Original Research

From the Departments of Psychology (Ms. Zhao, Drs. Russell, Wesley, and Porter) and Hematology (Mss. Johnson and Pullen, Dr. Hankins), St. Jude Children’s Research Hospital, Memphis, TN.

 

Abstract

• Background: Children with sickle cell disease (SCD) are surviving into adulthood. Mastery of disease knowledge may facilitate treatment continuity in adult care.
• Objective: To assess the accuracy and extent of medical history knowledge among adolescents with SCD through the use of a personal health record (PHR) form.
• Methods: 68 adolescent patients with SCD (52.9% male; mean age, 16.8 years; 100% African American) completed a PHR listing significant prior medical events (eg, disease complications, diagnostic evaluations, treatments). Responses were compared against participants’ electronic medical record. An agreement percentage was calculated to determine accuracy of knowledge.
• Results: Most adolescents correctly reported their sickle cell genotype (100%), usage of penicillin (97.1%), prior hospitalizations (96.5%), history of prior blood transfusions (93.8%), usage of hydroxyurea (88.2%), and allergies (85.2%). Fewer adolescents accurately reported usage of opioids (52.9%), prior acute chest syndrome events (50.9%), baseline hemoglobin (41.8%), and hepatitis (43.3%), pneumovax (30.2%), and menactra (14.5%) vaccinations.
• Conclusion: Adolescents are aware of most but not all aspects of their medical history. The present findings can inform areas of knowledge deficits. Future targeted interventions for transition education and preparation may be tailored based on individual disease knowledge.

 

Sickle cell disease (SCD) is a genetic disorder characterized by abnormal sickle hemoglobin resulting in chronic hemolytic anemia and vaso-occlusion [1]. More than 95% of children with SCD in the United States survive into adulthood; however, young adults (YAs) are at risk for mortality shortly after transfer to adult health care [2–5]. Specifically, YAs with SCD (ages 18 to 30) have increased hospital utilization, emergency department visits, and mortality compared to other age-groups [4–7]. During this critical period, transition preparation that includes improving disease literacy and ensuring medical history knowledge may be necessary for optimal outcomes.

In the extant YA literature, significant gaps in medical history knowledge during the transition period were observed in pediatric cancer and inflammatory bowel disease patients [8,9]. YAs often require multidisciplinary management of their chronic disease complications [10]. Therefore, possessing comprehensive knowledge of personal health history may facilitate communication with different adult care providers and promote continuity of care. In the SCD transition literature, transition readiness measures have been developed to assess several aspects of knowledge, including medical and disease knowledge; however, these measures are primarily self-reported perceptions of knowledge and do not evaluate the accuracy of knowledge [11,12]. The current pilot study addresses this gap with the aim of assessing medical history knowledge accuracy in adolescents with SCD.

Methods

Participants

From March 2011 to January 2014, adolescents (aged 15–18 years) with SCD (any genotype) were approached during their regular health maintenance visits by hematology social workers. They were invited to complete the Personal Health Record (PHR) as an implementation effort of transition preparation within our pediatric SCD program.

Personal Health Record

The PHR was developed through literature review and discussions with area adult hematologists. The form was modeled after first visit intake forms used in adult hematology clinics. It was reviewed by the hematology medical team and the institution’s patient education committee. Prior to implementation, the form was piloted to obtain patient feedback on format and content. The PHR consists of 33 questions with 168 possible items/data points covering 12 domains: personal information (eg, contact information, SCD genotype), health provider information, personal health history (ie, health diagnoses), blood transfusion history, sickle cell pain events, hospitalization history in the previous year, diagnostic testing history (eg, laboratory tests), current medications, immunizations, advance directives, resource information (eg, disability benefits), and activities of daily living. Some questions required patients to check “Yes” or “No” (eg, “Have you been hospitalized in the past year? Have you received flu vaccine?”) while some required a written response (eg, “What medicines do you currently take?”).

Adolescents were instructed to complete the PHR independently without the help of their caregivers. After completing the form, the social worker reviewed the answers and/or asked participants’ perspectives about communicating health information to providers. A copy of the completed PHR was provided to the adolescent to promote continued education regarding medical history knowledge. The retrospective review of the PHR answers and participants’ characteristics was approved by the institutional review board with a waiver of consent from participants.

Statistical Methods

PHR answers were compared with each individual’s electronic medical record (EMR) for accuracy of responses. PHR responses were considered accurate only if they matched the information in the EMR. PHR items absent in the EMR were not coded (inability to verify the accuracy of responses) to capture the most accurate depiction of adolescents’ medical history knowledge. Coding was checked by at least 2 coders for response accuracy. Due to lack of EMR information for certain items, we could not verify the accuracy of many PHR items. Therefore, only items with at least 75% of data verified (across all patients who completed the PHR) were included in subsequent analyses.

Using SPSS (version 18), an agreement percentage was calculated for each patient across verifiable items and used as the primary outcome measure of knowledge accuracy. We used t tests to investigate gender or genotype differences in medical history accuracy. To examine genotype differences, we stratified the sample by SCD genotype: HbSS/Sβ⁰ thalassemia and HbSC/Sβ+ thalassemia [13].

 

Results

Patient Characteristics

During the period of analysis, there were 95 eligible adolescents with SCD; all were approached, and 68 (71.6%) completed the PHR. Reasons for non-completion included recurrent missed visits, lack of time during the visit, or refusal. Of the 68 who completed the PHR, all were African American, 52.9% were male, and their mean age was 16.8 (± 0.9; range, 15–18) years (Table). Completion of the PHR took on average 15 minutes.

Knowledge Accuracy Among Adolescents with SCD

Seventeen items in 6 PHR domains had the highest number of data points (at least 75% verified), and therefore were the only items that could be analyzed. Analyzed items included information about sickle cell genotype, eye doctor care, comorbid health issues (eg, asthma), allergies, hospitalizations, surgeries, transfusions, acute chest syndrome (ACS) episodes, eye problems, baseline hemoglobin level, and vaccination history as well as adolescents’ knowledge of current medications, including hydroxyurea, penicillin, and opioid pain medications.

The accuracy of knowledge for select items is presented in the Figure. Adolescents were accurate reporters of SCD genotype (100%), hospitalizations in the previous year (96.5%), transfusion history (93.8%), and allergies (85.2%). Knowledge deficits included previous diagnosis of ACS (50.9%), baseline hemoglobin levels (41.8%), and hepatitis (43.3%), pneumovax (30.2%), and menactra (14.5%) vaccinations. Regarding current medications, adolescents were more accurate at reporting penicillin (97.1%) and hydroxyurea (88.2%) utilization, but less accurate regarding opioid pain medications (52.9%). No participants were able to report their health history with 100% accuracy.

Gender was not significantly associated with overall accuracy (P = 0.36). A significant difference was found in sickle genotype such that individuals with HbSC/Sβ+ thalassemia genotype (mean number of items, 8.23; SD = 1.70) were more accurate reporters of their medical history than those with HbSS/Sβ⁰ thalassemia genotype (mean number of items, 7.14; SD = 1.75; t(65) = –2.59, P = 0.01). Specifically, those with HbSS/Sβ⁰ thalassemia genotype were significantly less accurate reporters of vaccination history (meningococcus t(60) = 3.55, P = 0.001; pneumococcus t(60) = 2.46,  P = 0.02; hepatitis t(64) = 2.18, P = 0.03, eye problems t(62) = 3.62; P = 0.001, and surgical history t(62) = 2.14, P = 0.04).

 

Discussion

In the present study, we utilized the PHR to assess the accuracy of medical history knowledge of adolescents with SCD preparing to transition to adult care. Most adolescents were accurate reporters of important disease-relevant information (eg, genotype, transfusion history, hydroxyurea use), which may be a result of these topics being frequently discussed or recently encountered. For example, 97% of adolescents accurately reported penicillin use which may be related to our program’s emphasis on infection prevention education. However, disease knowledge of immunization history, prior ACS events, and opioid medication use might have been more difficult to recall due to the long interval from their occurrence until the completion of the PHR. Further, frequent changes in opioid medication use may have impacted the accuracy of adolescents’ answers with EMR data.

Individuals with HbSC/Sβ+ thalassemia genotype were more accurate reporters of their medical history, but the magnitude of difference was not large. These individuals tend to have fewer health issues and therefore less health information to recall, leading to higher accuracy. Furthermore, evidence demonstrates that individuals with HbSS/Sβ⁰ thalassemia genotype are at greater risk for cerebrovascular events and subsequent cognitive deficits [14], leading to more memory deficits and difficulty understanding and retaining health information [15]. The results suggest that patient health literacy, or an individual’s capacity to understand basic health information [16], may be a mediating factor in assessing for transition readiness. This is especially important given SCD risk for cognitive deficits [17].

Only 17 PHR items were analyzed due to conservative selection of items. Thus the present findings are not representative of the entire medical history. Additionally, the accuracy of medical history knowledge results may be limited by conservatism with abstracting information from the EMR (PHR information was considered accurate if it matched the information found in their EMR). Finally, we did not systematically assess the feasibility and utility of the PHR; ongoing participant feedback would aid in improving the PHR tool and implementation. It would be important to validate the PHR in a larger sample. However, our study is the first to our knowledge to systematically evaluate medical history knowledge among youth with SCD.

Conclusion and Practice Implications

The present study demonstrates that use of the PHR during regular health maintenance visits can help identify gaps in knowledge among adolescents with SCD who are approaching transfer to adult care. Sufficient knowledge of one’s medical history is an important aspect in transition preparation as it can facilitate the communication of medical information, thereby ensuring continuity of care [18,19]. The PHR could be used to teach medical history knowledge, assess a patient’s level of transition readiness at different time points, and identify areas for further targeted intervention. Knowledge tools, such as the PHR, can be investigated prospectively to assess the association of disease literacy and clinical outcomes, serving as a possible predictive instrument for transition health outcomes.

 

Corresponding author: Jerlym S. Porter, PhD, MPH, St. Jude Children’s Research Hospital, Dept. of Psychology, 262 Danny Thomas Pl., Mail Stop 740, Memphis, TN 38105, [email protected].

Funding/support: This work was supported in part by HRSA grant 6 U1EMC19331-03-02 (PI: Hankins).

Financial disclosures: None.

Author contributions: conception and design, MJ, AP, KMW, JSH, JSP; analysis and interpretation of data, MSZ, KMR, JSP; drafting of article, MSZ, JSP; critical revision of the article, MSZ, MJ, AP, KMW, JSH, JSP; provision of study materials or patients, MJ, AP; statistical expertise, KMR; obtaining of funding, JSH; collection and assembly of data, MSZ, MJ, AP, KMR, KMW.

From the Departments of Psychology (Ms. Zhao, Drs. Russell, Wesley, and Porter) and Hematology (Mss. Johnson and Pullen, Dr. Hankins), St. Jude Children’s Research Hospital, Memphis, TN.

 

Abstract

• Background: Children with sickle cell disease (SCD) are surviving into adulthood. Mastery of disease knowledge may facilitate treatment continuity in adult care.
• Objective: To assess the accuracy and extent of medical history knowledge among adolescents with SCD through the use of a personal health record (PHR) form.
• Methods: 68 adolescent patients with SCD (52.9% male; mean age, 16.8 years; 100% African American) completed a PHR listing significant prior medical events (eg, disease complications, diagnostic evaluations, treatments). Responses were compared against participants’ electronic medical record. An agreement percentage was calculated to determine accuracy of knowledge.
• Results: Most adolescents correctly reported their sickle cell genotype (100%), usage of penicillin (97.1%), prior hospitalizations (96.5%), history of prior blood transfusions (93.8%), usage of hydroxyurea (88.2%), and allergies (85.2%). Fewer adolescents accurately reported usage of opioids (52.9%), prior acute chest syndrome events (50.9%), baseline hemoglobin (41.8%), and hepatitis (43.3%), pneumovax (30.2%), and menactra (14.5%) vaccinations.
• Conclusion: Adolescents are aware of most but not all aspects of their medical history. The present findings can inform areas of knowledge deficits. Future targeted interventions for transition education and preparation may be tailored based on individual disease knowledge.

 

Sickle cell disease (SCD) is a genetic disorder characterized by abnormal sickle hemoglobin resulting in chronic hemolytic anemia and vaso-occlusion [1]. More than 95% of children with SCD in the United States survive into adulthood; however, young adults (YAs) are at risk for mortality shortly after transfer to adult health care [2–5]. Specifically, YAs with SCD (ages 18 to 30) have increased hospital utilization, emergency department visits, and mortality compared to other age-groups [4–7]. During this critical period, transition preparation that includes improving disease literacy and ensuring medical history knowledge may be necessary for optimal outcomes.

In the extant YA literature, significant gaps in medical history knowledge during the transition period were observed in pediatric cancer and inflammatory bowel disease patients [8,9]. YAs often require multidisciplinary management of their chronic disease complications [10]. Therefore, possessing comprehensive knowledge of personal health history may facilitate communication with different adult care providers and promote continuity of care. In the SCD transition literature, transition readiness measures have been developed to assess several aspects of knowledge, including medical and disease knowledge; however, these measures are primarily self-reported perceptions of knowledge and do not evaluate the accuracy of knowledge [11,12]. The current pilot study addresses this gap with the aim of assessing medical history knowledge accuracy in adolescents with SCD.

Methods

Participants

From March 2011 to January 2014, adolescents (aged 15–18 years) with SCD (any genotype) were approached during their regular health maintenance visits by hematology social workers. They were invited to complete the Personal Health Record (PHR) as an implementation effort of transition preparation within our pediatric SCD program.

Personal Health Record

The PHR was developed through literature review and discussions with area adult hematologists. The form was modeled after first visit intake forms used in adult hematology clinics. It was reviewed by the hematology medical team and the institution’s patient education committee. Prior to implementation, the form was piloted to obtain patient feedback on format and content. The PHR consists of 33 questions with 168 possible items/data points covering 12 domains: personal information (eg, contact information, SCD genotype), health provider information, personal health history (ie, health diagnoses), blood transfusion history, sickle cell pain events, hospitalization history in the previous year, diagnostic testing history (eg, laboratory tests), current medications, immunizations, advance directives, resource information (eg, disability benefits), and activities of daily living. Some questions required patients to check “Yes” or “No” (eg, “Have you been hospitalized in the past year? Have you received flu vaccine?”) while some required a written response (eg, “What medicines do you currently take?”).

Adolescents were instructed to complete the PHR independently without the help of their caregivers. After completing the form, the social worker reviewed the answers and/or asked participants’ perspectives about communicating health information to providers. A copy of the completed PHR was provided to the adolescent to promote continued education regarding medical history knowledge. The retrospective review of the PHR answers and participants’ characteristics was approved by the institutional review board with a waiver of consent from participants.

Statistical Methods

PHR answers were compared with each individual’s electronic medical record (EMR) for accuracy of responses. PHR responses were considered accurate only if they matched the information in the EMR. PHR items absent in the EMR were not coded (inability to verify the accuracy of responses) to capture the most accurate depiction of adolescents’ medical history knowledge. Coding was checked by at least 2 coders for response accuracy. Due to lack of EMR information for certain items, we could not verify the accuracy of many PHR items. Therefore, only items with at least 75% of data verified (across all patients who completed the PHR) were included in subsequent analyses.

Using SPSS (version 18), an agreement percentage was calculated for each patient across verifiable items and used as the primary outcome measure of knowledge accuracy. We used t tests to investigate gender or genotype differences in medical history accuracy. To examine genotype differences, we stratified the sample by SCD genotype: HbSS/Sβ⁰ thalassemia and HbSC/Sβ+ thalassemia [13].

 

Results

Patient Characteristics

During the period of analysis, there were 95 eligible adolescents with SCD; all were approached, and 68 (71.6%) completed the PHR. Reasons for non-completion included recurrent missed visits, lack of time during the visit, or refusal. Of the 68 who completed the PHR, all were African American, 52.9% were male, and their mean age was 16.8 (± 0.9; range, 15–18) years (Table). Completion of the PHR took on average 15 minutes.

Knowledge Accuracy Among Adolescents with SCD

Seventeen items in 6 PHR domains had the highest number of data points (at least 75% verified), and therefore were the only items that could be analyzed. Analyzed items included information about sickle cell genotype, eye doctor care, comorbid health issues (eg, asthma), allergies, hospitalizations, surgeries, transfusions, acute chest syndrome (ACS) episodes, eye problems, baseline hemoglobin level, and vaccination history as well as adolescents’ knowledge of current medications, including hydroxyurea, penicillin, and opioid pain medications.

The accuracy of knowledge for select items is presented in the Figure. Adolescents were accurate reporters of SCD genotype (100%), hospitalizations in the previous year (96.5%), transfusion history (93.8%), and allergies (85.2%). Knowledge deficits included previous diagnosis of ACS (50.9%), baseline hemoglobin levels (41.8%), and hepatitis (43.3%), pneumovax (30.2%), and menactra (14.5%) vaccinations. Regarding current medications, adolescents were more accurate at reporting penicillin (97.1%) and hydroxyurea (88.2%) utilization, but less accurate regarding opioid pain medications (52.9%). No participants were able to report their health history with 100% accuracy.

Gender was not significantly associated with overall accuracy (P = 0.36). A significant difference was found in sickle genotype such that individuals with HbSC/Sβ+ thalassemia genotype (mean number of items, 8.23; SD = 1.70) were more accurate reporters of their medical history than those with HbSS/Sβ⁰ thalassemia genotype (mean number of items, 7.14; SD = 1.75; t(65) = –2.59, P = 0.01). Specifically, those with HbSS/Sβ⁰ thalassemia genotype were significantly less accurate reporters of vaccination history (meningococcus t(60) = 3.55, P = 0.001; pneumococcus t(60) = 2.46,  P = 0.02; hepatitis t(64) = 2.18, P = 0.03, eye problems t(62) = 3.62; P = 0.001, and surgical history t(62) = 2.14, P = 0.04).

 

Discussion

In the present study, we utilized the PHR to assess the accuracy of medical history knowledge of adolescents with SCD preparing to transition to adult care. Most adolescents were accurate reporters of important disease-relevant information (eg, genotype, transfusion history, hydroxyurea use), which may be a result of these topics being frequently discussed or recently encountered. For example, 97% of adolescents accurately reported penicillin use which may be related to our program’s emphasis on infection prevention education. However, disease knowledge of immunization history, prior ACS events, and opioid medication use might have been more difficult to recall due to the long interval from their occurrence until the completion of the PHR. Further, frequent changes in opioid medication use may have impacted the accuracy of adolescents’ answers with EMR data.

