® campaign issued a recommendation to use troponin I or T to diagnose AMI, and not to test for myoglobin or creatine kinase-MB (CK-MB).² This recommendation was in line with guidelines from the American Heart Association and the American College of Cardiology, which recommended that myoglobin and CK-MB are not useful and offer no benefit for the diagnosis of acute coronary syndrome.³ Some institutions have developed interventions to promote troponin-only testing, reporting substantial cost savings and no negative consequences.^4,5

Despite these successes, it is likely that institutions vary with respect to the adoption of the Choosing Wisely^® troponin-only testing recommendation.⁶ Implementing this recommendation requires both promoting clinician behavior change and a strong institutional culture of high-value care.⁷ Understanding the variation across institutions of troponin-only testing could inform how to promote high-value care recommendations nationwide. We aimed to describe patterns of troponin, myoglobin, and CK-MB testing in a sample of academic teaching hospitals before and after the Choosing Wisely^® recommendation.

METHODS

Troponin, myoglobin, and CK-MB ordering data were extracted from Vizient’s (formerly University HealthSystem Consortium, Chicago, IL) Clinical Database/Resource Manager (CDB/RM^®) for all patients with a principal discharge diagnosis of AMI at all hospitals reporting all 36 months from the fourth quarter of 2013 through the third quarter of 2016. This period includes time both before and after the Choosing Wisely^® recommendation, which was released in the first quarter of 2015. Vizient’s CDB/RM contains ordering data for 300 academic medical centers and their affiliated hospitals and includes the discharge diagnoses for patients cared for by these institutions. Only patients with a principal discharge diagnosis of AMI were included because the Choosing Wisely^® recommendation is specific with regard to troponin-only testing for the diagnosis of AMI. Patients with a principal diagnosis code for subcategories of myocardial ischemia (eg, stable angina, unstable angina) were not included because of the large number of diagnosis codes for these subcategories (more than 100 in the International Classification of Diseases, Ninth Revision and the International Classification of Diseases, Tenth Revision) and because the variation in their use across institutions within the dataset limited the utility of using these codes to consistently and accurately identify patients with myocardial ischemia. Moreover, the diagnosis of AMI encompasses the subcategories of myocardial ischemia.⁸

Hospital rates of ordering cardiac biomarkers (troponin-only or troponin and myoglobin/CK-MB) were determined overall for the entire study period and for each quarter of the study period based on the total patients with a discharge diagnosis of AMI. For each quarter of the 12 study quarters, all the hospitals were divided into tertiles based on their rate of troponin-only testing per discharge diagnosis of AMI. Hospitals were then classified into 3 groups based on their tertile ranking over the full 12 study quarters. The first group included hospitals whose rate of troponin-only testing placed them in the top tertile for each and all quarters throughout the study period. The second group included hospitals whose troponin-only testing rate placed them in the bottom tertile for each and all quarters throughout the study period. The third group included hospitals whose troponin-only testing rate each quarter led to either an increase or decrease in their tertile ranking throughout the study period. χ² tests were used to test for bivariate associations among hospitals based on their rate of troponin-only testing and hospital size (number of beds), their regional geographic location, the volume of AMI patients seen at the hospital, whether the primary physician during the hospitalization was a cardiologist or other provider, and the hospitals’ quality ratings. Quality rating was based on an internal Vizient rating and the “Best Hospitals for Cardiology and Heart Surgery Rankings” as published in the US News & World Report.⁹ The Vizient quality rating is based on a composite score that combines scores from the domains of quality (hospital quality incentive scores), safety (patient safety indicators), patient-centeredness (Hospital Consumer Assessment of Healthcare Providers and Systems Hospital Survey), and equity (distribution of care by race/ethnicity, gender, and age). Simple slopes were calculated to determine the rate of change in troponin-only testing for each study quarter, and Student t tests were used to compare the rates of change of these simple slopes across study quarters.

 

RESULTS

Of the 300 hospitals in Vizient’s CDB/RM, 91 (30%, 91/300) had full reporting of data throughout the study period. These hospitals had a total of 106,954 inpatient discharges with a principal diagnosis of AMI during the study period. The overall rates of troponin-only testing for AMI discharges by hospital varied from 0% to 87.4% (Figure 1). The mean rate of troponin-only testing across all patients with a discharge diagnosis of AMI was 29.2% at the start of the study (fourth quarter of 2013) and 53.5% at the end of the study (third quarter 2016; Supplemental Figure). Nineteen hospitals (21%, 19/91; 27,973 discharges) had high rates of troponin-only testing for AMI and were in the top tertile of all hospitals throughout the study period. Thirty-four hospitals (37%, 34/91; 35,080 discharges) ordered both troponin and myoglobin/CK-MB tests to diagnose AMI, and they were in the bottom tertile of all hospitals throughout the study period. In the 38 hospitals (42%, 38/91; 43,090 discharges) that were not in the top or bottom tertile for all study quarters, the rate of troponin-only testing for AMI increased at each hospital during each quarter of the study period (Table).

Pattern of Troponin-Only Testing by Hospital Size

Of the hospitals in the top tertile of troponin-only testing throughout the study period, the majority had ≥500 beds (13/19), but the highest rate of troponin-only testing was in hospitals that had <250 beds (n = 4, troponin-only testing rate of 82/100 patients). Additionally, in hospitals that improved their troponin-only testing during the study period, hospitals that had <500 beds had higher rates of troponin-only testing than did hospitals with ≥500 beds. The differences in the rates of troponin-only testing across the 3 groups of hospitals and hospital size were statistically significant (P < 0.0001; Table).

Pattern of Troponin-Only Testing by Geographic Region

The rate of troponin-only testing also varied and was statistically significantly different when comparing the 3 groups of hospitals across geographic regions of the country (P < 0.0001). Of the hospitals in the top tertile of troponin-only testing throughout the study period, the majority were in the Midwest (n = 6) and Mid-Atlantic (n = 5) regions. However, the rate of troponin-only testing for AMI in this group was highest in hospitals in the West (86/100 patients) and/or Southeast (75/100 patients) regions, although this rate was based on a small number of hospitals in these geographic areas (n = 1 in the West, n = 2 in the Southeast). Of hospitals in the bottom tertile of troponin-only testing throughout the study period, the majority were in the Mid-Atlantic region (n = 10). Hospitals that increased their troponin-only testing during the study period were predominantly in the Midwest (n = 12) and Mid-Atlantic regions (n = 11; Table), with the hospitals in the Midwest having the highest rate of troponin-only testing in this group.

Pattern of Troponin-Only Testing by Volume of AMI Patients

Of the hospitals in the top tertile of troponin-only testing during the study period, the majority cared for ≥1500 AMI patients (n = 9), but interestingly, among these hospitals, those caring for a smaller volume of AMI patients all had higher rates of troponin-only testing per 100 patients (P < 0.0001; Table). There was no other obvious pattern of troponin-only testing based on the volume of AMI patients cared for in hospitals in either the bottom tertile of troponin-only testing or hospitals that improved troponin-only testing during the study period.

Pattern of Troponin-Only Testing by Physician Type

Of the hospitals in the top tertile of troponin-only testing throughout the study period, those where a cardiologist cared for patients with AMI had higher rates of troponin-only testing (71/100 patients) than did hospitals where patients were cared for by a noncardiologist (60/100 patients). However, of the hospitals that improved their troponin-only testing during the study period, higher rates of troponin-only testing were seen in hospitals where patients were cared for by a noncardiologist (48/100 patients) compared with patients cared for by a cardiologist (34/100 patients; Table). These differences in hospital rates of troponin-only testing during the study period based on physician type were statistically significant (P < 0.0001; Table).

Pattern of Troponin-Only Testing by Quality Rating

Hospitals that were in the top tertile of troponin-only testing and were rated highly by Vizient’s quality rating or recognized as a top hospital by the US News & World Report had higher rates of troponin-only testing per 100 patients than did hospitals in the top tertile that were not ranked highly by Vizient’s quality rating or recognized as a top hospital by the US News & World Report. However, the majority of hospitals in the top tertile of troponin-only testing were not rated highly by Vizient (n = 15) or recognized as a top hospital by the US News & World Report (n = 16). The large majority of hospitals in the bottom tertile of troponin-only testing were not recognized as high-quality hospitals by Vizient (n = 32) or the US News & World Report (n = 31). Of the hospitals that improved their troponin-only testing during the study period, the majority were not recognized as high-quality hospitals by Vizient (n = 33) or the US News & World Report (n = 36), but among this group, those hospitals recognized by Vizient as high quality (n = 5) had the highest rate of troponin-only testing (57/100 patients). The differences in the rate of troponin-only testing across the different groups of hospitals and quality ratings were statistically significant (P < 0.0001; Table).

 

The Effect of Choosing Wisely^® on Troponin-Only Testing

While in many institutions the rates of troponin-only testing were increasing before the Choosing Wisely^® recommendation was released in 2015, the release of the recommendation was associated with a significant increase in the rate of troponin-only testing in the institutions that were in the bottom tertile of troponin-only testing prior to the release of the recommendation but moved to the top tertile after the release of the recommendation (n = 5). The slope percentage of the rate of change of the 5 hospitals that went from the bottom tertile to the top tertile after the release of the Choosing Wisely^® recommendation was 5.7%. Additionally, the Choosing Wisely^® recommendation was associated with an accelerated rate of troponin-only testing in hospitals moving from the bottom tertile before the release of the recommendation to the middle tertile after the recommendation (n = 15; slope = 3.2%) and in hospitals moving from the middle tertile before the release of the recommendation to the top tertile after (n = 6; slope = 2.4%) (Figure 2). For all of these hospitals (n = 26), the increased rate of troponin-only testing in the study quarter after the Choosing Wisely^® recommendation was statistically significantly higher and different from the rate of troponin-only testing in all other study quarters, except for the period between 2014 quarter 3 and quarter 4 (P = 0.08), the period between 2015 quarter 2 and quarter 3 (P = 0.18), and 2015 quarter 3 and quarter 4 (P = 0.06), where the effect did not quite reach statistical significance (Figure 3).

DISCUSSION

In a broad sample of academic teaching hospitals, there was an overall increase in the rate of troponin-only testing starting from the fourth quarter of 2013 through the third quarter of 2016. However, there was wide variation in the adoption of troponin-only testing for AMI across institutions. Our study identified several high-performing hospitals where the rate of troponin-only testing was high prior to and after the Choosing Wisely^® troponin-only recommendation. Additionally, we identified several poor-performing hospitals, which even after the release of the Choosing Wisely® recommendation continue to order both troponin and myoglobin/CK-MB tests for the diagnosis of AMI. Lastly, we identified several hospitals in which the release of the Choosing Wisely® recommendation was associated with a significant increase in the rate of troponin-only testing for the diagnosis of AMI. 
The high-performing hospitals in our sample that were in the top tertile of troponin-only testing throughout the study period are “early adopters,” having already instituted troponin-only testing before the release of the Choosing Wisely^® troponin-only recommendation. These hospitals vary in size, geographic region of the country, volume of AMI patients cared for, whether AMI patients are cared for by a cardiologist or other provider, and quality rating. Interestingly, in these hospitals, AMI patients admitted under the care of a cardiologist had higher rates of troponin-only testing than when admitted under another physician type. This is perhaps not surprising given that cardiologists would be the most likely to be aware of the data supporting troponin-only testing prior to the Choosing Wisely^® recommendation and the most likely to institute interventions to promote troponin-only testing and disseminate this knowledge across their institution. These institutions and their practice of troponin-only testing before the Choosing Wisely^® recommendation represent the idea of positive deviance,¹⁰ whereby they had identified troponin-only testing as a superior strategy and instituted successful initiatives to reduce the use of unnecessary myoglobin and CK-MB testing before their peer hospitals and the release of the Choosing Wisely^® recommendation. Further efforts to explore and understand the additional factors that define the hospitals that had high rates of troponin-only testing prior to the Choosing Wisely^® recommendation may be helpful to understanding the necessary culture and institutional factors that can promote high-value care.

In the hospitals that demonstrated increasing adoption of troponin-only testing, there are several interesting patterns. First, among these hospitals, smaller hospitals tended to have higher overall rates of troponin-only testing per 100 patients than larger hospitals. Additionally, the hospitals with the highest rates were located in the Midwest region. These hospitals may be learning from and following the high-performing institutions observed in our data that are also located in the Midwest. Additionally, among the hospitals that significantly increased their rate of troponin-only testing, we see that the Choosing Wisely^® recommendation appeared to facilitate accelerated adoption of troponin-only testing. In these institutions, it is likely that the impact of Choosing Wisely^® was significant because there was attention to high-value care and already an existing movement underway to institute such high-value practices. For example, natural champions, leadership, infrastructure, and a supportive culture may all be prerequisites for Choosing Wisely^® recommendations to become institutionally adopted.

Lastly, in the hospitals that have continued to order myoglobin and CK-MB, future work is needed to understand and overcome barriers to adopting high-value care practices.

There are several limitations to this study. First, because this was an observational study, we cannot prove a causal relationship between the Choosing Wisely^® recommendation and the increased rates of troponin-only testing. Additionally, the Vizient CDB/RM contains reporting data for a limited number of academic medical centers only, and therefore, these results may not represent practices at nonacademic or even other academic medical centers. Our study only included patients with a principal discharge diagnosis of AMI because the Choosing Wisely^® recommendation to order troponin-only is specific for diagnosing patients with AMI. However, it is possible that the Choosing Wisely^® recommendation also has affected provider ordering in patients with diagnoses such as chest pain or angina, and these affects would not be captured in our study. Lastly, because instituting high-value care practices take time, our follow-up time may not have been long enough to capture improvement in troponin-only testing at institutions responding to and attempting to adhere to the Choosing Wisely^® recommendation to order troponin-only testing for patients with AMI.

 

Disclosure 

No other individuals besides the authors contributed to this work. This project was not funded or supported by any external grant or agency. Dr. Prochaska’s institute received funding from the Agency for Research Healthcare and Quality for a K12 Career Development Grant (AHRQ K12 HS023007) outside the submitted work. Dr. Hohmann and Dr Modes have nothing to disclose. Dr. Arora receives financial compensation as a member of the Board of Directors for the American Board of Internal Medicine and has received grant funding from the ABIM Foundation. She also receives royalties from McGraw Hill.

Evidence suggests that troponin-only testing is the superior strategy to diagnose acute myocardial infarction (AMI).¹ Because of this, in February 2015, the Choosing Wisely^® campaign issued a recommendation to use troponin I or T to diagnose AMI, and not to test for myoglobin or creatine kinase-MB (CK-MB).² This recommendation was in line with guidelines from the American Heart Association and the American College of Cardiology, which recommended that myoglobin and CK-MB are not useful and offer no benefit for the diagnosis of acute coronary syndrome.³ Some institutions have developed interventions to promote troponin-only testing, reporting substantial cost savings and no negative consequences.^4,5

Despite these successes, it is likely that institutions vary with respect to the adoption of the Choosing Wisely^® troponin-only testing recommendation.⁶ Implementing this recommendation requires both promoting clinician behavior change and a strong institutional culture of high-value care.⁷ Understanding the variation across institutions of troponin-only testing could inform how to promote high-value care recommendations nationwide. We aimed to describe patterns of troponin, myoglobin, and CK-MB testing in a sample of academic teaching hospitals before and after the Choosing Wisely^® recommendation.

METHODS

Troponin, myoglobin, and CK-MB ordering data were extracted from Vizient’s (formerly University HealthSystem Consortium, Chicago, IL) Clinical Database/Resource Manager (CDB/RM^®) for all patients with a principal discharge diagnosis of AMI at all hospitals reporting all 36 months from the fourth quarter of 2013 through the third quarter of 2016. This period includes time both before and after the Choosing Wisely^® recommendation, which was released in the first quarter of 2015. Vizient’s CDB/RM contains ordering data for 300 academic medical centers and their affiliated hospitals and includes the discharge diagnoses for patients cared for by these institutions. Only patients with a principal discharge diagnosis of AMI were included because the Choosing Wisely^® recommendation is specific with regard to troponin-only testing for the diagnosis of AMI. Patients with a principal diagnosis code for subcategories of myocardial ischemia (eg, stable angina, unstable angina) were not included because of the large number of diagnosis codes for these subcategories (more than 100 in the International Classification of Diseases, Ninth Revision and the International Classification of Diseases, Tenth Revision) and because the variation in their use across institutions within the dataset limited the utility of using these codes to consistently and accurately identify patients with myocardial ischemia. Moreover, the diagnosis of AMI encompasses the subcategories of myocardial ischemia.⁸

Hospital rates of ordering cardiac biomarkers (troponin-only or troponin and myoglobin/CK-MB) were determined overall for the entire study period and for each quarter of the study period based on the total patients with a discharge diagnosis of AMI. For each quarter of the 12 study quarters, all the hospitals were divided into tertiles based on their rate of troponin-only testing per discharge diagnosis of AMI. Hospitals were then classified into 3 groups based on their tertile ranking over the full 12 study quarters. The first group included hospitals whose rate of troponin-only testing placed them in the top tertile for each and all quarters throughout the study period. The second group included hospitals whose troponin-only testing rate placed them in the bottom tertile for each and all quarters throughout the study period. The third group included hospitals whose troponin-only testing rate each quarter led to either an increase or decrease in their tertile ranking throughout the study period. χ² tests were used to test for bivariate associations among hospitals based on their rate of troponin-only testing and hospital size (number of beds), their regional geographic location, the volume of AMI patients seen at the hospital, whether the primary physician during the hospitalization was a cardiologist or other provider, and the hospitals’ quality ratings. Quality rating was based on an internal Vizient rating and the “Best Hospitals for Cardiology and Heart Surgery Rankings” as published in the US News & World Report.⁹ The Vizient quality rating is based on a composite score that combines scores from the domains of quality (hospital quality incentive scores), safety (patient safety indicators), patient-centeredness (Hospital Consumer Assessment of Healthcare Providers and Systems Hospital Survey), and equity (distribution of care by race/ethnicity, gender, and age). Simple slopes were calculated to determine the rate of change in troponin-only testing for each study quarter, and Student t tests were used to compare the rates of change of these simple slopes across study quarters.

 

RESULTS

Of the 300 hospitals in Vizient’s CDB/RM, 91 (30%, 91/300) had full reporting of data throughout the study period. These hospitals had a total of 106,954 inpatient discharges with a principal diagnosis of AMI during the study period. The overall rates of troponin-only testing for AMI discharges by hospital varied from 0% to 87.4% (Figure 1). The mean rate of troponin-only testing across all patients with a discharge diagnosis of AMI was 29.2% at the start of the study (fourth quarter of 2013) and 53.5% at the end of the study (third quarter 2016; Supplemental Figure). Nineteen hospitals (21%, 19/91; 27,973 discharges) had high rates of troponin-only testing for AMI and were in the top tertile of all hospitals throughout the study period. Thirty-four hospitals (37%, 34/91; 35,080 discharges) ordered both troponin and myoglobin/CK-MB tests to diagnose AMI, and they were in the bottom tertile of all hospitals throughout the study period. In the 38 hospitals (42%, 38/91; 43,090 discharges) that were not in the top or bottom tertile for all study quarters, the rate of troponin-only testing for AMI increased at each hospital during each quarter of the study period (Table).

Pattern of Troponin-Only Testing by Hospital Size

Of the hospitals in the top tertile of troponin-only testing throughout the study period, the majority had ≥500 beds (13/19), but the highest rate of troponin-only testing was in hospitals that had <250 beds (n = 4, troponin-only testing rate of 82/100 patients). Additionally, in hospitals that improved their troponin-only testing during the study period, hospitals that had <500 beds had higher rates of troponin-only testing than did hospitals with ≥500 beds. The differences in the rates of troponin-only testing across the 3 groups of hospitals and hospital size were statistically significant (P < 0.0001; Table).

Pattern of Troponin-Only Testing by Geographic Region

The rate of troponin-only testing also varied and was statistically significantly different when comparing the 3 groups of hospitals across geographic regions of the country (P < 0.0001). Of the hospitals in the top tertile of troponin-only testing throughout the study period, the majority were in the Midwest (n = 6) and Mid-Atlantic (n = 5) regions. However, the rate of troponin-only testing for AMI in this group was highest in hospitals in the West (86/100 patients) and/or Southeast (75/100 patients) regions, although this rate was based on a small number of hospitals in these geographic areas (n = 1 in the West, n = 2 in the Southeast). Of hospitals in the bottom tertile of troponin-only testing throughout the study period, the majority were in the Mid-Atlantic region (n = 10). Hospitals that increased their troponin-only testing during the study period were predominantly in the Midwest (n = 12) and Mid-Atlantic regions (n = 11; Table), with the hospitals in the Midwest having the highest rate of troponin-only testing in this group.

Pattern of Troponin-Only Testing by Volume of AMI Patients

Of the hospitals in the top tertile of troponin-only testing during the study period, the majority cared for ≥1500 AMI patients (n = 9), but interestingly, among these hospitals, those caring for a smaller volume of AMI patients all had higher rates of troponin-only testing per 100 patients (P < 0.0001; Table). There was no other obvious pattern of troponin-only testing based on the volume of AMI patients cared for in hospitals in either the bottom tertile of troponin-only testing or hospitals that improved troponin-only testing during the study period.

Pattern of Troponin-Only Testing by Physician Type

Of the hospitals in the top tertile of troponin-only testing throughout the study period, those where a cardiologist cared for patients with AMI had higher rates of troponin-only testing (71/100 patients) than did hospitals where patients were cared for by a noncardiologist (60/100 patients). However, of the hospitals that improved their troponin-only testing during the study period, higher rates of troponin-only testing were seen in hospitals where patients were cared for by a noncardiologist (48/100 patients) compared with patients cared for by a cardiologist (34/100 patients; Table). These differences in hospital rates of troponin-only testing during the study period based on physician type were statistically significant (P < 0.0001; Table).

Pattern of Troponin-Only Testing by Quality Rating

Hospitals that were in the top tertile of troponin-only testing and were rated highly by Vizient’s quality rating or recognized as a top hospital by the US News & World Report had higher rates of troponin-only testing per 100 patients than did hospitals in the top tertile that were not ranked highly by Vizient’s quality rating or recognized as a top hospital by the US News & World Report. However, the majority of hospitals in the top tertile of troponin-only testing were not rated highly by Vizient (n = 15) or recognized as a top hospital by the US News & World Report (n = 16). The large majority of hospitals in the bottom tertile of troponin-only testing were not recognized as high-quality hospitals by Vizient (n = 32) or the US News & World Report (n = 31). Of the hospitals that improved their troponin-only testing during the study period, the majority were not recognized as high-quality hospitals by Vizient (n = 33) or the US News & World Report (n = 36), but among this group, those hospitals recognized by Vizient as high quality (n = 5) had the highest rate of troponin-only testing (57/100 patients). The differences in the rate of troponin-only testing across the different groups of hospitals and quality ratings were statistically significant (P < 0.0001; Table).

 

The Effect of Choosing Wisely^® on Troponin-Only Testing

While in many institutions the rates of troponin-only testing were increasing before the Choosing Wisely^® recommendation was released in 2015, the release of the recommendation was associated with a significant increase in the rate of troponin-only testing in the institutions that were in the bottom tertile of troponin-only testing prior to the release of the recommendation but moved to the top tertile after the release of the recommendation (n = 5). The slope percentage of the rate of change of the 5 hospitals that went from the bottom tertile to the top tertile after the release of the Choosing Wisely^® recommendation was 5.7%. Additionally, the Choosing Wisely^® recommendation was associated with an accelerated rate of troponin-only testing in hospitals moving from the bottom tertile before the release of the recommendation to the middle tertile after the recommendation (n = 15; slope = 3.2%) and in hospitals moving from the middle tertile before the release of the recommendation to the top tertile after (n = 6; slope = 2.4%) (Figure 2). For all of these hospitals (n = 26), the increased rate of troponin-only testing in the study quarter after the Choosing Wisely^® recommendation was statistically significantly higher and different from the rate of troponin-only testing in all other study quarters, except for the period between 2014 quarter 3 and quarter 4 (P = 0.08), the period between 2015 quarter 2 and quarter 3 (P = 0.18), and 2015 quarter 3 and quarter 4 (P = 0.06), where the effect did not quite reach statistical significance (Figure 3).

DISCUSSION

In a broad sample of academic teaching hospitals, there was an overall increase in the rate of troponin-only testing starting from the fourth quarter of 2013 through the third quarter of 2016. However, there was wide variation in the adoption of troponin-only testing for AMI across institutions. Our study identified several high-performing hospitals where the rate of troponin-only testing was high prior to and after the Choosing Wisely^® troponin-only recommendation. Additionally, we identified several poor-performing hospitals, which even after the release of the Choosing Wisely® recommendation continue to order both troponin and myoglobin/CK-MB tests for the diagnosis of AMI. Lastly, we identified several hospitals in which the release of the Choosing Wisely® recommendation was associated with a significant increase in the rate of troponin-only testing for the diagnosis of AMI. 
The high-performing hospitals in our sample that were in the top tertile of troponin-only testing throughout the study period are “early adopters,” having already instituted troponin-only testing before the release of the Choosing Wisely^® troponin-only recommendation. These hospitals vary in size, geographic region of the country, volume of AMI patients cared for, whether AMI patients are cared for by a cardiologist or other provider, and quality rating. Interestingly, in these hospitals, AMI patients admitted under the care of a cardiologist had higher rates of troponin-only testing than when admitted under another physician type. This is perhaps not surprising given that cardiologists would be the most likely to be aware of the data supporting troponin-only testing prior to the Choosing Wisely^® recommendation and the most likely to institute interventions to promote troponin-only testing and disseminate this knowledge across their institution. These institutions and their practice of troponin-only testing before the Choosing Wisely^® recommendation represent the idea of positive deviance,¹⁰ whereby they had identified troponin-only testing as a superior strategy and instituted successful initiatives to reduce the use of unnecessary myoglobin and CK-MB testing before their peer hospitals and the release of the Choosing Wisely^® recommendation. Further efforts to explore and understand the additional factors that define the hospitals that had high rates of troponin-only testing prior to the Choosing Wisely^® recommendation may be helpful to understanding the necessary culture and institutional factors that can promote high-value care.

In the hospitals that demonstrated increasing adoption of troponin-only testing, there are several interesting patterns. First, among these hospitals, smaller hospitals tended to have higher overall rates of troponin-only testing per 100 patients than larger hospitals. Additionally, the hospitals with the highest rates were located in the Midwest region. These hospitals may be learning from and following the high-performing institutions observed in our data that are also located in the Midwest. Additionally, among the hospitals that significantly increased their rate of troponin-only testing, we see that the Choosing Wisely^® recommendation appeared to facilitate accelerated adoption of troponin-only testing. In these institutions, it is likely that the impact of Choosing Wisely^® was significant because there was attention to high-value care and already an existing movement underway to institute such high-value practices. For example, natural champions, leadership, infrastructure, and a supportive culture may all be prerequisites for Choosing Wisely^® recommendations to become institutionally adopted.

Lastly, in the hospitals that have continued to order myoglobin and CK-MB, future work is needed to understand and overcome barriers to adopting high-value care practices.

There are several limitations to this study. First, because this was an observational study, we cannot prove a causal relationship between the Choosing Wisely^® recommendation and the increased rates of troponin-only testing. Additionally, the Vizient CDB/RM contains reporting data for a limited number of academic medical centers only, and therefore, these results may not represent practices at nonacademic or even other academic medical centers. Our study only included patients with a principal discharge diagnosis of AMI because the Choosing Wisely^® recommendation to order troponin-only is specific for diagnosing patients with AMI. However, it is possible that the Choosing Wisely^® recommendation also has affected provider ordering in patients with diagnoses such as chest pain or angina, and these affects would not be captured in our study. Lastly, because instituting high-value care practices take time, our follow-up time may not have been long enough to capture improvement in troponin-only testing at institutions responding to and attempting to adhere to the Choosing Wisely^® recommendation to order troponin-only testing for patients with AMI.

 

Disclosure 

No other individuals besides the authors contributed to this work. This project was not funded or supported by any external grant or agency. Dr. Prochaska’s institute received funding from the Agency for Research Healthcare and Quality for a K12 Career Development Grant (AHRQ K12 HS023007) outside the submitted work. Dr. Hohmann and Dr Modes have nothing to disclose. Dr. Arora receives financial compensation as a member of the Board of Directors for the American Board of Internal Medicine and has received grant funding from the ABIM Foundation. She also receives royalties from McGraw Hill.

Given the frequency, potential preventability, and costs associated with hospital readmissions, reducing readmissions is a priority in efforts to improve the quality and value of healthcare.^1,2 State and national bodies have created diverse initiatives to facilitate improvements in hospital discharge practices and reduce 30-day readmission rates across payers.^3-5 For example, the Agency for Healthcare Research and Quality (AHRQ) and the Institute for Healthcare Improvement have published tools for improving discharge practices.^6,7 Medicare instituted financial penalties for hospitals with higher-than-expected readmission rates for acute myocardial infarction (AMI), heart failure (HF), and pneumonia in 2012, while private payers and Medicaid programs have established their own policies.^8-13 Furthermore, private payers and Medicaid programs shifted toward capitated and value-based reimbursement models in which readmissions lead to financial losses for hospitals.^14,15 Accordingly, hospitals have implemented diverse interventions to reduce readmissions.^16,17 From 2009 to 2013, 30-day readmissions declined among privately insured adults (from 12.4% to 11.7%), Medicare patients (from 22.0% to 20.0%), and uninsured individuals (11.5% to 11.0%) but climbed among patients with Medicaid (from 19.8% to 20.5%) after index admissions for AMI, HF, pneumonia, or chronic obstructive pulmonary disease.¹⁸

To date, research, policies, and quality improvement interventions have largely focused on improvements to one aspect of the system of care—that provided in the inpatient setting—among older adults with Medicare. Yet, inpatient readmissions may underestimate how often patients return to the hospital because patients can be placed under observation or stabilized and discharged from the emergency department (ED) instead of being readmitted. Observation and ED visits are less costly to payers than inpatient admissions.¹⁹ Thus, information about utilization of inpatient, observation, and ED visits within 30 days of hospital discharge may be more informative than inpatient readmissions alone. However, little is known about trends in returns to the hospital for observation and ED visits and whether such trends vary by payer.