Individuals with HbSC/Sβ+ thalassemia genotype were more accurate reporters of their medical history, but the magnitude of difference was not large. These individuals tend to have fewer health issues and therefore less health information to recall, leading to higher accuracy. Furthermore, evidence demonstrates that individuals with HbSS/Sβ⁰ thalassemia genotype are at greater risk for cerebrovascular events and subsequent cognitive deficits [14], leading to more memory deficits and difficulty understanding and retaining health information [15]. The results suggest that patient health literacy, or an individual’s capacity to understand basic health information [16], may be a mediating factor in assessing for transition readiness. This is especially important given SCD risk for cognitive deficits [17].

Only 17 PHR items were analyzed due to conservative selection of items. Thus the present findings are not representative of the entire medical history. Additionally, the accuracy of medical history knowledge results may be limited by conservatism with abstracting information from the EMR (PHR information was considered accurate if it matched the information found in their EMR). Finally, we did not systematically assess the feasibility and utility of the PHR; ongoing participant feedback would aid in improving the PHR tool and implementation. It would be important to validate the PHR in a larger sample. However, our study is the first to our knowledge to systematically evaluate medical history knowledge among youth with SCD.

Conclusion and Practice Implications

The present study demonstrates that use of the PHR during regular health maintenance visits can help identify gaps in knowledge among adolescents with SCD who are approaching transfer to adult care. Sufficient knowledge of one’s medical history is an important aspect in transition preparation as it can facilitate the communication of medical information, thereby ensuring continuity of care [18,19]. The PHR could be used to teach medical history knowledge, assess a patient’s level of transition readiness at different time points, and identify areas for further targeted intervention. Knowledge tools, such as the PHR, can be investigated prospectively to assess the association of disease literacy and clinical outcomes, serving as a possible predictive instrument for transition health outcomes.

 

Corresponding author: Jerlym S. Porter, PhD, MPH, St. Jude Children’s Research Hospital, Dept. of Psychology, 262 Danny Thomas Pl., Mail Stop 740, Memphis, TN 38105, [email protected].

Funding/support: This work was supported in part by HRSA grant 6 U1EMC19331-03-02 (PI: Hankins).

Financial disclosures: None.

Author contributions: conception and design, MJ, AP, KMW, JSH, JSP; analysis and interpretation of data, MSZ, KMR, JSP; drafting of article, MSZ, JSP; critical revision of the article, MSZ, MJ, AP, KMW, JSH, JSP; provision of study materials or patients, MJ, AP; statistical expertise, KMR; obtaining of funding, JSH; collection and assembly of data, MSZ, MJ, AP, KMR, KMW.

From the Departments of Psychology (Ms. Zhao, Drs. Russell, Wesley, and Porter) and Hematology (Mss. Johnson and Pullen, Dr. Hankins), St. Jude Children’s Research Hospital, Memphis, TN.

 

Abstract

• Background: Children with sickle cell disease (SCD) are surviving into adulthood. Mastery of disease knowledge may facilitate treatment continuity in adult care.
• Objective: To assess the accuracy and extent of medical history knowledge among adolescents with SCD through the use of a personal health record (PHR) form.
• Methods: 68 adolescent patients with SCD (52.9% male; mean age, 16.8 years; 100% African American) completed a PHR listing significant prior medical events (eg, disease complications, diagnostic evaluations, treatments). Responses were compared against participants’ electronic medical record. An agreement percentage was calculated to determine accuracy of knowledge.
• Results: Most adolescents correctly reported their sickle cell genotype (100%), usage of penicillin (97.1%), prior hospitalizations (96.5%), history of prior blood transfusions (93.8%), usage of hydroxyurea (88.2%), and allergies (85.2%). Fewer adolescents accurately reported usage of opioids (52.9%), prior acute chest syndrome events (50.9%), baseline hemoglobin (41.8%), and hepatitis (43.3%), pneumovax (30.2%), and menactra (14.5%) vaccinations.
• Conclusion: Adolescents are aware of most but not all aspects of their medical history. The present findings can inform areas of knowledge deficits. Future targeted interventions for transition education and preparation may be tailored based on individual disease knowledge.

 

Sickle cell disease (SCD) is a genetic disorder characterized by abnormal sickle hemoglobin resulting in chronic hemolytic anemia and vaso-occlusion [1]. More than 95% of children with SCD in the United States survive into adulthood; however, young adults (YAs) are at risk for mortality shortly after transfer to adult health care [2–5]. Specifically, YAs with SCD (ages 18 to 30) have increased hospital utilization, emergency department visits, and mortality compared to other age-groups [4–7]. During this critical period, transition preparation that includes improving disease literacy and ensuring medical history knowledge may be necessary for optimal outcomes.

In the extant YA literature, significant gaps in medical history knowledge during the transition period were observed in pediatric cancer and inflammatory bowel disease patients [8,9]. YAs often require multidisciplinary management of their chronic disease complications [10]. Therefore, possessing comprehensive knowledge of personal health history may facilitate communication with different adult care providers and promote continuity of care. In the SCD transition literature, transition readiness measures have been developed to assess several aspects of knowledge, including medical and disease knowledge; however, these measures are primarily self-reported perceptions of knowledge and do not evaluate the accuracy of knowledge [11,12]. The current pilot study addresses this gap with the aim of assessing medical history knowledge accuracy in adolescents with SCD.

Methods

Participants

From March 2011 to January 2014, adolescents (aged 15–18 years) with SCD (any genotype) were approached during their regular health maintenance visits by hematology social workers. They were invited to complete the Personal Health Record (PHR) as an implementation effort of transition preparation within our pediatric SCD program.

Personal Health Record

The PHR was developed through literature review and discussions with area adult hematologists. The form was modeled after first visit intake forms used in adult hematology clinics. It was reviewed by the hematology medical team and the institution’s patient education committee. Prior to implementation, the form was piloted to obtain patient feedback on format and content. The PHR consists of 33 questions with 168 possible items/data points covering 12 domains: personal information (eg, contact information, SCD genotype), health provider information, personal health history (ie, health diagnoses), blood transfusion history, sickle cell pain events, hospitalization history in the previous year, diagnostic testing history (eg, laboratory tests), current medications, immunizations, advance directives, resource information (eg, disability benefits), and activities of daily living. Some questions required patients to check “Yes” or “No” (eg, “Have you been hospitalized in the past year? Have you received flu vaccine?”) while some required a written response (eg, “What medicines do you currently take?”).

Adolescents were instructed to complete the PHR independently without the help of their caregivers. After completing the form, the social worker reviewed the answers and/or asked participants’ perspectives about communicating health information to providers. A copy of the completed PHR was provided to the adolescent to promote continued education regarding medical history knowledge. The retrospective review of the PHR answers and participants’ characteristics was approved by the institutional review board with a waiver of consent from participants.

Statistical Methods

PHR answers were compared with each individual’s electronic medical record (EMR) for accuracy of responses. PHR responses were considered accurate only if they matched the information in the EMR. PHR items absent in the EMR were not coded (inability to verify the accuracy of responses) to capture the most accurate depiction of adolescents’ medical history knowledge. Coding was checked by at least 2 coders for response accuracy. Due to lack of EMR information for certain items, we could not verify the accuracy of many PHR items. Therefore, only items with at least 75% of data verified (across all patients who completed the PHR) were included in subsequent analyses.

Using SPSS (version 18), an agreement percentage was calculated for each patient across verifiable items and used as the primary outcome measure of knowledge accuracy. We used t tests to investigate gender or genotype differences in medical history accuracy. To examine genotype differences, we stratified the sample by SCD genotype: HbSS/Sβ⁰ thalassemia and HbSC/Sβ+ thalassemia [13].

 

Results

Patient Characteristics

During the period of analysis, there were 95 eligible adolescents with SCD; all were approached, and 68 (71.6%) completed the PHR. Reasons for non-completion included recurrent missed visits, lack of time during the visit, or refusal. Of the 68 who completed the PHR, all were African American, 52.9% were male, and their mean age was 16.8 (± 0.9; range, 15–18) years (Table). Completion of the PHR took on average 15 minutes.

Knowledge Accuracy Among Adolescents with SCD

Seventeen items in 6 PHR domains had the highest number of data points (at least 75% verified), and therefore were the only items that could be analyzed. Analyzed items included information about sickle cell genotype, eye doctor care, comorbid health issues (eg, asthma), allergies, hospitalizations, surgeries, transfusions, acute chest syndrome (ACS) episodes, eye problems, baseline hemoglobin level, and vaccination history as well as adolescents’ knowledge of current medications, including hydroxyurea, penicillin, and opioid pain medications.

The accuracy of knowledge for select items is presented in the Figure. Adolescents were accurate reporters of SCD genotype (100%), hospitalizations in the previous year (96.5%), transfusion history (93.8%), and allergies (85.2%). Knowledge deficits included previous diagnosis of ACS (50.9%), baseline hemoglobin levels (41.8%), and hepatitis (43.3%), pneumovax (30.2%), and menactra (14.5%) vaccinations. Regarding current medications, adolescents were more accurate at reporting penicillin (97.1%) and hydroxyurea (88.2%) utilization, but less accurate regarding opioid pain medications (52.9%). No participants were able to report their health history with 100% accuracy.

Gender was not significantly associated with overall accuracy (P = 0.36). A significant difference was found in sickle genotype such that individuals with HbSC/Sβ+ thalassemia genotype (mean number of items, 8.23; SD = 1.70) were more accurate reporters of their medical history than those with HbSS/Sβ⁰ thalassemia genotype (mean number of items, 7.14; SD = 1.75; t(65) = –2.59, P = 0.01). Specifically, those with HbSS/Sβ⁰ thalassemia genotype were significantly less accurate reporters of vaccination history (meningococcus t(60) = 3.55, P = 0.001; pneumococcus t(60) = 2.46,  P = 0.02; hepatitis t(64) = 2.18, P = 0.03, eye problems t(62) = 3.62; P = 0.001, and surgical history t(62) = 2.14, P = 0.04).

 

Discussion

In the present study, we utilized the PHR to assess the accuracy of medical history knowledge of adolescents with SCD preparing to transition to adult care. Most adolescents were accurate reporters of important disease-relevant information (eg, genotype, transfusion history, hydroxyurea use), which may be a result of these topics being frequently discussed or recently encountered. For example, 97% of adolescents accurately reported penicillin use which may be related to our program’s emphasis on infection prevention education. However, disease knowledge of immunization history, prior ACS events, and opioid medication use might have been more difficult to recall due to the long interval from their occurrence until the completion of the PHR. Further, frequent changes in opioid medication use may have impacted the accuracy of adolescents’ answers with EMR data.

Individuals with HbSC/Sβ+ thalassemia genotype were more accurate reporters of their medical history, but the magnitude of difference was not large. These individuals tend to have fewer health issues and therefore less health information to recall, leading to higher accuracy. Furthermore, evidence demonstrates that individuals with HbSS/Sβ⁰ thalassemia genotype are at greater risk for cerebrovascular events and subsequent cognitive deficits [14], leading to more memory deficits and difficulty understanding and retaining health information [15]. The results suggest that patient health literacy, or an individual’s capacity to understand basic health information [16], may be a mediating factor in assessing for transition readiness. This is especially important given SCD risk for cognitive deficits [17].

Only 17 PHR items were analyzed due to conservative selection of items. Thus the present findings are not representative of the entire medical history. Additionally, the accuracy of medical history knowledge results may be limited by conservatism with abstracting information from the EMR (PHR information was considered accurate if it matched the information found in their EMR). Finally, we did not systematically assess the feasibility and utility of the PHR; ongoing participant feedback would aid in improving the PHR tool and implementation. It would be important to validate the PHR in a larger sample. However, our study is the first to our knowledge to systematically evaluate medical history knowledge among youth with SCD.

Conclusion and Practice Implications

The present study demonstrates that use of the PHR during regular health maintenance visits can help identify gaps in knowledge among adolescents with SCD who are approaching transfer to adult care. Sufficient knowledge of one’s medical history is an important aspect in transition preparation as it can facilitate the communication of medical information, thereby ensuring continuity of care [18,19]. The PHR could be used to teach medical history knowledge, assess a patient’s level of transition readiness at different time points, and identify areas for further targeted intervention. Knowledge tools, such as the PHR, can be investigated prospectively to assess the association of disease literacy and clinical outcomes, serving as a possible predictive instrument for transition health outcomes.

 

Corresponding author: Jerlym S. Porter, PhD, MPH, St. Jude Children’s Research Hospital, Dept. of Psychology, 262 Danny Thomas Pl., Mail Stop 740, Memphis, TN 38105, [email protected].

Funding/support: This work was supported in part by HRSA grant 6 U1EMC19331-03-02 (PI: Hankins).

Financial disclosures: None.

Author contributions: conception and design, MJ, AP, KMW, JSH, JSP; analysis and interpretation of data, MSZ, KMR, JSP; drafting of article, MSZ, JSP; critical revision of the article, MSZ, MJ, AP, KMW, JSH, JSP; provision of study materials or patients, MJ, AP; statistical expertise, KMR; obtaining of funding, JSH; collection and assembly of data, MSZ, MJ, AP, KMR, KMW.

Shoulder arthroplasty (SA), including total SA (TSA) and reverse TSA, is an effective surgical treatment for fracture and primary or secondary degenerative disease of the shoulder.¹ Over the past few decades, use of SA has increased dramatically, from about 5000 cases in 1990 to 7000 in 2000 and more than 26,000 in 2008.^1,2

Complications associated with SA generally are classified as perioperative (occurring during the operative index) or long-term (postdischarge).³ Long-term complications include implant loosening, instability, revision, infection, rotator cuff tear, neural injury, and deltoid detachment.^1,4,5 Perioperative complications, which are less commonly reported, include intraoperative fracture, infection, neural injury, venous thromboembolic events (VTEs, including pulmonary embolism [PE] and deep vein thrombosis [DVT]), transfusion, and death.^3,6-10

SA is an attractive treatment option for patients with rheumatoid arthritis (RA), as the effects of pain on these patients are greater in the shoulder joint than in any other joint.¹¹ Patients with RA pose unique orthopedic surgical challenges, including any combination of decreased bone mineralization, poor capsular tissue integrity, and osteonecrosis.^3,12 In addition, RA patients may be taking immunosuppressive medications that have severe side effects, and they may require multiple surgeries.^12,13 These factors predispose patients with RA to complications that include infection and wound dehiscence.^3,5,12-14

The complex nature of RA has prompted investigators to examine outcome measures in this patient group. Hambright and colleagues³ used the Nationwide Inpatient Sample (NIS) to examine perioperative outcomes in RA patients who underwent TSA between 1988 and 2005.³ They found that TSA patients with RA had shorter and less costly hospital stays and were more likely to have a routine discharge.³ Using the same patient population drawn from the period 2006–2011, we conducted a study to determine if this unexpected trend persists as the number of TSAs and quality of postoperative care continue to increase. Given the potential for anemia of chronic disease and the systemic inflammatory nature of RA, we hypothesized that the perioperative complication profile of RA patients would be worse than that of non-RA patients.

Materials and Methods

NIS data were acquired for the period 2006–2011. The NIS is the largest publicly available all-payer inpatient database, with a random 20% sample of about 1000 US hospitals accounting for 7 to 8 million inpatient stays. The database supplies weights used to estimate national totals, at about 35 million inpatient visits per year. NIS inpatient data are limited to the operative index. Postdischarge information is not available. The NIS is managed by the Healthcare Cost and Utilization Project, which is sponsored by the Agency for Healthcare Research and Quality. The quality of NIS data is assessed and validated by an independent contractor. NIS data have been widely used to examine perioperative outcomes.^15-17

NIS data cover patient and hospital demographics, hospital length of stay (LOS), discharge status, payer information, charges, and perioperative outcomes and procedure/diagnosis codes (ICD-9; International Classification of Diseases, Ninth Revision¹⁸).

As our Institutional Review Board (IRB) reviewed the database and determined the project was not human subject research, IRB involvement was not required. This study paralleled successful efforts with similar RA and non-RA patients who had shoulder and elbow surgery.^3,19 SA patients were identified by ICD-9 procedure code 81.80, but this code does not specify whether the prosthesis was unconstrained, semiconstrained, or constrained. ICD-9 coding also does not specify whether the TSA was traditional or reverse. Patients with RA were identified by ICD-9 diagnosis codes 714.0, 714.1, and 714.2. Patients without one of these codes were placed in the non-RA cohort. Patients with codes associated with pathologic fractures secondary to metastatic cancer or bone malignant neoplasm as a secondary or primary diagnosis and patients who had revision surgery indicated by code 81.83 were excluded, as they have a disproportionately higher comorbidity burden.

After each cohort was defined, demographic data (age, sex, race, income quartile based on ZIP postal code) were compared, as were data on primary payer, hospital demographics, LOS (≤5 days, defined as perioperative index), discharge type, inflation-adjusted charges in 2014 dollars based on the Consumer Price Indexes (http://www.bls.gov/cpi/), and mortality. Perioperative complications—respiratory, gastrointestinal, genitourinary, accidental puncture/laceration, central nervous system, wound dehiscence, device-related (including embolism, fibrosis, hemorrhage, pain, stenosis, or thrombus caused by any device, implant, or graft), cardiac, hematoma/seroma, acute respiratory distress syndrome, postoperative shock, VTE, postoperative infection complications, and intraoperative transfusions—were considered using ICD-9 codes (996.X-999.X and 99.X, respectively).²⁰ Although commonly used to determine perioperative comorbidity burden using ICD-9 coding, the modified Charlson index was not considered because RA is a component of the index and would therefore bias the variable.^3,21

Statistical analyses, including χ² tests and 2-sample t tests, were performed for categorical and continuous variables, respectively. P < .05 was considered significant. Fisher exact test was used for cohorts with fewer than 5 occurrences. Multivariate logistic regression models were then calculated to determine the effect of RA on different outcomes and complications, with age, race, sex, hospital region, hospital type, number of hospital beds, primary payer, and hospital ownership as covariates. Statistical analyses were performed using the R statistical programming language.22

 

Results

Of the 34,970 patients who underwent SA between 2006 and 2011, 1674 (4.8%) had a diagnosis of RA and 33,296 (95.2%) did not. On average, patients with RA tended to be younger than patients without RA (66.4 vs 69.1 years; P < .001), and a larger percentage of RA patients were female (75.5% vs 54.4%; P < .001). Compared with non-RA patients, RA patients comprised a different ethnic group and had a different expected primary payer (P < .001). SA patients with and without RA did not differ in income quartile based on ZIP code, total number of hospital beds, hospital region, or hospital teaching status (P = .34, .78, .59, and .82, respectively) (Table 1).