Our objective was to assess whether changes have occurred in rates of total 30-day, all-cause, unplanned returns to the hospital among adults with index admissions for AMI, HF, and pneumonia in which returns to the hospital included inpatient readmissions, observation visits, and ED visits. We also assessed whether changes in the rate of hospital inpatient readmissions coincided with changes in rates of returns for ED or observation visits. To examine the effects of readmission policies implemented by diverse payers and broad changes to the health system following the Affordable Care Act, we compared data from 201 hospitals in 4 states in 2009 and 2010 with data from the same hospitals for 2013 and 2014.

METHODS

Data Sources, Populations, and Study Variables

We used Healthcare Cost and Utilization Project (HCUP) State Inpatient Databases, State Emergency Department Databases, and State Ambulatory Surgery and Services Databases from Georgia, Nebraska, South Carolina, and Tennessee. These states comprise 7% of the US population and were the only states with data that included all observation and ED visits as well as encrypted patient identification numbers that permitted linkage across facilities and hospitals.²⁰

Index admissions for patients aged 18 years and older were eligible if they occurred at nonfederal general medical/surgical hospitals (excluding critical access hospitals) that had at least 1 index admission per target condition per year and at least 5 inpatient, observation, and ED visits for any condition per year.

We classified patients into the following 4 populations by age and insurance coverage: 18 to 64 years with private insurance, 65 years and older with Medicare (excluding younger adults with Medicare), 18 to 64 years with Medicaid, and 18 to 64 years without insurance. We identified patients aged 65 years and older with Medicare by using the primary or secondary expected payer for the index admission. This group included patients who were dually eligible for Medicare and Medicaid. If Medicare was not the primary or secondary payer, we used the primary payer to identify Medicaid, privately insured, and uninsured patients aged 18 to 64 years. None of the states expanded Medicaid coverage during the years studied.

The primary outcome of interest was the rate of having 1 or more all-cause, unplanned return(s) to an acute care hospital within 30 days of discharge after an index admission for AMI, HF, and pneumonia as defined by a modified version of Centers for Medicare & Medicaid Services’ readmission metrics.^21,22 We examined total return rates as well as rates for inpatient, observation, and ED care. We also examined the leading diagnoses associated with returns to the hospital. For each index admission, we included only 1 return visit, giving priority to inpatient readmissions, then observation visits, and then ED visits.

The HCUP databases are consistent with the definition of limited data sets under the Health Insurance Portability and Accountability Act Privacy Rule and contain no direct patient identifiers. The AHRQ Institutional Review Board considers research using HCUP data to have exempt status.

 

Statistical Analysis

To compare rates at which patients returned to the hospital during 2 cohort periods (2009 and 2010 vs 2013 and 2014), we used coarsened exact matching, a well-established matching technique for balancing covariates between 2 populations of patients that may be related to the outcome.²³ For observational datasets, coarsened exact matching is preferable to traditional matching because it enables the investigator to assess balance between the 2 populations, select the desired degree of balance, and eliminate observations for which comparable matches cannot be found.

We assembled sets of index admissions in each study period that were similar with respect to payer, primary diagnosis, and other factors. Matching variables included the patient’s age group, sex, and Elixhauser Comorbidity Index²⁴ (in deciles), as well as the hospital’s ratio of observation visits relative to inpatient admissions in 2009 and 2010 combined (in quartiles; see supplementary Appendix). For Medicare beneficiaries, we also matched on dual enrollment in Medicaid.

We conducted the matching process separately for each target condition and payer population. First, we grouped index admissions in both periods into strata defined by all possible combinations of the matching variables and allowing one-to-many random matching within strata. We then dropped records in any strata for which there were no records in 1 of the time periods. Finally, we calculated weights based on the size of each stratum. We used these weights to account for the different numbers of index admissions in each stratum between the 2 study periods. For example, if a stratum contained 10 index admissions in 2009 and 2010 combined and 20 in 2013 and 2014 combined, an admission weighed double in the earlier period. After weighting, the index admissions in each period (2009 and 2010; 2013 and 2014) had similar characteristics (Table 1).

After matching and weighting, we compared the percentage of index admissions for which patients returned to the hospital and the primary diagnoses at the return visit between the 2 study periods using 2-sided χ² tests (P < 0.05). Analyses were conducted by using SAS software (version 9.4; SAS Institute Inc., Cary, NC).

RESULTS

There were 423,503 eligible index admissions for AMI, HF, and pneumonia in the 2 periods combined; 422,840 (99.8%) were successfully matched and included in this analysis. After matching weights were applied, there were few statistically significant differences across the 2 time periods (see Table 1 and supplementary Appendix).

From 2009 and 2010 to 2013 and 2014, the percentage of patients hospitalized for AMI, HF, and pneumonia who had only observation or ED visits when they returned to the hospital increased from 41.4% to 46.7% among patients with private insurance (P < 0.001), from 27.8% to 32.1% among older patients with Medicare (P < 0.001), from 39.5% to 41.8% among patients with Medicaid (P = 0.03), and from 49.2% to 52.8% among patients without insurance (P = 0.004; Table 1). The percentage of returns to the hospital for observation increased across all payers (P < 0.001); in 2013 and 2014 combined, observation visits ranged from 6.8% of hospital returns among patients with Medicare to 11.1% among patients with private insurance. The percentage of returns to the hospital for an ED visit increased among patients with private insurance (P = 0.02) and Medicare (P < 0.001); in 2013 and 2014, ED visits ranged from 25.3% of returns to the hospital among patients with Medicare to 42.9% among uninsured patients.

The increases in 30-day observation and ED visits coincided with reductions in inpatient readmissions among patients with private insurance and Medicare and contributed to growth in total returns to the hospital among patients with Medicaid or no insurance (Figure 1).

Among privately insured individuals, a decline in inpatient readmissions (from 8.9% to 8.2%; P = 0.001) coincided with increases in observation visits (from 1.2% to 1.7%; P < 0.001) and ED visits (from 5.1% to 5.5%; P = 0.02), leading to a stable rate of approximately 15% at which patients with AMI, HF, or pneumonia returned to the hospital during both periods (P = 0.45). Among Medicare patients, inpatient readmissions declined from 18.3% to 16.9% (P < 0.001), while observation visits and ED visits increased (from 1.2% to 1.7% and 5.8% to 6.3%, respectively; P < 0.001), leading to a small net decrease in total returns to the hospital (25.3% vs 25.0%; P = 0.04). Among Medicaid recipients, inpatient readmissions were unchanged (18.7%; P = 0.93), but an increase in observation visits (from 2.0% to 2.7%; P < 0.001) and a nonsignificant increase in ED visits (from 10.3% to 10.7%; P = 0.26) led to a rise in total revisits (31.0% vs 32.1%; P = 0.04). Among uninsured adults, inpatient readmissions were stable (around 9.5%; P = 0.76), while there was a rise in observation visits (1.3% vs 2.0%; P < 0.001) and ED visits (8.0% vs 8.6%; P = 0.04), yielding an increase in total revisits (18.8% vs 20.1%; P = 0.004).

Figure 2

shows individual differences for each of the 3 target conditions between 2009 and 2010 versus 2013 and 2014 by payer. Overall, rates at which patients returned to the hospital within 30 days remained stable, with 3 exceptions. For patients with private insurance, total returns to the hospital rose for pneumonia (14.8% vs 15.9%; P = 0.02). For seniors with Medicare, total returns to the hospital declined for pneumonia (from 24.1% to 23.5%; P = 0.03). Among the uninsured, total returns to the hospital rose for AMI (15.5% vs 17.2%; P = 0.02).

Patients initially hospitalized for HF and pneumonia who returned to the hospital within 30 days often returned for the same conditions (Table 2).

Reasons for returning to the hospital were similar in the 2 periods (2009 and 2010; 2013 and 2014) across the 3 target conditions. However, when patients returned to the hospital in 2013 and 2014 with the same diagnosis as the index admission, they were less likely to be readmitted and more likely to be placed under observation than in 2009 and 2010.

 

DISCUSSION

Matching index admissions for AMI, HF, or pneumonia in 201 hospitals in 2009 and 2010 with those in 2013 and 2014, we observed that increases in observation and ED visits coincided with reductions in inpatient readmissions among patients with private insurance and Medicare and contributed to growth in total returns to the hospital among patients with Medicaid or no insurance. Among patients with private insurance and Medicare, inpatient readmissions declined significantly for all 3 target conditions, but total returns to the hospital remained constant for AMI and HF, rose for privately insured patients with pneumonia, and declined modestly for Medicare patients with pneumonia. Inpatient readmissions were unchanged for adults aged 18 to 64 years with Medicaid or no insurance, but total returns to the hospital increased significantly, reaching 32% among those with Medicaid.

These findings add to recent literature, which has primarily emphasized inpatient readmissions among Medicare beneficiaries with several exceptions. A prior analysis indicates that readmissions have declined among diverse payer populations nationally.¹⁸ Gerhardt et al²⁵ found that from 2011 to 2012, all-cause 30-day readmissions declined among fee-for-service (FFS) Medicare beneficiaries following any index admission, while ED revisits remained stable and observation revisits increased slightly. Evaluating the CMS Hospital Readmission Reductions Program (HRRP), Zuckerman et al¹⁷ reported that from 2007 to 2015, inpatient readmissions declined among FFS Medicare beneficiaries aged 65 years and older who were hospitalized with AMI, HF, or pneumonia, while returns to the hospital for observation rose approximately 2%; ED visits were not included. We found that Medicare (FFS and Medicare Advantage) patients with AMI and HF returned to the hospital with the same frequency in 2009 and 2010 as in 2013 and 2014, and those patients with pneumonia returned slightly less often. In aggregate, declines in inpatient readmissions in the 4 states we studied coincided with increases in observation and ED care. Moreover, these shifts occurred not only among Medicare beneficiaries but also among privately insured adults, Medicaid recipients, and the uninsured.

Three factors may have contributed to these apparent shifts from readmissions to observation and ED visits. First, some authors have suggested that hospitals may reduce readmissions by intentionally placing some of the patients who return to the hospital under observation instead of admitting them.^17,26 If true, hospitals with greater declines in readmissions would have larger increases in observation revisits. Zuckerman et al¹⁷ found no correlation among Medicare beneficiaries between hospital-level trends in observation revisits and readmissions, but returns to observation rose more rapidly for AMI, HF, and pneumonia (compared with other conditions) during long term follow-up than during the HRRP implementation period. Other authors have documented that declines in readmissions have been greatest at hospitals with the highest baseline readmission rates,^27,28 and hospitals with lower readmission rates have more observation return visits.²⁹

Second, shifts from inpatient readmissions to return visits for observation may reflect unintentional rather than intentional changes in the services provided. Clinical practice patterns are evolving such that patients who present to the hospital for acute care increasingly are placed under observation or discharged from the ED instead of being admitted, regardless of whether they recently were hospitalized.³⁰ Inpatient admissions, which are strongly correlated with readmission rates,^28,31 are declining nationally,³² and both observation and ED visits are rising.^33-35 Although little is known about effects on health outcomes and patient out-of-pocket costs,shifts from inpatient admissions to observation and ED visits reduce costs to payers.^36,37

Third, instead of substitution, more patients may be returning for lower-acuity conditions that can be treated in the ED or under observation. Hospitals are implementing diverse and multifaceted interventions to reduce readmissions that can involve assessing patient needs and the risk for readmission, educating patients about self-care and risks after discharge, reconciling medication, scheduling follow-up visits, and monitoring patients through telephone calls and home nursing visits.^26,38,39 Although the intent may be to reduce patients’ need to return to the hospital, interventions that educate patients about risks after discharge may lower the threshold at which they find symptoms worrisome enough to return. This could increase lower-acuity return visits. We found that reasons for returning were similar in 2009 and 2010 versus 2013 and 2014, but we did not examine acuity of illness at the time of return.

Other areas of concern are the high rates at which Medicaid patients are returning to the hospital and the increases in rates of returns among Medicaid patients and the uninsured. Individuals in these disadvantaged populations may be having difficulty accessing ambulatory care or may be turning to the ED more often for lower acuity problems that arise after discharge. In 3 of the 4 states we studied, 15% to 16% of adults live in poverty and 10% to 30% live in primary care health professional shortage areas.^40,41 Given the implications for patient outcomes and costs, trends among these populations warrant further scrutiny.^42,43

This analysis has several limitations. Data were from 4 states, but trends in readmissions are similar nationally. From 2010 through 2015, the all-condition readmission rate declined by 8% among Medicare beneficiaries nationally and by 6.1% in South Carolina, 7.4% in Georgia, 8.3% in Nebraska, and 8.7% in Tennessee.⁴⁴ We report trends across hospitals and did not examine hospital-level revisits. Therefore, further research is needed to determine whether these findings are related to co-occurring trends, intentional substitution, or other factors.

In conclusion, measuring inpatient readmissions without accounting for return visits to the ED and observation underestimates the rate at which patients return to the hospital following an inpatient hospitalization. Because of growth in observation and ED visits, trends in the total rates at which patients return to the hospital can differ from trends in inpatient readmissions. In the 4 states we studied, total return rates were particularly high and rising among patients with Medicaid and lower, but also rising, among the uninsured. Policy analysts and researchers should investigate the factors contributing to growth in readmissions in these vulnerable populations and determine whether similar trends are occurring nationwide. Hospitalists play critical roles in admitting and discharging inpatients, caring for patients under observation, and implementing quality improvement programs. Irrespective of payer, hospitalists’ efforts to improve the quality and value of care should include observation and ED visits as well as inpatient readmissions.

 

Acknowledgments

The authors gratefully acknowledge Minya Sheng, M.S. (Truven Health Analytics) for assistance in programming and data management and Linda Lee, Ph.D. (Truven Health Analytics) for providing editorial review of the manuscript. We also wish to acknowledge the 4 HCUP Partner organizations that contributed to the HCUP State Databases used in this study: Georgia Hospital Association, Nebraska Hospital Association, South Carolina Revenue and Fiscal Affairs Office, and Tennessee Hospital Association.

Disclosure

Funding for this study was provided by the AHRQ Center for Delivery, Organization, and Markets, HCUP (Contract No. HHSA-290-2013-00002-C). The views expressed in this article are those of the authors and do not necessarily reflect those of the AHRQ or the U.S. Department of Health and Human Services. The authors have no conflicts of interest or financial disclosures to declare.

Given the frequency, potential preventability, and costs associated with hospital readmissions, reducing readmissions is a priority in efforts to improve the quality and value of healthcare.^1,2 State and national bodies have created diverse initiatives to facilitate improvements in hospital discharge practices and reduce 30-day readmission rates across payers.^3-5 For example, the Agency for Healthcare Research and Quality (AHRQ) and the Institute for Healthcare Improvement have published tools for improving discharge practices.^6,7 Medicare instituted financial penalties for hospitals with higher-than-expected readmission rates for acute myocardial infarction (AMI), heart failure (HF), and pneumonia in 2012, while private payers and Medicaid programs have established their own policies.^8-13 Furthermore, private payers and Medicaid programs shifted toward capitated and value-based reimbursement models in which readmissions lead to financial losses for hospitals.^14,15 Accordingly, hospitals have implemented diverse interventions to reduce readmissions.^16,17 From 2009 to 2013, 30-day readmissions declined among privately insured adults (from 12.4% to 11.7%), Medicare patients (from 22.0% to 20.0%), and uninsured individuals (11.5% to 11.0%) but climbed among patients with Medicaid (from 19.8% to 20.5%) after index admissions for AMI, HF, pneumonia, or chronic obstructive pulmonary disease.¹⁸

To date, research, policies, and quality improvement interventions have largely focused on improvements to one aspect of the system of care—that provided in the inpatient setting—among older adults with Medicare. Yet, inpatient readmissions may underestimate how often patients return to the hospital because patients can be placed under observation or stabilized and discharged from the emergency department (ED) instead of being readmitted. Observation and ED visits are less costly to payers than inpatient admissions.¹⁹ Thus, information about utilization of inpatient, observation, and ED visits within 30 days of hospital discharge may be more informative than inpatient readmissions alone. However, little is known about trends in returns to the hospital for observation and ED visits and whether such trends vary by payer.

Our objective was to assess whether changes have occurred in rates of total 30-day, all-cause, unplanned returns to the hospital among adults with index admissions for AMI, HF, and pneumonia in which returns to the hospital included inpatient readmissions, observation visits, and ED visits. We also assessed whether changes in the rate of hospital inpatient readmissions coincided with changes in rates of returns for ED or observation visits. To examine the effects of readmission policies implemented by diverse payers and broad changes to the health system following the Affordable Care Act, we compared data from 201 hospitals in 4 states in 2009 and 2010 with data from the same hospitals for 2013 and 2014.

METHODS

Data Sources, Populations, and Study Variables

We used Healthcare Cost and Utilization Project (HCUP) State Inpatient Databases, State Emergency Department Databases, and State Ambulatory Surgery and Services Databases from Georgia, Nebraska, South Carolina, and Tennessee. These states comprise 7% of the US population and were the only states with data that included all observation and ED visits as well as encrypted patient identification numbers that permitted linkage across facilities and hospitals.²⁰

Index admissions for patients aged 18 years and older were eligible if they occurred at nonfederal general medical/surgical hospitals (excluding critical access hospitals) that had at least 1 index admission per target condition per year and at least 5 inpatient, observation, and ED visits for any condition per year.

We classified patients into the following 4 populations by age and insurance coverage: 18 to 64 years with private insurance, 65 years and older with Medicare (excluding younger adults with Medicare), 18 to 64 years with Medicaid, and 18 to 64 years without insurance. We identified patients aged 65 years and older with Medicare by using the primary or secondary expected payer for the index admission. This group included patients who were dually eligible for Medicare and Medicaid. If Medicare was not the primary or secondary payer, we used the primary payer to identify Medicaid, privately insured, and uninsured patients aged 18 to 64 years. None of the states expanded Medicaid coverage during the years studied.

The primary outcome of interest was the rate of having 1 or more all-cause, unplanned return(s) to an acute care hospital within 30 days of discharge after an index admission for AMI, HF, and pneumonia as defined by a modified version of Centers for Medicare & Medicaid Services’ readmission metrics.^21,22 We examined total return rates as well as rates for inpatient, observation, and ED care. We also examined the leading diagnoses associated with returns to the hospital. For each index admission, we included only 1 return visit, giving priority to inpatient readmissions, then observation visits, and then ED visits.

The HCUP databases are consistent with the definition of limited data sets under the Health Insurance Portability and Accountability Act Privacy Rule and contain no direct patient identifiers. The AHRQ Institutional Review Board considers research using HCUP data to have exempt status.

 

Statistical Analysis

To compare rates at which patients returned to the hospital during 2 cohort periods (2009 and 2010 vs 2013 and 2014), we used coarsened exact matching, a well-established matching technique for balancing covariates between 2 populations of patients that may be related to the outcome.²³ For observational datasets, coarsened exact matching is preferable to traditional matching because it enables the investigator to assess balance between the 2 populations, select the desired degree of balance, and eliminate observations for which comparable matches cannot be found.

We assembled sets of index admissions in each study period that were similar with respect to payer, primary diagnosis, and other factors. Matching variables included the patient’s age group, sex, and Elixhauser Comorbidity Index²⁴ (in deciles), as well as the hospital’s ratio of observation visits relative to inpatient admissions in 2009 and 2010 combined (in quartiles; see supplementary Appendix). For Medicare beneficiaries, we also matched on dual enrollment in Medicaid.

We conducted the matching process separately for each target condition and payer population. First, we grouped index admissions in both periods into strata defined by all possible combinations of the matching variables and allowing one-to-many random matching within strata. We then dropped records in any strata for which there were no records in 1 of the time periods. Finally, we calculated weights based on the size of each stratum. We used these weights to account for the different numbers of index admissions in each stratum between the 2 study periods. For example, if a stratum contained 10 index admissions in 2009 and 2010 combined and 20 in 2013 and 2014 combined, an admission weighed double in the earlier period. After weighting, the index admissions in each period (2009 and 2010; 2013 and 2014) had similar characteristics (Table 1).

After matching and weighting, we compared the percentage of index admissions for which patients returned to the hospital and the primary diagnoses at the return visit between the 2 study periods using 2-sided χ² tests (P < 0.05). Analyses were conducted by using SAS software (version 9.4; SAS Institute Inc., Cary, NC).

RESULTS

There were 423,503 eligible index admissions for AMI, HF, and pneumonia in the 2 periods combined; 422,840 (99.8%) were successfully matched and included in this analysis. After matching weights were applied, there were few statistically significant differences across the 2 time periods (see Table 1 and supplementary Appendix).

From 2009 and 2010 to 2013 and 2014, the percentage of patients hospitalized for AMI, HF, and pneumonia who had only observation or ED visits when they returned to the hospital increased from 41.4% to 46.7% among patients with private insurance (P < 0.001), from 27.8% to 32.1% among older patients with Medicare (P < 0.001), from 39.5% to 41.8% among patients with Medicaid (P = 0.03), and from 49.2% to 52.8% among patients without insurance (P = 0.004; Table 1). The percentage of returns to the hospital for observation increased across all payers (P < 0.001); in 2013 and 2014 combined, observation visits ranged from 6.8% of hospital returns among patients with Medicare to 11.1% among patients with private insurance. The percentage of returns to the hospital for an ED visit increased among patients with private insurance (P = 0.02) and Medicare (P < 0.001); in 2013 and 2014, ED visits ranged from 25.3% of returns to the hospital among patients with Medicare to 42.9% among uninsured patients.

The increases in 30-day observation and ED visits coincided with reductions in inpatient readmissions among patients with private insurance and Medicare and contributed to growth in total returns to the hospital among patients with Medicaid or no insurance (Figure 1).

Among privately insured individuals, a decline in inpatient readmissions (from 8.9% to 8.2%; P = 0.001) coincided with increases in observation visits (from 1.2% to 1.7%; P < 0.001) and ED visits (from 5.1% to 5.5%; P = 0.02), leading to a stable rate of approximately 15% at which patients with AMI, HF, or pneumonia returned to the hospital during both periods (P = 0.45). Among Medicare patients, inpatient readmissions declined from 18.3% to 16.9% (P < 0.001), while observation visits and ED visits increased (from 1.2% to 1.7% and 5.8% to 6.3%, respectively; P < 0.001), leading to a small net decrease in total returns to the hospital (25.3% vs 25.0%; P = 0.04). Among Medicaid recipients, inpatient readmissions were unchanged (18.7%; P = 0.93), but an increase in observation visits (from 2.0% to 2.7%; P < 0.001) and a nonsignificant increase in ED visits (from 10.3% to 10.7%; P = 0.26) led to a rise in total revisits (31.0% vs 32.1%; P = 0.04). Among uninsured adults, inpatient readmissions were stable (around 9.5%; P = 0.76), while there was a rise in observation visits (1.3% vs 2.0%; P < 0.001) and ED visits (8.0% vs 8.6%; P = 0.04), yielding an increase in total revisits (18.8% vs 20.1%; P = 0.004).

Figure 2

shows individual differences for each of the 3 target conditions between 2009 and 2010 versus 2013 and 2014 by payer. Overall, rates at which patients returned to the hospital within 30 days remained stable, with 3 exceptions. For patients with private insurance, total returns to the hospital rose for pneumonia (14.8% vs 15.9%; P = 0.02). For seniors with Medicare, total returns to the hospital declined for pneumonia (from 24.1% to 23.5%; P = 0.03). Among the uninsured, total returns to the hospital rose for AMI (15.5% vs 17.2%; P = 0.02).

Patients initially hospitalized for HF and pneumonia who returned to the hospital within 30 days often returned for the same conditions (Table 2).

Reasons for returning to the hospital were similar in the 2 periods (2009 and 2010; 2013 and 2014) across the 3 target conditions. However, when patients returned to the hospital in 2013 and 2014 with the same diagnosis as the index admission, they were less likely to be readmitted and more likely to be placed under observation than in 2009 and 2010.

 

DISCUSSION

Matching index admissions for AMI, HF, or pneumonia in 201 hospitals in 2009 and 2010 with those in 2013 and 2014, we observed that increases in observation and ED visits coincided with reductions in inpatient readmissions among patients with private insurance and Medicare and contributed to growth in total returns to the hospital among patients with Medicaid or no insurance. Among patients with private insurance and Medicare, inpatient readmissions declined significantly for all 3 target conditions, but total returns to the hospital remained constant for AMI and HF, rose for privately insured patients with pneumonia, and declined modestly for Medicare patients with pneumonia. Inpatient readmissions were unchanged for adults aged 18 to 64 years with Medicaid or no insurance, but total returns to the hospital increased significantly, reaching 32% among those with Medicaid.

These findings add to recent literature, which has primarily emphasized inpatient readmissions among Medicare beneficiaries with several exceptions. A prior analysis indicates that readmissions have declined among diverse payer populations nationally.¹⁸ Gerhardt et al²⁵ found that from 2011 to 2012, all-cause 30-day readmissions declined among fee-for-service (FFS) Medicare beneficiaries following any index admission, while ED revisits remained stable and observation revisits increased slightly. Evaluating the CMS Hospital Readmission Reductions Program (HRRP), Zuckerman et al¹⁷ reported that from 2007 to 2015, inpatient readmissions declined among FFS Medicare beneficiaries aged 65 years and older who were hospitalized with AMI, HF, or pneumonia, while returns to the hospital for observation rose approximately 2%; ED visits were not included. We found that Medicare (FFS and Medicare Advantage) patients with AMI and HF returned to the hospital with the same frequency in 2009 and 2010 as in 2013 and 2014, and those patients with pneumonia returned slightly less often. In aggregate, declines in inpatient readmissions in the 4 states we studied coincided with increases in observation and ED care. Moreover, these shifts occurred not only among Medicare beneficiaries but also among privately insured adults, Medicaid recipients, and the uninsured.

Three factors may have contributed to these apparent shifts from readmissions to observation and ED visits. First, some authors have suggested that hospitals may reduce readmissions by intentionally placing some of the patients who return to the hospital under observation instead of admitting them.^17,26 If true, hospitals with greater declines in readmissions would have larger increases in observation revisits. Zuckerman et al¹⁷ found no correlation among Medicare beneficiaries between hospital-level trends in observation revisits and readmissions, but returns to observation rose more rapidly for AMI, HF, and pneumonia (compared with other conditions) during long term follow-up than during the HRRP implementation period. Other authors have documented that declines in readmissions have been greatest at hospitals with the highest baseline readmission rates,^27,28 and hospitals with lower readmission rates have more observation return visits.²⁹

Second, shifts from inpatient readmissions to return visits for observation may reflect unintentional rather than intentional changes in the services provided. Clinical practice patterns are evolving such that patients who present to the hospital for acute care increasingly are placed under observation or discharged from the ED instead of being admitted, regardless of whether they recently were hospitalized.³⁰ Inpatient admissions, which are strongly correlated with readmission rates,^28,31 are declining nationally,³² and both observation and ED visits are rising.^33-35 Although little is known about effects on health outcomes and patient out-of-pocket costs,shifts from inpatient admissions to observation and ED visits reduce costs to payers.^36,37

Third, instead of substitution, more patients may be returning for lower-acuity conditions that can be treated in the ED or under observation. Hospitals are implementing diverse and multifaceted interventions to reduce readmissions that can involve assessing patient needs and the risk for readmission, educating patients about self-care and risks after discharge, reconciling medication, scheduling follow-up visits, and monitoring patients through telephone calls and home nursing visits.^26,38,39 Although the intent may be to reduce patients’ need to return to the hospital, interventions that educate patients about risks after discharge may lower the threshold at which they find symptoms worrisome enough to return. This could increase lower-acuity return visits. We found that reasons for returning were similar in 2009 and 2010 versus 2013 and 2014, but we did not examine acuity of illness at the time of return.

Other areas of concern are the high rates at which Medicaid patients are returning to the hospital and the increases in rates of returns among Medicaid patients and the uninsured. Individuals in these disadvantaged populations may be having difficulty accessing ambulatory care or may be turning to the ED more often for lower acuity problems that arise after discharge. In 3 of the 4 states we studied, 15% to 16% of adults live in poverty and 10% to 30% live in primary care health professional shortage areas.^40,41 Given the implications for patient outcomes and costs, trends among these populations warrant further scrutiny.^42,43

This analysis has several limitations. Data were from 4 states, but trends in readmissions are similar nationally. From 2010 through 2015, the all-condition readmission rate declined by 8% among Medicare beneficiaries nationally and by 6.1% in South Carolina, 7.4% in Georgia, 8.3% in Nebraska, and 8.7% in Tennessee.⁴⁴ We report trends across hospitals and did not examine hospital-level revisits. Therefore, further research is needed to determine whether these findings are related to co-occurring trends, intentional substitution, or other factors.

In conclusion, measuring inpatient readmissions without accounting for return visits to the ED and observation underestimates the rate at which patients return to the hospital following an inpatient hospitalization. Because of growth in observation and ED visits, trends in the total rates at which patients return to the hospital can differ from trends in inpatient readmissions. In the 4 states we studied, total return rates were particularly high and rising among patients with Medicaid and lower, but also rising, among the uninsured. Policy analysts and researchers should investigate the factors contributing to growth in readmissions in these vulnerable populations and determine whether similar trends are occurring nationwide. Hospitalists play critical roles in admitting and discharging inpatients, caring for patients under observation, and implementing quality improvement programs. Irrespective of payer, hospitalists’ efforts to improve the quality and value of care should include observation and ED visits as well as inpatient readmissions.

 

Acknowledgments

The authors gratefully acknowledge Minya Sheng, M.S. (Truven Health Analytics) for assistance in programming and data management and Linda Lee, Ph.D. (Truven Health Analytics) for providing editorial review of the manuscript. We also wish to acknowledge the 4 HCUP Partner organizations that contributed to the HCUP State Databases used in this study: Georgia Hospital Association, Nebraska Hospital Association, South Carolina Revenue and Fiscal Affairs Office, and Tennessee Hospital Association.

Disclosure

Funding for this study was provided by the AHRQ Center for Delivery, Organization, and Markets, HCUP (Contract No. HHSA-290-2013-00002-C). The views expressed in this article are those of the authors and do not necessarily reflect those of the AHRQ or the U.S. Department of Health and Human Services. The authors have no conflicts of interest or financial disclosures to declare.