LOS was significantly (P < .001) statistically longer for RA patients (2.196 days) than for non-RA patients (2.085 days). RA patients were significantly less likely to be discharged home (63.0% vs 67.6%; P < .001). (Routine discharge was defined as discharge home, whereas nonroutine discharge was defined as discharge to a short-term hospital, skilled nursing facility, intermediate care, another type of facility, home health care, against medical advice, or death.) In addition, inflation-adjusted charges associated with SA were significantly higher (P = .018) for RA patients ($54,284) than for non-RA patients ($52,663) (Table 1).

Regarding the rates of complications that occurred during the perioperative index, there were no significant differences between RA and non-RA cohorts. These complications included respiratory, gastrointestinal, genitourinary, accidental puncture/laceration, central nervous system, wound dehiscence, device-related, cardiac, hematoma/seroma, acute respiratory distress syndrome, postoperative shock, VTE, and postoperative infection (Table 2). In addition, there was no significant difference in mortality between the groups (P = .48).

In TSA, blood transfusions were more likely (P < .001) to be given to RA patients (9.00%) than to non-RA patients (6.16%). Multivariate regression analyses were performed with age, race, sex, hospital region, hospital type, number of hospital beds, primary payer, and hospital ownership as covariates. These analyses revealed that transfusion (P < .001), discharge type (P = .002), total inflation-adjusted charges (P < .001), and LOS (P = .047) remained significant (Table 3).

Discussion

Large national databases like NIS allow study of uncommon medical occurrences and help delineate risks and trends that otherwise might be indeterminable. Although it has been suggested that patients with RA may have poorer long-term outcomes after SA, the perioperative risk profile indicates that TSA is well tolerated in RA patients during the operative index.^3,23-25

The data on this study’s 34,970 patients, drawn from the period 2006–2011, demonstrated no significant differences in safety profile with respect to the 14 perioperative complications and outcomes examined, except blood transfusion rate. Rates of postoperative infection (RA, 0.24%; non-RA, 0.14%; P = .303), VTE (RA, 0.30%; non-RA, 0.25%; P = .905), and transfusion (RA, 9.00%; non-RA, 6.16%; P < .001) are of particular interest because of the severity of these situations.

Postoperative infection is a potentially serious complication and often occurs secondary to diabetes, RA, lupus erythematosus, prior surgery, or a nosocomial or remote source.¹ The often costly treatment options include antibiotic suppression, irrigation and debridement with implant retention, 1-stage exchange with antibiotic-impregnated cement fixation, staged reimplantation, resection arthroplasty, arthrodesis, and amputation.¹ The overall 0.14% infection rate determined in this study is lower than the 0.7% reported for SA patients in the literature.¹ Given the nature of the NIS database, this rate underestimates the true postoperative infection rate, as any infection that occurred after the perioperative period is not captured.²⁶ The present study’s perioperative infection rates (RA, 0.24%; non-RA, 0.14%) for the period 2006–2011 are comparable to the rates (RA, 0.17%; non-RA, 0.24%) reported by Hambright and colleagues3 for the same patient population over the preceding, 18-year period (1988–2005) and similarly do not significantly differ between groups. Although infection is uncommon in the immediate perioperative period, the ICD-9 codes used refer specifically to infection resulting from surgery and do not represent concomitant infection.

VTEs, which include PEs and DVTs, are rare but potentially life-threatening surgical complications.^27,28 Mechanical prophylaxis and chemical prophylaxis have been recommended for major orthopedic surgery, particularly lower extremity surgery, such as total hip arthroplasty (THA) and total knee arthroplasty (TKA).^28,29 In the present study, VTE rates were low, 0.30% (RA) and 0.25% (non-RA), and not significantly different in bivariate or multivariate analyses. These rates are comparable to those found in other national-database SA studies.²⁸ VTEs that occur outside the index hospital admission are not captured in this database. Therefore, the rates in the present study may be lower than the true incidence after SA. Mortality secondary to VTE usually occurs within 24 hours but may occur up to 90 days after surgery. DVT rates, on the other hand, are difficult to evaluate because of differences in screening practices.^27,28,30,31

 

That RA patients were more likely than non-RA patients to receive perioperative blood transfusions supports prior findings that SA patients with RA were more likely than SA patients with osteoarthritis (OA) to receive perioperative blood transfusions.⁸ RA patients have been shown to have high rates of anemia of chronic disease, ranging from 22% to 77%.³² During joint replacement, these patients often require transfusions.^32,33 However, these findings differ from prior findings of no differences between RA and non-RA patients in the same patient population during the period 1988–2005.³ This difference may be a product of the constantly changing transfusion guidelines and increased use; transfusion rates increased 140% between 1997 and 2007, making transfusions the fastest growing common procedure in the United States during that time.³⁴ There was no difference between RA and non-RA patients in household income (as determined by ZIP code analysis), number of hospital beds, hospital region, or hospital teaching status. Compared with non-RA patients, RA patients were more likely to be younger, female, and of a difference race and to have a different expected primary payer (P < .001).These findings are consistent with previous findings in the literature.³ In the present SA study, however, RA patients were more likely than non-RA patients to have longer LOS, higher inflation-adjusted hospital charges, and nonroutine discharge. These findings deviate from those of the study covering the preceding 18 years (1988–2005).³ Despite the findings of a changing environment of care for RA patients, by Hambright and colleagues³ and Weiss and colleagues,³⁵ the trend appears to have shifted. Both groups had shorter average LOS than either group from the preceding 18 years.³ Although statistically significant in bivariate analysis, the difference in LOS between the 2 groups differed by an average of 0.11 day (2 hours 24 minutes) and was not clinically relevant.

In addition, the higher charges for patients with RA represent a deviation from the preceding 18 years.³ Other studies have also shown that RA is associated with increased cost in TSA.³⁶ Patients with RA often have rotator cuff pathology, indicating reverse SA may be used more frequently.^37,38 The increased implant cost associated with reverse SA may account for the increased costs in RA patients.³⁹ As mentioned, TSA type is not captured in the NIS database. In addition, that RA patients were less likely than non-RA patients to have routine discharge may indicate RA cases are more complex because of their complications.^1,5,14,40 A recent study of complications in RA patients (1163 who underwent THA, 2692 who underwent TKA) found that THA patients with RA were significantly more likely than THA patients with OA to dislocate, and TKA patients with RA were significantly more likely than TKA patients with OA to develop an infection after surgery.⁴¹ Postoperative dislocation has been shown to increase hospital costs in other orthopedic procedures.⁴² Also, during TSA, patients with RA are more likely than patients with OA to receive intraoperative blood transfusions.⁸ These complications—combined with the fact that RA is a chronic, progressive, systemic inflammatory disease that can affect soft tissue and blood vessel wall healing and is associated with medications having potential side effects—could contribute to the apparent increased hospital charges and LOS.^3,12,13,43 Factors that include surgeon preference, impact of primary payer, and hospital practice may also affect final charges. Total charges in the NIS database include administrative fees, hospital costs, device-related costs, operating room costs, and ancillary staff costs. Total charges do not include professional fees and differ from the total cost that represents the amount reimbursed by the payer. Charges tend to correlate with but overestimate the total costs.⁴⁴

This study had several important limitations. As mentioned, only events that occur during the operative admission are captured in the NIS database, and thus postoperative complications or serious adverse events that lead to readmission cannot be identified. In addition, outpatient TSAs are not captured in the NIS database, and thus inclusion of only inpatient procedures yields higher average LOS and total charges.⁴⁵ Given the limited granularity of ICD-9 coding, this study could not determine RA severity, estimated blood loss, length of surgery, complication severity, type of TSA procedure/prosthesis, or cause of death. Although commonly used to determine comorbidity burden, the modified Charlson index could not be used, and therefore could not be entered as a covariate in multivariate analysis. Furthermore, the NIS database does not include imaging or patient-reported outcomes information, such as improvements in pain or function, which are of crucial importance in considering surgery.

Conclusion

Our findings corroborated findings that the demographics and the perioperative safety profile for TSA were similar for patients with and without RA. The risk for complications or death in the perioperative period was low. Compared with non-RA patients, RA patients had significantly higher charges and longer LOS and were less likely to be discharged home after surgery. The 0.11-day difference in LOS, though statistically significant, was not clinically relevant. These findings differ from those for the preceding, 18-year period (1988–2005). Future research should focus on the causes of these changes.

Shoulder arthroplasty (SA), including total SA (TSA) and reverse TSA, is an effective surgical treatment for fracture and primary or secondary degenerative disease of the shoulder.¹ Over the past few decades, use of SA has increased dramatically, from about 5000 cases in 1990 to 7000 in 2000 and more than 26,000 in 2008.^1,2

Complications associated with SA generally are classified as perioperative (occurring during the operative index) or long-term (postdischarge).³ Long-term complications include implant loosening, instability, revision, infection, rotator cuff tear, neural injury, and deltoid detachment.^1,4,5 Perioperative complications, which are less commonly reported, include intraoperative fracture, infection, neural injury, venous thromboembolic events (VTEs, including pulmonary embolism [PE] and deep vein thrombosis [DVT]), transfusion, and death.^3,6-10

SA is an attractive treatment option for patients with rheumatoid arthritis (RA), as the effects of pain on these patients are greater in the shoulder joint than in any other joint.¹¹ Patients with RA pose unique orthopedic surgical challenges, including any combination of decreased bone mineralization, poor capsular tissue integrity, and osteonecrosis.^3,12 In addition, RA patients may be taking immunosuppressive medications that have severe side effects, and they may require multiple surgeries.^12,13 These factors predispose patients with RA to complications that include infection and wound dehiscence.^3,5,12-14

The complex nature of RA has prompted investigators to examine outcome measures in this patient group. Hambright and colleagues³ used the Nationwide Inpatient Sample (NIS) to examine perioperative outcomes in RA patients who underwent TSA between 1988 and 2005.³ They found that TSA patients with RA had shorter and less costly hospital stays and were more likely to have a routine discharge.³ Using the same patient population drawn from the period 2006–2011, we conducted a study to determine if this unexpected trend persists as the number of TSAs and quality of postoperative care continue to increase. Given the potential for anemia of chronic disease and the systemic inflammatory nature of RA, we hypothesized that the perioperative complication profile of RA patients would be worse than that of non-RA patients.

Materials and Methods

NIS data were acquired for the period 2006–2011. The NIS is the largest publicly available all-payer inpatient database, with a random 20% sample of about 1000 US hospitals accounting for 7 to 8 million inpatient stays. The database supplies weights used to estimate national totals, at about 35 million inpatient visits per year. NIS inpatient data are limited to the operative index. Postdischarge information is not available. The NIS is managed by the Healthcare Cost and Utilization Project, which is sponsored by the Agency for Healthcare Research and Quality. The quality of NIS data is assessed and validated by an independent contractor. NIS data have been widely used to examine perioperative outcomes.^15-17

NIS data cover patient and hospital demographics, hospital length of stay (LOS), discharge status, payer information, charges, and perioperative outcomes and procedure/diagnosis codes (ICD-9; International Classification of Diseases, Ninth Revision¹⁸).

As our Institutional Review Board (IRB) reviewed the database and determined the project was not human subject research, IRB involvement was not required. This study paralleled successful efforts with similar RA and non-RA patients who had shoulder and elbow surgery.^3,19 SA patients were identified by ICD-9 procedure code 81.80, but this code does not specify whether the prosthesis was unconstrained, semiconstrained, or constrained. ICD-9 coding also does not specify whether the TSA was traditional or reverse. Patients with RA were identified by ICD-9 diagnosis codes 714.0, 714.1, and 714.2. Patients without one of these codes were placed in the non-RA cohort. Patients with codes associated with pathologic fractures secondary to metastatic cancer or bone malignant neoplasm as a secondary or primary diagnosis and patients who had revision surgery indicated by code 81.83 were excluded, as they have a disproportionately higher comorbidity burden.

After each cohort was defined, demographic data (age, sex, race, income quartile based on ZIP postal code) were compared, as were data on primary payer, hospital demographics, LOS (≤5 days, defined as perioperative index), discharge type, inflation-adjusted charges in 2014 dollars based on the Consumer Price Indexes (http://www.bls.gov/cpi/), and mortality. Perioperative complications—respiratory, gastrointestinal, genitourinary, accidental puncture/laceration, central nervous system, wound dehiscence, device-related (including embolism, fibrosis, hemorrhage, pain, stenosis, or thrombus caused by any device, implant, or graft), cardiac, hematoma/seroma, acute respiratory distress syndrome, postoperative shock, VTE, postoperative infection complications, and intraoperative transfusions—were considered using ICD-9 codes (996.X-999.X and 99.X, respectively).²⁰ Although commonly used to determine perioperative comorbidity burden using ICD-9 coding, the modified Charlson index was not considered because RA is a component of the index and would therefore bias the variable.^3,21

Statistical analyses, including χ² tests and 2-sample t tests, were performed for categorical and continuous variables, respectively. P < .05 was considered significant. Fisher exact test was used for cohorts with fewer than 5 occurrences. Multivariate logistic regression models were then calculated to determine the effect of RA on different outcomes and complications, with age, race, sex, hospital region, hospital type, number of hospital beds, primary payer, and hospital ownership as covariates. Statistical analyses were performed using the R statistical programming language.22

 

Results

Of the 34,970 patients who underwent SA between 2006 and 2011, 1674 (4.8%) had a diagnosis of RA and 33,296 (95.2%) did not. On average, patients with RA tended to be younger than patients without RA (66.4 vs 69.1 years; P < .001), and a larger percentage of RA patients were female (75.5% vs 54.4%; P < .001). Compared with non-RA patients, RA patients comprised a different ethnic group and had a different expected primary payer (P < .001). SA patients with and without RA did not differ in income quartile based on ZIP code, total number of hospital beds, hospital region, or hospital teaching status (P = .34, .78, .59, and .82, respectively) (Table 1).

LOS was significantly (P < .001) statistically longer for RA patients (2.196 days) than for non-RA patients (2.085 days). RA patients were significantly less likely to be discharged home (63.0% vs 67.6%; P < .001). (Routine discharge was defined as discharge home, whereas nonroutine discharge was defined as discharge to a short-term hospital, skilled nursing facility, intermediate care, another type of facility, home health care, against medical advice, or death.) In addition, inflation-adjusted charges associated with SA were significantly higher (P = .018) for RA patients ($54,284) than for non-RA patients ($52,663) (Table 1).

Regarding the rates of complications that occurred during the perioperative index, there were no significant differences between RA and non-RA cohorts. These complications included respiratory, gastrointestinal, genitourinary, accidental puncture/laceration, central nervous system, wound dehiscence, device-related, cardiac, hematoma/seroma, acute respiratory distress syndrome, postoperative shock, VTE, and postoperative infection (Table 2). In addition, there was no significant difference in mortality between the groups (P = .48).

In TSA, blood transfusions were more likely (P < .001) to be given to RA patients (9.00%) than to non-RA patients (6.16%). Multivariate regression analyses were performed with age, race, sex, hospital region, hospital type, number of hospital beds, primary payer, and hospital ownership as covariates. These analyses revealed that transfusion (P < .001), discharge type (P = .002), total inflation-adjusted charges (P < .001), and LOS (P = .047) remained significant (Table 3).

Discussion

Large national databases like NIS allow study of uncommon medical occurrences and help delineate risks and trends that otherwise might be indeterminable. Although it has been suggested that patients with RA may have poorer long-term outcomes after SA, the perioperative risk profile indicates that TSA is well tolerated in RA patients during the operative index.^3,23-25

The data on this study’s 34,970 patients, drawn from the period 2006–2011, demonstrated no significant differences in safety profile with respect to the 14 perioperative complications and outcomes examined, except blood transfusion rate. Rates of postoperative infection (RA, 0.24%; non-RA, 0.14%; P = .303), VTE (RA, 0.30%; non-RA, 0.25%; P = .905), and transfusion (RA, 9.00%; non-RA, 6.16%; P < .001) are of particular interest because of the severity of these situations.

Postoperative infection is a potentially serious complication and often occurs secondary to diabetes, RA, lupus erythematosus, prior surgery, or a nosocomial or remote source.¹ The often costly treatment options include antibiotic suppression, irrigation and debridement with implant retention, 1-stage exchange with antibiotic-impregnated cement fixation, staged reimplantation, resection arthroplasty, arthrodesis, and amputation.¹ The overall 0.14% infection rate determined in this study is lower than the 0.7% reported for SA patients in the literature.¹ Given the nature of the NIS database, this rate underestimates the true postoperative infection rate, as any infection that occurred after the perioperative period is not captured.²⁶ The present study’s perioperative infection rates (RA, 0.24%; non-RA, 0.14%) for the period 2006–2011 are comparable to the rates (RA, 0.17%; non-RA, 0.24%) reported by Hambright and colleagues3 for the same patient population over the preceding, 18-year period (1988–2005) and similarly do not significantly differ between groups. Although infection is uncommon in the immediate perioperative period, the ICD-9 codes used refer specifically to infection resulting from surgery and do not represent concomitant infection.

VTEs, which include PEs and DVTs, are rare but potentially life-threatening surgical complications.^27,28 Mechanical prophylaxis and chemical prophylaxis have been recommended for major orthopedic surgery, particularly lower extremity surgery, such as total hip arthroplasty (THA) and total knee arthroplasty (TKA).^28,29 In the present study, VTE rates were low, 0.30% (RA) and 0.25% (non-RA), and not significantly different in bivariate or multivariate analyses. These rates are comparable to those found in other national-database SA studies.²⁸ VTEs that occur outside the index hospital admission are not captured in this database. Therefore, the rates in the present study may be lower than the true incidence after SA. Mortality secondary to VTE usually occurs within 24 hours but may occur up to 90 days after surgery. DVT rates, on the other hand, are difficult to evaluate because of differences in screening practices.^27,28,30,31

 

That RA patients were more likely than non-RA patients to receive perioperative blood transfusions supports prior findings that SA patients with RA were more likely than SA patients with osteoarthritis (OA) to receive perioperative blood transfusions.⁸ RA patients have been shown to have high rates of anemia of chronic disease, ranging from 22% to 77%.³² During joint replacement, these patients often require transfusions.^32,33 However, these findings differ from prior findings of no differences between RA and non-RA patients in the same patient population during the period 1988–2005.³ This difference may be a product of the constantly changing transfusion guidelines and increased use; transfusion rates increased 140% between 1997 and 2007, making transfusions the fastest growing common procedure in the United States during that time.³⁴ There was no difference between RA and non-RA patients in household income (as determined by ZIP code analysis), number of hospital beds, hospital region, or hospital teaching status. Compared with non-RA patients, RA patients were more likely to be younger, female, and of a difference race and to have a different expected primary payer (P < .001).These findings are consistent with previous findings in the literature.³ In the present SA study, however, RA patients were more likely than non-RA patients to have longer LOS, higher inflation-adjusted hospital charges, and nonroutine discharge. These findings deviate from those of the study covering the preceding 18 years (1988–2005).³ Despite the findings of a changing environment of care for RA patients, by Hambright and colleagues³ and Weiss and colleagues,³⁵ the trend appears to have shifted. Both groups had shorter average LOS than either group from the preceding 18 years.³ Although statistically significant in bivariate analysis, the difference in LOS between the 2 groups differed by an average of 0.11 day (2 hours 24 minutes) and was not clinically relevant.

In addition, the higher charges for patients with RA represent a deviation from the preceding 18 years.³ Other studies have also shown that RA is associated with increased cost in TSA.³⁶ Patients with RA often have rotator cuff pathology, indicating reverse SA may be used more frequently.^37,38 The increased implant cost associated with reverse SA may account for the increased costs in RA patients.³⁹ As mentioned, TSA type is not captured in the NIS database. In addition, that RA patients were less likely than non-RA patients to have routine discharge may indicate RA cases are more complex because of their complications.^1,5,14,40 A recent study of complications in RA patients (1163 who underwent THA, 2692 who underwent TKA) found that THA patients with RA were significantly more likely than THA patients with OA to dislocate, and TKA patients with RA were significantly more likely than TKA patients with OA to develop an infection after surgery.⁴¹ Postoperative dislocation has been shown to increase hospital costs in other orthopedic procedures.⁴² Also, during TSA, patients with RA are more likely than patients with OA to receive intraoperative blood transfusions.⁸ These complications—combined with the fact that RA is a chronic, progressive, systemic inflammatory disease that can affect soft tissue and blood vessel wall healing and is associated with medications having potential side effects—could contribute to the apparent increased hospital charges and LOS.^3,12,13,43 Factors that include surgeon preference, impact of primary payer, and hospital practice may also affect final charges. Total charges in the NIS database include administrative fees, hospital costs, device-related costs, operating room costs, and ancillary staff costs. Total charges do not include professional fees and differ from the total cost that represents the amount reimbursed by the payer. Charges tend to correlate with but overestimate the total costs.⁴⁴

This study had several important limitations. As mentioned, only events that occur during the operative admission are captured in the NIS database, and thus postoperative complications or serious adverse events that lead to readmission cannot be identified. In addition, outpatient TSAs are not captured in the NIS database, and thus inclusion of only inpatient procedures yields higher average LOS and total charges.⁴⁵ Given the limited granularity of ICD-9 coding, this study could not determine RA severity, estimated blood loss, length of surgery, complication severity, type of TSA procedure/prosthesis, or cause of death. Although commonly used to determine comorbidity burden, the modified Charlson index could not be used, and therefore could not be entered as a covariate in multivariate analysis. Furthermore, the NIS database does not include imaging or patient-reported outcomes information, such as improvements in pain or function, which are of crucial importance in considering surgery.

Conclusion

Our findings corroborated findings that the demographics and the perioperative safety profile for TSA were similar for patients with and without RA. The risk for complications or death in the perioperative period was low. Compared with non-RA patients, RA patients had significantly higher charges and longer LOS and were less likely to be discharged home after surgery. The 0.11-day difference in LOS, though statistically significant, was not clinically relevant. These findings differ from those for the preceding, 18-year period (1988–2005). Future research should focus on the causes of these changes.

Shoulder arthroplasty (SA), including total SA (TSA) and reverse TSA, is an effective surgical treatment for fracture and primary or secondary degenerative disease of the shoulder.¹ Over the past few decades, use of SA has increased dramatically, from about 5000 cases in 1990 to 7000 in 2000 and more than 26,000 in 2008.^1,2

Complications associated with SA generally are classified as perioperative (occurring during the operative index) or long-term (postdischarge).³ Long-term complications include implant loosening, instability, revision, infection, rotator cuff tear, neural injury, and deltoid detachment.^1,4,5 Perioperative complications, which are less commonly reported, include intraoperative fracture, infection, neural injury, venous thromboembolic events (VTEs, including pulmonary embolism [PE] and deep vein thrombosis [DVT]), transfusion, and death.^3,6-10

SA is an attractive treatment option for patients with rheumatoid arthritis (RA), as the effects of pain on these patients are greater in the shoulder joint than in any other joint.¹¹ Patients with RA pose unique orthopedic surgical challenges, including any combination of decreased bone mineralization, poor capsular tissue integrity, and osteonecrosis.^3,12 In addition, RA patients may be taking immunosuppressive medications that have severe side effects, and they may require multiple surgeries.^12,13 These factors predispose patients with RA to complications that include infection and wound dehiscence.^3,5,12-14

The complex nature of RA has prompted investigators to examine outcome measures in this patient group. Hambright and colleagues³ used the Nationwide Inpatient Sample (NIS) to examine perioperative outcomes in RA patients who underwent TSA between 1988 and 2005.³ They found that TSA patients with RA had shorter and less costly hospital stays and were more likely to have a routine discharge.³ Using the same patient population drawn from the period 2006–2011, we conducted a study to determine if this unexpected trend persists as the number of TSAs and quality of postoperative care continue to increase. Given the potential for anemia of chronic disease and the systemic inflammatory nature of RA, we hypothesized that the perioperative complication profile of RA patients would be worse than that of non-RA patients.

Materials and Methods

NIS data were acquired for the period 2006–2011. The NIS is the largest publicly available all-payer inpatient database, with a random 20% sample of about 1000 US hospitals accounting for 7 to 8 million inpatient stays. The database supplies weights used to estimate national totals, at about 35 million inpatient visits per year. NIS inpatient data are limited to the operative index. Postdischarge information is not available. The NIS is managed by the Healthcare Cost and Utilization Project, which is sponsored by the Agency for Healthcare Research and Quality. The quality of NIS data is assessed and validated by an independent contractor. NIS data have been widely used to examine perioperative outcomes.^15-17

NIS data cover patient and hospital demographics, hospital length of stay (LOS), discharge status, payer information, charges, and perioperative outcomes and procedure/diagnosis codes (ICD-9; International Classification of Diseases, Ninth Revision¹⁸).

As our Institutional Review Board (IRB) reviewed the database and determined the project was not human subject research, IRB involvement was not required. This study paralleled successful efforts with similar RA and non-RA patients who had shoulder and elbow surgery.^3,19 SA patients were identified by ICD-9 procedure code 81.80, but this code does not specify whether the prosthesis was unconstrained, semiconstrained, or constrained. ICD-9 coding also does not specify whether the TSA was traditional or reverse. Patients with RA were identified by ICD-9 diagnosis codes 714.0, 714.1, and 714.2. Patients without one of these codes were placed in the non-RA cohort. Patients with codes associated with pathologic fractures secondary to metastatic cancer or bone malignant neoplasm as a secondary or primary diagnosis and patients who had revision surgery indicated by code 81.83 were excluded, as they have a disproportionately higher comorbidity burden.

After each cohort was defined, demographic data (age, sex, race, income quartile based on ZIP postal code) were compared, as were data on primary payer, hospital demographics, LOS (≤5 days, defined as perioperative index), discharge type, inflation-adjusted charges in 2014 dollars based on the Consumer Price Indexes (http://www.bls.gov/cpi/), and mortality. Perioperative complications—respiratory, gastrointestinal, genitourinary, accidental puncture/laceration, central nervous system, wound dehiscence, device-related (including embolism, fibrosis, hemorrhage, pain, stenosis, or thrombus caused by any device, implant, or graft), cardiac, hematoma/seroma, acute respiratory distress syndrome, postoperative shock, VTE, postoperative infection complications, and intraoperative transfusions—were considered using ICD-9 codes (996.X-999.X and 99.X, respectively).²⁰ Although commonly used to determine perioperative comorbidity burden using ICD-9 coding, the modified Charlson index was not considered because RA is a component of the index and would therefore bias the variable.^3,21

Statistical analyses, including χ² tests and 2-sample t tests, were performed for categorical and continuous variables, respectively. P < .05 was considered significant. Fisher exact test was used for cohorts with fewer than 5 occurrences. Multivariate logistic regression models were then calculated to determine the effect of RA on different outcomes and complications, with age, race, sex, hospital region, hospital type, number of hospital beds, primary payer, and hospital ownership as covariates. Statistical analyses were performed using the R statistical programming language.22

 

Results

Of the 34,970 patients who underwent SA between 2006 and 2011, 1674 (4.8%) had a diagnosis of RA and 33,296 (95.2%) did not. On average, patients with RA tended to be younger than patients without RA (66.4 vs 69.1 years; P < .001), and a larger percentage of RA patients were female (75.5% vs 54.4%; P < .001). Compared with non-RA patients, RA patients comprised a different ethnic group and had a different expected primary payer (P < .001). SA patients with and without RA did not differ in income quartile based on ZIP code, total number of hospital beds, hospital region, or hospital teaching status (P = .34, .78, .59, and .82, respectively) (Table 1).

LOS was significantly (P < .001) statistically longer for RA patients (2.196 days) than for non-RA patients (2.085 days). RA patients were significantly less likely to be discharged home (63.0% vs 67.6%; P < .001). (Routine discharge was defined as discharge home, whereas nonroutine discharge was defined as discharge to a short-term hospital, skilled nursing facility, intermediate care, another type of facility, home health care, against medical advice, or death.) In addition, inflation-adjusted charges associated with SA were significantly higher (P = .018) for RA patients ($54,284) than for non-RA patients ($52,663) (Table 1).

Regarding the rates of complications that occurred during the perioperative index, there were no significant differences between RA and non-RA cohorts. These complications included respiratory, gastrointestinal, genitourinary, accidental puncture/laceration, central nervous system, wound dehiscence, device-related, cardiac, hematoma/seroma, acute respiratory distress syndrome, postoperative shock, VTE, and postoperative infection (Table 2). In addition, there was no significant difference in mortality between the groups (P = .48).

In TSA, blood transfusions were more likely (P < .001) to be given to RA patients (9.00%) than to non-RA patients (6.16%). Multivariate regression analyses were performed with age, race, sex, hospital region, hospital type, number of hospital beds, primary payer, and hospital ownership as covariates. These analyses revealed that transfusion (P < .001), discharge type (P = .002), total inflation-adjusted charges (P < .001), and LOS (P = .047) remained significant (Table 3).

Discussion

Large national databases like NIS allow study of uncommon medical occurrences and help delineate risks and trends that otherwise might be indeterminable. Although it has been suggested that patients with RA may have poorer long-term outcomes after SA, the perioperative risk profile indicates that TSA is well tolerated in RA patients during the operative index.^3,23-25

The data on this study’s 34,970 patients, drawn from the period 2006–2011, demonstrated no significant differences in safety profile with respect to the 14 perioperative complications and outcomes examined, except blood transfusion rate. Rates of postoperative infection (RA, 0.24%; non-RA, 0.14%; P = .303), VTE (RA, 0.30%; non-RA, 0.25%; P = .905), and transfusion (RA, 9.00%; non-RA, 6.16%; P < .001) are of particular interest because of the severity of these situations.

Postoperative infection is a potentially serious complication and often occurs secondary to diabetes, RA, lupus erythematosus, prior surgery, or a nosocomial or remote source.¹ The often costly treatment options include antibiotic suppression, irrigation and debridement with implant retention, 1-stage exchange with antibiotic-impregnated cement fixation, staged reimplantation, resection arthroplasty, arthrodesis, and amputation.¹ The overall 0.14% infection rate determined in this study is lower than the 0.7% reported for SA patients in the literature.¹ Given the nature of the NIS database, this rate underestimates the true postoperative infection rate, as any infection that occurred after the perioperative period is not captured.²⁶ The present study’s perioperative infection rates (RA, 0.24%; non-RA, 0.14%) for the period 2006–2011 are comparable to the rates (RA, 0.17%; non-RA, 0.24%) reported by Hambright and colleagues3 for the same patient population over the preceding, 18-year period (1988–2005) and similarly do not significantly differ between groups. Although infection is uncommon in the immediate perioperative period, the ICD-9 codes used refer specifically to infection resulting from surgery and do not represent concomitant infection.

VTEs, which include PEs and DVTs, are rare but potentially life-threatening surgical complications.^27,28 Mechanical prophylaxis and chemical prophylaxis have been recommended for major orthopedic surgery, particularly lower extremity surgery, such as total hip arthroplasty (THA) and total knee arthroplasty (TKA).^28,29 In the present study, VTE rates were low, 0.30% (RA) and 0.25% (non-RA), and not significantly different in bivariate or multivariate analyses. These rates are comparable to those found in other national-database SA studies.²⁸ VTEs that occur outside the index hospital admission are not captured in this database. Therefore, the rates in the present study may be lower than the true incidence after SA. Mortality secondary to VTE usually occurs within 24 hours but may occur up to 90 days after surgery. DVT rates, on the other hand, are difficult to evaluate because of differences in screening practices.^27,28,30,31

 

That RA patients were more likely than non-RA patients to receive perioperative blood transfusions supports prior findings that SA patients with RA were more likely than SA patients with osteoarthritis (OA) to receive perioperative blood transfusions.⁸ RA patients have been shown to have high rates of anemia of chronic disease, ranging from 22% to 77%.³² During joint replacement, these patients often require transfusions.^32,33 However, these findings differ from prior findings of no differences between RA and non-RA patients in the same patient population during the period 1988–2005.³ This difference may be a product of the constantly changing transfusion guidelines and increased use; transfusion rates increased 140% between 1997 and 2007, making transfusions the fastest growing common procedure in the United States during that time.³⁴ There was no difference between RA and non-RA patients in household income (as determined by ZIP code analysis), number of hospital beds, hospital region, or hospital teaching status. Compared with non-RA patients, RA patients were more likely to be younger, female, and of a difference race and to have a different expected primary payer (P < .001).These findings are consistent with previous findings in the literature.³ In the present SA study, however, RA patients were more likely than non-RA patients to have longer LOS, higher inflation-adjusted hospital charges, and nonroutine discharge. These findings deviate from those of the study covering the preceding 18 years (1988–2005).³ Despite the findings of a changing environment of care for RA patients, by Hambright and colleagues³ and Weiss and colleagues,³⁵ the trend appears to have shifted. Both groups had shorter average LOS than either group from the preceding 18 years.³ Although statistically significant in bivariate analysis, the difference in LOS between the 2 groups differed by an average of 0.11 day (2 hours 24 minutes) and was not clinically relevant.

In addition, the higher charges for patients with RA represent a deviation from the preceding 18 years.³ Other studies have also shown that RA is associated with increased cost in TSA.³⁶ Patients with RA often have rotator cuff pathology, indicating reverse SA may be used more frequently.^37,38 The increased implant cost associated with reverse SA may account for the increased costs in RA patients.³⁹ As mentioned, TSA type is not captured in the NIS database. In addition, that RA patients were less likely than non-RA patients to have routine discharge may indicate RA cases are more complex because of their complications.^1,5,14,40 A recent study of complications in RA patients (1163 who underwent THA, 2692 who underwent TKA) found that THA patients with RA were significantly more likely than THA patients with OA to dislocate, and TKA patients with RA were significantly more likely than TKA patients with OA to develop an infection after surgery.⁴¹ Postoperative dislocation has been shown to increase hospital costs in other orthopedic procedures.⁴² Also, during TSA, patients with RA are more likely than patients with OA to receive intraoperative blood transfusions.⁸ These complications—combined with the fact that RA is a chronic, progressive, systemic inflammatory disease that can affect soft tissue and blood vessel wall healing and is associated with medications having potential side effects—could contribute to the apparent increased hospital charges and LOS.^3,12,13,43 Factors that include surgeon preference, impact of primary payer, and hospital practice may also affect final charges. Total charges in the NIS database include administrative fees, hospital costs, device-related costs, operating room costs, and ancillary staff costs. Total charges do not include professional fees and differ from the total cost that represents the amount reimbursed by the payer. Charges tend to correlate with but overestimate the total costs.⁴⁴

This study had several important limitations. As mentioned, only events that occur during the operative admission are captured in the NIS database, and thus postoperative complications or serious adverse events that lead to readmission cannot be identified. In addition, outpatient TSAs are not captured in the NIS database, and thus inclusion of only inpatient procedures yields higher average LOS and total charges.⁴⁵ Given the limited granularity of ICD-9 coding, this study could not determine RA severity, estimated blood loss, length of surgery, complication severity, type of TSA procedure/prosthesis, or cause of death. Although commonly used to determine comorbidity burden, the modified Charlson index could not be used, and therefore could not be entered as a covariate in multivariate analysis. Furthermore, the NIS database does not include imaging or patient-reported outcomes information, such as improvements in pain or function, which are of crucial importance in considering surgery.

Conclusion

Our findings corroborated findings that the demographics and the perioperative safety profile for TSA were similar for patients with and without RA. The risk for complications or death in the perioperative period was low. Compared with non-RA patients, RA patients had significantly higher charges and longer LOS and were less likely to be discharged home after surgery. The 0.11-day difference in LOS, though statistically significant, was not clinically relevant. These findings differ from those for the preceding, 18-year period (1988–2005). Future research should focus on the causes of these changes.

The patient experience has become increasingly important to healthcare in the United States. It is now a metric used commonly to determine physician compensation and accounts for nearly 30% of the Centers for Medicare and Medicaid Services' (CMS) Value‐Based Purchasing (VBP) reimbursement for fiscal years 2015 and 2016.[1, 2]

In April 2015, CMS added a 5‐star patient experience score to its Hospital Compare website in an attempt to address the Affordable Care Act's call for transparent and easily understandable public reporting.[3] A hospital's principal score is the Summary Star Rating, which is based on responses to the Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) survey. The formulas used to calculate Summary Star Ratings have been reported by CMS.[4]

Studies published over the past decade suggest that gender, age, education level, length of hospital stay, travel distance, and other factors may influence patient satisfaction.[5, 6, 7, 8] One study utilizing a national dataset suggested that higher patient satisfaction was associated with greater inpatient healthcare utilization and higher healthcare expenditures.[9] It is therefore possible that emphasizing patient experience scores could adversely impact healthcare resource utilization. However, positive patient experience may also be an important independent dimension of quality for patients and correlate with improved clinical outcomes.[10]

We know of no literature describing patient factors associated with the Summary Star Rating. Given that this rating is now used as a standard metric by which patient experience can be compared across more than 3,500 hospitals,[11] data describing the association between patient‐level factors and the Summary Star Rating may provide hospitals with an opportunity to target improvement efforts. We aimed to determine the degree to which resource utilization is associated with a satisfaction score based on the Summary Star Rating methodology.