Given the frequency, potential preventability, and costs associated with hospital readmissions, reducing readmissions is a priority in efforts to improve the quality and value of healthcare.^1,2 State and national bodies have created diverse initiatives to facilitate improvements in hospital discharge practices and reduce 30-day readmission rates across payers.^3-5 For example, the Agency for Healthcare Research and Quality (AHRQ) and the Institute for Healthcare Improvement have published tools for improving discharge practices.^6,7 Medicare instituted financial penalties for hospitals with higher-than-expected readmission rates for acute myocardial infarction (AMI), heart failure (HF), and pneumonia in 2012, while private payers and Medicaid programs have established their own policies.^8-13 Furthermore, private payers and Medicaid programs shifted toward capitated and value-based reimbursement models in which readmissions lead to financial losses for hospitals.^14,15 Accordingly, hospitals have implemented diverse interventions to reduce readmissions.^16,17 From 2009 to 2013, 30-day readmissions declined among privately insured adults (from 12.4% to 11.7%), Medicare patients (from 22.0% to 20.0%), and uninsured individuals (11.5% to 11.0%) but climbed among patients with Medicaid (from 19.8% to 20.5%) after index admissions for AMI, HF, pneumonia, or chronic obstructive pulmonary disease.¹⁸

To date, research, policies, and quality improvement interventions have largely focused on improvements to one aspect of the system of care—that provided in the inpatient setting—among older adults with Medicare. Yet, inpatient readmissions may underestimate how often patients return to the hospital because patients can be placed under observation or stabilized and discharged from the emergency department (ED) instead of being readmitted. Observation and ED visits are less costly to payers than inpatient admissions.¹⁹ Thus, information about utilization of inpatient, observation, and ED visits within 30 days of hospital discharge may be more informative than inpatient readmissions alone. However, little is known about trends in returns to the hospital for observation and ED visits and whether such trends vary by payer.

Our objective was to assess whether changes have occurred in rates of total 30-day, all-cause, unplanned returns to the hospital among adults with index admissions for AMI, HF, and pneumonia in which returns to the hospital included inpatient readmissions, observation visits, and ED visits. We also assessed whether changes in the rate of hospital inpatient readmissions coincided with changes in rates of returns for ED or observation visits. To examine the effects of readmission policies implemented by diverse payers and broad changes to the health system following the Affordable Care Act, we compared data from 201 hospitals in 4 states in 2009 and 2010 with data from the same hospitals for 2013 and 2014.

METHODS

Data Sources, Populations, and Study Variables

We used Healthcare Cost and Utilization Project (HCUP) State Inpatient Databases, State Emergency Department Databases, and State Ambulatory Surgery and Services Databases from Georgia, Nebraska, South Carolina, and Tennessee. These states comprise 7% of the US population and were the only states with data that included all observation and ED visits as well as encrypted patient identification numbers that permitted linkage across facilities and hospitals.²⁰

Index admissions for patients aged 18 years and older were eligible if they occurred at nonfederal general medical/surgical hospitals (excluding critical access hospitals) that had at least 1 index admission per target condition per year and at least 5 inpatient, observation, and ED visits for any condition per year.

We classified patients into the following 4 populations by age and insurance coverage: 18 to 64 years with private insurance, 65 years and older with Medicare (excluding younger adults with Medicare), 18 to 64 years with Medicaid, and 18 to 64 years without insurance. We identified patients aged 65 years and older with Medicare by using the primary or secondary expected payer for the index admission. This group included patients who were dually eligible for Medicare and Medicaid. If Medicare was not the primary or secondary payer, we used the primary payer to identify Medicaid, privately insured, and uninsured patients aged 18 to 64 years. None of the states expanded Medicaid coverage during the years studied.

The primary outcome of interest was the rate of having 1 or more all-cause, unplanned return(s) to an acute care hospital within 30 days of discharge after an index admission for AMI, HF, and pneumonia as defined by a modified version of Centers for Medicare & Medicaid Services’ readmission metrics.^21,22 We examined total return rates as well as rates for inpatient, observation, and ED care. We also examined the leading diagnoses associated with returns to the hospital. For each index admission, we included only 1 return visit, giving priority to inpatient readmissions, then observation visits, and then ED visits.

The HCUP databases are consistent with the definition of limited data sets under the Health Insurance Portability and Accountability Act Privacy Rule and contain no direct patient identifiers. The AHRQ Institutional Review Board considers research using HCUP data to have exempt status.

 

Statistical Analysis

To compare rates at which patients returned to the hospital during 2 cohort periods (2009 and 2010 vs 2013 and 2014), we used coarsened exact matching, a well-established matching technique for balancing covariates between 2 populations of patients that may be related to the outcome.²³ For observational datasets, coarsened exact matching is preferable to traditional matching because it enables the investigator to assess balance between the 2 populations, select the desired degree of balance, and eliminate observations for which comparable matches cannot be found.

We assembled sets of index admissions in each study period that were similar with respect to payer, primary diagnosis, and other factors. Matching variables included the patient’s age group, sex, and Elixhauser Comorbidity Index²⁴ (in deciles), as well as the hospital’s ratio of observation visits relative to inpatient admissions in 2009 and 2010 combined (in quartiles; see supplementary Appendix). For Medicare beneficiaries, we also matched on dual enrollment in Medicaid.

We conducted the matching process separately for each target condition and payer population. First, we grouped index admissions in both periods into strata defined by all possible combinations of the matching variables and allowing one-to-many random matching within strata. We then dropped records in any strata for which there were no records in 1 of the time periods. Finally, we calculated weights based on the size of each stratum. We used these weights to account for the different numbers of index admissions in each stratum between the 2 study periods. For example, if a stratum contained 10 index admissions in 2009 and 2010 combined and 20 in 2013 and 2014 combined, an admission weighed double in the earlier period. After weighting, the index admissions in each period (2009 and 2010; 2013 and 2014) had similar characteristics (Table 1).

After matching and weighting, we compared the percentage of index admissions for which patients returned to the hospital and the primary diagnoses at the return visit between the 2 study periods using 2-sided χ² tests (P < 0.05). Analyses were conducted by using SAS software (version 9.4; SAS Institute Inc., Cary, NC).

RESULTS

There were 423,503 eligible index admissions for AMI, HF, and pneumonia in the 2 periods combined; 422,840 (99.8%) were successfully matched and included in this analysis. After matching weights were applied, there were few statistically significant differences across the 2 time periods (see Table 1 and supplementary Appendix).

From 2009 and 2010 to 2013 and 2014, the percentage of patients hospitalized for AMI, HF, and pneumonia who had only observation or ED visits when they returned to the hospital increased from 41.4% to 46.7% among patients with private insurance (P < 0.001), from 27.8% to 32.1% among older patients with Medicare (P < 0.001), from 39.5% to 41.8% among patients with Medicaid (P = 0.03), and from 49.2% to 52.8% among patients without insurance (P = 0.004; Table 1). The percentage of returns to the hospital for observation increased across all payers (P < 0.001); in 2013 and 2014 combined, observation visits ranged from 6.8% of hospital returns among patients with Medicare to 11.1% among patients with private insurance. The percentage of returns to the hospital for an ED visit increased among patients with private insurance (P = 0.02) and Medicare (P < 0.001); in 2013 and 2014, ED visits ranged from 25.3% of returns to the hospital among patients with Medicare to 42.9% among uninsured patients.

The increases in 30-day observation and ED visits coincided with reductions in inpatient readmissions among patients with private insurance and Medicare and contributed to growth in total returns to the hospital among patients with Medicaid or no insurance (Figure 1).

Among privately insured individuals, a decline in inpatient readmissions (from 8.9% to 8.2%; P = 0.001) coincided with increases in observation visits (from 1.2% to 1.7%; P < 0.001) and ED visits (from 5.1% to 5.5%; P = 0.02), leading to a stable rate of approximately 15% at which patients with AMI, HF, or pneumonia returned to the hospital during both periods (P = 0.45). Among Medicare patients, inpatient readmissions declined from 18.3% to 16.9% (P < 0.001), while observation visits and ED visits increased (from 1.2% to 1.7% and 5.8% to 6.3%, respectively; P < 0.001), leading to a small net decrease in total returns to the hospital (25.3% vs 25.0%; P = 0.04). Among Medicaid recipients, inpatient readmissions were unchanged (18.7%; P = 0.93), but an increase in observation visits (from 2.0% to 2.7%; P < 0.001) and a nonsignificant increase in ED visits (from 10.3% to 10.7%; P = 0.26) led to a rise in total revisits (31.0% vs 32.1%; P = 0.04). Among uninsured adults, inpatient readmissions were stable (around 9.5%; P = 0.76), while there was a rise in observation visits (1.3% vs 2.0%; P < 0.001) and ED visits (8.0% vs 8.6%; P = 0.04), yielding an increase in total revisits (18.8% vs 20.1%; P = 0.004).

Figure 2

shows individual differences for each of the 3 target conditions between 2009 and 2010 versus 2013 and 2014 by payer. Overall, rates at which patients returned to the hospital within 30 days remained stable, with 3 exceptions. For patients with private insurance, total returns to the hospital rose for pneumonia (14.8% vs 15.9%; P = 0.02). For seniors with Medicare, total returns to the hospital declined for pneumonia (from 24.1% to 23.5%; P = 0.03). Among the uninsured, total returns to the hospital rose for AMI (15.5% vs 17.2%; P = 0.02).

Patients initially hospitalized for HF and pneumonia who returned to the hospital within 30 days often returned for the same conditions (Table 2).

Reasons for returning to the hospital were similar in the 2 periods (2009 and 2010; 2013 and 2014) across the 3 target conditions. However, when patients returned to the hospital in 2013 and 2014 with the same diagnosis as the index admission, they were less likely to be readmitted and more likely to be placed under observation than in 2009 and 2010.

 

DISCUSSION

Matching index admissions for AMI, HF, or pneumonia in 201 hospitals in 2009 and 2010 with those in 2013 and 2014, we observed that increases in observation and ED visits coincided with reductions in inpatient readmissions among patients with private insurance and Medicare and contributed to growth in total returns to the hospital among patients with Medicaid or no insurance. Among patients with private insurance and Medicare, inpatient readmissions declined significantly for all 3 target conditions, but total returns to the hospital remained constant for AMI and HF, rose for privately insured patients with pneumonia, and declined modestly for Medicare patients with pneumonia. Inpatient readmissions were unchanged for adults aged 18 to 64 years with Medicaid or no insurance, but total returns to the hospital increased significantly, reaching 32% among those with Medicaid.

These findings add to recent literature, which has primarily emphasized inpatient readmissions among Medicare beneficiaries with several exceptions. A prior analysis indicates that readmissions have declined among diverse payer populations nationally.¹⁸ Gerhardt et al²⁵ found that from 2011 to 2012, all-cause 30-day readmissions declined among fee-for-service (FFS) Medicare beneficiaries following any index admission, while ED revisits remained stable and observation revisits increased slightly. Evaluating the CMS Hospital Readmission Reductions Program (HRRP), Zuckerman et al¹⁷ reported that from 2007 to 2015, inpatient readmissions declined among FFS Medicare beneficiaries aged 65 years and older who were hospitalized with AMI, HF, or pneumonia, while returns to the hospital for observation rose approximately 2%; ED visits were not included. We found that Medicare (FFS and Medicare Advantage) patients with AMI and HF returned to the hospital with the same frequency in 2009 and 2010 as in 2013 and 2014, and those patients with pneumonia returned slightly less often. In aggregate, declines in inpatient readmissions in the 4 states we studied coincided with increases in observation and ED care. Moreover, these shifts occurred not only among Medicare beneficiaries but also among privately insured adults, Medicaid recipients, and the uninsured.

Three factors may have contributed to these apparent shifts from readmissions to observation and ED visits. First, some authors have suggested that hospitals may reduce readmissions by intentionally placing some of the patients who return to the hospital under observation instead of admitting them.^17,26 If true, hospitals with greater declines in readmissions would have larger increases in observation revisits. Zuckerman et al¹⁷ found no correlation among Medicare beneficiaries between hospital-level trends in observation revisits and readmissions, but returns to observation rose more rapidly for AMI, HF, and pneumonia (compared with other conditions) during long term follow-up than during the HRRP implementation period. Other authors have documented that declines in readmissions have been greatest at hospitals with the highest baseline readmission rates,^27,28 and hospitals with lower readmission rates have more observation return visits.²⁹

Second, shifts from inpatient readmissions to return visits for observation may reflect unintentional rather than intentional changes in the services provided. Clinical practice patterns are evolving such that patients who present to the hospital for acute care increasingly are placed under observation or discharged from the ED instead of being admitted, regardless of whether they recently were hospitalized.³⁰ Inpatient admissions, which are strongly correlated with readmission rates,^28,31 are declining nationally,³² and both observation and ED visits are rising.^33-35 Although little is known about effects on health outcomes and patient out-of-pocket costs,shifts from inpatient admissions to observation and ED visits reduce costs to payers.^36,37

Third, instead of substitution, more patients may be returning for lower-acuity conditions that can be treated in the ED or under observation. Hospitals are implementing diverse and multifaceted interventions to reduce readmissions that can involve assessing patient needs and the risk for readmission, educating patients about self-care and risks after discharge, reconciling medication, scheduling follow-up visits, and monitoring patients through telephone calls and home nursing visits.^26,38,39 Although the intent may be to reduce patients’ need to return to the hospital, interventions that educate patients about risks after discharge may lower the threshold at which they find symptoms worrisome enough to return. This could increase lower-acuity return visits. We found that reasons for returning were similar in 2009 and 2010 versus 2013 and 2014, but we did not examine acuity of illness at the time of return.

Other areas of concern are the high rates at which Medicaid patients are returning to the hospital and the increases in rates of returns among Medicaid patients and the uninsured. Individuals in these disadvantaged populations may be having difficulty accessing ambulatory care or may be turning to the ED more often for lower acuity problems that arise after discharge. In 3 of the 4 states we studied, 15% to 16% of adults live in poverty and 10% to 30% live in primary care health professional shortage areas.^40,41 Given the implications for patient outcomes and costs, trends among these populations warrant further scrutiny.^42,43

This analysis has several limitations. Data were from 4 states, but trends in readmissions are similar nationally. From 2010 through 2015, the all-condition readmission rate declined by 8% among Medicare beneficiaries nationally and by 6.1% in South Carolina, 7.4% in Georgia, 8.3% in Nebraska, and 8.7% in Tennessee.⁴⁴ We report trends across hospitals and did not examine hospital-level revisits. Therefore, further research is needed to determine whether these findings are related to co-occurring trends, intentional substitution, or other factors.

In conclusion, measuring inpatient readmissions without accounting for return visits to the ED and observation underestimates the rate at which patients return to the hospital following an inpatient hospitalization. Because of growth in observation and ED visits, trends in the total rates at which patients return to the hospital can differ from trends in inpatient readmissions. In the 4 states we studied, total return rates were particularly high and rising among patients with Medicaid and lower, but also rising, among the uninsured. Policy analysts and researchers should investigate the factors contributing to growth in readmissions in these vulnerable populations and determine whether similar trends are occurring nationwide. Hospitalists play critical roles in admitting and discharging inpatients, caring for patients under observation, and implementing quality improvement programs. Irrespective of payer, hospitalists’ efforts to improve the quality and value of care should include observation and ED visits as well as inpatient readmissions.

 

Acknowledgments

The authors gratefully acknowledge Minya Sheng, M.S. (Truven Health Analytics) for assistance in programming and data management and Linda Lee, Ph.D. (Truven Health Analytics) for providing editorial review of the manuscript. We also wish to acknowledge the 4 HCUP Partner organizations that contributed to the HCUP State Databases used in this study: Georgia Hospital Association, Nebraska Hospital Association, South Carolina Revenue and Fiscal Affairs Office, and Tennessee Hospital Association.

Disclosure

Funding for this study was provided by the AHRQ Center for Delivery, Organization, and Markets, HCUP (Contract No. HHSA-290-2013-00002-C). The views expressed in this article are those of the authors and do not necessarily reflect those of the AHRQ or the U.S. Department of Health and Human Services. The authors have no conflicts of interest or financial disclosures to declare.

Hospitalist physicians care for an increasing proportion of general medicine inpatients and request a significant share of all subspecialty consultations.¹ Subspecialty consultation in inpatient care is increasing,^2,3 and effective hospitalist–consulting service interactions may affect team communication, patient care, and hospitalist learning. Therefore, enhancing hospitalist–consulting service interactions may have a broad-reaching, positive impact. Researchers in previous studies have explored resident–fellow consult interactions in the inpatient and emergency department settings as well as attending-to-attending consultation in the outpatient setting.^4-7 However, to our knowledge, hospitalist–consulting team interactions have not been previously described. In academic medical centers, hospitalists are attending physicians who interact with both fellows (supervised by attending consultants) and directly with subspecialty attendings. Therefore, the exploration of the hospitalist–consultant interaction requires an evaluation of hospitalist–fellow and hospitalist–subspecialty attending interactions. The hospitalist–fellow interaction in particular is unique because it represents an unusual dynamic, in which an attending physician is primarily communicating with a trainee when requesting assistance with patient care.⁸ In order to explore hospitalist–consultant interactions (herein, the term “consultant” includes both fellow and attending consultants), we conducted a survey study in which we examine hospitalist practices and attitudes regarding consultation, with a specific focus on hospitalist consultation with internal medicine subspecialty consult services. In addition, we compared fellow–hospitalist and attending–hospitalist interactions and explored barriers to and facilitating factors of an effective hospitalist–consultant relationship.

METHODS

Survey Development

The survey instrument was developed by the authors based on findings of prior studies in which researchers examined consultation.^2-6,9-16 The survey contained 31 questions (supplementary Appendix A) and evaluated 4 domains of the use of medical subspecialty consultation in direct patient care: (1) current consultation practices, (2) preferences regarding consultants, (3) barriers to and facilitating factors of effective consultation (both with respect to hospitalist learning and patient care), and (4) a comparison between hospitalist–fellow and hospitalist–subspecialty attending interactions. An evaluation of current consultation practices included a focus on communication methods (eg, in person, over the phone, through paging, or notes) because these have been found to be important during consultation.^5,6,9,15,16 In order to explore hospitalist preferences regarding consult interactions and investigate perceptions of barriers to and facilitating factors of effective consultation, questions were developed based on previous literature, including our qualitative work examining resident–fellow interactions during consultation.^4-6,9,12 We compared hospitalist consultation experiences among attending and fellow consultants because the interaction in which an attending hospitalist physician is primarily communicating with a trainee may differ from a consultation between a hospitalist attending and a subspecialty attending.⁸ Participants were asked to exclude their experiences when working on teaching services, during which students or housestaff often interact with consultants. The survey was cognitively tested with both hospitalist and non-hospitalist attending physicians not participating in the study and was revised by the authors using an iterative approach.

Study Participants

Hospitalist attending physicians at University of Texas Southwestern (UTSW) Medical Center, Emory University School of Medicine, Massachusetts General Hospital (MGH), and the Medical University of South Carolina (MUSC) were eligible to participate in the study. Consult team structures at each institution were composed of either a subspecialist-attending-only or a fellow-and-subspecialty-attending team. Fellows at all institutions are supervised by a subspecialty attending when performing consultations. Respondents who self-identified as nurse practitioners or physician assistants were excluded from the analysis. Hospitalists employed by the Veterans Affairs hospital system were also excluded. The study was approved by the institutional review boards of UTSW, Emory, MUSC, and MGH.

The survey was anonymous and administered to all hospitalists at participating institutions via a web-based survey tool (Qualtrics, Provo, UT). Participants were eligible to enter a raffle for a $500 gift card, and completion of the survey was not required for entry into the raffle.

 

Statistics

Results were summarized using the mean with standard deviation for continuous variables and the frequency with percentage for categorical variables after excluding missing values. All analyses were conducted using SAS version 9.4 (SAS Institute, Cary, NC). A 2-sided P value of ≤0.05 was considered statistically significant.

RESULTS

Of a possible 261 respondents, 122 (46.7%) participated in the survey. Missing values for survey responses ranged from 0% to 21.3%, with a mean of 15.2%. Demographic characteristics are shown in Table 1. Respondents had a mean age of 37.7 years and had worked as attending hospitalists for an average of 5.6 years. The majority of respondents (86.1%) practiced in academic medical centers, with the remaining working in satellite community hospitals. Respondents reported working daytime shifts 74.1% of the time on average and being on inpatient, direct-care services without house-staff 70.5% of the time.

Current Consultation Practices

Current consultation practices and descriptions of hospitalist–consultant communication are shown in Table 2. Forty percent of respondents requested 0-1 consults per day, while 51.7% requested 2-3 per day. The most common reasons for requesting a consultation were assistance with treatment (48.5%), assistance with diagnosis (25.7%), and request for a procedure (21.8%). When asked whether the frequency of consultation is changing, slightly more hospitalists felt that their personal use of consultation was increasing as compared to those who felt that it was decreasing (38.5% vs 30.3%, respectively).

An exploration of communication practices during consultation revealed that hospitalists most often interacted with fellows rather than attending physicians (81.4%). However, even when a fellow performs a consult and communicates with a hospitalist, a subspecialty attending is involved in the care of the patient, although he or she may not communicate directly with the hospitalist. Respondents indicated that they most often communicated a consult request to the consultant by phone (76.2%). Pushback from consultants (defined as perceived reluctance or resistance to perform the consult for any reason) was perceived as common, with 64.4% of hospitalists indicating that they experience pushback at least “sometimes” (3 on a 5-point Likert scale) and 22.1% reporting that pushback was “frequent” or occurred “most of the time”. Follow-up interactions (defined as communication of recommendations after the consultant evaluated the patient) infrequently occurred through in-person communication, with 90.9% reporting that this occurred in less than half of consultations. Communication by phone was most common, with 61.2% reporting that it occurred at least half the time, and 86% of respondents reported that communication by paging only occurred at least “sometimes”. Consultation was commonly seen as a valuable educational experience, with 56.9% of hospitalists indicating that they learned from at least half of consults.

Hospitalist Preferences

Eighty-six percent of respondents agreed that consultants should be required to communicate their recommendations either in person or over the phone. Eighty-three percent of hospitalists agreed that they would like to receive more teaching from the consulting services, and 74.0% agreed that consultants should attempt to teach hospitalists during consult interactions regardless of whether the hospitalist initiates the teaching–learning interaction.

Barriers to and Facilitating Factors of Effective Consultation

Participants reported that multiple factors affected patient care and their own learning during inpatient consultation (Figure 1). Consultant pushback, high hospitalist clinical workload, a perception that consultants had limited time, and minimal in-person interactions were all seen as factors that negatively affected the consult interaction. These generally affected both learning and patient care. Conversely, working on an interesting clinical case, more hospitalist free time, positive interaction with the consultant, and having previously worked with the consultant positively affected both learning and patient care (Figure 1).

Fellow Versus Attending Interactions

Respondents indicated that interacting directly with the consult attending was superior to hospitalist–fellow interactions in all aspects of care but particularly with respect to pushback, confidence in recommendations, professionalism, and hospitalist learning (Figure 2).

DISCUSSION

To our knowledge, this is the first study to describe hospitalist attending practices, attitudes, and perceptions of internal medicine subspecialty consultation. Our findings, which focus on the interaction between hospitalists and internal medicine subspecialty attendings and fellows, outline the hospitalist perspective on consultant interactions and identify a number of factors that are amenable to intervention. We found that hospitalists perceive the consult interaction to be important for patient care and a valuable opportunity for their own learning. In-person communication was seen as an important component of effective consultation but was reported to occur in a minority of consultations. We demonstrate that hospitalist–subspecialty attending consult interactions are perceived more positively than hospitalist–fellow interactions. Finally, we describe barriers and facilitating factors that may inform future interventions targeting this important interaction.

 

Effective communication between consultants and the primary team is critical for both patient care and teaching interactions.^4-7 Pushback on consultation was reported to be the most significant barrier to hospitalist learning and had a major impact on patient care. Because hospitalists are attending physicians, we hypothesized that they may perceive pushback from fellows less frequently than residents.⁴ However, in our study, hospitalists reported pushback to be relatively frequent in their daily practice. Moreover, hospitalists reported a strong preference for in-person interactions with consultants, but our study demonstrated that such interactions are relatively infrequent. Researchers in studies of resident–fellow consult interactions have noted similar findings, suggesting that hospitalists and internal medicine residents face similar challenges during consultation.^4-6 Hospitalists reported that positive interpersonal interactions and personal familiarity with the consultant positively affected the consult interaction. Most importantly, these effects were perceived to affect both hospitalist learning and patient care, suggesting the importance of interpersonal interactions in consultative medicine.

In an era of increasing clinical workload, the consult interaction represents an important workplace-based learning opportunity.⁴ Centered on a consult question, the hospitalist–consultant interaction embodies a teachable moment and can be an efficient opportunity to learn because both parties are familiar with the patient. Indeed, survey respondents reported that they frequently learned from consultation, and there was a strong preference for more teaching from consultants in this setting. However, the hospitalist–fellow consult interaction is unique because attending hospitalists are frequently communicating with fellow trainees, which could limit fellows’ confidence in their role as teachers and hospitalists’ perception of their role as learners. Our study identifies a number of barriers and facilitating factors (including communication, pushback, familiarity, and clinical workload) that affect the hospitalist–consultant teaching interaction and may be amenable to intervention.

Hospitalists expressed a consistent preference for interacting with attending subspecialists compared to clinical fellows during consultation. Preference for interaction with attendings was strongest in the areas of pushback, confidence in recommendations, professionalism, and learning from consultation. Some of the factors that relate to consult service structure and fellow experience, such as timeliness of consultation and confidence in recommendations, may not be amenable to intervention. For instance, fellows must first see and then staff the consult with their attending prior to leaving formal recommendations, which makes their communication less timely than that of attending physicians, when they are the primary consultant. However, aspects of the hospitalist–consultant interaction (such as professionalism, ease of communication, and pushback) should not be affected by the difference in experience between fellows and attending physicians. The reasons for such perceptions deserve further exploration; however, differences in incentive structures, workload, and communication skills between fellows and attending consultants may be potential explanations.

Our findings suggest that interventions aimed at enhancing hospitalist–consultant interactions focus on enhancing direct communication and teaching while limiting the perception of pushback. A number of interventions that are primarily focused on instituting a systematic approach to requesting consultation have shown an improvement in resident and medical student consult communication^17,18 as well as resident–fellow teaching interactions.⁹ However, it is not clear whether these interventions would be effective given that hospitalists have more experience communicating with consultants than trainees. Given the unique nature of the hospitalist–consultant interaction, multiple barriers may need to be addressed in order to have a significant impact. Efforts to increase direct communication, such as a mechanism for hospitalists to make and request in-person or direct verbal communication about a particular consultation during the consult request, can help consultants prioritize direct communication with hospitalists for specific patients. Familiarizing fellows with hospitalist workflow and the locations of hospitalist workrooms also may promote in-person communication. Fellowship training can focus on enhancing fellow teaching and communication skills,^19-22 particularly as they relate to hospitalists. Fellows in particular may benefit because the hospitalist–fellow teaching interaction may be bidirectional, with hospitalists having expertise in systems practice and quality efforts that can inform fellows’ practice. Furthermore, interacting with hospitalists is an opportunity for fellows to practice professional interactions, which will be critical to their careers. Increasing familiarity between fellows and hospitalists through joint events may also serve to enhance the interaction. Finally, enabling hospitalists to provide feedback to fellows stands to benefit both parties because multisource feedback is an important tool in assessing trainee competence and improving performance.²³ However, we should note that because our study focused on hospitalist perceptions, an exploration of subspecialty fellows’ and attendings’ perceptions of the hospitalist–consultant interaction would provide additional, important data for shaping interventions.

Strengths of our study include the inclusion of multiple study sites, which may increase generalizability; however, our study has several limitations. The incomplete response rate reduces both generalizability and statistical power and may have created selection or nonresponder bias. However, low response rates occur commonly when surveying medical professionals, and our results are consistent with many prior hospitalist survey studies.^24-26 Further, we conducted our study at a single time point; therefore, we could not evaluate the effect of fellow experience on hospitalist perceptions. However, we conducted our study in the second half of the academic year, when fellows had already gained considerable experience in the consultation setting. We did not capture participants’ institutional affiliations; therefore, a subgroup analysis by institution could not be performed. Additionally, our study reflects hospitalist perception rather than objectively measured communication practices between hospitalists and consultants, and it does not include the perspective of subspecialists. The specific needs of nurse practitioners and physicians’ assistants, who were excluded from this study, should also be evaluated in future research. Lastly, this is a hypothesis-generating study and should be replicated in a national cohort.

 

CONCLUSION

The hospitalists represented in our sample population perceived the consult interaction to be important for patient care and a valuable opportunity for their own learning. Participants expressed that they would like to increase direct communication with consultants and enhance consultant–hospitalist teaching interactions. Multiple barriers to effective hospitalist–consultant interactions (including communication, pushback, and hospitalist–consultant familiarity) are amenable to intervention.

Disclosure

The authors have no financial disclosures or conflicts of interest.

Hospitalist physicians care for an increasing proportion of general medicine inpatients and request a significant share of all subspecialty consultations.¹ Subspecialty consultation in inpatient care is increasing,^2,3 and effective hospitalist–consulting service interactions may affect team communication, patient care, and hospitalist learning. Therefore, enhancing hospitalist–consulting service interactions may have a broad-reaching, positive impact. Researchers in previous studies have explored resident–fellow consult interactions in the inpatient and emergency department settings as well as attending-to-attending consultation in the outpatient setting.^4-7 However, to our knowledge, hospitalist–consulting team interactions have not been previously described. In academic medical centers, hospitalists are attending physicians who interact with both fellows (supervised by attending consultants) and directly with subspecialty attendings. Therefore, the exploration of the hospitalist–consultant interaction requires an evaluation of hospitalist–fellow and hospitalist–subspecialty attending interactions. The hospitalist–fellow interaction in particular is unique because it represents an unusual dynamic, in which an attending physician is primarily communicating with a trainee when requesting assistance with patient care.⁸ In order to explore hospitalist–consultant interactions (herein, the term “consultant” includes both fellow and attending consultants), we conducted a survey study in which we examine hospitalist practices and attitudes regarding consultation, with a specific focus on hospitalist consultation with internal medicine subspecialty consult services. In addition, we compared fellow–hospitalist and attending–hospitalist interactions and explored barriers to and facilitating factors of an effective hospitalist–consultant relationship.