METHODS

The study was conducted at the University of Rochester Medical Center (URMC), an 830‐bed tertiary care center in upstate New York. This was a retrospective review of all HCAHPS surveys returned to URMC over a 27‐month period from January 1, 2012 to April 1, 2014. URMC follows the standard CMS process for determining which patients receive surveys as follows. During the study timeframe, HCAHPS surveys were mailed to patients 18 years of age and older who had an inpatient stay spanning at least 1 midnight. Surveys were mailed within 5 days of discharge, and were generally returned within 6 weeks. URMC did not utilize telephone or email surveys during the study period. Surveys were not sent to patients who (1) were transferred to another facility, (2) were discharged to hospice, (3) died during the hospitalization, (4) received psychiatric or rehabilitative services during the hospitalization, (5) had an international address, and/or (6) were prisoners.

The survey vendor (Press Ganey, South Bend, IN) for URMC provided raw data for returned surveys with patient answers to questions. Administrative and billing databases were used to add demographic and clinical data for the corresponding hospitalization to the dataset. These data included age, gender, payer status (public, private, self, charity), length of stay, number of attendings who saw the patient (based on encounters documented in the electronic medical record (EMR)), all discharge International Classification of Diseases, 9th Revision (ICD‐9) diagnoses for the hospitalization, total charges for the hospitalization, and intensive care unit (ICU) utilization as evidenced by a documented encounter with a member of the Division of Critical Care/Pulmonary Medicine.

CMS analyzes surveys within 1 of 3 clinical service categories (medical, surgical, or obstetrics/gynecology) based on the discharging service. To parallel this approach, each returned survey was placed into 1 of these categories based on the clinical service of the discharging physician. Patients placed in the obstetrics/gynecology category (n = 1317, 13%) will be analyzed in a future analysis given inherent differences in patient characteristics that require evaluation of other variables.

Approximations of CMS Summary Star Rating

The HCAHPS survey is a multiple‐choice questionnaire that includes several domains of patient satisfaction. Respondents are asked to rate areas of satisfaction with their hospital experience on a Likert scale. CMS uses a weighted average of Likert responses to a subset of HCAHPS questions to calculate a hospital's raw score in 11 domains, as well as an overall raw summary score. CMS then adjusts each raw score for differences between hospitals (eg, clustering, improvement over time, method of survey) to determine a hospital's star rating in each domain and an overall Summary Star Rating (the Summary Star Rating is the primary factor by which consumers can compare hospitals).[4] Because our data were from a single hospital system, the between‐hospital scoring adjustments utilized by CMS were not applicable. Instead, we calculated the raw scores exactly as CMS does prior to the adjustments. Thus, our scores reflect the scores that CMS would have given URMC during the study period prior to standardized adjustments; we refer to this as the raw satisfaction rating (RSR). We calculated an RSR for every eligible survey. The RSR was calculated as a continuous variable from 0 (lowest) to 1 (highest). Detailed explanation of our RSR calculation is available in the Supporting Information in the online version of this article.

Statistical Analysis

All analyses were performed in aggregate and by service (medical vs surgical). Categorical variables were summarized using frequencies with percentages. Comparisons across levels of categorical variables were performed with the ² test. We report bivariate associations between the independent variables and RSRs in the top decile using unadjusted odds ratios (ORs) with 95% confidence intervals (CIs). Similarly, multivariable logistic regression was used for adjusted analyses. For the variables of severity of illness and resource intensity, the group with the lowest illness severity and lowest resource use served as the reference groups. We modeled patients without an ICU encounter and with an ICU encounter separately.

Charges, number of unique attendings encountered, and lengths of stay were highly correlated, and likely various measures of the same underlying construct of resource intensity, and therefore could not be entered into our models simultaneously. We combined these into a resource intensity score using factor analysis with a varimax rotation, and extracted factor scores for a single factor (supported by a scree plot). We then placed patients into 4 groups based on the distribution of the factor scores: low (25th percentile), moderate (25th50th percentile), major (50th75th percentile), and extreme (>75th percentile).

We used the Charlson‐Deyo comorbidity score as our disease severity index.[12] The index uses ICD‐9 diagnoses with points assigned for the impact of each diagnosis on morbidity and the points summed to an overall score. This provides a measure of disease severity for a patient based on the number of diagnoses and relative mortality of the individual diagnoses. Scores were categorized as 0 (representing no major illness burden), 1 to 3, 4 to 6, and >6.

All statistical analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC), and P values 0.05 were considered statistically significant. This study was approved by the institutional review board at the University of Rochester Medical Center.

RESULTS

Our initial search identified 10,007 returned surveys (29% of eligible patients returned surveys during the study period). Of these, 5059 (51%) were categorized as medical, 3630 (36%) as surgical, and 1317 (13%) as obstetrics/gynecology. One survey did not have the service of the discharging physician recorded and was excluded. Cohort demographics and relationship to RSRs in the top decile for the 8689 medical and surgical patients can be found in Table 1. The most common discharge diagnosis‐related groups (DRGs) for medical patients were 247, percutaneous cardiovascular procedure with drug‐eluding stent without major complications or comorbidities (MCC) (3.8%); 871, septicemia or severe sepsis without mechanical ventilation >96 hours with MCC (2.7%); and 392, esophagitis, gastroenteritis, and miscellaneous digestive disorders with MCC (2.3%). The most common DRGs for surgical patients were 460, spinal fusion except cervical without MCC (3.5%); 328, stomach, esophageal and duodenal procedure without complication or comorbidities or MCC (3.3%); and 491, back and neck procedure excluding spinal fusion without complication or comorbidities or MCC (3.1%).

Table 1.Cohort Demographics and Raw Satisfaction Ratings in the Top Decile

	Overall				Medical				Surgical
	Total	90th	Top Decile	P	Total	90th	Top Decile	P	Total	90th	Top Decile	P
NOTE: Data are presented as no. (%). Percentages may not add up to 100 due to rounding. Abbreviations: ICU, intensive care unit. *Calculated using the Charlson‐Deyo index; smaller values indicate less severity. Low = $10,000; medium = $10,000$40,000; high = >$40,000.
Overall	8,689	7,789 (90)	900 (10)		5,059	4,646 (92)	413 (8)		3,630	3,143 (87)	487 (13)
Age, y
30	419 (5)	371 (89)	48 (12)	0.001	218 (4)	208 (95)	10 (5)	0.001	201 (6)	163 (81)	38 (19)	0.001
3049	1,029 (12)	902 (88)	127 (12)		533 (11)	482 (90)	51 (10)		496 (14)	420 (85)	76 (15)
5069	3,911 (45)	3,450 (88)	461 (12)		2,136 (42)	1,930 (90)	206 (10)		1,775 (49)	1,520 (86)	255 (14)
>69	3,330 (38)	3,066 (92)	264 (8)		2,172 (43)	2,026 (93)	146 (7)		1,158 (32)	1,040 (90)	118 (10)
Gender
Male	4,640 (53)	4,142 (89)	498 (11)	0.220	2,596 (51)	2,379 (92)	217 (8)	0.602	2,044 (56)	1,763 (86)	281 (14)	0.506
Female	4,049 (47)	3,647 (90)	402 (10)		2,463 (49)	2,267 (92)	196 (8)		1,586 (44)	1,380 (87)	206 (13)
ICU encounter
No	7,122 (82)	6,441 (90)	681 (10)	0.001	4,547 (90)	4,193 (92)	354 (8)	0.001	2,575 (71)	2,248 (87)	327 (13)	0.048
Yes	1,567 (18)	1,348 (86)	219 (14)		512 (10)	453 (89)	59 (12)		1,055 (29)	895 (85)	160 (15)
Payer
Public	5,564 (64)	5,036 (91)	528 (10)	0.001	3,424 (68)	3,161 (92)	263 (8)	0.163	2,140 (59)	1,875 (88)	265 (12)	0.148
Private	3,064 (35)	2,702 (88)	362 (12)		1,603 (32)	1,458 (91)	145 (9)		1,461 (40)	1,244 (85)	217 (15)
Charity	45 (1)	37 (82)	8 (18)		25 (1)	21 (84)	4 (16)		20 (1)	16 (80)	4 (20)
Self	16 (0)	14 (88)	2 (13)		7 (0)	6 (86)	1 (14)		9 (0)	8 (89)	1 (11)
Length of stay, d
3	3,156 (36)	2,930 (93)	226 (7)	0.001	1,961 (39)	1,865 (95)	96 (5)	0.001	1,195 (33)	1,065 (89)	130 (11)	0.001
36	3,330 (38)	2,959 (89)	371 (11)		1,867 (37)	1,702 (91)	165 (9)		1,463 (40)	1,257 (86)	206 (14)
>6	2,203 (25)	1,900 (86)	303 (14)		1,231 (24)	1,079 (88)	152 (12)		972 (27)	821 (85)	151 (16)
No. of attendings
4	3,959 (46)	3,615 (91)	344 (9)	0.001	2,307 (46)	2,160 (94)	147 (6)	0.001	1,652 (46)	1,455 (88)	197 (12)	0.052
46	3,067 (35)	2,711 (88)	356 (12)		1,836 (36)	1,663 (91)	173 (9)		1,231 (34)	1,048 (85)	183 (15)
>6	1,663 (19)	1,463 (88)	200 (12)		916 (18)	823 (90)	93 (10)		747 (21)	640 (86)	107 (14)
Severity index*
0 (lowest)	2,812 (32)	2,505 (89)	307 (11)	0.272	1,273 (25)	1,185 (93)	88 (7)	0.045	1,539 (42)	1,320 (86)	219 (14)	0.261
13	4,253 (49)	3,827 (90)	426 (10)		2,604 (52)	2,395 (92)	209 (8)		1,649 (45)	1,432 (87)	217 (13)
46	1163 (13)	1,052 (91)	111 (10)		849 (17)	770 (91)	79 (9)		314 (9)	282 (90)	32 (10)
>6 (highest)	461 (5)	405 (88)	56 (12)		333 (7)	296 (89)	37 (11)		128 (4)	109 (85)	19 (15)
Charges,
Low	1,820 (21)	1,707 (94)	113 (6)	0.001	1,426 (28)	1,357 (95)	69 (5)	0.001	394 (11)	350 (89)	44 (11)	0.007
Medium	5,094 (59)	4,581 (90)	513 (10)		2,807 (56)	2,582 (92)	225 (8)		2,287 (63)	1,999 (87)	288 (13)
High	1,775 (20)	1,501 (85)	274 (15)		826 (16)	707 (86)	119 (14)		949 (26)	794 (84)	155 (16)

Unadjusted analysis of medical and surgical patients identified significant associations of several variables with a top decile RSR (Table 2). Patients with longer lengths of stay (OR: 2.07, 95% CI: 1.72‐2.48), more attendings (OR: 1.44, 95% CI: 1.19‐1.73), and higher hospital charges (OR: 2.76, 95% CI: 2.19‐3.47) were more likely to report an RSR in the top decile. Patients without an ICU encounter (OR: 0.65, 95% CI: 0.55‐0.77) and on a medical service (OR: 0.57, 95% CI: 0.5‐ 0.66) were less likely to report an RSR in the top decile. Several associations were identified in only the medical or surgical cohorts. In the medical cohort, patients with the highest illness severity index (OR: 1.68, 95% CI: 1.12‐ 2.52) and with 7 different attending physicians (OR: 1.66, 95% CI: 1.27‐2.18) were more likely to report RSRs in the top decile. In the surgical cohort, patients 30 years of age (OR: 2.05, 95% CI 1.38‐3.07) were more likely to report an RSR in the top decile than patients >69 years of age. Insurance payer category and gender were not significantly associated with top decile RSRs.

Table 2.Bivariate Comparisons of Associations Between Top Decile Satisfaction Ratings and Reference Levels

	Overall		Medical		Surgical
	Odds Ratio (95% CI)	P	Odds Ratio (95% CI)	P	Odds Ratio (95% CI)	P
NOTE: Abbreviations: CI, confidence interval; ICU, intensive care unit; Ref, reference. *Calculated using the Charlson‐Deyo index; smaller values indicate less severity. Low = $10,000; medium = $10,000$40,000; high = >$40,000.
Age, y
30	1.5 (1.082.08)	0.014	0.67 (0.351.29)	0.227	2.05 (1.383.07)	0.001
3049	1.64 (1.312.05)	.001	1.47 (1.052.05)	0.024	1.59 (1.172.17)	0.003
5069	1.55 (1.321.82)	.001	1.48 (1.191.85)	0.001	1.48 (1.171.86)	0.001
>69	Ref		Ref		Ref
Gender
Male	1.09 (0.951.25)	0.220	1.06 (0.861.29)	0.602	1.07 (0.881.3)	0.506
Female	Ref		Ref		Ref
ICU encounter
No	0.65 (0.550.77)	0.001	0.65 (0.480.87)	0.004	0.81 (0.661)	0.048
Yes	Ref		Ref		Ref
Payer
Public	0.73 (0.173.24)	0.683	0.5 (0.064.16)	0.521	1.13 (0.149.08)	0.908
Private	0.94 (0.214.14)	0.933	0.6 (0.074.99)	0.634	1.4 (0.1711.21)	0.754
Charity	1.51 (0.298.02)	0.626	1.14 (0.1112.25)	0.912	2 (0.1920.97)	0.563
Self	Ref		Ref		Ref
Length of stay, d
3	Ref		Ref		Ref
36	1.63 (1.371.93)	0.001	1.88 (1.452.44)	0.001	1.34 (1.061.7)	0.014
>6	2.07 (1.722.48)	0.001	2.74 (2.13.57)	0.001	1.51 (1.171.94)	0.001
No. of attendings
4	Ref		Ref		Ref
46	1.38 (1.181.61)	0.001	1.53 (1.221.92)	0.001	1.29 (1.041.6)	0.021
>6	1.44 (1.191.73)	0.001	1.66 (1.272.18)	0.001	1.23 (0.961.59)	0.102
Severity index*
0 (lowest)	Ref		Ref		Ref
13	0.91 (0.781.06)	0.224	1.18 (0.911.52)	0.221	0.91 (0.751.12)	0.380
46	0.86 (0.681.08)	0.200	1.38 (1.011.9)	0.046	0.68 (0.461.01)	0.058
>6 (highest)	1.13 (0.831.53)	0.436	1.68 (1.122.52)	0.012	1.05 (0.631.75)	0.849
Charges
Low	Ref		Ref		Ref
Medium	1.69 (1.372.09)	0.001	1.71 (1.32.26)	0.001	1.15 (0.821.61)	0.428
High	2.76 (2.193.47)	0.001	3.31 (2.434.51)	0.001	1.55 (1.092.22)	0.016
Service
Medical	0.57 (0.50.66)	0.001
Surgical	Ref

Multivariable modeling (Table 3) for all patients without an ICU encounter suggested that (1) patients aged 30 years, 30 to 49 years, and 50 to 69 years were more likely to report top decile RSRs when compared to patients 70 years and older (OR: 1.61, 95% CI: 1.09‐2.36; OR: 1.44, 95% CI: 1.08‐1.93; and OR: 1.39, 95% CI: 1.13‐1.71, respectively) and (2), when compared to patients with extreme resource intensity scores, patients with higher resource intensity scores were more likely to report top decile RSRs (moderate [OR: 1.42, 95% CI: 1.11‐1.83], major [OR: 1.56, 95% CI: 1.22‐2.01], and extreme [OR: 2.29, 95% CI: 1.8‐2.92]. These results were relatively consistent within medical and surgical subgroups (Table 3).

Table 3.Multivariable Logistic Regression Model for Top Decile Raw Satisfaction Ratings for Patients on the General Wards*

	Overall		Medical		Surgical
	Odds Ratio (95% CI)	P	Odds Ratio (95% CI)	P	Odds Ratio (95% CI)	P
NOTE: Abbreviations: CI, confidence interval; Ref, reference. *Excludes the 1,567 patients who had an intensive care unit encounter. Calculated using the Charlson‐Deyo index. Component variables include length of stay, number of attendings, and charges
Age, y
30	1.61 (1.092.36)	0.016	0.82 (0.41.7)	0.596	2.31 (1.393.82)	0.001
3049	1.44 (1.081.93)	0.014	1.55 (1.032.32)	0.034	1.41 (0.912.17)	0.120
5069	1.39 (1.131.71)	0.002	1.44 (1.11.88)	0.008	1.39 (11.93)	0.049
>69	Ref		Ref		Ref
Sex
Male	1 (0.851.17)	0.964	1 (0.81.25)	0.975	0.99 (0.791.26)	0.965
Female	Ref		Ref		Ref
Payer
Public	0.62 (0.142.8)	0.531	0.42 (0.053.67)	0.432	1.03 (0.128.59)	0.978
Private	0.67 (0.153.02)	0.599	0.42 (0.053.67)	0.434	1.17 (0.149.69)	0.884
Charity	1.54 (0.288.41)	0.620	1 (0.0911.13)	0.999	2.56 (0.2328.25)	0.444
Self	Ref		Ref		Ref
Severity index
0 (lowest)	Ref		Ref		Ref
13	1.07 (0.891.29)	0.485	1.18 (0.881.58)	0.267	1 (0.781.29)	0.986
46	1.14 (0.861.51)	0.377	1.42 (0.992.04)	0.056	0.6 (0.331.1)	0.100
>6 (highest)	1.31 (0.911.9)	0.150	1.47 (0.932.33)	0.097	1.1 (0.542.21)	0.795
Resource intensity score
Low	Ref		Ref		Ref
Moderate	1.42 (1.111.83)	0.006	1.6 (1.112.3)	0.011	0.94 (0.661.34)	0.722
Major	1.56 (1.222.01)	0.001	1.69 (1.182.43)	0.004	1.28 (0.911.8)	0.151
Extreme	2.29 (1.82.92)	0.001	2.72 (1.943.82)	0.001	1.63 (1.172.26)	0.004
Service
Medical	0.59 (0.50.69)	0.001
Surgical	Ref

In those with at least 1 ICU attending encounter (see Supporting Table 1 in the online version of this article), no variables demonstrated significant association with top decile RSRs in the overall group or in the medical subgroup. For surgical patients with at least 1 ICU attending encounter (see Supporting Table 1 in the online version of this article), patients aged 30 to 49 and 50 to 69 years were more likely to provide top decile RSRs (OR: 1.93, 95% CI: 1.08‐3.46 and OR: 1.65, 95% CI 1.07‐2.53, respectively). Resource intensity was not significantly associated with top decile RSRs.

DISCUSSION

Our analysis suggests that, for patients on the general care floors, resource utilization is associated with the RSR and, therefore, potentially the CMS Summary Star Rating. Adjusting for severity of illness, patients with higher resource utilization were more likely to report top decile RSRs.

Prior data regarding utilization and satisfaction are mixed. In a 2‐year, prospective, national examination, patients in the highest quartile of patient satisfaction had increased healthcare and prescription drug expenditures as well as increased rates of hospitalization when compared with patients in the lowest quartile of patient satisfaction.[9] However, a recent national study of surgical administrative databases suggested hospitals with high patient satisfaction provided more efficient care.[13]

One reason for the conflicting data may be that large, national evaluations are unable to control for between‐hospital confounders (ie, hospital quality of care). By capturing all eligible returned surveys at 1 institution, our design allowed us to collect granular data. We found that in 1 hospital setting, patient population, facilities, and food services, patients receiving more clinical resources generally assigned higher ratings than patients receiving less.

It is possible that utilization is a proxy for serious illness, and that patients with serious illness receive more attention during hospitalization and are more satisfied when discharged in a good state of health. However, we did adjust for severity of illness in our model using the Charlson‐Deyo index and we suggest that, other factors being equal, hospitals with higher per‐patient expenditures may be assigned higher Summary Star Ratings.

CMS has recently implemented a number of metrics designed to decrease healthcare costs by improving quality, safety, and efficiency. Concurrently, CMS has also prioritized patient experience. The Summary Star Rating was created to provide healthcare consumers with an easy way to compare the patient experience between hospitals[4]; however, our data suggest that this metric may be at odds with inpatient cost savings and efficiency metrics.

Per‐patient spending becomes particularly salient when considering that in fiscal year 2016, CMS' hospital VBP reimbursement will include 2 metrics: an efficiency outcome measure labeled Medicare spending per beneficiary, and a patient experience outcome measure based on HCAHPS survey dimensions.[2] Together, these 2 metrics will comprise nearly half of the total VBP performance score used to determine reimbursement. Although our data suggest that these 2 VBP metrics may be correlated, it should be noted that we measured inpatient hospital charges, whereas the CMS efficiency outcome measure includes costs for episode of care spanning 3 days prior to hospitalization to 30 days after hospitalization.

Patient expectations likely play a role in satisfaction.[14, 15, 16] In an outpatient setting, physician fulfillment of patient requests has been associated with positive patient evaluations of care.[17] However, patients appear to value education, shared decision making, and provider empathy more than testing and intervention.[14, 18, 19, 20, 21, 22, 23] Perhaps, in the absence of the former attributes, patients use additional resource expenditure as a proxy.

It is not clear that higher resource expenditure improves outcomes. A landmark study of nearly 1 million Medicare enrollees by Fisher et al. suggests that, although Medicare patients in higher‐spending regions receive more care than those in lower‐spending regions, this does not result in better health outcomes, specifically with regard to mortality.[24, 25] Patients who live in areas of high hospital capacity use the hospital more frequently than do patients in areas of low hospital capacity, but this does not appear to result in improved mortality rates.[26] In fact, physicians in areas of high healthcare capacity report more difficulty maintaining high‐quality patient relationships and feel less able to provide high‐quality care than physicians in lower‐capacity areas.[27]

We hypothesize the cause of the association between resource utilization and patient satisfaction could be that patients (1) perceive that a doctor who allows them to stay longer in the hospital or who performs additional testing cares more about their well‐being and (2) that these patients feel more strongly that their concerns are being heard and addressed by their physicians. A systematic review of primary care patients identified many studies that found a positive association between meeting patient expectations and satisfaction with care, but also suggested that although patients frequently expect information, physicians misperceive this as an expectation of specific action.[28] A separate systematic review found that patient education in the form of decision aides can help patients develop more reasonable expectations and reduce utilization of certain discretionary procedures such as elective surgeries and prostate‐specific antigen testing.[29]

We did not specifically address clinical outcomes in our analysis because the clinical outcomes on which CMS currently adjusts VBP reimbursement focus on 30‐day mortality for specific diagnoses, nosocomial infections, and iatrogenic events.[30] Our data include only returned surveys from living patients, and it is likely that 30‐day mortality was similar throughout all subsets of patients. Additionally, the nosocomial and iatrogenic outcome measures used by CMS are sufficiently rare on the general floors and are unlikely to have significantly influenced our results.[31]

Our study has several strengths. Nearly all medical and surgical patient surveys returned during the study period were included, and therefore our calculations are likely to accurately reflect the Summary Star Rating that would have been assigned for the period. Second, the large sample size helps attenuate potential differences in commonly used outcome metrics. Third, by adjusting for a variety of demographic and clinical variables, we were able to decrease the likelihood of unidentified confounders.

Notably, we identified 38 (0.4%) surveys returned for patients under 18 years of age at admission. These surveys were included in our analysis because, to the best of our knowledge, they would have existed in the pool of surveys CMS could have used to assign a Summary Star Rating.

Our study also has limitations. First, geographically diverse data are needed to ensure generalizability. Second, we used the Charlson‐Deyo Comorbidity Index to describe the degree of illness for each patient. This index represents a patient's total illness burden but may not describe the relative severity of the patient's current illness relative to another patient. Third, we selected variables we felt were most likely to be associated with patient experience, but unidentified confounding remains possible. Fourth, attendings caring for ICU patients fall within the Division of Critical Care/Pulmonary Medicine. Therefore, we may have inadvertently placed patients into the ICU cohort who received a pulmonary/critical care consult on the general floors. Fifth, our data describe associations only for patients who returned surveys. Although there may be inherent biases in patients who return surveys, HCAHPS survey responses are used by CMS to determine a hospital's overall satisfaction score.

CONCLUSION

For patients who return HCAHPS surveys, resource utilization may be positively associated with a hospital's Summary Star Rating. These data suggest that hospitals with higher per‐patient expenditures may receive higher Summary Star Ratings, which could result in hospitals with higher per‐patient resource utilization appearing more attractive to healthcare consumers. Future studies should attempt to confirm our findings at other institutions and to determine causative factors.

Acknowledgements

The authors thank Jason Machan, PhD (Department of Orthopedics and Surgery, Warren Alpert Medical School, Brown University, Providence, Rhode Island) for his help with study design, and Ms. Brenda Foster (data analyst, University of Rochester Medical Center, Rochester, NY) for her help with data collection.

Disclosures: Nothing to report.

The patient experience has become increasingly important to healthcare in the United States. It is now a metric used commonly to determine physician compensation and accounts for nearly 30% of the Centers for Medicare and Medicaid Services' (CMS) Value‐Based Purchasing (VBP) reimbursement for fiscal years 2015 and 2016.[1, 2]

In April 2015, CMS added a 5‐star patient experience score to its Hospital Compare website in an attempt to address the Affordable Care Act's call for transparent and easily understandable public reporting.[3] A hospital's principal score is the Summary Star Rating, which is based on responses to the Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) survey. The formulas used to calculate Summary Star Ratings have been reported by CMS.[4]

Studies published over the past decade suggest that gender, age, education level, length of hospital stay, travel distance, and other factors may influence patient satisfaction.[5, 6, 7, 8] One study utilizing a national dataset suggested that higher patient satisfaction was associated with greater inpatient healthcare utilization and higher healthcare expenditures.[9] It is therefore possible that emphasizing patient experience scores could adversely impact healthcare resource utilization. However, positive patient experience may also be an important independent dimension of quality for patients and correlate with improved clinical outcomes.[10]

We know of no literature describing patient factors associated with the Summary Star Rating. Given that this rating is now used as a standard metric by which patient experience can be compared across more than 3,500 hospitals,[11] data describing the association between patient‐level factors and the Summary Star Rating may provide hospitals with an opportunity to target improvement efforts. We aimed to determine the degree to which resource utilization is associated with a satisfaction score based on the Summary Star Rating methodology.

METHODS

The study was conducted at the University of Rochester Medical Center (URMC), an 830‐bed tertiary care center in upstate New York. This was a retrospective review of all HCAHPS surveys returned to URMC over a 27‐month period from January 1, 2012 to April 1, 2014. URMC follows the standard CMS process for determining which patients receive surveys as follows. During the study timeframe, HCAHPS surveys were mailed to patients 18 years of age and older who had an inpatient stay spanning at least 1 midnight. Surveys were mailed within 5 days of discharge, and were generally returned within 6 weeks. URMC did not utilize telephone or email surveys during the study period. Surveys were not sent to patients who (1) were transferred to another facility, (2) were discharged to hospice, (3) died during the hospitalization, (4) received psychiatric or rehabilitative services during the hospitalization, (5) had an international address, and/or (6) were prisoners.

The survey vendor (Press Ganey, South Bend, IN) for URMC provided raw data for returned surveys with patient answers to questions. Administrative and billing databases were used to add demographic and clinical data for the corresponding hospitalization to the dataset. These data included age, gender, payer status (public, private, self, charity), length of stay, number of attendings who saw the patient (based on encounters documented in the electronic medical record (EMR)), all discharge International Classification of Diseases, 9th Revision (ICD‐9) diagnoses for the hospitalization, total charges for the hospitalization, and intensive care unit (ICU) utilization as evidenced by a documented encounter with a member of the Division of Critical Care/Pulmonary Medicine.

CMS analyzes surveys within 1 of 3 clinical service categories (medical, surgical, or obstetrics/gynecology) based on the discharging service. To parallel this approach, each returned survey was placed into 1 of these categories based on the clinical service of the discharging physician. Patients placed in the obstetrics/gynecology category (n = 1317, 13%) will be analyzed in a future analysis given inherent differences in patient characteristics that require evaluation of other variables.

Approximations of CMS Summary Star Rating

The HCAHPS survey is a multiple‐choice questionnaire that includes several domains of patient satisfaction. Respondents are asked to rate areas of satisfaction with their hospital experience on a Likert scale. CMS uses a weighted average of Likert responses to a subset of HCAHPS questions to calculate a hospital's raw score in 11 domains, as well as an overall raw summary score. CMS then adjusts each raw score for differences between hospitals (eg, clustering, improvement over time, method of survey) to determine a hospital's star rating in each domain and an overall Summary Star Rating (the Summary Star Rating is the primary factor by which consumers can compare hospitals).[4] Because our data were from a single hospital system, the between‐hospital scoring adjustments utilized by CMS were not applicable. Instead, we calculated the raw scores exactly as CMS does prior to the adjustments. Thus, our scores reflect the scores that CMS would have given URMC during the study period prior to standardized adjustments; we refer to this as the raw satisfaction rating (RSR). We calculated an RSR for every eligible survey. The RSR was calculated as a continuous variable from 0 (lowest) to 1 (highest). Detailed explanation of our RSR calculation is available in the Supporting Information in the online version of this article.

Statistical Analysis

All analyses were performed in aggregate and by service (medical vs surgical). Categorical variables were summarized using frequencies with percentages. Comparisons across levels of categorical variables were performed with the ² test. We report bivariate associations between the independent variables and RSRs in the top decile using unadjusted odds ratios (ORs) with 95% confidence intervals (CIs). Similarly, multivariable logistic regression was used for adjusted analyses. For the variables of severity of illness and resource intensity, the group with the lowest illness severity and lowest resource use served as the reference groups. We modeled patients without an ICU encounter and with an ICU encounter separately.

Charges, number of unique attendings encountered, and lengths of stay were highly correlated, and likely various measures of the same underlying construct of resource intensity, and therefore could not be entered into our models simultaneously. We combined these into a resource intensity score using factor analysis with a varimax rotation, and extracted factor scores for a single factor (supported by a scree plot). We then placed patients into 4 groups based on the distribution of the factor scores: low (25th percentile), moderate (25th50th percentile), major (50th75th percentile), and extreme (>75th percentile).

We used the Charlson‐Deyo comorbidity score as our disease severity index.[12] The index uses ICD‐9 diagnoses with points assigned for the impact of each diagnosis on morbidity and the points summed to an overall score. This provides a measure of disease severity for a patient based on the number of diagnoses and relative mortality of the individual diagnoses. Scores were categorized as 0 (representing no major illness burden), 1 to 3, 4 to 6, and >6.

All statistical analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC), and P values 0.05 were considered statistically significant. This study was approved by the institutional review board at the University of Rochester Medical Center.

RESULTS

Our initial search identified 10,007 returned surveys (29% of eligible patients returned surveys during the study period). Of these, 5059 (51%) were categorized as medical, 3630 (36%) as surgical, and 1317 (13%) as obstetrics/gynecology. One survey did not have the service of the discharging physician recorded and was excluded. Cohort demographics and relationship to RSRs in the top decile for the 8689 medical and surgical patients can be found in Table 1. The most common discharge diagnosis‐related groups (DRGs) for medical patients were 247, percutaneous cardiovascular procedure with drug‐eluding stent without major complications or comorbidities (MCC) (3.8%); 871, septicemia or severe sepsis without mechanical ventilation >96 hours with MCC (2.7%); and 392, esophagitis, gastroenteritis, and miscellaneous digestive disorders with MCC (2.3%). The most common DRGs for surgical patients were 460, spinal fusion except cervical without MCC (3.5%); 328, stomach, esophageal and duodenal procedure without complication or comorbidities or MCC (3.3%); and 491, back and neck procedure excluding spinal fusion without complication or comorbidities or MCC (3.1%).

Table 1.Cohort Demographics and Raw Satisfaction Ratings in the Top Decile

	Overall				Medical				Surgical
	Total	90th	Top Decile	P	Total	90th	Top Decile	P	Total	90th	Top Decile	P
NOTE: Data are presented as no. (%). Percentages may not add up to 100 due to rounding. Abbreviations: ICU, intensive care unit. *Calculated using the Charlson‐Deyo index; smaller values indicate less severity. Low = $10,000; medium = $10,000$40,000; high = >$40,000.
Overall	8,689	7,789 (90)	900 (10)		5,059	4,646 (92)	413 (8)		3,630	3,143 (87)	487 (13)
Age, y
30	419 (5)	371 (89)	48 (12)	0.001	218 (4)	208 (95)	10 (5)	0.001	201 (6)	163 (81)	38 (19)	0.001
3049	1,029 (12)	902 (88)	127 (12)		533 (11)	482 (90)	51 (10)		496 (14)	420 (85)	76 (15)
5069	3,911 (45)	3,450 (88)	461 (12)		2,136 (42)	1,930 (90)	206 (10)		1,775 (49)	1,520 (86)	255 (14)
>69	3,330 (38)	3,066 (92)	264 (8)		2,172 (43)	2,026 (93)	146 (7)		1,158 (32)	1,040 (90)	118 (10)
Gender
Male	4,640 (53)	4,142 (89)	498 (11)	0.220	2,596 (51)	2,379 (92)	217 (8)	0.602	2,044 (56)	1,763 (86)	281 (14)	0.506
Female	4,049 (47)	3,647 (90)	402 (10)		2,463 (49)	2,267 (92)	196 (8)		1,586 (44)	1,380 (87)	206 (13)
ICU encounter
No	7,122 (82)	6,441 (90)	681 (10)	0.001	4,547 (90)	4,193 (92)	354 (8)	0.001	2,575 (71)	2,248 (87)	327 (13)	0.048
Yes	1,567 (18)	1,348 (86)	219 (14)		512 (10)	453 (89)	59 (12)		1,055 (29)	895 (85)	160 (15)
Payer
Public	5,564 (64)	5,036 (91)	528 (10)	0.001	3,424 (68)	3,161 (92)	263 (8)	0.163	2,140 (59)	1,875 (88)	265 (12)	0.148
Private	3,064 (35)	2,702 (88)	362 (12)		1,603 (32)	1,458 (91)	145 (9)		1,461 (40)	1,244 (85)	217 (15)
Charity	45 (1)	37 (82)	8 (18)		25 (1)	21 (84)	4 (16)		20 (1)	16 (80)	4 (20)
Self	16 (0)	14 (88)	2 (13)		7 (0)	6 (86)	1 (14)		9 (0)	8 (89)	1 (11)
Length of stay, d
3	3,156 (36)	2,930 (93)	226 (7)	0.001	1,961 (39)	1,865 (95)	96 (5)	0.001	1,195 (33)	1,065 (89)	130 (11)	0.001
36	3,330 (38)	2,959 (89)	371 (11)		1,867 (37)	1,702 (91)	165 (9)		1,463 (40)	1,257 (86)	206 (14)
>6	2,203 (25)	1,900 (86)	303 (14)		1,231 (24)	1,079 (88)	152 (12)		972 (27)	821 (85)	151 (16)
No. of attendings
4	3,959 (46)	3,615 (91)	344 (9)	0.001	2,307 (46)	2,160 (94)	147 (6)	0.001	1,652 (46)	1,455 (88)	197 (12)	0.052
46	3,067 (35)	2,711 (88)	356 (12)		1,836 (36)	1,663 (91)	173 (9)		1,231 (34)	1,048 (85)	183 (15)
>6	1,663 (19)	1,463 (88)	200 (12)		916 (18)	823 (90)	93 (10)		747 (21)	640 (86)	107 (14)
Severity index*
0 (lowest)	2,812 (32)	2,505 (89)	307 (11)	0.272	1,273 (25)	1,185 (93)	88 (7)	0.045	1,539 (42)	1,320 (86)	219 (14)	0.261
13	4,253 (49)	3,827 (90)	426 (10)		2,604 (52)	2,395 (92)	209 (8)		1,649 (45)	1,432 (87)	217 (13)
46	1163 (13)	1,052 (91)	111 (10)		849 (17)	770 (91)	79 (9)		314 (9)	282 (90)	32 (10)
>6 (highest)	461 (5)	405 (88)	56 (12)		333 (7)	296 (89)	37 (11)		128 (4)	109 (85)	19 (15)
Charges,
Low	1,820 (21)	1,707 (94)	113 (6)	0.001	1,426 (28)	1,357 (95)	69 (5)	0.001	394 (11)	350 (89)	44 (11)	0.007
Medium	5,094 (59)	4,581 (90)	513 (10)		2,807 (56)	2,582 (92)	225 (8)		2,287 (63)	1,999 (87)	288 (13)
High	1,775 (20)	1,501 (85)	274 (15)		826 (16)	707 (86)	119 (14)		949 (26)	794 (84)	155 (16)

Unadjusted analysis of medical and surgical patients identified significant associations of several variables with a top decile RSR (Table 2). Patients with longer lengths of stay (OR: 2.07, 95% CI: 1.72‐2.48), more attendings (OR: 1.44, 95% CI: 1.19‐1.73), and higher hospital charges (OR: 2.76, 95% CI: 2.19‐3.47) were more likely to report an RSR in the top decile. Patients without an ICU encounter (OR: 0.65, 95% CI: 0.55‐0.77) and on a medical service (OR: 0.57, 95% CI: 0.5‐ 0.66) were less likely to report an RSR in the top decile. Several associations were identified in only the medical or surgical cohorts. In the medical cohort, patients with the highest illness severity index (OR: 1.68, 95% CI: 1.12‐ 2.52) and with 7 different attending physicians (OR: 1.66, 95% CI: 1.27‐2.18) were more likely to report RSRs in the top decile. In the surgical cohort, patients 30 years of age (OR: 2.05, 95% CI 1.38‐3.07) were more likely to report an RSR in the top decile than patients >69 years of age. Insurance payer category and gender were not significantly associated with top decile RSRs.