METHODS

Survey Development

The survey instrument was developed by the authors based on findings of prior studies in which researchers examined consultation.^2-6,9-16 The survey contained 31 questions (supplementary Appendix A) and evaluated 4 domains of the use of medical subspecialty consultation in direct patient care: (1) current consultation practices, (2) preferences regarding consultants, (3) barriers to and facilitating factors of effective consultation (both with respect to hospitalist learning and patient care), and (4) a comparison between hospitalist–fellow and hospitalist–subspecialty attending interactions. An evaluation of current consultation practices included a focus on communication methods (eg, in person, over the phone, through paging, or notes) because these have been found to be important during consultation.^5,6,9,15,16 In order to explore hospitalist preferences regarding consult interactions and investigate perceptions of barriers to and facilitating factors of effective consultation, questions were developed based on previous literature, including our qualitative work examining resident–fellow interactions during consultation.^4-6,9,12 We compared hospitalist consultation experiences among attending and fellow consultants because the interaction in which an attending hospitalist physician is primarily communicating with a trainee may differ from a consultation between a hospitalist attending and a subspecialty attending.⁸ Participants were asked to exclude their experiences when working on teaching services, during which students or housestaff often interact with consultants. The survey was cognitively tested with both hospitalist and non-hospitalist attending physicians not participating in the study and was revised by the authors using an iterative approach.

Study Participants

Hospitalist attending physicians at University of Texas Southwestern (UTSW) Medical Center, Emory University School of Medicine, Massachusetts General Hospital (MGH), and the Medical University of South Carolina (MUSC) were eligible to participate in the study. Consult team structures at each institution were composed of either a subspecialist-attending-only or a fellow-and-subspecialty-attending team. Fellows at all institutions are supervised by a subspecialty attending when performing consultations. Respondents who self-identified as nurse practitioners or physician assistants were excluded from the analysis. Hospitalists employed by the Veterans Affairs hospital system were also excluded. The study was approved by the institutional review boards of UTSW, Emory, MUSC, and MGH.

The survey was anonymous and administered to all hospitalists at participating institutions via a web-based survey tool (Qualtrics, Provo, UT). Participants were eligible to enter a raffle for a $500 gift card, and completion of the survey was not required for entry into the raffle.

 

Statistics

Results were summarized using the mean with standard deviation for continuous variables and the frequency with percentage for categorical variables after excluding missing values. All analyses were conducted using SAS version 9.4 (SAS Institute, Cary, NC). A 2-sided P value of ≤0.05 was considered statistically significant.

RESULTS

Of a possible 261 respondents, 122 (46.7%) participated in the survey. Missing values for survey responses ranged from 0% to 21.3%, with a mean of 15.2%. Demographic characteristics are shown in Table 1. Respondents had a mean age of 37.7 years and had worked as attending hospitalists for an average of 5.6 years. The majority of respondents (86.1%) practiced in academic medical centers, with the remaining working in satellite community hospitals. Respondents reported working daytime shifts 74.1% of the time on average and being on inpatient, direct-care services without house-staff 70.5% of the time.

Current Consultation Practices

Current consultation practices and descriptions of hospitalist–consultant communication are shown in Table 2. Forty percent of respondents requested 0-1 consults per day, while 51.7% requested 2-3 per day. The most common reasons for requesting a consultation were assistance with treatment (48.5%), assistance with diagnosis (25.7%), and request for a procedure (21.8%). When asked whether the frequency of consultation is changing, slightly more hospitalists felt that their personal use of consultation was increasing as compared to those who felt that it was decreasing (38.5% vs 30.3%, respectively).

An exploration of communication practices during consultation revealed that hospitalists most often interacted with fellows rather than attending physicians (81.4%). However, even when a fellow performs a consult and communicates with a hospitalist, a subspecialty attending is involved in the care of the patient, although he or she may not communicate directly with the hospitalist. Respondents indicated that they most often communicated a consult request to the consultant by phone (76.2%). Pushback from consultants (defined as perceived reluctance or resistance to perform the consult for any reason) was perceived as common, with 64.4% of hospitalists indicating that they experience pushback at least “sometimes” (3 on a 5-point Likert scale) and 22.1% reporting that pushback was “frequent” or occurred “most of the time”. Follow-up interactions (defined as communication of recommendations after the consultant evaluated the patient) infrequently occurred through in-person communication, with 90.9% reporting that this occurred in less than half of consultations. Communication by phone was most common, with 61.2% reporting that it occurred at least half the time, and 86% of respondents reported that communication by paging only occurred at least “sometimes”. Consultation was commonly seen as a valuable educational experience, with 56.9% of hospitalists indicating that they learned from at least half of consults.

Hospitalist Preferences

Eighty-six percent of respondents agreed that consultants should be required to communicate their recommendations either in person or over the phone. Eighty-three percent of hospitalists agreed that they would like to receive more teaching from the consulting services, and 74.0% agreed that consultants should attempt to teach hospitalists during consult interactions regardless of whether the hospitalist initiates the teaching–learning interaction.

Barriers to and Facilitating Factors of Effective Consultation

Participants reported that multiple factors affected patient care and their own learning during inpatient consultation (Figure 1). Consultant pushback, high hospitalist clinical workload, a perception that consultants had limited time, and minimal in-person interactions were all seen as factors that negatively affected the consult interaction. These generally affected both learning and patient care. Conversely, working on an interesting clinical case, more hospitalist free time, positive interaction with the consultant, and having previously worked with the consultant positively affected both learning and patient care (Figure 1).

Fellow Versus Attending Interactions

Respondents indicated that interacting directly with the consult attending was superior to hospitalist–fellow interactions in all aspects of care but particularly with respect to pushback, confidence in recommendations, professionalism, and hospitalist learning (Figure 2).

DISCUSSION

To our knowledge, this is the first study to describe hospitalist attending practices, attitudes, and perceptions of internal medicine subspecialty consultation. Our findings, which focus on the interaction between hospitalists and internal medicine subspecialty attendings and fellows, outline the hospitalist perspective on consultant interactions and identify a number of factors that are amenable to intervention. We found that hospitalists perceive the consult interaction to be important for patient care and a valuable opportunity for their own learning. In-person communication was seen as an important component of effective consultation but was reported to occur in a minority of consultations. We demonstrate that hospitalist–subspecialty attending consult interactions are perceived more positively than hospitalist–fellow interactions. Finally, we describe barriers and facilitating factors that may inform future interventions targeting this important interaction.

 

Effective communication between consultants and the primary team is critical for both patient care and teaching interactions.^4-7 Pushback on consultation was reported to be the most significant barrier to hospitalist learning and had a major impact on patient care. Because hospitalists are attending physicians, we hypothesized that they may perceive pushback from fellows less frequently than residents.⁴ However, in our study, hospitalists reported pushback to be relatively frequent in their daily practice. Moreover, hospitalists reported a strong preference for in-person interactions with consultants, but our study demonstrated that such interactions are relatively infrequent. Researchers in studies of resident–fellow consult interactions have noted similar findings, suggesting that hospitalists and internal medicine residents face similar challenges during consultation.^4-6 Hospitalists reported that positive interpersonal interactions and personal familiarity with the consultant positively affected the consult interaction. Most importantly, these effects were perceived to affect both hospitalist learning and patient care, suggesting the importance of interpersonal interactions in consultative medicine.

In an era of increasing clinical workload, the consult interaction represents an important workplace-based learning opportunity.⁴ Centered on a consult question, the hospitalist–consultant interaction embodies a teachable moment and can be an efficient opportunity to learn because both parties are familiar with the patient. Indeed, survey respondents reported that they frequently learned from consultation, and there was a strong preference for more teaching from consultants in this setting. However, the hospitalist–fellow consult interaction is unique because attending hospitalists are frequently communicating with fellow trainees, which could limit fellows’ confidence in their role as teachers and hospitalists’ perception of their role as learners. Our study identifies a number of barriers and facilitating factors (including communication, pushback, familiarity, and clinical workload) that affect the hospitalist–consultant teaching interaction and may be amenable to intervention.

Hospitalists expressed a consistent preference for interacting with attending subspecialists compared to clinical fellows during consultation. Preference for interaction with attendings was strongest in the areas of pushback, confidence in recommendations, professionalism, and learning from consultation. Some of the factors that relate to consult service structure and fellow experience, such as timeliness of consultation and confidence in recommendations, may not be amenable to intervention. For instance, fellows must first see and then staff the consult with their attending prior to leaving formal recommendations, which makes their communication less timely than that of attending physicians, when they are the primary consultant. However, aspects of the hospitalist–consultant interaction (such as professionalism, ease of communication, and pushback) should not be affected by the difference in experience between fellows and attending physicians. The reasons for such perceptions deserve further exploration; however, differences in incentive structures, workload, and communication skills between fellows and attending consultants may be potential explanations.

Our findings suggest that interventions aimed at enhancing hospitalist–consultant interactions focus on enhancing direct communication and teaching while limiting the perception of pushback. A number of interventions that are primarily focused on instituting a systematic approach to requesting consultation have shown an improvement in resident and medical student consult communication^17,18 as well as resident–fellow teaching interactions.⁹ However, it is not clear whether these interventions would be effective given that hospitalists have more experience communicating with consultants than trainees. Given the unique nature of the hospitalist–consultant interaction, multiple barriers may need to be addressed in order to have a significant impact. Efforts to increase direct communication, such as a mechanism for hospitalists to make and request in-person or direct verbal communication about a particular consultation during the consult request, can help consultants prioritize direct communication with hospitalists for specific patients. Familiarizing fellows with hospitalist workflow and the locations of hospitalist workrooms also may promote in-person communication. Fellowship training can focus on enhancing fellow teaching and communication skills,^19-22 particularly as they relate to hospitalists. Fellows in particular may benefit because the hospitalist–fellow teaching interaction may be bidirectional, with hospitalists having expertise in systems practice and quality efforts that can inform fellows’ practice. Furthermore, interacting with hospitalists is an opportunity for fellows to practice professional interactions, which will be critical to their careers. Increasing familiarity between fellows and hospitalists through joint events may also serve to enhance the interaction. Finally, enabling hospitalists to provide feedback to fellows stands to benefit both parties because multisource feedback is an important tool in assessing trainee competence and improving performance.²³ However, we should note that because our study focused on hospitalist perceptions, an exploration of subspecialty fellows’ and attendings’ perceptions of the hospitalist–consultant interaction would provide additional, important data for shaping interventions.

Strengths of our study include the inclusion of multiple study sites, which may increase generalizability; however, our study has several limitations. The incomplete response rate reduces both generalizability and statistical power and may have created selection or nonresponder bias. However, low response rates occur commonly when surveying medical professionals, and our results are consistent with many prior hospitalist survey studies.^24-26 Further, we conducted our study at a single time point; therefore, we could not evaluate the effect of fellow experience on hospitalist perceptions. However, we conducted our study in the second half of the academic year, when fellows had already gained considerable experience in the consultation setting. We did not capture participants’ institutional affiliations; therefore, a subgroup analysis by institution could not be performed. Additionally, our study reflects hospitalist perception rather than objectively measured communication practices between hospitalists and consultants, and it does not include the perspective of subspecialists. The specific needs of nurse practitioners and physicians’ assistants, who were excluded from this study, should also be evaluated in future research. Lastly, this is a hypothesis-generating study and should be replicated in a national cohort.

 

CONCLUSION

The hospitalists represented in our sample population perceived the consult interaction to be important for patient care and a valuable opportunity for their own learning. Participants expressed that they would like to increase direct communication with consultants and enhance consultant–hospitalist teaching interactions. Multiple barriers to effective hospitalist–consultant interactions (including communication, pushback, and hospitalist–consultant familiarity) are amenable to intervention.

Disclosure

The authors have no financial disclosures or conflicts of interest.

Hospitalist physicians care for an increasing proportion of general medicine inpatients and request a significant share of all subspecialty consultations.¹ Subspecialty consultation in inpatient care is increasing,^2,3 and effective hospitalist–consulting service interactions may affect team communication, patient care, and hospitalist learning. Therefore, enhancing hospitalist–consulting service interactions may have a broad-reaching, positive impact. Researchers in previous studies have explored resident–fellow consult interactions in the inpatient and emergency department settings as well as attending-to-attending consultation in the outpatient setting.^4-7 However, to our knowledge, hospitalist–consulting team interactions have not been previously described. In academic medical centers, hospitalists are attending physicians who interact with both fellows (supervised by attending consultants) and directly with subspecialty attendings. Therefore, the exploration of the hospitalist–consultant interaction requires an evaluation of hospitalist–fellow and hospitalist–subspecialty attending interactions. The hospitalist–fellow interaction in particular is unique because it represents an unusual dynamic, in which an attending physician is primarily communicating with a trainee when requesting assistance with patient care.⁸ In order to explore hospitalist–consultant interactions (herein, the term “consultant” includes both fellow and attending consultants), we conducted a survey study in which we examine hospitalist practices and attitudes regarding consultation, with a specific focus on hospitalist consultation with internal medicine subspecialty consult services. In addition, we compared fellow–hospitalist and attending–hospitalist interactions and explored barriers to and facilitating factors of an effective hospitalist–consultant relationship.

METHODS

Survey Development

The survey instrument was developed by the authors based on findings of prior studies in which researchers examined consultation.^2-6,9-16 The survey contained 31 questions (supplementary Appendix A) and evaluated 4 domains of the use of medical subspecialty consultation in direct patient care: (1) current consultation practices, (2) preferences regarding consultants, (3) barriers to and facilitating factors of effective consultation (both with respect to hospitalist learning and patient care), and (4) a comparison between hospitalist–fellow and hospitalist–subspecialty attending interactions. An evaluation of current consultation practices included a focus on communication methods (eg, in person, over the phone, through paging, or notes) because these have been found to be important during consultation.^5,6,9,15,16 In order to explore hospitalist preferences regarding consult interactions and investigate perceptions of barriers to and facilitating factors of effective consultation, questions were developed based on previous literature, including our qualitative work examining resident–fellow interactions during consultation.^4-6,9,12 We compared hospitalist consultation experiences among attending and fellow consultants because the interaction in which an attending hospitalist physician is primarily communicating with a trainee may differ from a consultation between a hospitalist attending and a subspecialty attending.⁸ Participants were asked to exclude their experiences when working on teaching services, during which students or housestaff often interact with consultants. The survey was cognitively tested with both hospitalist and non-hospitalist attending physicians not participating in the study and was revised by the authors using an iterative approach.

Study Participants

Hospitalist attending physicians at University of Texas Southwestern (UTSW) Medical Center, Emory University School of Medicine, Massachusetts General Hospital (MGH), and the Medical University of South Carolina (MUSC) were eligible to participate in the study. Consult team structures at each institution were composed of either a subspecialist-attending-only or a fellow-and-subspecialty-attending team. Fellows at all institutions are supervised by a subspecialty attending when performing consultations. Respondents who self-identified as nurse practitioners or physician assistants were excluded from the analysis. Hospitalists employed by the Veterans Affairs hospital system were also excluded. The study was approved by the institutional review boards of UTSW, Emory, MUSC, and MGH.

The survey was anonymous and administered to all hospitalists at participating institutions via a web-based survey tool (Qualtrics, Provo, UT). Participants were eligible to enter a raffle for a $500 gift card, and completion of the survey was not required for entry into the raffle.

 

Statistics

Results were summarized using the mean with standard deviation for continuous variables and the frequency with percentage for categorical variables after excluding missing values. All analyses were conducted using SAS version 9.4 (SAS Institute, Cary, NC). A 2-sided P value of ≤0.05 was considered statistically significant.

RESULTS

Of a possible 261 respondents, 122 (46.7%) participated in the survey. Missing values for survey responses ranged from 0% to 21.3%, with a mean of 15.2%. Demographic characteristics are shown in Table 1. Respondents had a mean age of 37.7 years and had worked as attending hospitalists for an average of 5.6 years. The majority of respondents (86.1%) practiced in academic medical centers, with the remaining working in satellite community hospitals. Respondents reported working daytime shifts 74.1% of the time on average and being on inpatient, direct-care services without house-staff 70.5% of the time.

Current Consultation Practices

Current consultation practices and descriptions of hospitalist–consultant communication are shown in Table 2. Forty percent of respondents requested 0-1 consults per day, while 51.7% requested 2-3 per day. The most common reasons for requesting a consultation were assistance with treatment (48.5%), assistance with diagnosis (25.7%), and request for a procedure (21.8%). When asked whether the frequency of consultation is changing, slightly more hospitalists felt that their personal use of consultation was increasing as compared to those who felt that it was decreasing (38.5% vs 30.3%, respectively).

An exploration of communication practices during consultation revealed that hospitalists most often interacted with fellows rather than attending physicians (81.4%). However, even when a fellow performs a consult and communicates with a hospitalist, a subspecialty attending is involved in the care of the patient, although he or she may not communicate directly with the hospitalist. Respondents indicated that they most often communicated a consult request to the consultant by phone (76.2%). Pushback from consultants (defined as perceived reluctance or resistance to perform the consult for any reason) was perceived as common, with 64.4% of hospitalists indicating that they experience pushback at least “sometimes” (3 on a 5-point Likert scale) and 22.1% reporting that pushback was “frequent” or occurred “most of the time”. Follow-up interactions (defined as communication of recommendations after the consultant evaluated the patient) infrequently occurred through in-person communication, with 90.9% reporting that this occurred in less than half of consultations. Communication by phone was most common, with 61.2% reporting that it occurred at least half the time, and 86% of respondents reported that communication by paging only occurred at least “sometimes”. Consultation was commonly seen as a valuable educational experience, with 56.9% of hospitalists indicating that they learned from at least half of consults.

Hospitalist Preferences

Eighty-six percent of respondents agreed that consultants should be required to communicate their recommendations either in person or over the phone. Eighty-three percent of hospitalists agreed that they would like to receive more teaching from the consulting services, and 74.0% agreed that consultants should attempt to teach hospitalists during consult interactions regardless of whether the hospitalist initiates the teaching–learning interaction.

Barriers to and Facilitating Factors of Effective Consultation

Participants reported that multiple factors affected patient care and their own learning during inpatient consultation (Figure 1). Consultant pushback, high hospitalist clinical workload, a perception that consultants had limited time, and minimal in-person interactions were all seen as factors that negatively affected the consult interaction. These generally affected both learning and patient care. Conversely, working on an interesting clinical case, more hospitalist free time, positive interaction with the consultant, and having previously worked with the consultant positively affected both learning and patient care (Figure 1).

Fellow Versus Attending Interactions

Respondents indicated that interacting directly with the consult attending was superior to hospitalist–fellow interactions in all aspects of care but particularly with respect to pushback, confidence in recommendations, professionalism, and hospitalist learning (Figure 2).

DISCUSSION

To our knowledge, this is the first study to describe hospitalist attending practices, attitudes, and perceptions of internal medicine subspecialty consultation. Our findings, which focus on the interaction between hospitalists and internal medicine subspecialty attendings and fellows, outline the hospitalist perspective on consultant interactions and identify a number of factors that are amenable to intervention. We found that hospitalists perceive the consult interaction to be important for patient care and a valuable opportunity for their own learning. In-person communication was seen as an important component of effective consultation but was reported to occur in a minority of consultations. We demonstrate that hospitalist–subspecialty attending consult interactions are perceived more positively than hospitalist–fellow interactions. Finally, we describe barriers and facilitating factors that may inform future interventions targeting this important interaction.

 

Effective communication between consultants and the primary team is critical for both patient care and teaching interactions.^4-7 Pushback on consultation was reported to be the most significant barrier to hospitalist learning and had a major impact on patient care. Because hospitalists are attending physicians, we hypothesized that they may perceive pushback from fellows less frequently than residents.⁴ However, in our study, hospitalists reported pushback to be relatively frequent in their daily practice. Moreover, hospitalists reported a strong preference for in-person interactions with consultants, but our study demonstrated that such interactions are relatively infrequent. Researchers in studies of resident–fellow consult interactions have noted similar findings, suggesting that hospitalists and internal medicine residents face similar challenges during consultation.^4-6 Hospitalists reported that positive interpersonal interactions and personal familiarity with the consultant positively affected the consult interaction. Most importantly, these effects were perceived to affect both hospitalist learning and patient care, suggesting the importance of interpersonal interactions in consultative medicine.

In an era of increasing clinical workload, the consult interaction represents an important workplace-based learning opportunity.⁴ Centered on a consult question, the hospitalist–consultant interaction embodies a teachable moment and can be an efficient opportunity to learn because both parties are familiar with the patient. Indeed, survey respondents reported that they frequently learned from consultation, and there was a strong preference for more teaching from consultants in this setting. However, the hospitalist–fellow consult interaction is unique because attending hospitalists are frequently communicating with fellow trainees, which could limit fellows’ confidence in their role as teachers and hospitalists’ perception of their role as learners. Our study identifies a number of barriers and facilitating factors (including communication, pushback, familiarity, and clinical workload) that affect the hospitalist–consultant teaching interaction and may be amenable to intervention.

Hospitalists expressed a consistent preference for interacting with attending subspecialists compared to clinical fellows during consultation. Preference for interaction with attendings was strongest in the areas of pushback, confidence in recommendations, professionalism, and learning from consultation. Some of the factors that relate to consult service structure and fellow experience, such as timeliness of consultation and confidence in recommendations, may not be amenable to intervention. For instance, fellows must first see and then staff the consult with their attending prior to leaving formal recommendations, which makes their communication less timely than that of attending physicians, when they are the primary consultant. However, aspects of the hospitalist–consultant interaction (such as professionalism, ease of communication, and pushback) should not be affected by the difference in experience between fellows and attending physicians. The reasons for such perceptions deserve further exploration; however, differences in incentive structures, workload, and communication skills between fellows and attending consultants may be potential explanations.

Our findings suggest that interventions aimed at enhancing hospitalist–consultant interactions focus on enhancing direct communication and teaching while limiting the perception of pushback. A number of interventions that are primarily focused on instituting a systematic approach to requesting consultation have shown an improvement in resident and medical student consult communication^17,18 as well as resident–fellow teaching interactions.⁹ However, it is not clear whether these interventions would be effective given that hospitalists have more experience communicating with consultants than trainees. Given the unique nature of the hospitalist–consultant interaction, multiple barriers may need to be addressed in order to have a significant impact. Efforts to increase direct communication, such as a mechanism for hospitalists to make and request in-person or direct verbal communication about a particular consultation during the consult request, can help consultants prioritize direct communication with hospitalists for specific patients. Familiarizing fellows with hospitalist workflow and the locations of hospitalist workrooms also may promote in-person communication. Fellowship training can focus on enhancing fellow teaching and communication skills,^19-22 particularly as they relate to hospitalists. Fellows in particular may benefit because the hospitalist–fellow teaching interaction may be bidirectional, with hospitalists having expertise in systems practice and quality efforts that can inform fellows’ practice. Furthermore, interacting with hospitalists is an opportunity for fellows to practice professional interactions, which will be critical to their careers. Increasing familiarity between fellows and hospitalists through joint events may also serve to enhance the interaction. Finally, enabling hospitalists to provide feedback to fellows stands to benefit both parties because multisource feedback is an important tool in assessing trainee competence and improving performance.²³ However, we should note that because our study focused on hospitalist perceptions, an exploration of subspecialty fellows’ and attendings’ perceptions of the hospitalist–consultant interaction would provide additional, important data for shaping interventions.

Strengths of our study include the inclusion of multiple study sites, which may increase generalizability; however, our study has several limitations. The incomplete response rate reduces both generalizability and statistical power and may have created selection or nonresponder bias. However, low response rates occur commonly when surveying medical professionals, and our results are consistent with many prior hospitalist survey studies.^24-26 Further, we conducted our study at a single time point; therefore, we could not evaluate the effect of fellow experience on hospitalist perceptions. However, we conducted our study in the second half of the academic year, when fellows had already gained considerable experience in the consultation setting. We did not capture participants’ institutional affiliations; therefore, a subgroup analysis by institution could not be performed. Additionally, our study reflects hospitalist perception rather than objectively measured communication practices between hospitalists and consultants, and it does not include the perspective of subspecialists. The specific needs of nurse practitioners and physicians’ assistants, who were excluded from this study, should also be evaluated in future research. Lastly, this is a hypothesis-generating study and should be replicated in a national cohort.

 

CONCLUSION

The hospitalists represented in our sample population perceived the consult interaction to be important for patient care and a valuable opportunity for their own learning. Participants expressed that they would like to increase direct communication with consultants and enhance consultant–hospitalist teaching interactions. Multiple barriers to effective hospitalist–consultant interactions (including communication, pushback, and hospitalist–consultant familiarity) are amenable to intervention.

Disclosure

The authors have no financial disclosures or conflicts of interest.

Take-Home Points

• Shared decision-making tools, such as predictive models, can help empower the patient to make decisions for or against surgery equipped with more information about the expected outcome.
• There is a role for preoperative collection of PROMs in the clinical decision-making process.
• Mental health state, as reported by the VR-12 MCS, is a significant predictor of postoperative pain and function as reported by the VAS pain and ASES function scores.
• A significant portion of the predictive ability of this model comes from the fact that at 1-year postoperatively, patients receiving a rTSA will on average have a 3.8 point lower on ASES function score than those receiving a TSA (P < .001, ω²=.083).
• Future studies to discern the role of different modalities to improve a patient’s emotional health preoperatively will be beneficial as the healthcare industry trends toward value based medicine collecting PROMs as part of reimbursement schemes.

Over the past few decades, decisions regarding patients’ care have gradually transitioned from a paternalistic model to a more cooperative exchange between patient and physician. Shared decision-making provides patients a measure of autonomy in making choices for their health and their future. Patient participation may mitigate uncertainty and discomfort during selection of a course of treatment, which may lead to increased satisfaction levels after surgery.¹ Moreover, shared decision-making may help patients better manage postoperative expectations through evidenced-based discussions of preoperative health levels and their corresponding outcomes. Patient-reported outcome measures (PROMs) use clinically sensitive and specific metrics to evaluate a patient’s self-reported pain, functional ability, and mental state.² These metrics are useful in setting patient expectations for potential outcomes of treatment options. Use of evidence-based clinical decision-making tools, such as PROM-based predictive models, can facilitate a collaborative decision-making environment for patient and physician. Given the present cost-containment era, and the need for preoperative metrics that can assist in predictive analysis of postoperative improvement, models are clearly valuable.

In attempts to help patients set well-informed and reasonable expectations, physicians have turned to PROMs to facilitate preoperative evidence-based discussions. Although PROMs have been in use for almost 30 years, only recently have they been used to create tools that can aid quantitatively in the surgical decision-making process.^2-6 Combining physical examination findings, imaging studies, comorbidities, and quantitative tools, such as this model, may enhance patients’ understanding of their preoperative condition and expected prognosis and thereby guide their surgical decisions.

We conducted a study to determine whether certain preoperative PROMs can predict 1-year postoperative visual analog scale (VAS) pain scores and American Shoulder and Elbow Surgeons (ASES) Function scores in total shoulder arthroplasty (TSA) and reverse TSA (rTSA). We hypothesized that preoperative mental health status as captured by Veterans RAND 12-Item Health Survey (VR-12) mental health component summary (MCS) score and preoperative VAS pain score would predict both VAS pain score and ASES Function score 1 year after surgery. Specifically, we hypothesized that a higher preoperative VR-12 MCS score would predict less pain and better function 1 year after surgery and that a higher preoperative VAS pain score would predict more pain and worse function 1 year after surgery.

Methods

This study was approved by the Institutional Review Board of Partners Healthcare. The study used the Surgical Outcome System (Arthrex), a comprehensive prospective database that stores preoperative and 1-year postoperative patient demographics and TSA-PROM data. Surveys are emailed to all enrolled patients before surgery and 1 year after surgery. As indicated by the Institutional Review Boards of all participating institutions, patients in the Surgical Outcome System have to sign a consent form to permit use of their responses in research.

The database includes patient data from 42 orthopedic surgeons across the United States. All primary TSAs and primary rTSAs in the database were included in this study, regardless of arthroplasty indication. Revisions were excluded. Also excluded were cases in which the 1-year postoperative questionnaire was not completed. Of the 1681 patients eligible for 1-year follow-up, 1225 (73%) completed the 1-year postoperative questionnaire. PROMs used in the study were VAS pain score, ASES Function score, VR-12 MCS score, and Single Assessment Numerical Evaluation (SANE). Unfortunately, not all surgeons use every measure in the 1-year postoperative questionnaire set. Thus, in our complete models, total number of observations was 1004 for modeling 1-year postoperative VAS pain scores and 986 for modeling 1-year postoperative ASES Function scores. 

Metrics

On VAS, pain is rated from 0 (no pain) to 10 (pain as bad as it can be). Tashjian and colleagues⁷ estimated that the minimal clinically important difference (MCID) for postoperative VAS pain scores was 1.4 in a cohort of 326 patients who had TSA, rTSA, or shoulder hemiarthroplasty. ASES Function score is scaled from 0 to 30, with 30 representing best function.⁸ Wong and colleagues⁹ identified an MCID of 6.5 for ASES Function scores in a cohort of 107 patients who had TSA or rTSA. SANE ratings range from 0% to 100%, with 100% indicating the patient’s shoulder was totally “normal.”10 VR-12 MCS scores appear on a logarithmic scale, with higher numbers representing better mental health. The population mean estimate for VR-12 MCS scores is 50.1 (SD, 11.49; overall possible range, –2.47 to 76.1).11 Our patient population’s scores ranged from 12.5 to 73.8.

Statistical Analysis

Simple bivariate and multivariate linear regressions were performed to evaluate the predictive value of each of the outlined PROMs. Our complete model controls for patient sex, age, and type of arthroplasty. Categorical variables were dummy-coded. Both 1-year postoperative VAS pain score and 1-year postoperative ASES Function score were investigated as dependent variables. Regression coefficients and P and ω² values are reported. Omega square represents how much of the variance in an outcome variable a model explains, like R², and ω² values can also be calculated for individual factors to see how much variance a given factor accounts for. For a simple relative risk calculation, we divided our cohort into 3 equal-sized groups based on preoperative VR-12 MCS scores and compared the risk that patients with scores in the top third (better mental health) would end up below certain ASES Total scores with the risk of patients with scores in the bottom third (worse mental health). All statistical analyses were performed with Stata (StataCorp).

Results

Table 1 lists summary statistics for the population used in these models.

Our complete model for predicting VAS pain score 1 year after surgery accounted for 8% of the variability in this pain score (ω² = .076), whereas our complete model for predicting ASES Function score 1 year after surgery accounted for 22% of the variability (ω² = .219). These models include preoperative scores for VAS pain, ASES Function, VR-12 MCS, SANE, age at time of surgery, sex, and type of arthroplasty as possible explanatory variables.

Predicting VAS Pain Score (Table 2)

Preoperative VAS pain score and VR-12 MCS score both predicted 1-year postoperative VAS pain score (P < .001). Preoperative ASES Function score did not predict pain 1 year after surgery. By contrast, higher preoperative VAS pain scores were associated with higher VAS pain scores 1 year after surgery. Higher preoperative VR-12 MCS scores were significantly associated with lower VAS pain scores 1 year after surgery, indicating that better preoperative mental health is significantly associated with better self-reported outcomes in terms of pain 1 year after surgery. These associations remained statistically significant when controlling for age at time of surgery, sex, and type of arthroplasty.

Preoperative VR-12 MCS score was more predictive of 1-year postoperative VAS pain score than preoperative VAS pain score was. In other words, preoperative VR-12 MCS score accounted for more variability in outcome for 1-year postoperative VAS pain score (2.4%; ω² = .023) than preoperative VAS pain score did (1.6%; ω² = .015). 

Predicting ASES Function Score (Table 3)

By contrast, preoperative VAS pain score did not predict 1-year postoperative ASES Function score. Preoperative ASES Function and VR-12 MCS scores both predicted 1-year postoperative ASES Function score (P < .001). Higher preoperative ASES Function scores were associated with higher 1-year postoperative ASES Function scores. In other words, reporting better shoulder function before surgery was associated with reporting better shoulder function after surgery.

An example gives a sense of the effect size associated with the coefficient for preoperative ASES Function score. Our model predicts that, compared with a patient who reports 5 points lower on preoperative ASES Function (which ranges from 0-30), a patient who reports 5 points higher will report on average about 1 point higher on 1-year postoperative ASES Function. As in the model for postoperative pain, these associations with preoperative function and mental health scores held when controlling for age, sex, and type of arthroplasty. 

As in the postoperative pain model, preoperative VR-12 MCS score was more predictive of 1-year postoperative ASES Function score than preoperative ASES Function score was. Preoperative VR-12 MCS score accounted for more of the variation that our model predicts (ω² = .029) than preoperative ASES Function score did (ω² = .020). We compared the risk that patients with high preoperative VR-12 MCS scores (top third of cohort) would end up with ASES Total scores below 70, below 80, or below 90 with the risk of patients with low preoperative VR-12 MCS scores (bottom third). Results appear in Table 4.

A significant part of the predictive ability of our model for postoperative ASES Function scores stems from the fact that a patient who undergoes rTSA (vs TSA) is predicted to have an ASES Function score 3.8 points lower 1 year after surgery (P < .001, ω² = .083). With type of arthroplasty controlled for, female sex is associated with an ASES Function score 1.6 points lower 1 year after surgery (P < .001, ω² = .016).

Preoperative SANE score did not predict 1-year postoperative VAS pain score or ASES Function score. In addition, when our complete model was run with 1-year postoperative SANE score as the dependent variable, preoperative SANE score did not predict 1-year postoperative SANE score. Our data provide no supporting evidence for the use of SANE scores for predictive modeling for shoulder arthroplasty.

Discussion

We prospectively gathered data to determine which factors would predict patient subjective outcomes of primary TSA and primary rTSA. We hypothesized that preoperative VR-12 MCS scores and preoperative VAS pain scores would predict postoperative pain and function as measured with those PROMs. Second, we hypothesized that better preoperative mental health (as measured with VR-12 MCS scores) would predict lower postoperative pain (VAS pain scores) and better postoperative function (ASES Function scores). Third, we hypothesized that higher preoperative pain (VAS pain scores) would correlate with higher postoperative pain (VAS pain scores) and worse postoperative function (ASES Function scores).

Our main goal is to provide patients and surgeons with a predictive model that generates insights into what patients can expect after surgery. This model is not intended to be a screening tool for operative indications, but a clinical tool for helping set expectations.

Our results showed that patients with more pain before surgery were more likely to have more pain 1 year after surgery—confirming the hypothesized relationship between pain before and after surgery. Contrary to the hypothesis, however, degree of pain before surgery was not associated with function 1 year after surgery. Our mental health hypothesis was confirmed: Patients with better preoperative mental health scores had on average less pain and better function 1 year after surgery. Not surprisingly, our model demonstrated that patients with better self-reported function before surgery had better self-reported function after surgery. Patient-reported function before surgery did not significantly affect how much pain the patient had 1 year after surgery. Although we did not hypothesize about the role of function in predicting 1-year outcomes, function is a significant factor to be considered when setting patient expectations regarding shoulder arthroplasty outcomes (Table 5).

Although the effect sizes of each analyzed factor are small, together our models for 1-year postoperative pain and function provide significant insight into patients’ likely outcomes 1 year after TSA and rTSA.

Table 6 and Table 7 list preoperative PROMs and baseline characteristics for 2 sample patients and the corresponding 1-year postoperative results they should expect according to our model. Patient 1 (Table 6) achieves a theoretical ASES Total score of 67, and patient 2 (Table 7) achieves a theoretical ASES Total score of 90. During discussion of surgical options, these patients should be counseled differently. If patient 1 expects a “normal” shoulder after surgery, he or she likely will be disappointed with the outcome. Tools such as those provided here can contribute to evidence-based discussions and well-informed decision making.

Many studies have found that mental health correlated with pain and function during recovery from orthopedic trauma.^12-18 For example, Wylie and colleagues¹⁹ found that preoperative mental health, as measured with the 36-Item Short Form Health Survey (SF-36) MCS score, predicted patient-reported pain and function in the setting of rotator cuff injury, regardless of treatment type (operative, nonoperative). Others have found that mental health may play a role in how patients report their pain and function on various PROMs.^20,21 Modalities for improving patients’ emotional health baseline may even become a preoperative requirement as the healthcare industry moves toward value-based medicine and collection of patient-related outcomes as part of reimbursement schemes. 

By contrast, some studies have found that preoperative mental health did not predict postoperative outcomes. For example, Kennedy and colleagues²² found that preoperative mental health (as measured with SF-36 MCS scores) did not predict functional outcome in patients with ankle arthritis treated with ankle arthroplasty or arthrodesis. Likewise, Styron and colleagues²³ found no correlation between preoperative mental health (SF-12 MCS scores) and postoperative mental health and function in TSA. Their findings contradict those of the present study and many other studies.^12-18 The contradiction in findings demonstrates the need for well-designed, sufficiently powered studies of the link between preoperative mental health and postoperative outcome. Our study, with its large sample and heterogeneous population, is a start.

Two other groups (Simmen and colleagues,¹⁸ Matsen and colleagues²⁴) have attempted to develop a model predicting outcomes of shoulder arthroplasty. Simmen and colleagues¹⁸ estimated the probability of “treatment success” 1 year after TSA. Their model had 4 factors predictive of patient outcomes. Previous shoulder surgery and age over 75 years were significantly associated with lower probability of success, whereas higher preoperative SF-36 MCS scores and higher preoperative DASH (Disabilities of the Arm, Shoulder, and Hand) Function scores were associated with higher probability of success. The authors deemed TSA successful if the patient achieved a Constant score of ≥80 out of 100. Their model predicts probability of TSA “success,” whereas our models predict particular outcome scores. Both their results and ours support the hypothesis that preoperative mental health and function scores can predict how well a patient fares after surgery. Simmen and colleagues18 based their model on a cohort of only 140 patients and reported a 33.6% success rate (47/140 surgeries).

Matsen and colleagues²⁴ used a 1-practice cohort of 337 patients who underwent different types of arthroplasties, including TSA, rTSA, hemiarthroplasty, and ream-and-run arthroplasty. Although their focus was not preoperative PROMs predicting postoperative PROMs, they used the Simple Shoulder Test (SST) baseline score as a predictive variable. They found that 6 baseline characteristics—American Society of Anesthesiologists class I, shoulder problem unrelated to work, no prior shoulder surgery, glenoid type other than A1, humeral head not superiorly displaced on anteroposterior radiograph, and lower baseline SST score—were statistically associated with better outcomes, and they developed a model driven by these characteristics. They urged other investigators to perform the same kind of analysis with larger patient populations from multiple practices. One of the strengths of our study is its large patient population. We collected data on 1004 patients for modeling 1-year postoperative VAS pain scores and 986 patients for modeling 1-year postoperative ASES Function scores.

Our study had several limitations. First, its data came from a 42-surgeon database, and there may be variations in how these surgeons enroll patients in the registry. If some surgeons did not enroll all their surgical patients, our sample could have been subject to selection bias. Second, in developing our model, we used only patient characteristics that were available in the database. On the other hand, the heterogeneity of the surgeon sample lended external validity to the model. A third limitation was that the model always predicts better pain and function outcomes after TSA than after rTSA. In other words, it does not consider whether TSA is appropriate for a particular patient. Instead, it predicts 1-year shoulder arthroplasty outcomes. 

Our goal here is not to provide outcomes information or a surgical screening tool, but to report on our use of a simple data-driven tool for setting expectations. When appropriate data become available, tools like this should be expanded to predict longer-term shoulder arthroplasty outcomes. We need more studies that combine preoperative PROMs, more baseline clinical and patient characteristics (following the Matsen and colleagues²⁴ model), and large sample sizes.

Conclusion

The educational models presented here can help patients and surgeons learn what to expect after surgery. These models reveal the value in collecting preoperative subjective PROMs and show how a quantitative tool can help facilitate shared decision-making and set patient expectations. Separately, the effect size of each factor is small, but together a patient’s preoperative VAS pain score, ASES Function score, VR-12 MCS score, age, sex, and type of arthroplasty can provide information predictive of the patient’s self-reported pain and function 1 year after surgery.

Take-Home Points

• Shared decision-making tools, such as predictive models, can help empower the patient to make decisions for or against surgery equipped with more information about the expected outcome.
• There is a role for preoperative collection of PROMs in the clinical decision-making process.
• Mental health state, as reported by the VR-12 MCS, is a significant predictor of postoperative pain and function as reported by the VAS pain and ASES function scores.
• A significant portion of the predictive ability of this model comes from the fact that at 1-year postoperatively, patients receiving a rTSA will on average have a 3.8 point lower on ASES function score than those receiving a TSA (P < .001, ω²=.083).
• Future studies to discern the role of different modalities to improve a patient’s emotional health preoperatively will be beneficial as the healthcare industry trends toward value based medicine collecting PROMs as part of reimbursement schemes.

Over the past few decades, decisions regarding patients’ care have gradually transitioned from a paternalistic model to a more cooperative exchange between patient and physician. Shared decision-making provides patients a measure of autonomy in making choices for their health and their future. Patient participation may mitigate uncertainty and discomfort during selection of a course of treatment, which may lead to increased satisfaction levels after surgery.¹ Moreover, shared decision-making may help patients better manage postoperative expectations through evidenced-based discussions of preoperative health levels and their corresponding outcomes. Patient-reported outcome measures (PROMs) use clinically sensitive and specific metrics to evaluate a patient’s self-reported pain, functional ability, and mental state.² These metrics are useful in setting patient expectations for potential outcomes of treatment options. Use of evidence-based clinical decision-making tools, such as PROM-based predictive models, can facilitate a collaborative decision-making environment for patient and physician. Given the present cost-containment era, and the need for preoperative metrics that can assist in predictive analysis of postoperative improvement, models are clearly valuable.

In attempts to help patients set well-informed and reasonable expectations, physicians have turned to PROMs to facilitate preoperative evidence-based discussions. Although PROMs have been in use for almost 30 years, only recently have they been used to create tools that can aid quantitatively in the surgical decision-making process.^2-6 Combining physical examination findings, imaging studies, comorbidities, and quantitative tools, such as this model, may enhance patients’ understanding of their preoperative condition and expected prognosis and thereby guide their surgical decisions.

We conducted a study to determine whether certain preoperative PROMs can predict 1-year postoperative visual analog scale (VAS) pain scores and American Shoulder and Elbow Surgeons (ASES) Function scores in total shoulder arthroplasty (TSA) and reverse TSA (rTSA). We hypothesized that preoperative mental health status as captured by Veterans RAND 12-Item Health Survey (VR-12) mental health component summary (MCS) score and preoperative VAS pain score would predict both VAS pain score and ASES Function score 1 year after surgery. Specifically, we hypothesized that a higher preoperative VR-12 MCS score would predict less pain and better function 1 year after surgery and that a higher preoperative VAS pain score would predict more pain and worse function 1 year after surgery.

Methods

This study was approved by the Institutional Review Board of Partners Healthcare. The study used the Surgical Outcome System (Arthrex), a comprehensive prospective database that stores preoperative and 1-year postoperative patient demographics and TSA-PROM data. Surveys are emailed to all enrolled patients before surgery and 1 year after surgery. As indicated by the Institutional Review Boards of all participating institutions, patients in the Surgical Outcome System have to sign a consent form to permit use of their responses in research.

The database includes patient data from 42 orthopedic surgeons across the United States. All primary TSAs and primary rTSAs in the database were included in this study, regardless of arthroplasty indication. Revisions were excluded. Also excluded were cases in which the 1-year postoperative questionnaire was not completed. Of the 1681 patients eligible for 1-year follow-up, 1225 (73%) completed the 1-year postoperative questionnaire. PROMs used in the study were VAS pain score, ASES Function score, VR-12 MCS score, and Single Assessment Numerical Evaluation (SANE). Unfortunately, not all surgeons use every measure in the 1-year postoperative questionnaire set. Thus, in our complete models, total number of observations was 1004 for modeling 1-year postoperative VAS pain scores and 986 for modeling 1-year postoperative ASES Function scores. 

Metrics

On VAS, pain is rated from 0 (no pain) to 10 (pain as bad as it can be). Tashjian and colleagues⁷ estimated that the minimal clinically important difference (MCID) for postoperative VAS pain scores was 1.4 in a cohort of 326 patients who had TSA, rTSA, or shoulder hemiarthroplasty. ASES Function score is scaled from 0 to 30, with 30 representing best function.⁸ Wong and colleagues⁹ identified an MCID of 6.5 for ASES Function scores in a cohort of 107 patients who had TSA or rTSA. SANE ratings range from 0% to 100%, with 100% indicating the patient’s shoulder was totally “normal.”10 VR-12 MCS scores appear on a logarithmic scale, with higher numbers representing better mental health. The population mean estimate for VR-12 MCS scores is 50.1 (SD, 11.49; overall possible range, –2.47 to 76.1).11 Our patient population’s scores ranged from 12.5 to 73.8.

Statistical Analysis

Simple bivariate and multivariate linear regressions were performed to evaluate the predictive value of each of the outlined PROMs. Our complete model controls for patient sex, age, and type of arthroplasty. Categorical variables were dummy-coded. Both 1-year postoperative VAS pain score and 1-year postoperative ASES Function score were investigated as dependent variables. Regression coefficients and P and ω² values are reported. Omega square represents how much of the variance in an outcome variable a model explains, like R², and ω² values can also be calculated for individual factors to see how much variance a given factor accounts for. For a simple relative risk calculation, we divided our cohort into 3 equal-sized groups based on preoperative VR-12 MCS scores and compared the risk that patients with scores in the top third (better mental health) would end up below certain ASES Total scores with the risk of patients with scores in the bottom third (worse mental health). All statistical analyses were performed with Stata (StataCorp).

Results

Table 1 lists summary statistics for the population used in these models.

Our complete model for predicting VAS pain score 1 year after surgery accounted for 8% of the variability in this pain score (ω² = .076), whereas our complete model for predicting ASES Function score 1 year after surgery accounted for 22% of the variability (ω² = .219). These models include preoperative scores for VAS pain, ASES Function, VR-12 MCS, SANE, age at time of surgery, sex, and type of arthroplasty as possible explanatory variables.

Predicting VAS Pain Score (Table 2)

Preoperative VAS pain score and VR-12 MCS score both predicted 1-year postoperative VAS pain score (P < .001). Preoperative ASES Function score did not predict pain 1 year after surgery. By contrast, higher preoperative VAS pain scores were associated with higher VAS pain scores 1 year after surgery. Higher preoperative VR-12 MCS scores were significantly associated with lower VAS pain scores 1 year after surgery, indicating that better preoperative mental health is significantly associated with better self-reported outcomes in terms of pain 1 year after surgery. These associations remained statistically significant when controlling for age at time of surgery, sex, and type of arthroplasty.

Preoperative VR-12 MCS score was more predictive of 1-year postoperative VAS pain score than preoperative VAS pain score was. In other words, preoperative VR-12 MCS score accounted for more variability in outcome for 1-year postoperative VAS pain score (2.4%; ω² = .023) than preoperative VAS pain score did (1.6%; ω² = .015). 

Predicting ASES Function Score (Table 3)

By contrast, preoperative VAS pain score did not predict 1-year postoperative ASES Function score. Preoperative ASES Function and VR-12 MCS scores both predicted 1-year postoperative ASES Function score (P < .001). Higher preoperative ASES Function scores were associated with higher 1-year postoperative ASES Function scores. In other words, reporting better shoulder function before surgery was associated with reporting better shoulder function after surgery.

An example gives a sense of the effect size associated with the coefficient for preoperative ASES Function score. Our model predicts that, compared with a patient who reports 5 points lower on preoperative ASES Function (which ranges from 0-30), a patient who reports 5 points higher will report on average about 1 point higher on 1-year postoperative ASES Function. As in the model for postoperative pain, these associations with preoperative function and mental health scores held when controlling for age, sex, and type of arthroplasty. 

As in the postoperative pain model, preoperative VR-12 MCS score was more predictive of 1-year postoperative ASES Function score than preoperative ASES Function score was. Preoperative VR-12 MCS score accounted for more of the variation that our model predicts (ω² = .029) than preoperative ASES Function score did (ω² = .020). We compared the risk that patients with high preoperative VR-12 MCS scores (top third of cohort) would end up with ASES Total scores below 70, below 80, or below 90 with the risk of patients with low preoperative VR-12 MCS scores (bottom third). Results appear in Table 4.

A significant part of the predictive ability of our model for postoperative ASES Function scores stems from the fact that a patient who undergoes rTSA (vs TSA) is predicted to have an ASES Function score 3.8 points lower 1 year after surgery (P < .001, ω² = .083). With type of arthroplasty controlled for, female sex is associated with an ASES Function score 1.6 points lower 1 year after surgery (P < .001, ω² = .016).

Preoperative SANE score did not predict 1-year postoperative VAS pain score or ASES Function score. In addition, when our complete model was run with 1-year postoperative SANE score as the dependent variable, preoperative SANE score did not predict 1-year postoperative SANE score. Our data provide no supporting evidence for the use of SANE scores for predictive modeling for shoulder arthroplasty.

Discussion

We prospectively gathered data to determine which factors would predict patient subjective outcomes of primary TSA and primary rTSA. We hypothesized that preoperative VR-12 MCS scores and preoperative VAS pain scores would predict postoperative pain and function as measured with those PROMs. Second, we hypothesized that better preoperative mental health (as measured with VR-12 MCS scores) would predict lower postoperative pain (VAS pain scores) and better postoperative function (ASES Function scores). Third, we hypothesized that higher preoperative pain (VAS pain scores) would correlate with higher postoperative pain (VAS pain scores) and worse postoperative function (ASES Function scores).

Our main goal is to provide patients and surgeons with a predictive model that generates insights into what patients can expect after surgery. This model is not intended to be a screening tool for operative indications, but a clinical tool for helping set expectations.

Our results showed that patients with more pain before surgery were more likely to have more pain 1 year after surgery—confirming the hypothesized relationship between pain before and after surgery. Contrary to the hypothesis, however, degree of pain before surgery was not associated with function 1 year after surgery. Our mental health hypothesis was confirmed: Patients with better preoperative mental health scores had on average less pain and better function 1 year after surgery. Not surprisingly, our model demonstrated that patients with better self-reported function before surgery had better self-reported function after surgery. Patient-reported function before surgery did not significantly affect how much pain the patient had 1 year after surgery. Although we did not hypothesize about the role of function in predicting 1-year outcomes, function is a significant factor to be considered when setting patient expectations regarding shoulder arthroplasty outcomes (Table 5).

Although the effect sizes of each analyzed factor are small, together our models for 1-year postoperative pain and function provide significant insight into patients’ likely outcomes 1 year after TSA and rTSA.

Table 6 and Table 7 list preoperative PROMs and baseline characteristics for 2 sample patients and the corresponding 1-year postoperative results they should expect according to our model. Patient 1 (Table 6) achieves a theoretical ASES Total score of 67, and patient 2 (Table 7) achieves a theoretical ASES Total score of 90. During discussion of surgical options, these patients should be counseled differently. If patient 1 expects a “normal” shoulder after surgery, he or she likely will be disappointed with the outcome. Tools such as those provided here can contribute to evidence-based discussions and well-informed decision making.

Many studies have found that mental health correlated with pain and function during recovery from orthopedic trauma.^12-18 For example, Wylie and colleagues¹⁹ found that preoperative mental health, as measured with the 36-Item Short Form Health Survey (SF-36) MCS score, predicted patient-reported pain and function in the setting of rotator cuff injury, regardless of treatment type (operative, nonoperative). Others have found that mental health may play a role in how patients report their pain and function on various PROMs.^20,21 Modalities for improving patients’ emotional health baseline may even become a preoperative requirement as the healthcare industry moves toward value-based medicine and collection of patient-related outcomes as part of reimbursement schemes. 

By contrast, some studies have found that preoperative mental health did not predict postoperative outcomes. For example, Kennedy and colleagues²² found that preoperative mental health (as measured with SF-36 MCS scores) did not predict functional outcome in patients with ankle arthritis treated with ankle arthroplasty or arthrodesis. Likewise, Styron and colleagues²³ found no correlation between preoperative mental health (SF-12 MCS scores) and postoperative mental health and function in TSA. Their findings contradict those of the present study and many other studies.^12-18 The contradiction in findings demonstrates the need for well-designed, sufficiently powered studies of the link between preoperative mental health and postoperative outcome. Our study, with its large sample and heterogeneous population, is a start.

Two other groups (Simmen and colleagues,¹⁸ Matsen and colleagues²⁴) have attempted to develop a model predicting outcomes of shoulder arthroplasty. Simmen and colleagues¹⁸ estimated the probability of “treatment success” 1 year after TSA. Their model had 4 factors predictive of patient outcomes. Previous shoulder surgery and age over 75 years were significantly associated with lower probability of success, whereas higher preoperative SF-36 MCS scores and higher preoperative DASH (Disabilities of the Arm, Shoulder, and Hand) Function scores were associated with higher probability of success. The authors deemed TSA successful if the patient achieved a Constant score of ≥80 out of 100. Their model predicts probability of TSA “success,” whereas our models predict particular outcome scores. Both their results and ours support the hypothesis that preoperative mental health and function scores can predict how well a patient fares after surgery. Simmen and colleagues18 based their model on a cohort of only 140 patients and reported a 33.6% success rate (47/140 surgeries).

Matsen and colleagues²⁴ used a 1-practice cohort of 337 patients who underwent different types of arthroplasties, including TSA, rTSA, hemiarthroplasty, and ream-and-run arthroplasty. Although their focus was not preoperative PROMs predicting postoperative PROMs, they used the Simple Shoulder Test (SST) baseline score as a predictive variable. They found that 6 baseline characteristics—American Society of Anesthesiologists class I, shoulder problem unrelated to work, no prior shoulder surgery, glenoid type other than A1, humeral head not superiorly displaced on anteroposterior radiograph, and lower baseline SST score—were statistically associated with better outcomes, and they developed a model driven by these characteristics. They urged other investigators to perform the same kind of analysis with larger patient populations from multiple practices. One of the strengths of our study is its large patient population. We collected data on 1004 patients for modeling 1-year postoperative VAS pain scores and 986 patients for modeling 1-year postoperative ASES Function scores.

Our study had several limitations. First, its data came from a 42-surgeon database, and there may be variations in how these surgeons enroll patients in the registry. If some surgeons did not enroll all their surgical patients, our sample could have been subject to selection bias. Second, in developing our model, we used only patient characteristics that were available in the database. On the other hand, the heterogeneity of the surgeon sample lended external validity to the model. A third limitation was that the model always predicts better pain and function outcomes after TSA than after rTSA. In other words, it does not consider whether TSA is appropriate for a particular patient. Instead, it predicts 1-year shoulder arthroplasty outcomes. 

Our goal here is not to provide outcomes information or a surgical screening tool, but to report on our use of a simple data-driven tool for setting expectations. When appropriate data become available, tools like this should be expanded to predict longer-term shoulder arthroplasty outcomes. We need more studies that combine preoperative PROMs, more baseline clinical and patient characteristics (following the Matsen and colleagues²⁴ model), and large sample sizes.

Conclusion

The educational models presented here can help patients and surgeons learn what to expect after surgery. These models reveal the value in collecting preoperative subjective PROMs and show how a quantitative tool can help facilitate shared decision-making and set patient expectations. Separately, the effect size of each factor is small, but together a patient’s preoperative VAS pain score, ASES Function score, VR-12 MCS score, age, sex, and type of arthroplasty can provide information predictive of the patient’s self-reported pain and function 1 year after surgery.

Take-Home Points

• Shared decision-making tools, such as predictive models, can help empower the patient to make decisions for or against surgery equipped with more information about the expected outcome.
• There is a role for preoperative collection of PROMs in the clinical decision-making process.
• Mental health state, as reported by the VR-12 MCS, is a significant predictor of postoperative pain and function as reported by the VAS pain and ASES function scores.
• A significant portion of the predictive ability of this model comes from the fact that at 1-year postoperatively, patients receiving a rTSA will on average have a 3.8 point lower on ASES function score than those receiving a TSA (P < .001, ω²=.083).
• Future studies to discern the role of different modalities to improve a patient’s emotional health preoperatively will be beneficial as the healthcare industry trends toward value based medicine collecting PROMs as part of reimbursement schemes.

Over the past few decades, decisions regarding patients’ care have gradually transitioned from a paternalistic model to a more cooperative exchange between patient and physician. Shared decision-making provides patients a measure of autonomy in making choices for their health and their future. Patient participation may mitigate uncertainty and discomfort during selection of a course of treatment, which may lead to increased satisfaction levels after surgery.¹ Moreover, shared decision-making may help patients better manage postoperative expectations through evidenced-based discussions of preoperative health levels and their corresponding outcomes. Patient-reported outcome measures (PROMs) use clinically sensitive and specific metrics to evaluate a patient’s self-reported pain, functional ability, and mental state.² These metrics are useful in setting patient expectations for potential outcomes of treatment options. Use of evidence-based clinical decision-making tools, such as PROM-based predictive models, can facilitate a collaborative decision-making environment for patient and physician. Given the present cost-containment era, and the need for preoperative metrics that can assist in predictive analysis of postoperative improvement, models are clearly valuable.

In attempts to help patients set well-informed and reasonable expectations, physicians have turned to PROMs to facilitate preoperative evidence-based discussions. Although PROMs have been in use for almost 30 years, only recently have they been used to create tools that can aid quantitatively in the surgical decision-making process.^2-6 Combining physical examination findings, imaging studies, comorbidities, and quantitative tools, such as this model, may enhance patients’ understanding of their preoperative condition and expected prognosis and thereby guide their surgical decisions.

We conducted a study to determine whether certain preoperative PROMs can predict 1-year postoperative visual analog scale (VAS) pain scores and American Shoulder and Elbow Surgeons (ASES) Function scores in total shoulder arthroplasty (TSA) and reverse TSA (rTSA). We hypothesized that preoperative mental health status as captured by Veterans RAND 12-Item Health Survey (VR-12) mental health component summary (MCS) score and preoperative VAS pain score would predict both VAS pain score and ASES Function score 1 year after surgery. Specifically, we hypothesized that a higher preoperative VR-12 MCS score would predict less pain and better function 1 year after surgery and that a higher preoperative VAS pain score would predict more pain and worse function 1 year after surgery.

Methods

This study was approved by the Institutional Review Board of Partners Healthcare. The study used the Surgical Outcome System (Arthrex), a comprehensive prospective database that stores preoperative and 1-year postoperative patient demographics and TSA-PROM data. Surveys are emailed to all enrolled patients before surgery and 1 year after surgery. As indicated by the Institutional Review Boards of all participating institutions, patients in the Surgical Outcome System have to sign a consent form to permit use of their responses in research.

The database includes patient data from 42 orthopedic surgeons across the United States. All primary TSAs and primary rTSAs in the database were included in this study, regardless of arthroplasty indication. Revisions were excluded. Also excluded were cases in which the 1-year postoperative questionnaire was not completed. Of the 1681 patients eligible for 1-year follow-up, 1225 (73%) completed the 1-year postoperative questionnaire. PROMs used in the study were VAS pain score, ASES Function score, VR-12 MCS score, and Single Assessment Numerical Evaluation (SANE). Unfortunately, not all surgeons use every measure in the 1-year postoperative questionnaire set. Thus, in our complete models, total number of observations was 1004 for modeling 1-year postoperative VAS pain scores and 986 for modeling 1-year postoperative ASES Function scores. 