Table 2.Bivariate Comparisons of Associations Between Top Decile Satisfaction Ratings and Reference Levels

	Overall		Medical		Surgical
	Odds Ratio (95% CI)	P	Odds Ratio (95% CI)	P	Odds Ratio (95% CI)	P
NOTE: Abbreviations: CI, confidence interval; ICU, intensive care unit; Ref, reference. *Calculated using the Charlson‐Deyo index; smaller values indicate less severity. Low = $10,000; medium = $10,000$40,000; high = >$40,000.
Age, y
30	1.5 (1.082.08)	0.014	0.67 (0.351.29)	0.227	2.05 (1.383.07)	0.001
3049	1.64 (1.312.05)	.001	1.47 (1.052.05)	0.024	1.59 (1.172.17)	0.003
5069	1.55 (1.321.82)	.001	1.48 (1.191.85)	0.001	1.48 (1.171.86)	0.001
>69	Ref		Ref		Ref
Gender
Male	1.09 (0.951.25)	0.220	1.06 (0.861.29)	0.602	1.07 (0.881.3)	0.506
Female	Ref		Ref		Ref
ICU encounter
No	0.65 (0.550.77)	0.001	0.65 (0.480.87)	0.004	0.81 (0.661)	0.048
Yes	Ref		Ref		Ref
Payer
Public	0.73 (0.173.24)	0.683	0.5 (0.064.16)	0.521	1.13 (0.149.08)	0.908
Private	0.94 (0.214.14)	0.933	0.6 (0.074.99)	0.634	1.4 (0.1711.21)	0.754
Charity	1.51 (0.298.02)	0.626	1.14 (0.1112.25)	0.912	2 (0.1920.97)	0.563
Self	Ref		Ref		Ref
Length of stay, d
3	Ref		Ref		Ref
36	1.63 (1.371.93)	0.001	1.88 (1.452.44)	0.001	1.34 (1.061.7)	0.014
>6	2.07 (1.722.48)	0.001	2.74 (2.13.57)	0.001	1.51 (1.171.94)	0.001
No. of attendings
4	Ref		Ref		Ref
46	1.38 (1.181.61)	0.001	1.53 (1.221.92)	0.001	1.29 (1.041.6)	0.021
>6	1.44 (1.191.73)	0.001	1.66 (1.272.18)	0.001	1.23 (0.961.59)	0.102
Severity index*
0 (lowest)	Ref		Ref		Ref
13	0.91 (0.781.06)	0.224	1.18 (0.911.52)	0.221	0.91 (0.751.12)	0.380
46	0.86 (0.681.08)	0.200	1.38 (1.011.9)	0.046	0.68 (0.461.01)	0.058
>6 (highest)	1.13 (0.831.53)	0.436	1.68 (1.122.52)	0.012	1.05 (0.631.75)	0.849
Charges
Low	Ref		Ref		Ref
Medium	1.69 (1.372.09)	0.001	1.71 (1.32.26)	0.001	1.15 (0.821.61)	0.428
High	2.76 (2.193.47)	0.001	3.31 (2.434.51)	0.001	1.55 (1.092.22)	0.016
Service
Medical	0.57 (0.50.66)	0.001
Surgical	Ref

Multivariable modeling (Table 3) for all patients without an ICU encounter suggested that (1) patients aged 30 years, 30 to 49 years, and 50 to 69 years were more likely to report top decile RSRs when compared to patients 70 years and older (OR: 1.61, 95% CI: 1.09‐2.36; OR: 1.44, 95% CI: 1.08‐1.93; and OR: 1.39, 95% CI: 1.13‐1.71, respectively) and (2), when compared to patients with extreme resource intensity scores, patients with higher resource intensity scores were more likely to report top decile RSRs (moderate [OR: 1.42, 95% CI: 1.11‐1.83], major [OR: 1.56, 95% CI: 1.22‐2.01], and extreme [OR: 2.29, 95% CI: 1.8‐2.92]. These results were relatively consistent within medical and surgical subgroups (Table 3).

Table 3.Multivariable Logistic Regression Model for Top Decile Raw Satisfaction Ratings for Patients on the General Wards*

	Overall		Medical		Surgical
	Odds Ratio (95% CI)	P	Odds Ratio (95% CI)	P	Odds Ratio (95% CI)	P
NOTE: Abbreviations: CI, confidence interval; Ref, reference. *Excludes the 1,567 patients who had an intensive care unit encounter. Calculated using the Charlson‐Deyo index. Component variables include length of stay, number of attendings, and charges
Age, y
30	1.61 (1.092.36)	0.016	0.82 (0.41.7)	0.596	2.31 (1.393.82)	0.001
3049	1.44 (1.081.93)	0.014	1.55 (1.032.32)	0.034	1.41 (0.912.17)	0.120
5069	1.39 (1.131.71)	0.002	1.44 (1.11.88)	0.008	1.39 (11.93)	0.049
>69	Ref		Ref		Ref
Sex
Male	1 (0.851.17)	0.964	1 (0.81.25)	0.975	0.99 (0.791.26)	0.965
Female	Ref		Ref		Ref
Payer
Public	0.62 (0.142.8)	0.531	0.42 (0.053.67)	0.432	1.03 (0.128.59)	0.978
Private	0.67 (0.153.02)	0.599	0.42 (0.053.67)	0.434	1.17 (0.149.69)	0.884
Charity	1.54 (0.288.41)	0.620	1 (0.0911.13)	0.999	2.56 (0.2328.25)	0.444
Self	Ref		Ref		Ref
Severity index
0 (lowest)	Ref		Ref		Ref
13	1.07 (0.891.29)	0.485	1.18 (0.881.58)	0.267	1 (0.781.29)	0.986
46	1.14 (0.861.51)	0.377	1.42 (0.992.04)	0.056	0.6 (0.331.1)	0.100
>6 (highest)	1.31 (0.911.9)	0.150	1.47 (0.932.33)	0.097	1.1 (0.542.21)	0.795
Resource intensity score
Low	Ref		Ref		Ref
Moderate	1.42 (1.111.83)	0.006	1.6 (1.112.3)	0.011	0.94 (0.661.34)	0.722
Major	1.56 (1.222.01)	0.001	1.69 (1.182.43)	0.004	1.28 (0.911.8)	0.151
Extreme	2.29 (1.82.92)	0.001	2.72 (1.943.82)	0.001	1.63 (1.172.26)	0.004
Service
Medical	0.59 (0.50.69)	0.001
Surgical	Ref

In those with at least 1 ICU attending encounter (see Supporting Table 1 in the online version of this article), no variables demonstrated significant association with top decile RSRs in the overall group or in the medical subgroup. For surgical patients with at least 1 ICU attending encounter (see Supporting Table 1 in the online version of this article), patients aged 30 to 49 and 50 to 69 years were more likely to provide top decile RSRs (OR: 1.93, 95% CI: 1.08‐3.46 and OR: 1.65, 95% CI 1.07‐2.53, respectively). Resource intensity was not significantly associated with top decile RSRs.

DISCUSSION

Our analysis suggests that, for patients on the general care floors, resource utilization is associated with the RSR and, therefore, potentially the CMS Summary Star Rating. Adjusting for severity of illness, patients with higher resource utilization were more likely to report top decile RSRs.

Prior data regarding utilization and satisfaction are mixed. In a 2‐year, prospective, national examination, patients in the highest quartile of patient satisfaction had increased healthcare and prescription drug expenditures as well as increased rates of hospitalization when compared with patients in the lowest quartile of patient satisfaction.[9] However, a recent national study of surgical administrative databases suggested hospitals with high patient satisfaction provided more efficient care.[13]

One reason for the conflicting data may be that large, national evaluations are unable to control for between‐hospital confounders (ie, hospital quality of care). By capturing all eligible returned surveys at 1 institution, our design allowed us to collect granular data. We found that in 1 hospital setting, patient population, facilities, and food services, patients receiving more clinical resources generally assigned higher ratings than patients receiving less.

It is possible that utilization is a proxy for serious illness, and that patients with serious illness receive more attention during hospitalization and are more satisfied when discharged in a good state of health. However, we did adjust for severity of illness in our model using the Charlson‐Deyo index and we suggest that, other factors being equal, hospitals with higher per‐patient expenditures may be assigned higher Summary Star Ratings.

CMS has recently implemented a number of metrics designed to decrease healthcare costs by improving quality, safety, and efficiency. Concurrently, CMS has also prioritized patient experience. The Summary Star Rating was created to provide healthcare consumers with an easy way to compare the patient experience between hospitals[4]; however, our data suggest that this metric may be at odds with inpatient cost savings and efficiency metrics.

Per‐patient spending becomes particularly salient when considering that in fiscal year 2016, CMS' hospital VBP reimbursement will include 2 metrics: an efficiency outcome measure labeled Medicare spending per beneficiary, and a patient experience outcome measure based on HCAHPS survey dimensions.[2] Together, these 2 metrics will comprise nearly half of the total VBP performance score used to determine reimbursement. Although our data suggest that these 2 VBP metrics may be correlated, it should be noted that we measured inpatient hospital charges, whereas the CMS efficiency outcome measure includes costs for episode of care spanning 3 days prior to hospitalization to 30 days after hospitalization.

Patient expectations likely play a role in satisfaction.[14, 15, 16] In an outpatient setting, physician fulfillment of patient requests has been associated with positive patient evaluations of care.[17] However, patients appear to value education, shared decision making, and provider empathy more than testing and intervention.[14, 18, 19, 20, 21, 22, 23] Perhaps, in the absence of the former attributes, patients use additional resource expenditure as a proxy.

It is not clear that higher resource expenditure improves outcomes. A landmark study of nearly 1 million Medicare enrollees by Fisher et al. suggests that, although Medicare patients in higher‐spending regions receive more care than those in lower‐spending regions, this does not result in better health outcomes, specifically with regard to mortality.[24, 25] Patients who live in areas of high hospital capacity use the hospital more frequently than do patients in areas of low hospital capacity, but this does not appear to result in improved mortality rates.[26] In fact, physicians in areas of high healthcare capacity report more difficulty maintaining high‐quality patient relationships and feel less able to provide high‐quality care than physicians in lower‐capacity areas.[27]

We hypothesize the cause of the association between resource utilization and patient satisfaction could be that patients (1) perceive that a doctor who allows them to stay longer in the hospital or who performs additional testing cares more about their well‐being and (2) that these patients feel more strongly that their concerns are being heard and addressed by their physicians. A systematic review of primary care patients identified many studies that found a positive association between meeting patient expectations and satisfaction with care, but also suggested that although patients frequently expect information, physicians misperceive this as an expectation of specific action.[28] A separate systematic review found that patient education in the form of decision aides can help patients develop more reasonable expectations and reduce utilization of certain discretionary procedures such as elective surgeries and prostate‐specific antigen testing.[29]

We did not specifically address clinical outcomes in our analysis because the clinical outcomes on which CMS currently adjusts VBP reimbursement focus on 30‐day mortality for specific diagnoses, nosocomial infections, and iatrogenic events.[30] Our data include only returned surveys from living patients, and it is likely that 30‐day mortality was similar throughout all subsets of patients. Additionally, the nosocomial and iatrogenic outcome measures used by CMS are sufficiently rare on the general floors and are unlikely to have significantly influenced our results.[31]

Our study has several strengths. Nearly all medical and surgical patient surveys returned during the study period were included, and therefore our calculations are likely to accurately reflect the Summary Star Rating that would have been assigned for the period. Second, the large sample size helps attenuate potential differences in commonly used outcome metrics. Third, by adjusting for a variety of demographic and clinical variables, we were able to decrease the likelihood of unidentified confounders.

Notably, we identified 38 (0.4%) surveys returned for patients under 18 years of age at admission. These surveys were included in our analysis because, to the best of our knowledge, they would have existed in the pool of surveys CMS could have used to assign a Summary Star Rating.

Our study also has limitations. First, geographically diverse data are needed to ensure generalizability. Second, we used the Charlson‐Deyo Comorbidity Index to describe the degree of illness for each patient. This index represents a patient's total illness burden but may not describe the relative severity of the patient's current illness relative to another patient. Third, we selected variables we felt were most likely to be associated with patient experience, but unidentified confounding remains possible. Fourth, attendings caring for ICU patients fall within the Division of Critical Care/Pulmonary Medicine. Therefore, we may have inadvertently placed patients into the ICU cohort who received a pulmonary/critical care consult on the general floors. Fifth, our data describe associations only for patients who returned surveys. Although there may be inherent biases in patients who return surveys, HCAHPS survey responses are used by CMS to determine a hospital's overall satisfaction score.

CONCLUSION

For patients who return HCAHPS surveys, resource utilization may be positively associated with a hospital's Summary Star Rating. These data suggest that hospitals with higher per‐patient expenditures may receive higher Summary Star Ratings, which could result in hospitals with higher per‐patient resource utilization appearing more attractive to healthcare consumers. Future studies should attempt to confirm our findings at other institutions and to determine causative factors.

Acknowledgements

The authors thank Jason Machan, PhD (Department of Orthopedics and Surgery, Warren Alpert Medical School, Brown University, Providence, Rhode Island) for his help with study design, and Ms. Brenda Foster (data analyst, University of Rochester Medical Center, Rochester, NY) for her help with data collection.

Disclosures: Nothing to report.

The patient experience has become increasingly important to healthcare in the United States. It is now a metric used commonly to determine physician compensation and accounts for nearly 30% of the Centers for Medicare and Medicaid Services' (CMS) Value‐Based Purchasing (VBP) reimbursement for fiscal years 2015 and 2016.[1, 2]

In April 2015, CMS added a 5‐star patient experience score to its Hospital Compare website in an attempt to address the Affordable Care Act's call for transparent and easily understandable public reporting.[3] A hospital's principal score is the Summary Star Rating, which is based on responses to the Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) survey. The formulas used to calculate Summary Star Ratings have been reported by CMS.[4]

Studies published over the past decade suggest that gender, age, education level, length of hospital stay, travel distance, and other factors may influence patient satisfaction.[5, 6, 7, 8] One study utilizing a national dataset suggested that higher patient satisfaction was associated with greater inpatient healthcare utilization and higher healthcare expenditures.[9] It is therefore possible that emphasizing patient experience scores could adversely impact healthcare resource utilization. However, positive patient experience may also be an important independent dimension of quality for patients and correlate with improved clinical outcomes.[10]

We know of no literature describing patient factors associated with the Summary Star Rating. Given that this rating is now used as a standard metric by which patient experience can be compared across more than 3,500 hospitals,[11] data describing the association between patient‐level factors and the Summary Star Rating may provide hospitals with an opportunity to target improvement efforts. We aimed to determine the degree to which resource utilization is associated with a satisfaction score based on the Summary Star Rating methodology.

METHODS

The study was conducted at the University of Rochester Medical Center (URMC), an 830‐bed tertiary care center in upstate New York. This was a retrospective review of all HCAHPS surveys returned to URMC over a 27‐month period from January 1, 2012 to April 1, 2014. URMC follows the standard CMS process for determining which patients receive surveys as follows. During the study timeframe, HCAHPS surveys were mailed to patients 18 years of age and older who had an inpatient stay spanning at least 1 midnight. Surveys were mailed within 5 days of discharge, and were generally returned within 6 weeks. URMC did not utilize telephone or email surveys during the study period. Surveys were not sent to patients who (1) were transferred to another facility, (2) were discharged to hospice, (3) died during the hospitalization, (4) received psychiatric or rehabilitative services during the hospitalization, (5) had an international address, and/or (6) were prisoners.

The survey vendor (Press Ganey, South Bend, IN) for URMC provided raw data for returned surveys with patient answers to questions. Administrative and billing databases were used to add demographic and clinical data for the corresponding hospitalization to the dataset. These data included age, gender, payer status (public, private, self, charity), length of stay, number of attendings who saw the patient (based on encounters documented in the electronic medical record (EMR)), all discharge International Classification of Diseases, 9th Revision (ICD‐9) diagnoses for the hospitalization, total charges for the hospitalization, and intensive care unit (ICU) utilization as evidenced by a documented encounter with a member of the Division of Critical Care/Pulmonary Medicine.

CMS analyzes surveys within 1 of 3 clinical service categories (medical, surgical, or obstetrics/gynecology) based on the discharging service. To parallel this approach, each returned survey was placed into 1 of these categories based on the clinical service of the discharging physician. Patients placed in the obstetrics/gynecology category (n = 1317, 13%) will be analyzed in a future analysis given inherent differences in patient characteristics that require evaluation of other variables.

Approximations of CMS Summary Star Rating

The HCAHPS survey is a multiple‐choice questionnaire that includes several domains of patient satisfaction. Respondents are asked to rate areas of satisfaction with their hospital experience on a Likert scale. CMS uses a weighted average of Likert responses to a subset of HCAHPS questions to calculate a hospital's raw score in 11 domains, as well as an overall raw summary score. CMS then adjusts each raw score for differences between hospitals (eg, clustering, improvement over time, method of survey) to determine a hospital's star rating in each domain and an overall Summary Star Rating (the Summary Star Rating is the primary factor by which consumers can compare hospitals).[4] Because our data were from a single hospital system, the between‐hospital scoring adjustments utilized by CMS were not applicable. Instead, we calculated the raw scores exactly as CMS does prior to the adjustments. Thus, our scores reflect the scores that CMS would have given URMC during the study period prior to standardized adjustments; we refer to this as the raw satisfaction rating (RSR). We calculated an RSR for every eligible survey. The RSR was calculated as a continuous variable from 0 (lowest) to 1 (highest). Detailed explanation of our RSR calculation is available in the Supporting Information in the online version of this article.

Statistical Analysis

All analyses were performed in aggregate and by service (medical vs surgical). Categorical variables were summarized using frequencies with percentages. Comparisons across levels of categorical variables were performed with the ² test. We report bivariate associations between the independent variables and RSRs in the top decile using unadjusted odds ratios (ORs) with 95% confidence intervals (CIs). Similarly, multivariable logistic regression was used for adjusted analyses. For the variables of severity of illness and resource intensity, the group with the lowest illness severity and lowest resource use served as the reference groups. We modeled patients without an ICU encounter and with an ICU encounter separately.