Metrics

On VAS, pain is rated from 0 (no pain) to 10 (pain as bad as it can be). Tashjian and colleagues⁷ estimated that the minimal clinically important difference (MCID) for postoperative VAS pain scores was 1.4 in a cohort of 326 patients who had TSA, rTSA, or shoulder hemiarthroplasty. ASES Function score is scaled from 0 to 30, with 30 representing best function.⁸ Wong and colleagues⁹ identified an MCID of 6.5 for ASES Function scores in a cohort of 107 patients who had TSA or rTSA. SANE ratings range from 0% to 100%, with 100% indicating the patient’s shoulder was totally “normal.”10 VR-12 MCS scores appear on a logarithmic scale, with higher numbers representing better mental health. The population mean estimate for VR-12 MCS scores is 50.1 (SD, 11.49; overall possible range, –2.47 to 76.1).11 Our patient population’s scores ranged from 12.5 to 73.8.

Statistical Analysis

Simple bivariate and multivariate linear regressions were performed to evaluate the predictive value of each of the outlined PROMs. Our complete model controls for patient sex, age, and type of arthroplasty. Categorical variables were dummy-coded. Both 1-year postoperative VAS pain score and 1-year postoperative ASES Function score were investigated as dependent variables. Regression coefficients and P and ω² values are reported. Omega square represents how much of the variance in an outcome variable a model explains, like R², and ω² values can also be calculated for individual factors to see how much variance a given factor accounts for. For a simple relative risk calculation, we divided our cohort into 3 equal-sized groups based on preoperative VR-12 MCS scores and compared the risk that patients with scores in the top third (better mental health) would end up below certain ASES Total scores with the risk of patients with scores in the bottom third (worse mental health). All statistical analyses were performed with Stata (StataCorp).

Results

Table 1 lists summary statistics for the population used in these models.

Our complete model for predicting VAS pain score 1 year after surgery accounted for 8% of the variability in this pain score (ω² = .076), whereas our complete model for predicting ASES Function score 1 year after surgery accounted for 22% of the variability (ω² = .219). These models include preoperative scores for VAS pain, ASES Function, VR-12 MCS, SANE, age at time of surgery, sex, and type of arthroplasty as possible explanatory variables.

Predicting VAS Pain Score (Table 2)

Preoperative VAS pain score and VR-12 MCS score both predicted 1-year postoperative VAS pain score (P < .001). Preoperative ASES Function score did not predict pain 1 year after surgery. By contrast, higher preoperative VAS pain scores were associated with higher VAS pain scores 1 year after surgery. Higher preoperative VR-12 MCS scores were significantly associated with lower VAS pain scores 1 year after surgery, indicating that better preoperative mental health is significantly associated with better self-reported outcomes in terms of pain 1 year after surgery. These associations remained statistically significant when controlling for age at time of surgery, sex, and type of arthroplasty.

Preoperative VR-12 MCS score was more predictive of 1-year postoperative VAS pain score than preoperative VAS pain score was. In other words, preoperative VR-12 MCS score accounted for more variability in outcome for 1-year postoperative VAS pain score (2.4%; ω² = .023) than preoperative VAS pain score did (1.6%; ω² = .015). 

Predicting ASES Function Score (Table 3)

By contrast, preoperative VAS pain score did not predict 1-year postoperative ASES Function score. Preoperative ASES Function and VR-12 MCS scores both predicted 1-year postoperative ASES Function score (P < .001). Higher preoperative ASES Function scores were associated with higher 1-year postoperative ASES Function scores. In other words, reporting better shoulder function before surgery was associated with reporting better shoulder function after surgery.

An example gives a sense of the effect size associated with the coefficient for preoperative ASES Function score. Our model predicts that, compared with a patient who reports 5 points lower on preoperative ASES Function (which ranges from 0-30), a patient who reports 5 points higher will report on average about 1 point higher on 1-year postoperative ASES Function. As in the model for postoperative pain, these associations with preoperative function and mental health scores held when controlling for age, sex, and type of arthroplasty. 

As in the postoperative pain model, preoperative VR-12 MCS score was more predictive of 1-year postoperative ASES Function score than preoperative ASES Function score was. Preoperative VR-12 MCS score accounted for more of the variation that our model predicts (ω² = .029) than preoperative ASES Function score did (ω² = .020). We compared the risk that patients with high preoperative VR-12 MCS scores (top third of cohort) would end up with ASES Total scores below 70, below 80, or below 90 with the risk of patients with low preoperative VR-12 MCS scores (bottom third). Results appear in Table 4.

A significant part of the predictive ability of our model for postoperative ASES Function scores stems from the fact that a patient who undergoes rTSA (vs TSA) is predicted to have an ASES Function score 3.8 points lower 1 year after surgery (P < .001, ω² = .083). With type of arthroplasty controlled for, female sex is associated with an ASES Function score 1.6 points lower 1 year after surgery (P < .001, ω² = .016).

Preoperative SANE score did not predict 1-year postoperative VAS pain score or ASES Function score. In addition, when our complete model was run with 1-year postoperative SANE score as the dependent variable, preoperative SANE score did not predict 1-year postoperative SANE score. Our data provide no supporting evidence for the use of SANE scores for predictive modeling for shoulder arthroplasty.

Discussion

We prospectively gathered data to determine which factors would predict patient subjective outcomes of primary TSA and primary rTSA. We hypothesized that preoperative VR-12 MCS scores and preoperative VAS pain scores would predict postoperative pain and function as measured with those PROMs. Second, we hypothesized that better preoperative mental health (as measured with VR-12 MCS scores) would predict lower postoperative pain (VAS pain scores) and better postoperative function (ASES Function scores). Third, we hypothesized that higher preoperative pain (VAS pain scores) would correlate with higher postoperative pain (VAS pain scores) and worse postoperative function (ASES Function scores).

Our main goal is to provide patients and surgeons with a predictive model that generates insights into what patients can expect after surgery. This model is not intended to be a screening tool for operative indications, but a clinical tool for helping set expectations.

Our results showed that patients with more pain before surgery were more likely to have more pain 1 year after surgery—confirming the hypothesized relationship between pain before and after surgery. Contrary to the hypothesis, however, degree of pain before surgery was not associated with function 1 year after surgery. Our mental health hypothesis was confirmed: Patients with better preoperative mental health scores had on average less pain and better function 1 year after surgery. Not surprisingly, our model demonstrated that patients with better self-reported function before surgery had better self-reported function after surgery. Patient-reported function before surgery did not significantly affect how much pain the patient had 1 year after surgery. Although we did not hypothesize about the role of function in predicting 1-year outcomes, function is a significant factor to be considered when setting patient expectations regarding shoulder arthroplasty outcomes (Table 5).

Although the effect sizes of each analyzed factor are small, together our models for 1-year postoperative pain and function provide significant insight into patients’ likely outcomes 1 year after TSA and rTSA.

Table 6 and Table 7 list preoperative PROMs and baseline characteristics for 2 sample patients and the corresponding 1-year postoperative results they should expect according to our model. Patient 1 (Table 6) achieves a theoretical ASES Total score of 67, and patient 2 (Table 7) achieves a theoretical ASES Total score of 90. During discussion of surgical options, these patients should be counseled differently. If patient 1 expects a “normal” shoulder after surgery, he or she likely will be disappointed with the outcome. Tools such as those provided here can contribute to evidence-based discussions and well-informed decision making.

Many studies have found that mental health correlated with pain and function during recovery from orthopedic trauma.^12-18 For example, Wylie and colleagues¹⁹ found that preoperative mental health, as measured with the 36-Item Short Form Health Survey (SF-36) MCS score, predicted patient-reported pain and function in the setting of rotator cuff injury, regardless of treatment type (operative, nonoperative). Others have found that mental health may play a role in how patients report their pain and function on various PROMs.^20,21 Modalities for improving patients’ emotional health baseline may even become a preoperative requirement as the healthcare industry moves toward value-based medicine and collection of patient-related outcomes as part of reimbursement schemes. 

By contrast, some studies have found that preoperative mental health did not predict postoperative outcomes. For example, Kennedy and colleagues²² found that preoperative mental health (as measured with SF-36 MCS scores) did not predict functional outcome in patients with ankle arthritis treated with ankle arthroplasty or arthrodesis. Likewise, Styron and colleagues²³ found no correlation between preoperative mental health (SF-12 MCS scores) and postoperative mental health and function in TSA. Their findings contradict those of the present study and many other studies.^12-18 The contradiction in findings demonstrates the need for well-designed, sufficiently powered studies of the link between preoperative mental health and postoperative outcome. Our study, with its large sample and heterogeneous population, is a start.

Two other groups (Simmen and colleagues,¹⁸ Matsen and colleagues²⁴) have attempted to develop a model predicting outcomes of shoulder arthroplasty. Simmen and colleagues¹⁸ estimated the probability of “treatment success” 1 year after TSA. Their model had 4 factors predictive of patient outcomes. Previous shoulder surgery and age over 75 years were significantly associated with lower probability of success, whereas higher preoperative SF-36 MCS scores and higher preoperative DASH (Disabilities of the Arm, Shoulder, and Hand) Function scores were associated with higher probability of success. The authors deemed TSA successful if the patient achieved a Constant score of ≥80 out of 100. Their model predicts probability of TSA “success,” whereas our models predict particular outcome scores. Both their results and ours support the hypothesis that preoperative mental health and function scores can predict how well a patient fares after surgery. Simmen and colleagues18 based their model on a cohort of only 140 patients and reported a 33.6% success rate (47/140 surgeries).

Matsen and colleagues²⁴ used a 1-practice cohort of 337 patients who underwent different types of arthroplasties, including TSA, rTSA, hemiarthroplasty, and ream-and-run arthroplasty. Although their focus was not preoperative PROMs predicting postoperative PROMs, they used the Simple Shoulder Test (SST) baseline score as a predictive variable. They found that 6 baseline characteristics—American Society of Anesthesiologists class I, shoulder problem unrelated to work, no prior shoulder surgery, glenoid type other than A1, humeral head not superiorly displaced on anteroposterior radiograph, and lower baseline SST score—were statistically associated with better outcomes, and they developed a model driven by these characteristics. They urged other investigators to perform the same kind of analysis with larger patient populations from multiple practices. One of the strengths of our study is its large patient population. We collected data on 1004 patients for modeling 1-year postoperative VAS pain scores and 986 patients for modeling 1-year postoperative ASES Function scores.

Our study had several limitations. First, its data came from a 42-surgeon database, and there may be variations in how these surgeons enroll patients in the registry. If some surgeons did not enroll all their surgical patients, our sample could have been subject to selection bias. Second, in developing our model, we used only patient characteristics that were available in the database. On the other hand, the heterogeneity of the surgeon sample lended external validity to the model. A third limitation was that the model always predicts better pain and function outcomes after TSA than after rTSA. In other words, it does not consider whether TSA is appropriate for a particular patient. Instead, it predicts 1-year shoulder arthroplasty outcomes. 

Our goal here is not to provide outcomes information or a surgical screening tool, but to report on our use of a simple data-driven tool for setting expectations. When appropriate data become available, tools like this should be expanded to predict longer-term shoulder arthroplasty outcomes. We need more studies that combine preoperative PROMs, more baseline clinical and patient characteristics (following the Matsen and colleagues²⁴ model), and large sample sizes.

Conclusion

The educational models presented here can help patients and surgeons learn what to expect after surgery. These models reveal the value in collecting preoperative subjective PROMs and show how a quantitative tool can help facilitate shared decision-making and set patient expectations. Separately, the effect size of each factor is small, but together a patient’s preoperative VAS pain score, ASES Function score, VR-12 MCS score, age, sex, and type of arthroplasty can provide information predictive of the patient’s self-reported pain and function 1 year after surgery.

Take-Home Points

• The authors have developed a total shoulder glenoid prosthesis that conforms with the humeral head in its center and is nonconforming on its peripheral edge.
• All clinical survey and range of motion parameters demonstrated statistically significant improvements at final follow-up.
• Only 3 shoulders (1.7%) required revision surgery.
• Eighty-six (63%) of 136 shoulders demonstrated no radiographic evidence of glenoid loosening.
• This is the first and largest study that evaluates the clinical and radiographic outcomes of this hybrid shoulder prosthesis.

Fixation of the glenoid component is the limiting factor in modern total shoulder arthroplasty (TSA). Glenoid loosening, the most common long-term complication, necessitates revision in up to 12% of patients.^1-4 By contrast, humeral component loosening is relatively uncommon, affecting as few as 0.34% of patients.⁵ Multiple long-term studies have found consistently high rates (45%-93%) of radiolucencies around the glenoid component.^3,6,7 Although their clinical significance has been debated, radiolucencies around the glenoid component raise concern about progressive loss of fixation.

Since TSA was introduced in the 1970s, complications with the glenoid component have been addressed with 2 different designs: conforming (congruent) and nonconforming. In a congruent articulation, the radii of curvature of the glenoid and humeral head components are identical, whereas they differ in a nonconforming model. Joint conformity is inversely related to glenohumeral translation.⁸ Neer’s original TSA was made congruent in order to limit translation and maximize the contact area. However, this design results in edge loading and a so-called rocking-horse phenomenon, which may lead to glenoid loosening.^9-13 Surgeons therefore have increasingly turned to nonconforming implants. In the nonconforming design, the radius of curvature of the humeral head is smaller than that of the glenoid. Although this design may reduce edge loading,¹⁴ it allows more translation and reduces the relative contact area of the glenohumeral joint. As a result, more contact stress is transmitted to the glenoid component, leading to polyethylene deformation and wear.^15,16

A desire to integrate the advantages of the 2 designs led to a novel glenoid implant design with variable conformity. This innovative component has a central conforming region and a peripheral nonconforming region or “translation zone” (Figure 1). 

Dual radii of curvature are designed to augment joint stability without increasing component wear. Biomechanical data have indicated that edge loading is not increased by having a central conforming region added to a nonconforming model.¹⁷ The clinical value of this prosthesis, however, has not been determined. Therefore, we conducted a study to describe the intermediate-term clinical and radiographic outcomes of TSAs that use a novel hybrid glenoid component.

Materials and Methods

This study was approved (protocol AAAD3473) by the Institutional Review Board of Columbia University and was conducted in compliance with Health Insurance Portability and Accountability Act (HIPAA) regulations.

Patient Selection

At Columbia University Medical Center, Dr. Bigliani performed 196 TSAs with a hybrid glenoid component (Bigliani-Flatow; Zimmer Biomet) in 169 patients between September 1998 and November 2007. All patients had received a diagnosis of primary glenohumeral arthritis as defined by Neer.¹⁸ Patients with previous surgery such as rotator cuff repair or subacromial decompression were included in our review, and patients with a nonprimary form of arthritis, such as rheumatoid, posttraumatic, or post-capsulorrhaphy arthritis, were excluded.

Operative Technique

For all surgeries, Dr. Bigliani performed a subscapularis tenotomy with regional anesthesia and a standard deltopectoral approach. A partial anterior capsulectomy was performed to increase the glenoid’s visibility. The inferior labrum was removed with a needle-tip bovie while the axillary nerve was being protected with a metal finger or narrow Darrach retractor. After reaming and trialing, the final glenoid component was cemented into place. Cement was placed only in the peg or keel holes and pressurized twice before final implantation. Of the 196 glenoid components, 168 (86%) were pegged and 28 (14%) keeled; in addition,190 of these components were all-polyethylene, whereas 6 had trabecular-metal backing. All glenoid components incorporated the hybrid design of dual radii of curvature. After the glenoid was cemented, the final humeral component was placed in 30° of retroversion. Whenever posterior wear was found, retroversion was reduced by 5° to 10°. The humeral prosthesis was cemented in cases (104/196, 53%) of poor bone quality or a large canal.

After surgery, the patient’s sling was fitted with an abduction pillow and a swathe, to be worn the first 24 hours, and the arm was passively ranged. Patients typically were discharged on postoperative day 2. Then, for 2 weeks, they followed an assisted passive range of motion (ROM) protocol, with limited external rotation, for promotion of subscapularis healing.

Clinical Outcomes

Dr. Bigliani assessed preoperative ROM in all planes. During initial evaluation, patients completed a questionnaire that consisted of the 36-Item Short Form Health Survey^19,20 (SF-36) and the American Shoulder and Elbow Surgeons²¹ (ASES) and Simple Shoulder Test²² (SST) surveys. Postoperative clinical data were collected from office follow-up visits, survey questionnaires, or both. Postoperative office data included ROM, subscapularis integrity testing (belly-press or lift-off), and any complications. Patients with <1 year of office follow-up were excluded. In addition, the same survey questionnaire that was used before surgery was mailed to all patients after surgery; then, for anyone who did not respond by mail, we attempted contact by telephone. Neer criteria were based on patients’ subjective assessment of each arm on a 3-point Likert scale (1 = very satisfied, 2 = satisfied, 3 = dissatisfied). Patients were also asked about any specific complications or revision operations since their index procedure.

Physical examination and office follow-up data were obtained for 129 patients (148/196 shoulders, 76% follow-up) at a mean of 3.7 years (range 1.0-10.2 years) after surgery. Surveys were completed by 117 patients (139/196 shoulders, 71% follow-up) at a mean of 5.1 years (range, 1.6-11.2 years) after surgery. Only 15 patients had neither 1 year of office follow-up nor a completed questionnaire. The remaining 154 patients (178/196 shoulders, 91% follow-up) had clinical follow-up with office, mail, or telephone questionnaire at a mean of 4.8 years (range, 1.0-11.2 years) after surgery. This cohort of patients was used to determine rates of surgical revisions, subscapularis tears, dislocations, and other complications.

Acromioplasty, performed in TSA patients who had subacromial impingement stemming from improved ROM, represented a second operation, and therefore the need for this surgery was deemed a complication as well. Figure 2 breaks down the 4 major study cohorts.

Radiographic Outcomes

Patients were included in the radiographic analysis if they had a shoulder radiograph at least 1 year after surgery. One hundred nineteen patients (136/196 shoulders, 69% follow-up) had radiographic follow-up at a mean of 3.7 years (range, 1.0-9.4 years) after surgery.

All radiographs were independently assessed by 2 blinded physicians who were not involved in the index procedure. Any disputed radiographs were reassessed by these physicians together, until consensus was reached. Radiographs were reviewed for the presence of glenoid lucencies around the pegs or keel and were scored using the system of Lazarus and colleagues²³ (Table 1). The humerus was assessed for total number of lucent lines in any of 8 periprosthetic zones, as described by Sperling and colleagues.²⁴

Statistical Analysis

Statistical analysis was performed with Stata Version 10.0. Paired t tests were used to compare preoperative and postoperative numerical data, including ROM and survey scores. We calculated 95% confidence intervals (CIs) and set statistical significance at P < .05. For qualitative measures, the Fisher exact test was used. Survivorship analysis was performed according to the Kaplan-Meier method, with right-censored data for no event or missing data.²⁵

Results

Clinical Analysis of Demographics

In demographics, the clinical and radiographic patient subgroups were similar to each other and to the overall study population (Table 2). Of 196 patients overall, 16 (8%) had a concomitant rotator cuff repair, and 27 (14%) underwent staged bilateral shoulder arthroplasties.

Clinical Analysis of ROM and Survey Scores

Operative shoulder ROM in forward elevation, external rotation at side, external rotation in abduction, and internal rotation all showed statistically significant (P < .001) improvement from before surgery to after surgery. Over 3.7 years, mean (SD) forward elevation improved from 107.3° (34.8°) to 159.0° (29.4°), external rotation at side improved from 20.4° (16.7°) to 49.4° (11.3°), and external rotation in abduction improved from 53.7° (24.3°) to 84.7° (9.1°). Internal rotation improved from a mean (SD) vertebral level of S1 (6.0 levels) to T9 (3.7 levels).

All validated survey scores also showed statistically significant (P < .001) improvement from before surgery to after surgery. Over 5.1 years, mean (SD) SF-36 scores improved from 64.9 (13.4) to 73.6 (17.1), ASES scores improved from 41.1 (22.5) to 82.7 (17.7), SST scores improved from 3.9 (2.8) to 9.7 (2.2), and visual analog scale pain scores improved from 5.6 (3.2) to 1.4 (2.1). Of 139 patients with follow-up, 130 (93.5%) were either satisfied or very satisfied with their TSA, and only 119 (86%) were either satisfied or very satisfied with the nonoperative shoulder.

Clinical Analysis of Postoperative Complications

Of the 178 shoulders evaluated for complications, 3 (1.7%) underwent revision surgery. Mean time to revision was 2.3 years (range, 1.5-3.9 years). Two revisions involved the glenoid component, and the third involved the humerus. In one of the glenoid cases, a 77-year-old woman fell and sustained a fracture at the base of the trabecular metal glenoid pegs; her component was revised to an all-polyethylene component, and she had no further complications. In the other glenoid case, a 73-year-old man’s all-polyethylene component loosened after 2 years and was revised to a trabecular metal implant, which loosened as well and was later converted to a hemiarthroplasty. In the humeral case, a 33-year-old man had his 4-year-old index TSA revised to a cemented stem and had no further complications.

Of the 148 patients with office follow-up, only 8 had a positive belly-press or lift-off test. Of all 178 clinical study shoulders, 10 (5.6%) had a subscapularis tear confirmed by magnetic resonance imaging or a physician. Of these 10 tears, 3 resulted from traumatic falls. Four of the 10 tears were managed nonoperatively, and the other 6 underwent surgical repair at a mean of 2.9 years (range, 0.3-7.8 years) after index TSA. In 2 of the 6 repair cases, a 46-mm humeral head had been used, and, in the other 4 cases, a 52-mm humeral head. Of the 6 repaired tears, 2 were massive, and 4 were isolated to the subscapularis. None of these 6 tears required a second repair. Seven (4%) of the 178 shoulders experienced a clinically significant posterosuperior subluxation or dislocation; 5 of the 7 were managed nonoperatively, and the other 2 underwent open capsular shift, at 0.5 year and 3.0 years, respectively. Table 3 lists the other postoperative complications that required surgery.

Table 4 compares the clinical and radiographic outcomes of patients who required subscapularis repair, capsular shift, or implant revision with the outcomes of all other study patients, and Figure 3 shows Kaplan-Meier survivorship.

Postoperative Radiographic Analysis

Glenoid Component. At a mean of 3.7 years (minimum, 1 year) after surgery, 86 (63%) of 136 radiographically evaluated shoulders showed no glenoid lucencies; the other 50 (37%) showed ≥1 lucency. Of the 136 shoulders, 33 (24%) had a Lazarus score of 1, 15 (11%) had a score of 2, and only 2 (2%) had a score of 3. None of the shoulders had a score of 4 or 5.

Humeral Component. Of the 136 shoulders, 91 (67%) showed no lucencies in any of the 8 humeral stem zones; the other 45 (33%) showed 1 to 3 lucencies. Thirty (22%) of the 136 shoulders had 1 stem lucency zone, 8 (6%) had 2, and 3 (2%) had 3. None of the shoulders had >3 periprosthetic zones with lucent lines.

Discussion

In this article, we describe a hybrid glenoid TSA component with dual radii of curvature. Its central portion is congruent with the humeral head, and its peripheral portion is noncongruent and larger. The most significant finding of our study is the low rate (1.1%) of glenoid component revision 4.8 years after surgery. This rate is the lowest that has been reported in a study of ≥100 patients. Overall implant survival appeared as an almost flat Kaplan-Meir curve. We attribute this low revision rate to improved biomechanics with the hybrid glenoid design. 

Symptomatic glenoid component loosening is the most common TSA complication.^1,26-28 In a review of 73 Neer TSAs, Cofield7 found glenoid radiolucencies in 71% of patients 3.8 years after surgery. Radiographic evidence of loosening, defined as component migration, or tilt, or a circumferential lucency 1.5 mm thick, was present in another 11% of patients, and 4.1% developed symptomatic loosening that required glenoid revision. In a study with 12.2-year follow-up, Torchia and colleagues³ found rates of 84% for glenoid radiolucencies, 44% for radiographic loosening, and 5.6% for symptomatic loosening that required revision. In a systematic review of studies with follow-up of ≥10 years, Bohsali and colleagues²⁷ found similar lucency and radiographic loosening rates and a 7% glenoid revision rate. These data suggest glenoid radiolucencies may progress to component loosening.

Degree of joint congruence is a key factor in glenoid loosening. Neer’s congruent design increases the contact area with concentric loading and reduces glenohumeral translation, which leads to reduced polyethylene wear and improved joint stability. In extreme arm positions, however, humeral head subluxation results in edge loading and a glenoid rocking-horse effect.^{9-13,17,29-31} Conversely, nonconforming implants allow increased glenohumeral translation without edge loading,¹⁴ though they also reduce the relative glenohumeral contact area and thus transmit more contact stress to the glenoid.^16,17 A hybrid glenoid component with central conforming and peripheral nonconforming zones may reduce the rocking-horse effect while maximizing ROM and joint stability. Wang and colleagues32 studied the biomechanical properties of this glenoid design and found that the addition of a central conforming region did not increase edge loading.

Additional results from our study support the efficacy of a hybrid glenoid component. Patients’ clinical outcomes improved significantly. At 5.1 years after surgery, 93.5% of patients were satisfied or very satisfied with their procedure and reported less satisfaction (86%) with the nonoperative shoulder. Also significant was the reduced number of radiolucencies. At 3.7 years after surgery, the overall percentage of shoulders with ≥1 glenoid radiolucency was 37%, considerably lower than the 82% reported by Cofield7 and the rates in more recent studies.^3,16,33-36 Of the 178 shoulders in our study, 10 (5.6%) had subscapularis tears, and 6 (3.4%) of 178 had these tears surgically repaired. This 3.4% compares favorably with the 5.9% (of 119 patients) found by Miller and colleagues37 28 months after surgery. Of our 178 shoulders, 27 (15.2%) had clinically significant postoperative complications; 18 (10.1%) of the 178 had these complications surgically treated, and 9 (5.1%) had them managed nonoperatively. Bohsali and colleagues²⁷ systematically reviewed 33 TSA studies and found a slightly higher complication rate (16.3%) 5.3 years after surgery. Furthermore, in our study, the 11 patients who underwent revision, capsular shift, or subscapularis repair had final outcomes comparable to those of the rest of our study population.

Our study had several potential weaknesses. First, its minimum clinical and radiographic follow-up was 1 year, whereas most long-term TSA series set a minimum of 2 years. We used 1 year because this was the first clinical study of the hybrid glenoid component design, and we wanted to maximize its sample size by reporting on intermediate-length outcomes. Even so, 93% (166/178) of our clinical patients and 83% (113/136) of our radiographic patients have had ≥2 years of follow-up, and we continue to follow all study patients for long-term outcomes. Another weakness of the study was its lack of a uniform group of patients with all the office, survey, complications, and radiographic data. Our retrospective study design made it difficult to obtain such a group without significantly reducing the sample size, so we divided patients into 4 data groups. A third potential weakness was the study’s variable method for collecting complications data. Rates of complications in the 178 shoulders were calculated from either office evaluation or patient self-report by mail or telephone. This data collection method is subject to recall bias, but mail and telephone contact was needed so the study would capture the large number of patients who had traveled to our institution for their surgery or had since moved away. Fourth, belly-press and lift-off tests were used in part to assess subscapularis function, but recent literature suggests post-TSA subscapularis assessment can be unreliable.³⁸ These tests may be positive in up to two-thirds of patients after 2 years.³⁹ Fifth, the generalizability of our findings to diagnoses such as rheumatoid and posttraumatic arthritis is limited. We had to restrict the study to patients with primary glenohumeral arthritis in order to minimize confounders.

This study’s main strength is its description of the clinical and radiographic outcomes of using a single prosthetic system in operations performed by a single surgeon in a large number of patients. This was the first and largest study evaluating the clinical and radiographic outcomes of this hybrid glenoid implant. Excluding patients with nonprimary arthritis allowed us to minimize potential confounding factors that affect patient outcomes. In conclusion, our study results showed the favorable clinical and radiographic outcomes of TSAs that have a hybrid glenoid component with dual radii of curvature. At a mean of 3.7 years after surgery, 63% of patients had no glenoid lucencies, and, at a mean of 4.8 years, only 1.7% of patients required revision. We continue to follow these patients to obtain long-term results of this innovative prosthesis.

Take-Home Points

• The authors have developed a total shoulder glenoid prosthesis that conforms with the humeral head in its center and is nonconforming on its peripheral edge.
• All clinical survey and range of motion parameters demonstrated statistically significant improvements at final follow-up.
• Only 3 shoulders (1.7%) required revision surgery.
• Eighty-six (63%) of 136 shoulders demonstrated no radiographic evidence of glenoid loosening.
• This is the first and largest study that evaluates the clinical and radiographic outcomes of this hybrid shoulder prosthesis.

Fixation of the glenoid component is the limiting factor in modern total shoulder arthroplasty (TSA). Glenoid loosening, the most common long-term complication, necessitates revision in up to 12% of patients.^1-4 By contrast, humeral component loosening is relatively uncommon, affecting as few as 0.34% of patients.⁵ Multiple long-term studies have found consistently high rates (45%-93%) of radiolucencies around the glenoid component.^3,6,7 Although their clinical significance has been debated, radiolucencies around the glenoid component raise concern about progressive loss of fixation.

Since TSA was introduced in the 1970s, complications with the glenoid component have been addressed with 2 different designs: conforming (congruent) and nonconforming. In a congruent articulation, the radii of curvature of the glenoid and humeral head components are identical, whereas they differ in a nonconforming model. Joint conformity is inversely related to glenohumeral translation.⁸ Neer’s original TSA was made congruent in order to limit translation and maximize the contact area. However, this design results in edge loading and a so-called rocking-horse phenomenon, which may lead to glenoid loosening.^9-13 Surgeons therefore have increasingly turned to nonconforming implants. In the nonconforming design, the radius of curvature of the humeral head is smaller than that of the glenoid. Although this design may reduce edge loading,¹⁴ it allows more translation and reduces the relative contact area of the glenohumeral joint. As a result, more contact stress is transmitted to the glenoid component, leading to polyethylene deformation and wear.^15,16

A desire to integrate the advantages of the 2 designs led to a novel glenoid implant design with variable conformity. This innovative component has a central conforming region and a peripheral nonconforming region or “translation zone” (Figure 1). 

Dual radii of curvature are designed to augment joint stability without increasing component wear. Biomechanical data have indicated that edge loading is not increased by having a central conforming region added to a nonconforming model.¹⁷ The clinical value of this prosthesis, however, has not been determined. Therefore, we conducted a study to describe the intermediate-term clinical and radiographic outcomes of TSAs that use a novel hybrid glenoid component.

Materials and Methods

This study was approved (protocol AAAD3473) by the Institutional Review Board of Columbia University and was conducted in compliance with Health Insurance Portability and Accountability Act (HIPAA) regulations.

Patient Selection

At Columbia University Medical Center, Dr. Bigliani performed 196 TSAs with a hybrid glenoid component (Bigliani-Flatow; Zimmer Biomet) in 169 patients between September 1998 and November 2007. All patients had received a diagnosis of primary glenohumeral arthritis as defined by Neer.¹⁸ Patients with previous surgery such as rotator cuff repair or subacromial decompression were included in our review, and patients with a nonprimary form of arthritis, such as rheumatoid, posttraumatic, or post-capsulorrhaphy arthritis, were excluded.

Operative Technique

For all surgeries, Dr. Bigliani performed a subscapularis tenotomy with regional anesthesia and a standard deltopectoral approach. A partial anterior capsulectomy was performed to increase the glenoid’s visibility. The inferior labrum was removed with a needle-tip bovie while the axillary nerve was being protected with a metal finger or narrow Darrach retractor. After reaming and trialing, the final glenoid component was cemented into place. Cement was placed only in the peg or keel holes and pressurized twice before final implantation. Of the 196 glenoid components, 168 (86%) were pegged and 28 (14%) keeled; in addition,190 of these components were all-polyethylene, whereas 6 had trabecular-metal backing. All glenoid components incorporated the hybrid design of dual radii of curvature. After the glenoid was cemented, the final humeral component was placed in 30° of retroversion. Whenever posterior wear was found, retroversion was reduced by 5° to 10°. The humeral prosthesis was cemented in cases (104/196, 53%) of poor bone quality or a large canal.

After surgery, the patient’s sling was fitted with an abduction pillow and a swathe, to be worn the first 24 hours, and the arm was passively ranged. Patients typically were discharged on postoperative day 2. Then, for 2 weeks, they followed an assisted passive range of motion (ROM) protocol, with limited external rotation, for promotion of subscapularis healing.

Clinical Outcomes

Dr. Bigliani assessed preoperative ROM in all planes. During initial evaluation, patients completed a questionnaire that consisted of the 36-Item Short Form Health Survey^19,20 (SF-36) and the American Shoulder and Elbow Surgeons²¹ (ASES) and Simple Shoulder Test²² (SST) surveys. Postoperative clinical data were collected from office follow-up visits, survey questionnaires, or both. Postoperative office data included ROM, subscapularis integrity testing (belly-press or lift-off), and any complications. Patients with <1 year of office follow-up were excluded. In addition, the same survey questionnaire that was used before surgery was mailed to all patients after surgery; then, for anyone who did not respond by mail, we attempted contact by telephone. Neer criteria were based on patients’ subjective assessment of each arm on a 3-point Likert scale (1 = very satisfied, 2 = satisfied, 3 = dissatisfied). Patients were also asked about any specific complications or revision operations since their index procedure.

Physical examination and office follow-up data were obtained for 129 patients (148/196 shoulders, 76% follow-up) at a mean of 3.7 years (range 1.0-10.2 years) after surgery. Surveys were completed by 117 patients (139/196 shoulders, 71% follow-up) at a mean of 5.1 years (range, 1.6-11.2 years) after surgery. Only 15 patients had neither 1 year of office follow-up nor a completed questionnaire. The remaining 154 patients (178/196 shoulders, 91% follow-up) had clinical follow-up with office, mail, or telephone questionnaire at a mean of 4.8 years (range, 1.0-11.2 years) after surgery. This cohort of patients was used to determine rates of surgical revisions, subscapularis tears, dislocations, and other complications.

Acromioplasty, performed in TSA patients who had subacromial impingement stemming from improved ROM, represented a second operation, and therefore the need for this surgery was deemed a complication as well. Figure 2 breaks down the 4 major study cohorts.

Radiographic Outcomes

Patients were included in the radiographic analysis if they had a shoulder radiograph at least 1 year after surgery. One hundred nineteen patients (136/196 shoulders, 69% follow-up) had radiographic follow-up at a mean of 3.7 years (range, 1.0-9.4 years) after surgery.

All radiographs were independently assessed by 2 blinded physicians who were not involved in the index procedure. Any disputed radiographs were reassessed by these physicians together, until consensus was reached. Radiographs were reviewed for the presence of glenoid lucencies around the pegs or keel and were scored using the system of Lazarus and colleagues²³ (Table 1). The humerus was assessed for total number of lucent lines in any of 8 periprosthetic zones, as described by Sperling and colleagues.²⁴

Statistical Analysis

Statistical analysis was performed with Stata Version 10.0. Paired t tests were used to compare preoperative and postoperative numerical data, including ROM and survey scores. We calculated 95% confidence intervals (CIs) and set statistical significance at P < .05. For qualitative measures, the Fisher exact test was used. Survivorship analysis was performed according to the Kaplan-Meier method, with right-censored data for no event or missing data.²⁵

Results

Clinical Analysis of Demographics

In demographics, the clinical and radiographic patient subgroups were similar to each other and to the overall study population (Table 2). Of 196 patients overall, 16 (8%) had a concomitant rotator cuff repair, and 27 (14%) underwent staged bilateral shoulder arthroplasties.

Clinical Analysis of ROM and Survey Scores

Operative shoulder ROM in forward elevation, external rotation at side, external rotation in abduction, and internal rotation all showed statistically significant (P < .001) improvement from before surgery to after surgery. Over 3.7 years, mean (SD) forward elevation improved from 107.3° (34.8°) to 159.0° (29.4°), external rotation at side improved from 20.4° (16.7°) to 49.4° (11.3°), and external rotation in abduction improved from 53.7° (24.3°) to 84.7° (9.1°). Internal rotation improved from a mean (SD) vertebral level of S1 (6.0 levels) to T9 (3.7 levels).

All validated survey scores also showed statistically significant (P < .001) improvement from before surgery to after surgery. Over 5.1 years, mean (SD) SF-36 scores improved from 64.9 (13.4) to 73.6 (17.1), ASES scores improved from 41.1 (22.5) to 82.7 (17.7), SST scores improved from 3.9 (2.8) to 9.7 (2.2), and visual analog scale pain scores improved from 5.6 (3.2) to 1.4 (2.1). Of 139 patients with follow-up, 130 (93.5%) were either satisfied or very satisfied with their TSA, and only 119 (86%) were either satisfied or very satisfied with the nonoperative shoulder.

Clinical Analysis of Postoperative Complications

Of the 178 shoulders evaluated for complications, 3 (1.7%) underwent revision surgery. Mean time to revision was 2.3 years (range, 1.5-3.9 years). Two revisions involved the glenoid component, and the third involved the humerus. In one of the glenoid cases, a 77-year-old woman fell and sustained a fracture at the base of the trabecular metal glenoid pegs; her component was revised to an all-polyethylene component, and she had no further complications. In the other glenoid case, a 73-year-old man’s all-polyethylene component loosened after 2 years and was revised to a trabecular metal implant, which loosened as well and was later converted to a hemiarthroplasty. In the humeral case, a 33-year-old man had his 4-year-old index TSA revised to a cemented stem and had no further complications.

Of the 148 patients with office follow-up, only 8 had a positive belly-press or lift-off test. Of all 178 clinical study shoulders, 10 (5.6%) had a subscapularis tear confirmed by magnetic resonance imaging or a physician. Of these 10 tears, 3 resulted from traumatic falls. Four of the 10 tears were managed nonoperatively, and the other 6 underwent surgical repair at a mean of 2.9 years (range, 0.3-7.8 years) after index TSA. In 2 of the 6 repair cases, a 46-mm humeral head had been used, and, in the other 4 cases, a 52-mm humeral head. Of the 6 repaired tears, 2 were massive, and 4 were isolated to the subscapularis. None of these 6 tears required a second repair. Seven (4%) of the 178 shoulders experienced a clinically significant posterosuperior subluxation or dislocation; 5 of the 7 were managed nonoperatively, and the other 2 underwent open capsular shift, at 0.5 year and 3.0 years, respectively. Table 3 lists the other postoperative complications that required surgery.

Table 4 compares the clinical and radiographic outcomes of patients who required subscapularis repair, capsular shift, or implant revision with the outcomes of all other study patients, and Figure 3 shows Kaplan-Meier survivorship.

Postoperative Radiographic Analysis

Glenoid Component. At a mean of 3.7 years (minimum, 1 year) after surgery, 86 (63%) of 136 radiographically evaluated shoulders showed no glenoid lucencies; the other 50 (37%) showed ≥1 lucency. Of the 136 shoulders, 33 (24%) had a Lazarus score of 1, 15 (11%) had a score of 2, and only 2 (2%) had a score of 3. None of the shoulders had a score of 4 or 5.

Humeral Component. Of the 136 shoulders, 91 (67%) showed no lucencies in any of the 8 humeral stem zones; the other 45 (33%) showed 1 to 3 lucencies. Thirty (22%) of the 136 shoulders had 1 stem lucency zone, 8 (6%) had 2, and 3 (2%) had 3. None of the shoulders had >3 periprosthetic zones with lucent lines.

Discussion

In this article, we describe a hybrid glenoid TSA component with dual radii of curvature. Its central portion is congruent with the humeral head, and its peripheral portion is noncongruent and larger. The most significant finding of our study is the low rate (1.1%) of glenoid component revision 4.8 years after surgery. This rate is the lowest that has been reported in a study of ≥100 patients. Overall implant survival appeared as an almost flat Kaplan-Meir curve. We attribute this low revision rate to improved biomechanics with the hybrid glenoid design. 

Symptomatic glenoid component loosening is the most common TSA complication.^1,26-28 In a review of 73 Neer TSAs, Cofield7 found glenoid radiolucencies in 71% of patients 3.8 years after surgery. Radiographic evidence of loosening, defined as component migration, or tilt, or a circumferential lucency 1.5 mm thick, was present in another 11% of patients, and 4.1% developed symptomatic loosening that required glenoid revision. In a study with 12.2-year follow-up, Torchia and colleagues³ found rates of 84% for glenoid radiolucencies, 44% for radiographic loosening, and 5.6% for symptomatic loosening that required revision. In a systematic review of studies with follow-up of ≥10 years, Bohsali and colleagues²⁷ found similar lucency and radiographic loosening rates and a 7% glenoid revision rate. These data suggest glenoid radiolucencies may progress to component loosening.

Degree of joint congruence is a key factor in glenoid loosening. Neer’s congruent design increases the contact area with concentric loading and reduces glenohumeral translation, which leads to reduced polyethylene wear and improved joint stability. In extreme arm positions, however, humeral head subluxation results in edge loading and a glenoid rocking-horse effect.^{9-13,17,29-31} Conversely, nonconforming implants allow increased glenohumeral translation without edge loading,¹⁴ though they also reduce the relative glenohumeral contact area and thus transmit more contact stress to the glenoid.^16,17 A hybrid glenoid component with central conforming and peripheral nonconforming zones may reduce the rocking-horse effect while maximizing ROM and joint stability. Wang and colleagues32 studied the biomechanical properties of this glenoid design and found that the addition of a central conforming region did not increase edge loading.

Additional results from our study support the efficacy of a hybrid glenoid component. Patients’ clinical outcomes improved significantly. At 5.1 years after surgery, 93.5% of patients were satisfied or very satisfied with their procedure and reported less satisfaction (86%) with the nonoperative shoulder. Also significant was the reduced number of radiolucencies. At 3.7 years after surgery, the overall percentage of shoulders with ≥1 glenoid radiolucency was 37%, considerably lower than the 82% reported by Cofield7 and the rates in more recent studies.^3,16,33-36 Of the 178 shoulders in our study, 10 (5.6%) had subscapularis tears, and 6 (3.4%) of 178 had these tears surgically repaired. This 3.4% compares favorably with the 5.9% (of 119 patients) found by Miller and colleagues37 28 months after surgery. Of our 178 shoulders, 27 (15.2%) had clinically significant postoperative complications; 18 (10.1%) of the 178 had these complications surgically treated, and 9 (5.1%) had them managed nonoperatively. Bohsali and colleagues²⁷ systematically reviewed 33 TSA studies and found a slightly higher complication rate (16.3%) 5.3 years after surgery. Furthermore, in our study, the 11 patients who underwent revision, capsular shift, or subscapularis repair had final outcomes comparable to those of the rest of our study population.

Our study had several potential weaknesses. First, its minimum clinical and radiographic follow-up was 1 year, whereas most long-term TSA series set a minimum of 2 years. We used 1 year because this was the first clinical study of the hybrid glenoid component design, and we wanted to maximize its sample size by reporting on intermediate-length outcomes. Even so, 93% (166/178) of our clinical patients and 83% (113/136) of our radiographic patients have had ≥2 years of follow-up, and we continue to follow all study patients for long-term outcomes. Another weakness of the study was its lack of a uniform group of patients with all the office, survey, complications, and radiographic data. Our retrospective study design made it difficult to obtain such a group without significantly reducing the sample size, so we divided patients into 4 data groups. A third potential weakness was the study’s variable method for collecting complications data. Rates of complications in the 178 shoulders were calculated from either office evaluation or patient self-report by mail or telephone. This data collection method is subject to recall bias, but mail and telephone contact was needed so the study would capture the large number of patients who had traveled to our institution for their surgery or had since moved away. Fourth, belly-press and lift-off tests were used in part to assess subscapularis function, but recent literature suggests post-TSA subscapularis assessment can be unreliable.³⁸ These tests may be positive in up to two-thirds of patients after 2 years.³⁹ Fifth, the generalizability of our findings to diagnoses such as rheumatoid and posttraumatic arthritis is limited. We had to restrict the study to patients with primary glenohumeral arthritis in order to minimize confounders.

This study’s main strength is its description of the clinical and radiographic outcomes of using a single prosthetic system in operations performed by a single surgeon in a large number of patients. This was the first and largest study evaluating the clinical and radiographic outcomes of this hybrid glenoid implant. Excluding patients with nonprimary arthritis allowed us to minimize potential confounding factors that affect patient outcomes. In conclusion, our study results showed the favorable clinical and radiographic outcomes of TSAs that have a hybrid glenoid component with dual radii of curvature. At a mean of 3.7 years after surgery, 63% of patients had no glenoid lucencies, and, at a mean of 4.8 years, only 1.7% of patients required revision. We continue to follow these patients to obtain long-term results of this innovative prosthesis.

Take-Home Points

• The authors have developed a total shoulder glenoid prosthesis that conforms with the humeral head in its center and is nonconforming on its peripheral edge.
• All clinical survey and range of motion parameters demonstrated statistically significant improvements at final follow-up.
• Only 3 shoulders (1.7%) required revision surgery.
• Eighty-six (63%) of 136 shoulders demonstrated no radiographic evidence of glenoid loosening.
• This is the first and largest study that evaluates the clinical and radiographic outcomes of this hybrid shoulder prosthesis.

Fixation of the glenoid component is the limiting factor in modern total shoulder arthroplasty (TSA). Glenoid loosening, the most common long-term complication, necessitates revision in up to 12% of patients.^1-4 By contrast, humeral component loosening is relatively uncommon, affecting as few as 0.34% of patients.⁵ Multiple long-term studies have found consistently high rates (45%-93%) of radiolucencies around the glenoid component.^3,6,7 Although their clinical significance has been debated, radiolucencies around the glenoid component raise concern about progressive loss of fixation.

Since TSA was introduced in the 1970s, complications with the glenoid component have been addressed with 2 different designs: conforming (congruent) and nonconforming. In a congruent articulation, the radii of curvature of the glenoid and humeral head components are identical, whereas they differ in a nonconforming model. Joint conformity is inversely related to glenohumeral translation.⁸ Neer’s original TSA was made congruent in order to limit translation and maximize the contact area. However, this design results in edge loading and a so-called rocking-horse phenomenon, which may lead to glenoid loosening.^9-13 Surgeons therefore have increasingly turned to nonconforming implants. In the nonconforming design, the radius of curvature of the humeral head is smaller than that of the glenoid. Although this design may reduce edge loading,¹⁴ it allows more translation and reduces the relative contact area of the glenohumeral joint. As a result, more contact stress is transmitted to the glenoid component, leading to polyethylene deformation and wear.^15,16

A desire to integrate the advantages of the 2 designs led to a novel glenoid implant design with variable conformity. This innovative component has a central conforming region and a peripheral nonconforming region or “translation zone” (Figure 1). 

Dual radii of curvature are designed to augment joint stability without increasing component wear. Biomechanical data have indicated that edge loading is not increased by having a central conforming region added to a nonconforming model.¹⁷ The clinical value of this prosthesis, however, has not been determined. Therefore, we conducted a study to describe the intermediate-term clinical and radiographic outcomes of TSAs that use a novel hybrid glenoid component.

Materials and Methods

This study was approved (protocol AAAD3473) by the Institutional Review Board of Columbia University and was conducted in compliance with Health Insurance Portability and Accountability Act (HIPAA) regulations.

Patient Selection

At Columbia University Medical Center, Dr. Bigliani performed 196 TSAs with a hybrid glenoid component (Bigliani-Flatow; Zimmer Biomet) in 169 patients between September 1998 and November 2007. All patients had received a diagnosis of primary glenohumeral arthritis as defined by Neer.¹⁸ Patients with previous surgery such as rotator cuff repair or subacromial decompression were included in our review, and patients with a nonprimary form of arthritis, such as rheumatoid, posttraumatic, or post-capsulorrhaphy arthritis, were excluded.

Operative Technique

For all surgeries, Dr. Bigliani performed a subscapularis tenotomy with regional anesthesia and a standard deltopectoral approach. A partial anterior capsulectomy was performed to increase the glenoid’s visibility. The inferior labrum was removed with a needle-tip bovie while the axillary nerve was being protected with a metal finger or narrow Darrach retractor. After reaming and trialing, the final glenoid component was cemented into place. Cement was placed only in the peg or keel holes and pressurized twice before final implantation. Of the 196 glenoid components, 168 (86%) were pegged and 28 (14%) keeled; in addition,190 of these components were all-polyethylene, whereas 6 had trabecular-metal backing. All glenoid components incorporated the hybrid design of dual radii of curvature. After the glenoid was cemented, the final humeral component was placed in 30° of retroversion. Whenever posterior wear was found, retroversion was reduced by 5° to 10°. The humeral prosthesis was cemented in cases (104/196, 53%) of poor bone quality or a large canal.

After surgery, the patient’s sling was fitted with an abduction pillow and a swathe, to be worn the first 24 hours, and the arm was passively ranged. Patients typically were discharged on postoperative day 2. Then, for 2 weeks, they followed an assisted passive range of motion (ROM) protocol, with limited external rotation, for promotion of subscapularis healing.

Clinical Outcomes

Dr. Bigliani assessed preoperative ROM in all planes. During initial evaluation, patients completed a questionnaire that consisted of the 36-Item Short Form Health Survey^19,20 (SF-36) and the American Shoulder and Elbow Surgeons²¹ (ASES) and Simple Shoulder Test²² (SST) surveys. Postoperative clinical data were collected from office follow-up visits, survey questionnaires, or both. Postoperative office data included ROM, subscapularis integrity testing (belly-press or lift-off), and any complications. Patients with <1 year of office follow-up were excluded. In addition, the same survey questionnaire that was used before surgery was mailed to all patients after surgery; then, for anyone who did not respond by mail, we attempted contact by telephone. Neer criteria were based on patients’ subjective assessment of each arm on a 3-point Likert scale (1 = very satisfied, 2 = satisfied, 3 = dissatisfied). Patients were also asked about any specific complications or revision operations since their index procedure.

Physical examination and office follow-up data were obtained for 129 patients (148/196 shoulders, 76% follow-up) at a mean of 3.7 years (range 1.0-10.2 years) after surgery. Surveys were completed by 117 patients (139/196 shoulders, 71% follow-up) at a mean of 5.1 years (range, 1.6-11.2 years) after surgery. Only 15 patients had neither 1 year of office follow-up nor a completed questionnaire. The remaining 154 patients (178/196 shoulders, 91% follow-up) had clinical follow-up with office, mail, or telephone questionnaire at a mean of 4.8 years (range, 1.0-11.2 years) after surgery. This cohort of patients was used to determine rates of surgical revisions, subscapularis tears, dislocations, and other complications.

Acromioplasty, performed in TSA patients who had subacromial impingement stemming from improved ROM, represented a second operation, and therefore the need for this surgery was deemed a complication as well. Figure 2 breaks down the 4 major study cohorts.

Radiographic Outcomes

Patients were included in the radiographic analysis if they had a shoulder radiograph at least 1 year after surgery. One hundred nineteen patients (136/196 shoulders, 69% follow-up) had radiographic follow-up at a mean of 3.7 years (range, 1.0-9.4 years) after surgery.

All radiographs were independently assessed by 2 blinded physicians who were not involved in the index procedure. Any disputed radiographs were reassessed by these physicians together, until consensus was reached. Radiographs were reviewed for the presence of glenoid lucencies around the pegs or keel and were scored using the system of Lazarus and colleagues²³ (Table 1). The humerus was assessed for total number of lucent lines in any of 8 periprosthetic zones, as described by Sperling and colleagues.²⁴

Statistical Analysis

Statistical analysis was performed with Stata Version 10.0. Paired t tests were used to compare preoperative and postoperative numerical data, including ROM and survey scores. We calculated 95% confidence intervals (CIs) and set statistical significance at P < .05. For qualitative measures, the Fisher exact test was used. Survivorship analysis was performed according to the Kaplan-Meier method, with right-censored data for no event or missing data.²⁵

Results

Clinical Analysis of Demographics

In demographics, the clinical and radiographic patient subgroups were similar to each other and to the overall study population (Table 2). Of 196 patients overall, 16 (8%) had a concomitant rotator cuff repair, and 27 (14%) underwent staged bilateral shoulder arthroplasties.

Clinical Analysis of ROM and Survey Scores

Operative shoulder ROM in forward elevation, external rotation at side, external rotation in abduction, and internal rotation all showed statistically significant (P < .001) improvement from before surgery to after surgery. Over 3.7 years, mean (SD) forward elevation improved from 107.3° (34.8°) to 159.0° (29.4°), external rotation at side improved from 20.4° (16.7°) to 49.4° (11.3°), and external rotation in abduction improved from 53.7° (24.3°) to 84.7° (9.1°). Internal rotation improved from a mean (SD) vertebral level of S1 (6.0 levels) to T9 (3.7 levels).

All validated survey scores also showed statistically significant (P < .001) improvement from before surgery to after surgery. Over 5.1 years, mean (SD) SF-36 scores improved from 64.9 (13.4) to 73.6 (17.1), ASES scores improved from 41.1 (22.5) to 82.7 (17.7), SST scores improved from 3.9 (2.8) to 9.7 (2.2), and visual analog scale pain scores improved from 5.6 (3.2) to 1.4 (2.1). Of 139 patients with follow-up, 130 (93.5%) were either satisfied or very satisfied with their TSA, and only 119 (86%) were either satisfied or very satisfied with the nonoperative shoulder.

Clinical Analysis of Postoperative Complications

Of the 178 shoulders evaluated for complications, 3 (1.7%) underwent revision surgery. Mean time to revision was 2.3 years (range, 1.5-3.9 years). Two revisions involved the glenoid component, and the third involved the humerus. In one of the glenoid cases, a 77-year-old woman fell and sustained a fracture at the base of the trabecular metal glenoid pegs; her component was revised to an all-polyethylene component, and she had no further complications. In the other glenoid case, a 73-year-old man’s all-polyethylene component loosened after 2 years and was revised to a trabecular metal implant, which loosened as well and was later converted to a hemiarthroplasty. In the humeral case, a 33-year-old man had his 4-year-old index TSA revised to a cemented stem and had no further complications.

Of the 148 patients with office follow-up, only 8 had a positive belly-press or lift-off test. Of all 178 clinical study shoulders, 10 (5.6%) had a subscapularis tear confirmed by magnetic resonance imaging or a physician. Of these 10 tears, 3 resulted from traumatic falls. Four of the 10 tears were managed nonoperatively, and the other 6 underwent surgical repair at a mean of 2.9 years (range, 0.3-7.8 years) after index TSA. In 2 of the 6 repair cases, a 46-mm humeral head had been used, and, in the other 4 cases, a 52-mm humeral head. Of the 6 repaired tears, 2 were massive, and 4 were isolated to the subscapularis. None of these 6 tears required a second repair. Seven (4%) of the 178 shoulders experienced a clinically significant posterosuperior subluxation or dislocation; 5 of the 7 were managed nonoperatively, and the other 2 underwent open capsular shift, at 0.5 year and 3.0 years, respectively. Table 3 lists the other postoperative complications that required surgery.

Table 4 compares the clinical and radiographic outcomes of patients who required subscapularis repair, capsular shift, or implant revision with the outcomes of all other study patients, and Figure 3 shows Kaplan-Meier survivorship.

Postoperative Radiographic Analysis

Glenoid Component. At a mean of 3.7 years (minimum, 1 year) after surgery, 86 (63%) of 136 radiographically evaluated shoulders showed no glenoid lucencies; the other 50 (37%) showed ≥1 lucency. Of the 136 shoulders, 33 (24%) had a Lazarus score of 1, 15 (11%) had a score of 2, and only 2 (2%) had a score of 3. None of the shoulders had a score of 4 or 5.

Humeral Component. Of the 136 shoulders, 91 (67%) showed no lucencies in any of the 8 humeral stem zones; the other 45 (33%) showed 1 to 3 lucencies. Thirty (22%) of the 136 shoulders had 1 stem lucency zone, 8 (6%) had 2, and 3 (2%) had 3. None of the shoulders had >3 periprosthetic zones with lucent lines.

Discussion

In this article, we describe a hybrid glenoid TSA component with dual radii of curvature. Its central portion is congruent with the humeral head, and its peripheral portion is noncongruent and larger. The most significant finding of our study is the low rate (1.1%) of glenoid component revision 4.8 years after surgery. This rate is the lowest that has been reported in a study of ≥100 patients. Overall implant survival appeared as an almost flat Kaplan-Meir curve. We attribute this low revision rate to improved biomechanics with the hybrid glenoid design. 

Symptomatic glenoid component loosening is the most common TSA complication.^1,26-28 In a review of 73 Neer TSAs, Cofield7 found glenoid radiolucencies in 71% of patients 3.8 years after surgery. Radiographic evidence of loosening, defined as component migration, or tilt, or a circumferential lucency 1.5 mm thick, was present in another 11% of patients, and 4.1% developed symptomatic loosening that required glenoid revision. In a study with 12.2-year follow-up, Torchia and colleagues³ found rates of 84% for glenoid radiolucencies, 44% for radiographic loosening, and 5.6% for symptomatic loosening that required revision. In a systematic review of studies with follow-up of ≥10 years, Bohsali and colleagues²⁷ found similar lucency and radiographic loosening rates and a 7% glenoid revision rate. These data suggest glenoid radiolucencies may progress to component loosening.

Degree of joint congruence is a key factor in glenoid loosening. Neer’s congruent design increases the contact area with concentric loading and reduces glenohumeral translation, which leads to reduced polyethylene wear and improved joint stability. In extreme arm positions, however, humeral head subluxation results in edge loading and a glenoid rocking-horse effect.^{9-13,17,29-31} Conversely, nonconforming implants allow increased glenohumeral translation without edge loading,¹⁴ though they also reduce the relative glenohumeral contact area and thus transmit more contact stress to the glenoid.^16,17 A hybrid glenoid component with central conforming and peripheral nonconforming zones may reduce the rocking-horse effect while maximizing ROM and joint stability. Wang and colleagues32 studied the biomechanical properties of this glenoid design and found that the addition of a central conforming region did not increase edge loading.

Additional results from our study support the efficacy of a hybrid glenoid component. Patients’ clinical outcomes improved significantly. At 5.1 years after surgery, 93.5% of patients were satisfied or very satisfied with their procedure and reported less satisfaction (86%) with the nonoperative shoulder. Also significant was the reduced number of radiolucencies. At 3.7 years after surgery, the overall percentage of shoulders with ≥1 glenoid radiolucency was 37%, considerably lower than the 82% reported by Cofield7 and the rates in more recent studies.^3,16,33-36 Of the 178 shoulders in our study, 10 (5.6%) had subscapularis tears, and 6 (3.4%) of 178 had these tears surgically repaired. This 3.4% compares favorably with the 5.9% (of 119 patients) found by Miller and colleagues37 28 months after surgery. Of our 178 shoulders, 27 (15.2%) had clinically significant postoperative complications; 18 (10.1%) of the 178 had these complications surgically treated, and 9 (5.1%) had them managed nonoperatively. Bohsali and colleagues²⁷ systematically reviewed 33 TSA studies and found a slightly higher complication rate (16.3%) 5.3 years after surgery. Furthermore, in our study, the 11 patients who underwent revision, capsular shift, or subscapularis repair had final outcomes comparable to those of the rest of our study population.

Our study had several potential weaknesses. First, its minimum clinical and radiographic follow-up was 1 year, whereas most long-term TSA series set a minimum of 2 years. We used 1 year because this was the first clinical study of the hybrid glenoid component design, and we wanted to maximize its sample size by reporting on intermediate-length outcomes. Even so, 93% (166/178) of our clinical patients and 83% (113/136) of our radiographic patients have had ≥2 years of follow-up, and we continue to follow all study patients for long-term outcomes. Another weakness of the study was its lack of a uniform group of patients with all the office, survey, complications, and radiographic data. Our retrospective study design made it difficult to obtain such a group without significantly reducing the sample size, so we divided patients into 4 data groups. A third potential weakness was the study’s variable method for collecting complications data. Rates of complications in the 178 shoulders were calculated from either office evaluation or patient self-report by mail or telephone. This data collection method is subject to recall bias, but mail and telephone contact was needed so the study would capture the large number of patients who had traveled to our institution for their surgery or had since moved away. Fourth, belly-press and lift-off tests were used in part to assess subscapularis function, but recent literature suggests post-TSA subscapularis assessment can be unreliable.³⁸ These tests may be positive in up to two-thirds of patients after 2 years.³⁹ Fifth, the generalizability of our findings to diagnoses such as rheumatoid and posttraumatic arthritis is limited. We had to restrict the study to patients with primary glenohumeral arthritis in order to minimize confounders.

This study’s main strength is its description of the clinical and radiographic outcomes of using a single prosthetic system in operations performed by a single surgeon in a large number of patients. This was the first and largest study evaluating the clinical and radiographic outcomes of this hybrid glenoid implant. Excluding patients with nonprimary arthritis allowed us to minimize potential confounding factors that affect patient outcomes. In conclusion, our study results showed the favorable clinical and radiographic outcomes of TSAs that have a hybrid glenoid component with dual radii of curvature. At a mean of 3.7 years after surgery, 63% of patients had no glenoid lucencies, and, at a mean of 4.8 years, only 1.7% of patients required revision. We continue to follow these patients to obtain long-term results of this innovative prosthesis.

Take-Home Points

• PFA improved knee function and pain scores in patients with isolated patellofemoral arthritis.
• The majority (84.2%) of patients undergoing PFA were female.
• Regardless of age or gender, 72.2% of patients returned to their desired preoperative activity after PFA, and 52.8% returned at the same or higher level.
• The rate of conversion from PFA to TKA was 6.3%.
• PFA is an alternative to TKA in active patients with isolated patellofemoral arthritis.

Compared with total knee arthroplasty (TKA), single-compartment knee arthroplasty may provide better physiologic function, faster recovery, and higher rates of return to activities in patients with unicompartmental knee disease.^1-3 In 1955, McKeever⁴ introduced patellar arthroplasty for surgical management of isolated patellofemoral arthritis. In 1979, Lubinus⁵ improved on the technique and design by adding a femoral component. Since then, implants and techniques have been developed to effect better clinical outcomes. Patellofemoral arthroplasty (PFA) has many advantages over TKA in the treatment of patellofemoral arthritis. PFA is less invasive, requires shorter tourniquet times, has faster recovery, and spares the tibiofemoral compartment, leaving more native bone for potential conversion to TKA. Regarding activity and function, the resurfacing arthroplasty (vs TKA) allows maintenance of nearly normal knee kinematics.

Despite these advantages, the broader orthopedic surgery community has only cautiously accepted PFA. The procedure has high complication rates. Persistent instability, malalignment, wear, impingement, and tibiofemoral arthritis progression can occur after PFA.⁶ Although first-generation PFA prostheses often failed because of mechanical problems, loosening, maltracking, or instability,⁷ the most common indication for PFA revision has been, according to a recent large retrospective study,⁸ unexplained pain. More than 10 to 15 years after PFA, tibiofemoral arthritis may be the primary mechanism of failure.⁹ Nevertheless, compared with standard TKA for isolated patellofemoral arthritis, modern PFA does not have significantly different clinical outcomes, including complication and revision rates.⁶Numerous patient factors influence functional prognosis before and after knee arthroplasty, regardless of surgical technique and implant used. Age, comorbidities, athletic status, mental health, pain, functional limitations, excessive caution, “artificial joint”–related worries, and rehabilitation protocol all influence function.¹⁰ Return to activity and other quality-of-life indices are important aspects of postoperative patient satisfaction.

Methods

We conducted a retrospective cohort study to describe functional status after PFA for patellofemoral arthritis. We identified 48 consecutive PFAs (39 patients) performed by a team of 2 orthopedic surgeons (specialists in treating patellofemoral pathology) between 2009 and 2014.

Three validated patient-reported outcome measures (PROMs) were used to determine preoperative (baseline) and postoperative functional status: Kujala score, Lysholm score, and International Knee Documentation Committee (IKDC) score. The Kujala score is a measure of knee function specific to the patellofemoral joint; the Lysholm score focuses on activities related to the knee; and the IKDC score is a general measure of knee function. Charts were reviewed to extract patients’ clinical data, including preoperative outcome scores, medical history, physical examination data, intraoperative characteristics, and postoperative course. By telephone, patients answered questions about their postoperative clinical course and completed final follow-up questionnaires. They were also asked which sporting or fitness activity they had preferred before surgery and whether they were able to return to that activity after surgery.

Statistical analysis included the study population’s descriptive statistics. Means and SDs were reported for continuous variables, and frequencies and percentages were reported for categorical variables. Paired t tests were used to analyze changes in PROM scores. For comparison of differences between characteristics of patients who did and did not return to their previous activity level, independent-samples t tests were used for continuous variables. Chi-square tests or Fisher exact tests were used to compare discrete variables. Statistical significance was set at P ≤ .05. All analyses were performed with SPSS Version 22.0 (IBM).

 

Results

Thirty-nine patients underwent PFA at our institution between 2009 and 2014. Mean age was 51.6 years. Of these patients, 84.2% were female, 28.6% had a body mass index of 30 kg/m² or higher, and 23.4% had PFA for posttraumatic arthritis related to prior patellofemoral instability. Table 1 lists the study cohort’s demographic data.

Table 2 lists self-reported activities limited by the affected knee before surgery, and Table 3 lists activity levels after surgery. Return to previous preferred activity was reported by 72.2% of patients, and 52.8% of patients reported returning to the same activity level or to a higher level. There were no differences in age (P = .978) or sex (P = .232) between patients who returned to the same or a higher activity level and patients who did not.

However, mean BMI was significantly (P = .016) higher in patients who returned to the same or a higher activity level (28.6 kg/m²) than in patients who did not (23.7 kg/m²). Although the rate of posttraumatic arthritis (26%) was higher than the rate of primary osteoarthritis (19%) in patients who returned to the same or a higher activity level, this difference was not statistically significant (P = .724).

Postoperative knee-specific PROM scores and general pain score (reported by the patient on a scale of 0-10) were statistically significantly improved (P < .001 for all measures) over preoperative scores (Table 4).

Mean follow-up was 26 months (range, 5-57 months). Kujala score improved a mean of 19.5 points; Lysholm score, 28.9 points; and IKDC score, 23.5 points. Mean general pain score improved from 6.3 before surgery to 2.8 after surgery. All PROM and pain score improvements were substantially larger than the minimal clinically important differences. Postoperative PROM scores and general pain score were significantly more improved in patients who returned to the same or a higher activity level than in patients who did not (P < .05 for all measures).

After surgery, 1 patient (2.6%) developed a pulmonary embolus, which was successfully identified and treated without incident. Five patients (10.4%) had another surgery on the same knee. Three patients (6.3%) underwent conversion to TKA: 1 for continued symptoms in the setting of newly diagnosed inflammatory arthritis, 1 for arthritic pain, and 1 for patellofemoral instability. Two patients (4.2%) underwent irrigation and débridement: 1 for hematoma and 1 for suspected (culture-negative) infection.

Discussion

Historically, the literature evaluating knee arthroplasty outcomes has focused on implant survivorship, pain relief, and patient satisfaction. Since the advent of partial knee arthroplasty options, more attention has been given to functional outcomes and return to activities after single-compartment knee resurfacing. TKA remains the gold standard by which newer, less invasive surgical options are measured. In a large prospective study, 97% of patients (age, >55 years) who had TKA for patellofemoral arthritis reported good or excellent clinical results, the majority being excellent.¹¹ Post-TKA functional status and activity levels may not be rated as highly. After TKA, many patients switch to lower impact sports or reduce or stop their participation in sports.¹² A small study of competitive adult tennis players found high levels of post-TKA satisfaction, ability to resume playing tennis, pain relief, and increased or continued enjoyment in playing.¹³ In a study of 355 patients (417 knees) who had underwent TKA, improvement in Knee Society function score showed a moderate correlation to an increase in weighted activity score (R = 0.362).¹⁴

Unicondylar knee arthroplasty (UKA) is becoming a popular treatment option for single-compartment tibiofemoral arthritis. A systematic review of 18 original studies of patients with knee osteoarthritis found that overall return to sports varied from 36% to 89% after TKA and from 75% to 100% after UKA.¹⁵ In another study, return-to-sports rates were similar for UKA (87%) and TKA (83%); the only significant difference was UKA patients returned quicker.¹⁶ The authors of a large meta-analysis conceded that significant heterogeneity of data prevented them from drawing definitive conclusions, but UKA patients seemed to return to low- and high-impact sports 2 weeks faster than their TKA counterparts.¹⁰ Overall, UKA and TKA patients (age, 51-71 years) had comparable return-to-sports rates at an average of 4 years after surgery.¹⁰ A smaller study corroborated faster return to sports for UKA over TKA patients and also found that, compared with TKA patients, UKA patients participated in sports more regularly and over a longer period.¹⁷ On the other hand, Walton and colleagues¹⁸ found similar return-to-sports rates but higher frequency of and satisfaction with sports participation in UKA over TKA patients.

A large retrospective study found no differences in rates of return to sports after TKA, UKA, patellar resurfacing, hip resurfacing, and total hip arthroplasty.¹⁹ Pain was the most common barrier to return. UKA patients who returned to sports tended to be younger than those who did not.²⁰ Naal and colleagues³ found that 95% of UKA patients returned to their activities—hiking, walking, cycling, and swimming being most common. Although 90.3% of patients said surgery maintained or improved their ability to participate in sports, participation in high-impact sports (eg, running) decreased after surgery.

Outcomes of PFA vary because of evolving patient selection, implant design, surgical technique, and return-to-activity expectations.^21,22 Most PFA outcome studies focus on implant survivorship, complication rates, and postoperative knee scores.^23-28 PFA studies focused on return to activities are limited. Kooijman and colleagues⁷ and Mertl and colleagues²⁹ reported good or excellent clinical results of PFA in 86% and 82% of patients, respectively. Neither study included a comprehensive analysis of postoperative functional status. Similarly, De Cloedt and colleagues³⁰ reported good PFA outcomes in 43% of patients with degenerative joint disease and in 83% of patients with instability. Specific activity status was not described. Dahm and colleagues³¹ and Farr and colleagues³² suggested postoperative pain resolution motivates some PFA patients not only to resume preoperative activities but to start participating in new, higher level activities after pain has subsided. However, the studies did not examine the characteristics of patients who returned to baseline activities and did not examine return-to-sports rates.

 

Study Strengths and Limitations

Our study focused on the PFA patient population of a surgical team of 2 fellowship-trained orthopedic surgeons (specialists in treating patellofemoral pathology). Although generalization of our findings to other surgeons and different implants may be limited, the study design standardized treatment in a way that makes these findings more reliable. The 100% follow-up strengthens these findings as well. Last, though the patient population was relatively small, it was consistent with or larger than the PFA patient groups studied previously.

Conclusion

In this study, PROM and pain scores were significantly improved after PFA. That almost 75% of patients returned to their preferred activities and >50% of patients returned at the same or a higher activity level provides useful information for preoperative discussions with patients who want to remain active after PFA. Prospective studies are needed to evaluate the longevity and durability of PFA, particularly in active patients.

Take-Home Points

• PFA improved knee function and pain scores in patients with isolated patellofemoral arthritis.
• The majority (84.2%) of patients undergoing PFA were female.
• Regardless of age or gender, 72.2% of patients returned to their desired preoperative activity after PFA, and 52.8% returned at the same or higher level.
• The rate of conversion from PFA to TKA was 6.3%.
• PFA is an alternative to TKA in active patients with isolated patellofemoral arthritis.

Compared with total knee arthroplasty (TKA), single-compartment knee arthroplasty may provide better physiologic function, faster recovery, and higher rates of return to activities in patients with unicompartmental knee disease.^1-3 In 1955, McKeever⁴ introduced patellar arthroplasty for surgical management of isolated patellofemoral arthritis. In 1979, Lubinus⁵ improved on the technique and design by adding a femoral component. Since then, implants and techniques have been developed to effect better clinical outcomes. Patellofemoral arthroplasty (PFA) has many advantages over TKA in the treatment of patellofemoral arthritis. PFA is less invasive, requires shorter tourniquet times, has faster recovery, and spares the tibiofemoral compartment, leaving more native bone for potential conversion to TKA. Regarding activity and function, the resurfacing arthroplasty (vs TKA) allows maintenance of nearly normal knee kinematics.

Despite these advantages, the broader orthopedic surgery community has only cautiously accepted PFA. The procedure has high complication rates. Persistent instability, malalignment, wear, impingement, and tibiofemoral arthritis progression can occur after PFA.⁶ Although first-generation PFA prostheses often failed because of mechanical problems, loosening, maltracking, or instability,⁷ the most common indication for PFA revision has been, according to a recent large retrospective study,⁸ unexplained pain. More than 10 to 15 years after PFA, tibiofemoral arthritis may be the primary mechanism of failure.⁹ Nevertheless, compared with standard TKA for isolated patellofemoral arthritis, modern PFA does not have significantly different clinical outcomes, including complication and revision rates.⁶Numerous patient factors influence functional prognosis before and after knee arthroplasty, regardless of surgical technique and implant used. Age, comorbidities, athletic status, mental health, pain, functional limitations, excessive caution, “artificial joint”–related worries, and rehabilitation protocol all influence function.¹⁰ Return to activity and other quality-of-life indices are important aspects of postoperative patient satisfaction.

Methods

We conducted a retrospective cohort study to describe functional status after PFA for patellofemoral arthritis. We identified 48 consecutive PFAs (39 patients) performed by a team of 2 orthopedic surgeons (specialists in treating patellofemoral pathology) between 2009 and 2014.

Three validated patient-reported outcome measures (PROMs) were used to determine preoperative (baseline) and postoperative functional status: Kujala score, Lysholm score, and International Knee Documentation Committee (IKDC) score. The Kujala score is a measure of knee function specific to the patellofemoral joint; the Lysholm score focuses on activities related to the knee; and the IKDC score is a general measure of knee function. Charts were reviewed to extract patients’ clinical data, including preoperative outcome scores, medical history, physical examination data, intraoperative characteristics, and postoperative course. By telephone, patients answered questions about their postoperative clinical course and completed final follow-up questionnaires. They were also asked which sporting or fitness activity they had preferred before surgery and whether they were able to return to that activity after surgery.

Statistical analysis included the study population’s descriptive statistics. Means and SDs were reported for continuous variables, and frequencies and percentages were reported for categorical variables. Paired t tests were used to analyze changes in PROM scores. For comparison of differences between characteristics of patients who did and did not return to their previous activity level, independent-samples t tests were used for continuous variables. Chi-square tests or Fisher exact tests were used to compare discrete variables. Statistical significance was set at P ≤ .05. All analyses were performed with SPSS Version 22.0 (IBM).

 

Results

Thirty-nine patients underwent PFA at our institution between 2009 and 2014. Mean age was 51.6 years. Of these patients, 84.2% were female, 28.6% had a body mass index of 30 kg/m² or higher, and 23.4% had PFA for posttraumatic arthritis related to prior patellofemoral instability. Table 1 lists the study cohort’s demographic data.

Table 2 lists self-reported activities limited by the affected knee before surgery, and Table 3 lists activity levels after surgery. Return to previous preferred activity was reported by 72.2% of patients, and 52.8% of patients reported returning to the same activity level or to a higher level. There were no differences in age (P = .978) or sex (P = .232) between patients who returned to the same or a higher activity level and patients who did not.

However, mean BMI was significantly (P = .016) higher in patients who returned to the same or a higher activity level (28.6 kg/m²) than in patients who did not (23.7 kg/m²). Although the rate of posttraumatic arthritis (26%) was higher than the rate of primary osteoarthritis (19%) in patients who returned to the same or a higher activity level, this difference was not statistically significant (P = .724).

Postoperative knee-specific PROM scores and general pain score (reported by the patient on a scale of 0-10) were statistically significantly improved (P < .001 for all measures) over preoperative scores (Table 4).

Mean follow-up was 26 months (range, 5-57 months). Kujala score improved a mean of 19.5 points; Lysholm score, 28.9 points; and IKDC score, 23.5 points. Mean general pain score improved from 6.3 before surgery to 2.8 after surgery. All PROM and pain score improvements were substantially larger than the minimal clinically important differences. Postoperative PROM scores and general pain score were significantly more improved in patients who returned to the same or a higher activity level than in patients who did not (P < .05 for all measures).

After surgery, 1 patient (2.6%) developed a pulmonary embolus, which was successfully identified and treated without incident. Five patients (10.4%) had another surgery on the same knee. Three patients (6.3%) underwent conversion to TKA: 1 for continued symptoms in the setting of newly diagnosed inflammatory arthritis, 1 for arthritic pain, and 1 for patellofemoral instability. Two patients (4.2%) underwent irrigation and débridement: 1 for hematoma and 1 for suspected (culture-negative) infection.

Discussion

Historically, the literature evaluating knee arthroplasty outcomes has focused on implant survivorship, pain relief, and patient satisfaction. Since the advent of partial knee arthroplasty options, more attention has been given to functional outcomes and return to activities after single-compartment knee resurfacing. TKA remains the gold standard by which newer, less invasive surgical options are measured. In a large prospective study, 97% of patients (age, >55 years) who had TKA for patellofemoral arthritis reported good or excellent clinical results, the majority being excellent.¹¹ Post-TKA functional status and activity levels may not be rated as highly. After TKA, many patients switch to lower impact sports or reduce or stop their participation in sports.¹² A small study of competitive adult tennis players found high levels of post-TKA satisfaction, ability to resume playing tennis, pain relief, and increased or continued enjoyment in playing.¹³ In a study of 355 patients (417 knees) who had underwent TKA, improvement in Knee Society function score showed a moderate correlation to an increase in weighted activity score (R = 0.362).¹⁴

Unicondylar knee arthroplasty (UKA) is becoming a popular treatment option for single-compartment tibiofemoral arthritis. A systematic review of 18 original studies of patients with knee osteoarthritis found that overall return to sports varied from 36% to 89% after TKA and from 75% to 100% after UKA.¹⁵ In another study, return-to-sports rates were similar for UKA (87%) and TKA (83%); the only significant difference was UKA patients returned quicker.¹⁶ The authors of a large meta-analysis conceded that significant heterogeneity of data prevented them from drawing definitive conclusions, but UKA patients seemed to return to low- and high-impact sports 2 weeks faster than their TKA counterparts.¹⁰ Overall, UKA and TKA patients (age, 51-71 years) had comparable return-to-sports rates at an average of 4 years after surgery.¹⁰ A smaller study corroborated faster return to sports for UKA over TKA patients and also found that, compared with TKA patients, UKA patients participated in sports more regularly and over a longer period.¹⁷ On the other hand, Walton and colleagues¹⁸ found similar return-to-sports rates but higher frequency of and satisfaction with sports participation in UKA over TKA patients.

A large retrospective study found no differences in rates of return to sports after TKA, UKA, patellar resurfacing, hip resurfacing, and total hip arthroplasty.¹⁹ Pain was the most common barrier to return. UKA patients who returned to sports tended to be younger than those who did not.²⁰ Naal and colleagues³ found that 95% of UKA patients returned to their activities—hiking, walking, cycling, and swimming being most common. Although 90.3% of patients said surgery maintained or improved their ability to participate in sports, participation in high-impact sports (eg, running) decreased after surgery.

Outcomes of PFA vary because of evolving patient selection, implant design, surgical technique, and return-to-activity expectations.^21,22 Most PFA outcome studies focus on implant survivorship, complication rates, and postoperative knee scores.^23-28 PFA studies focused on return to activities are limited. Kooijman and colleagues⁷ and Mertl and colleagues²⁹ reported good or excellent clinical results of PFA in 86% and 82% of patients, respectively. Neither study included a comprehensive analysis of postoperative functional status. Similarly, De Cloedt and colleagues³⁰ reported good PFA outcomes in 43% of patients with degenerative joint disease and in 83% of patients with instability. Specific activity status was not described. Dahm and colleagues³¹ and Farr and colleagues³² suggested postoperative pain resolution motivates some PFA patients not only to resume preoperative activities but to start participating in new, higher level activities after pain has subsided. However, the studies did not examine the characteristics of patients who returned to baseline activities and did not examine return-to-sports rates.

 

Study Strengths and Limitations

Our study focused on the PFA patient population of a surgical team of 2 fellowship-trained orthopedic surgeons (specialists in treating patellofemoral pathology). Although generalization of our findings to other surgeons and different implants may be limited, the study design standardized treatment in a way that makes these findings more reliable. The 100% follow-up strengthens these findings as well. Last, though the patient population was relatively small, it was consistent with or larger than the PFA patient groups studied previously.

Conclusion

In this study, PROM and pain scores were significantly improved after PFA. That almost 75% of patients returned to their preferred activities and >50% of patients returned at the same or a higher activity level provides useful information for preoperative discussions with patients who want to remain active after PFA. Prospective studies are needed to evaluate the longevity and durability of PFA, particularly in active patients.

Take-Home Points

• PFA improved knee function and pain scores in patients with isolated patellofemoral arthritis.
• The majority (84.2%) of patients undergoing PFA were female.
• Regardless of age or gender, 72.2% of patients returned to their desired preoperative activity after PFA, and 52.8% returned at the same or higher level.
• The rate of conversion from PFA to TKA was 6.3%.
• PFA is an alternative to TKA in active patients with isolated patellofemoral arthritis.

Compared with total knee arthroplasty (TKA), single-compartment knee arthroplasty may provide better physiologic function, faster recovery, and higher rates of return to activities in patients with unicompartmental knee disease.^1-3 In 1955, McKeever⁴ introduced patellar arthroplasty for surgical management of isolated patellofemoral arthritis. In 1979, Lubinus⁵ improved on the technique and design by adding a femoral component. Since then, implants and techniques have been developed to effect better clinical outcomes. Patellofemoral arthroplasty (PFA) has many advantages over TKA in the treatment of patellofemoral arthritis. PFA is less invasive, requires shorter tourniquet times, has faster recovery, and spares the tibiofemoral compartment, leaving more native bone for potential conversion to TKA. Regarding activity and function, the resurfacing arthroplasty (vs TKA) allows maintenance of nearly normal knee kinematics.

Despite these advantages, the broader orthopedic surgery community has only cautiously accepted PFA. The procedure has high complication rates. Persistent instability, malalignment, wear, impingement, and tibiofemoral arthritis progression can occur after PFA.⁶ Although first-generation PFA prostheses often failed because of mechanical problems, loosening, maltracking, or instability,⁷ the most common indication for PFA revision has been, according to a recent large retrospective study,⁸ unexplained pain. More than 10 to 15 years after PFA, tibiofemoral arthritis may be the primary mechanism of failure.⁹ Nevertheless, compared with standard TKA for isolated patellofemoral arthritis, modern PFA does not have significantly different clinical outcomes, including complication and revision rates.⁶Numerous patient factors influence functional prognosis before and after knee arthroplasty, regardless of surgical technique and implant used. Age, comorbidities, athletic status, mental health, pain, functional limitations, excessive caution, “artificial joint”–related worries, and rehabilitation protocol all influence function.¹⁰ Return to activity and other quality-of-life indices are important aspects of postoperative patient satisfaction.

Methods

We conducted a retrospective cohort study to describe functional status after PFA for patellofemoral arthritis. We identified 48 consecutive PFAs (39 patients) performed by a team of 2 orthopedic surgeons (specialists in treating patellofemoral pathology) between 2009 and 2014.

Three validated patient-reported outcome measures (PROMs) were used to determine preoperative (baseline) and postoperative functional status: Kujala score, Lysholm score, and International Knee Documentation Committee (IKDC) score. The Kujala score is a measure of knee function specific to the patellofemoral joint; the Lysholm score focuses on activities related to the knee; and the IKDC score is a general measure of knee function. Charts were reviewed to extract patients’ clinical data, including preoperative outcome scores, medical history, physical examination data, intraoperative characteristics, and postoperative course. By telephone, patients answered questions about their postoperative clinical course and completed final follow-up questionnaires. They were also asked which sporting or fitness activity they had preferred before surgery and whether they were able to return to that activity after surgery.

Statistical analysis included the study population’s descriptive statistics. Means and SDs were reported for continuous variables, and frequencies and percentages were reported for categorical variables. Paired t tests were used to analyze changes in PROM scores. For comparison of differences between characteristics of patients who did and did not return to their previous activity level, independent-samples t tests were used for continuous variables. Chi-square tests or Fisher exact tests were used to compare discrete variables. Statistical significance was set at P ≤ .05. All analyses were performed with SPSS Version 22.0 (IBM).

 

Results

Thirty-nine patients underwent PFA at our institution between 2009 and 2014. Mean age was 51.6 years. Of these patients, 84.2% were female, 28.6% had a body mass index of 30 kg/m² or higher, and 23.4% had PFA for posttraumatic arthritis related to prior patellofemoral instability. Table 1 lists the study cohort’s demographic data.

Table 2 lists self-reported activities limited by the affected knee before surgery, and Table 3 lists activity levels after surgery. Return to previous preferred activity was reported by 72.2% of patients, and 52.8% of patients reported returning to the same activity level or to a higher level. There were no differences in age (P = .978) or sex (P = .232) between patients who returned to the same or a higher activity level and patients who did not.

However, mean BMI was significantly (P = .016) higher in patients who returned to the same or a higher activity level (28.6 kg/m²) than in patients who did not (23.7 kg/m²). Although the rate of posttraumatic arthritis (26%) was higher than the rate of primary osteoarthritis (19%) in patients who returned to the same or a higher activity level, this difference was not statistically significant (P = .724).

Postoperative knee-specific PROM scores and general pain score (reported by the patient on a scale of 0-10) were statistically significantly improved (P < .001 for all measures) over preoperative scores (Table 4).

Mean follow-up was 26 months (range, 5-57 months). Kujala score improved a mean of 19.5 points; Lysholm score, 28.9 points; and IKDC score, 23.5 points. Mean general pain score improved from 6.3 before surgery to 2.8 after surgery. All PROM and pain score improvements were substantially larger than the minimal clinically important differences. Postoperative PROM scores and general pain score were significantly more improved in patients who returned to the same or a higher activity level than in patients who did not (P < .05 for all measures).

After surgery, 1 patient (2.6%) developed a pulmonary embolus, which was successfully identified and treated without incident. Five patients (10.4%) had another surgery on the same knee. Three patients (6.3%) underwent conversion to TKA: 1 for continued symptoms in the setting of newly diagnosed inflammatory arthritis, 1 for arthritic pain, and 1 for patellofemoral instability. Two patients (4.2%) underwent irrigation and débridement: 1 for hematoma and 1 for suspected (culture-negative) infection.

Discussion

Historically, the literature evaluating knee arthroplasty outcomes has focused on implant survivorship, pain relief, and patient satisfaction. Since the advent of partial knee arthroplasty options, more attention has been given to functional outcomes and return to activities after single-compartment knee resurfacing. TKA remains the gold standard by which newer, less invasive surgical options are measured. In a large prospective study, 97% of patients (age, >55 years) who had TKA for patellofemoral arthritis reported good or excellent clinical results, the majority being excellent.¹¹ Post-TKA functional status and activity levels may not be rated as highly. After TKA, many patients switch to lower impact sports or reduce or stop their participation in sports.¹² A small study of competitive adult tennis players found high levels of post-TKA satisfaction, ability to resume playing tennis, pain relief, and increased or continued enjoyment in playing.¹³ In a study of 355 patients (417 knees) who had underwent TKA, improvement in Knee Society function score showed a moderate correlation to an increase in weighted activity score (R = 0.362).¹⁴

Unicondylar knee arthroplasty (UKA) is becoming a popular treatment option for single-compartment tibiofemoral arthritis. A systematic review of 18 original studies of patients with knee osteoarthritis found that overall return to sports varied from 36% to 89% after TKA and from 75% to 100% after UKA.¹⁵ In another study, return-to-sports rates were similar for UKA (87%) and TKA (83%); the only significant difference was UKA patients returned quicker.¹⁶ The authors of a large meta-analysis conceded that significant heterogeneity of data prevented them from drawing definitive conclusions, but UKA patients seemed to return to low- and high-impact sports 2 weeks faster than their TKA counterparts.¹⁰ Overall, UKA and TKA patients (age, 51-71 years) had comparable return-to-sports rates at an average of 4 years after surgery.¹⁰ A smaller study corroborated faster return to sports for UKA over TKA patients and also found that, compared with TKA patients, UKA patients participated in sports more regularly and over a longer period.¹⁷ On the other hand, Walton and colleagues¹⁸ found similar return-to-sports rates but higher frequency of and satisfaction with sports participation in UKA over TKA patients.

A large retrospective study found no differences in rates of return to sports after TKA, UKA, patellar resurfacing, hip resurfacing, and total hip arthroplasty.¹⁹ Pain was the most common barrier to return. UKA patients who returned to sports tended to be younger than those who did not.²⁰ Naal and colleagues³ found that 95% of UKA patients returned to their activities—hiking, walking, cycling, and swimming being most common. Although 90.3% of patients said surgery maintained or improved their ability to participate in sports, participation in high-impact sports (eg, running) decreased after surgery.

Outcomes of PFA vary because of evolving patient selection, implant design, surgical technique, and return-to-activity expectations.^21,22 Most PFA outcome studies focus on implant survivorship, complication rates, and postoperative knee scores.^23-28 PFA studies focused on return to activities are limited. Kooijman and colleagues⁷ and Mertl and colleagues²⁹ reported good or excellent clinical results of PFA in 86% and 82% of patients, respectively. Neither study included a comprehensive analysis of postoperative functional status. Similarly, De Cloedt and colleagues³⁰ reported good PFA outcomes in 43% of patients with degenerative joint disease and in 83% of patients with instability. Specific activity status was not described. Dahm and colleagues³¹ and Farr and colleagues³² suggested postoperative pain resolution motivates some PFA patients not only to resume preoperative activities but to start participating in new, higher level activities after pain has subsided. However, the studies did not examine the characteristics of patients who returned to baseline activities and did not examine return-to-sports rates.

 

Study Strengths and Limitations

Our study focused on the PFA patient population of a surgical team of 2 fellowship-trained orthopedic surgeons (specialists in treating patellofemoral pathology). Although generalization of our findings to other surgeons and different implants may be limited, the study design standardized treatment in a way that makes these findings more reliable. The 100% follow-up strengthens these findings as well. Last, though the patient population was relatively small, it was consistent with or larger than the PFA patient groups studied previously.

Conclusion

In this study, PROM and pain scores were significantly improved after PFA. That almost 75% of patients returned to their preferred activities and >50% of patients returned at the same or a higher activity level provides useful information for preoperative discussions with patients who want to remain active after PFA. Prospective studies are needed to evaluate the longevity and durability of PFA, particularly in active patients.