Charges, number of unique attendings encountered, and lengths of stay were highly correlated, and likely various measures of the same underlying construct of resource intensity, and therefore could not be entered into our models simultaneously. We combined these into a resource intensity score using factor analysis with a varimax rotation, and extracted factor scores for a single factor (supported by a scree plot). We then placed patients into 4 groups based on the distribution of the factor scores: low (25th percentile), moderate (25th50th percentile), major (50th75th percentile), and extreme (>75th percentile).

We used the Charlson‐Deyo comorbidity score as our disease severity index.[12] The index uses ICD‐9 diagnoses with points assigned for the impact of each diagnosis on morbidity and the points summed to an overall score. This provides a measure of disease severity for a patient based on the number of diagnoses and relative mortality of the individual diagnoses. Scores were categorized as 0 (representing no major illness burden), 1 to 3, 4 to 6, and >6.

All statistical analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC), and P values 0.05 were considered statistically significant. This study was approved by the institutional review board at the University of Rochester Medical Center.

RESULTS

Our initial search identified 10,007 returned surveys (29% of eligible patients returned surveys during the study period). Of these, 5059 (51%) were categorized as medical, 3630 (36%) as surgical, and 1317 (13%) as obstetrics/gynecology. One survey did not have the service of the discharging physician recorded and was excluded. Cohort demographics and relationship to RSRs in the top decile for the 8689 medical and surgical patients can be found in Table 1. The most common discharge diagnosis‐related groups (DRGs) for medical patients were 247, percutaneous cardiovascular procedure with drug‐eluding stent without major complications or comorbidities (MCC) (3.8%); 871, septicemia or severe sepsis without mechanical ventilation >96 hours with MCC (2.7%); and 392, esophagitis, gastroenteritis, and miscellaneous digestive disorders with MCC (2.3%). The most common DRGs for surgical patients were 460, spinal fusion except cervical without MCC (3.5%); 328, stomach, esophageal and duodenal procedure without complication or comorbidities or MCC (3.3%); and 491, back and neck procedure excluding spinal fusion without complication or comorbidities or MCC (3.1%).

Table 1.Cohort Demographics and Raw Satisfaction Ratings in the Top Decile

	Overall				Medical				Surgical
	Total	90th	Top Decile	P	Total	90th	Top Decile	P	Total	90th	Top Decile	P
NOTE: Data are presented as no. (%). Percentages may not add up to 100 due to rounding. Abbreviations: ICU, intensive care unit. *Calculated using the Charlson‐Deyo index; smaller values indicate less severity. Low = $10,000; medium = $10,000$40,000; high = >$40,000.
Overall	8,689	7,789 (90)	900 (10)		5,059	4,646 (92)	413 (8)		3,630	3,143 (87)	487 (13)
Age, y
30	419 (5)	371 (89)	48 (12)	0.001	218 (4)	208 (95)	10 (5)	0.001	201 (6)	163 (81)	38 (19)	0.001
3049	1,029 (12)	902 (88)	127 (12)		533 (11)	482 (90)	51 (10)		496 (14)	420 (85)	76 (15)
5069	3,911 (45)	3,450 (88)	461 (12)		2,136 (42)	1,930 (90)	206 (10)		1,775 (49)	1,520 (86)	255 (14)
>69	3,330 (38)	3,066 (92)	264 (8)		2,172 (43)	2,026 (93)	146 (7)		1,158 (32)	1,040 (90)	118 (10)
Gender
Male	4,640 (53)	4,142 (89)	498 (11)	0.220	2,596 (51)	2,379 (92)	217 (8)	0.602	2,044 (56)	1,763 (86)	281 (14)	0.506
Female	4,049 (47)	3,647 (90)	402 (10)		2,463 (49)	2,267 (92)	196 (8)		1,586 (44)	1,380 (87)	206 (13)
ICU encounter
No	7,122 (82)	6,441 (90)	681 (10)	0.001	4,547 (90)	4,193 (92)	354 (8)	0.001	2,575 (71)	2,248 (87)	327 (13)	0.048
Yes	1,567 (18)	1,348 (86)	219 (14)		512 (10)	453 (89)	59 (12)		1,055 (29)	895 (85)	160 (15)
Payer
Public	5,564 (64)	5,036 (91)	528 (10)	0.001	3,424 (68)	3,161 (92)	263 (8)	0.163	2,140 (59)	1,875 (88)	265 (12)	0.148
Private	3,064 (35)	2,702 (88)	362 (12)		1,603 (32)	1,458 (91)	145 (9)		1,461 (40)	1,244 (85)	217 (15)
Charity	45 (1)	37 (82)	8 (18)		25 (1)	21 (84)	4 (16)		20 (1)	16 (80)	4 (20)
Self	16 (0)	14 (88)	2 (13)		7 (0)	6 (86)	1 (14)		9 (0)	8 (89)	1 (11)
Length of stay, d
3	3,156 (36)	2,930 (93)	226 (7)	0.001	1,961 (39)	1,865 (95)	96 (5)	0.001	1,195 (33)	1,065 (89)	130 (11)	0.001
36	3,330 (38)	2,959 (89)	371 (11)		1,867 (37)	1,702 (91)	165 (9)		1,463 (40)	1,257 (86)	206 (14)
>6	2,203 (25)	1,900 (86)	303 (14)		1,231 (24)	1,079 (88)	152 (12)		972 (27)	821 (85)	151 (16)
No. of attendings
4	3,959 (46)	3,615 (91)	344 (9)	0.001	2,307 (46)	2,160 (94)	147 (6)	0.001	1,652 (46)	1,455 (88)	197 (12)	0.052
46	3,067 (35)	2,711 (88)	356 (12)		1,836 (36)	1,663 (91)	173 (9)		1,231 (34)	1,048 (85)	183 (15)
>6	1,663 (19)	1,463 (88)	200 (12)		916 (18)	823 (90)	93 (10)		747 (21)	640 (86)	107 (14)
Severity index*
0 (lowest)	2,812 (32)	2,505 (89)	307 (11)	0.272	1,273 (25)	1,185 (93)	88 (7)	0.045	1,539 (42)	1,320 (86)	219 (14)	0.261
13	4,253 (49)	3,827 (90)	426 (10)		2,604 (52)	2,395 (92)	209 (8)		1,649 (45)	1,432 (87)	217 (13)
46	1163 (13)	1,052 (91)	111 (10)		849 (17)	770 (91)	79 (9)		314 (9)	282 (90)	32 (10)
>6 (highest)	461 (5)	405 (88)	56 (12)		333 (7)	296 (89)	37 (11)		128 (4)	109 (85)	19 (15)
Charges,
Low	1,820 (21)	1,707 (94)	113 (6)	0.001	1,426 (28)	1,357 (95)	69 (5)	0.001	394 (11)	350 (89)	44 (11)	0.007
Medium	5,094 (59)	4,581 (90)	513 (10)		2,807 (56)	2,582 (92)	225 (8)		2,287 (63)	1,999 (87)	288 (13)
High	1,775 (20)	1,501 (85)	274 (15)		826 (16)	707 (86)	119 (14)		949 (26)	794 (84)	155 (16)

Unadjusted analysis of medical and surgical patients identified significant associations of several variables with a top decile RSR (Table 2). Patients with longer lengths of stay (OR: 2.07, 95% CI: 1.72‐2.48), more attendings (OR: 1.44, 95% CI: 1.19‐1.73), and higher hospital charges (OR: 2.76, 95% CI: 2.19‐3.47) were more likely to report an RSR in the top decile. Patients without an ICU encounter (OR: 0.65, 95% CI: 0.55‐0.77) and on a medical service (OR: 0.57, 95% CI: 0.5‐ 0.66) were less likely to report an RSR in the top decile. Several associations were identified in only the medical or surgical cohorts. In the medical cohort, patients with the highest illness severity index (OR: 1.68, 95% CI: 1.12‐ 2.52) and with 7 different attending physicians (OR: 1.66, 95% CI: 1.27‐2.18) were more likely to report RSRs in the top decile. In the surgical cohort, patients 30 years of age (OR: 2.05, 95% CI 1.38‐3.07) were more likely to report an RSR in the top decile than patients >69 years of age. Insurance payer category and gender were not significantly associated with top decile RSRs.

Table 2.Bivariate Comparisons of Associations Between Top Decile Satisfaction Ratings and Reference Levels

	Overall		Medical		Surgical
	Odds Ratio (95% CI)	P	Odds Ratio (95% CI)	P	Odds Ratio (95% CI)	P
NOTE: Abbreviations: CI, confidence interval; ICU, intensive care unit; Ref, reference. *Calculated using the Charlson‐Deyo index; smaller values indicate less severity. Low = $10,000; medium = $10,000$40,000; high = >$40,000.
Age, y
30	1.5 (1.082.08)	0.014	0.67 (0.351.29)	0.227	2.05 (1.383.07)	0.001
3049	1.64 (1.312.05)	.001	1.47 (1.052.05)	0.024	1.59 (1.172.17)	0.003
5069	1.55 (1.321.82)	.001	1.48 (1.191.85)	0.001	1.48 (1.171.86)	0.001
>69	Ref		Ref		Ref
Gender
Male	1.09 (0.951.25)	0.220	1.06 (0.861.29)	0.602	1.07 (0.881.3)	0.506
Female	Ref		Ref		Ref
ICU encounter
No	0.65 (0.550.77)	0.001	0.65 (0.480.87)	0.004	0.81 (0.661)	0.048
Yes	Ref		Ref		Ref
Payer
Public	0.73 (0.173.24)	0.683	0.5 (0.064.16)	0.521	1.13 (0.149.08)	0.908
Private	0.94 (0.214.14)	0.933	0.6 (0.074.99)	0.634	1.4 (0.1711.21)	0.754
Charity	1.51 (0.298.02)	0.626	1.14 (0.1112.25)	0.912	2 (0.1920.97)	0.563
Self	Ref		Ref		Ref
Length of stay, d
3	Ref		Ref		Ref
36	1.63 (1.371.93)	0.001	1.88 (1.452.44)	0.001	1.34 (1.061.7)	0.014
>6	2.07 (1.722.48)	0.001	2.74 (2.13.57)	0.001	1.51 (1.171.94)	0.001
No. of attendings
4	Ref		Ref		Ref
46	1.38 (1.181.61)	0.001	1.53 (1.221.92)	0.001	1.29 (1.041.6)	0.021
>6	1.44 (1.191.73)	0.001	1.66 (1.272.18)	0.001	1.23 (0.961.59)	0.102
Severity index*
0 (lowest)	Ref		Ref		Ref
13	0.91 (0.781.06)	0.224	1.18 (0.911.52)	0.221	0.91 (0.751.12)	0.380
46	0.86 (0.681.08)	0.200	1.38 (1.011.9)	0.046	0.68 (0.461.01)	0.058
>6 (highest)	1.13 (0.831.53)	0.436	1.68 (1.122.52)	0.012	1.05 (0.631.75)	0.849
Charges
Low	Ref		Ref		Ref
Medium	1.69 (1.372.09)	0.001	1.71 (1.32.26)	0.001	1.15 (0.821.61)	0.428
High	2.76 (2.193.47)	0.001	3.31 (2.434.51)	0.001	1.55 (1.092.22)	0.016
Service
Medical	0.57 (0.50.66)	0.001
Surgical	Ref

Multivariable modeling (Table 3) for all patients without an ICU encounter suggested that (1) patients aged 30 years, 30 to 49 years, and 50 to 69 years were more likely to report top decile RSRs when compared to patients 70 years and older (OR: 1.61, 95% CI: 1.09‐2.36; OR: 1.44, 95% CI: 1.08‐1.93; and OR: 1.39, 95% CI: 1.13‐1.71, respectively) and (2), when compared to patients with extreme resource intensity scores, patients with higher resource intensity scores were more likely to report top decile RSRs (moderate [OR: 1.42, 95% CI: 1.11‐1.83], major [OR: 1.56, 95% CI: 1.22‐2.01], and extreme [OR: 2.29, 95% CI: 1.8‐2.92]. These results were relatively consistent within medical and surgical subgroups (Table 3).

Table 3.Multivariable Logistic Regression Model for Top Decile Raw Satisfaction Ratings for Patients on the General Wards*

	Overall		Medical		Surgical
	Odds Ratio (95% CI)	P	Odds Ratio (95% CI)	P	Odds Ratio (95% CI)	P
NOTE: Abbreviations: CI, confidence interval; Ref, reference. *Excludes the 1,567 patients who had an intensive care unit encounter. Calculated using the Charlson‐Deyo index. Component variables include length of stay, number of attendings, and charges
Age, y
30	1.61 (1.092.36)	0.016	0.82 (0.41.7)	0.596	2.31 (1.393.82)	0.001
3049	1.44 (1.081.93)	0.014	1.55 (1.032.32)	0.034	1.41 (0.912.17)	0.120
5069	1.39 (1.131.71)	0.002	1.44 (1.11.88)	0.008	1.39 (11.93)	0.049
>69	Ref		Ref		Ref
Sex
Male	1 (0.851.17)	0.964	1 (0.81.25)	0.975	0.99 (0.791.26)	0.965
Female	Ref		Ref		Ref
Payer
Public	0.62 (0.142.8)	0.531	0.42 (0.053.67)	0.432	1.03 (0.128.59)	0.978
Private	0.67 (0.153.02)	0.599	0.42 (0.053.67)	0.434	1.17 (0.149.69)	0.884
Charity	1.54 (0.288.41)	0.620	1 (0.0911.13)	0.999	2.56 (0.2328.25)	0.444
Self	Ref		Ref		Ref
Severity index
0 (lowest)	Ref		Ref		Ref
13	1.07 (0.891.29)	0.485	1.18 (0.881.58)	0.267	1 (0.781.29)	0.986
46	1.14 (0.861.51)	0.377	1.42 (0.992.04)	0.056	0.6 (0.331.1)	0.100
>6 (highest)	1.31 (0.911.9)	0.150	1.47 (0.932.33)	0.097	1.1 (0.542.21)	0.795
Resource intensity score
Low	Ref		Ref		Ref
Moderate	1.42 (1.111.83)	0.006	1.6 (1.112.3)	0.011	0.94 (0.661.34)	0.722
Major	1.56 (1.222.01)	0.001	1.69 (1.182.43)	0.004	1.28 (0.911.8)	0.151
Extreme	2.29 (1.82.92)	0.001	2.72 (1.943.82)	0.001	1.63 (1.172.26)	0.004
Service
Medical	0.59 (0.50.69)	0.001
Surgical	Ref

In those with at least 1 ICU attending encounter (see Supporting Table 1 in the online version of this article), no variables demonstrated significant association with top decile RSRs in the overall group or in the medical subgroup. For surgical patients with at least 1 ICU attending encounter (see Supporting Table 1 in the online version of this article), patients aged 30 to 49 and 50 to 69 years were more likely to provide top decile RSRs (OR: 1.93, 95% CI: 1.08‐3.46 and OR: 1.65, 95% CI 1.07‐2.53, respectively). Resource intensity was not significantly associated with top decile RSRs.

DISCUSSION

Our analysis suggests that, for patients on the general care floors, resource utilization is associated with the RSR and, therefore, potentially the CMS Summary Star Rating. Adjusting for severity of illness, patients with higher resource utilization were more likely to report top decile RSRs.

Prior data regarding utilization and satisfaction are mixed. In a 2‐year, prospective, national examination, patients in the highest quartile of patient satisfaction had increased healthcare and prescription drug expenditures as well as increased rates of hospitalization when compared with patients in the lowest quartile of patient satisfaction.[9] However, a recent national study of surgical administrative databases suggested hospitals with high patient satisfaction provided more efficient care.[13]

One reason for the conflicting data may be that large, national evaluations are unable to control for between‐hospital confounders (ie, hospital quality of care). By capturing all eligible returned surveys at 1 institution, our design allowed us to collect granular data. We found that in 1 hospital setting, patient population, facilities, and food services, patients receiving more clinical resources generally assigned higher ratings than patients receiving less.

It is possible that utilization is a proxy for serious illness, and that patients with serious illness receive more attention during hospitalization and are more satisfied when discharged in a good state of health. However, we did adjust for severity of illness in our model using the Charlson‐Deyo index and we suggest that, other factors being equal, hospitals with higher per‐patient expenditures may be assigned higher Summary Star Ratings.

CMS has recently implemented a number of metrics designed to decrease healthcare costs by improving quality, safety, and efficiency. Concurrently, CMS has also prioritized patient experience. The Summary Star Rating was created to provide healthcare consumers with an easy way to compare the patient experience between hospitals[4]; however, our data suggest that this metric may be at odds with inpatient cost savings and efficiency metrics.

Per‐patient spending becomes particularly salient when considering that in fiscal year 2016, CMS' hospital VBP reimbursement will include 2 metrics: an efficiency outcome measure labeled Medicare spending per beneficiary, and a patient experience outcome measure based on HCAHPS survey dimensions.[2] Together, these 2 metrics will comprise nearly half of the total VBP performance score used to determine reimbursement. Although our data suggest that these 2 VBP metrics may be correlated, it should be noted that we measured inpatient hospital charges, whereas the CMS efficiency outcome measure includes costs for episode of care spanning 3 days prior to hospitalization to 30 days after hospitalization.

Patient expectations likely play a role in satisfaction.[14, 15, 16] In an outpatient setting, physician fulfillment of patient requests has been associated with positive patient evaluations of care.[17] However, patients appear to value education, shared decision making, and provider empathy more than testing and intervention.[14, 18, 19, 20, 21, 22, 23] Perhaps, in the absence of the former attributes, patients use additional resource expenditure as a proxy.

It is not clear that higher resource expenditure improves outcomes. A landmark study of nearly 1 million Medicare enrollees by Fisher et al. suggests that, although Medicare patients in higher‐spending regions receive more care than those in lower‐spending regions, this does not result in better health outcomes, specifically with regard to mortality.[24, 25] Patients who live in areas of high hospital capacity use the hospital more frequently than do patients in areas of low hospital capacity, but this does not appear to result in improved mortality rates.[26] In fact, physicians in areas of high healthcare capacity report more difficulty maintaining high‐quality patient relationships and feel less able to provide high‐quality care than physicians in lower‐capacity areas.[27]

We hypothesize the cause of the association between resource utilization and patient satisfaction could be that patients (1) perceive that a doctor who allows them to stay longer in the hospital or who performs additional testing cares more about their well‐being and (2) that these patients feel more strongly that their concerns are being heard and addressed by their physicians. A systematic review of primary care patients identified many studies that found a positive association between meeting patient expectations and satisfaction with care, but also suggested that although patients frequently expect information, physicians misperceive this as an expectation of specific action.[28] A separate systematic review found that patient education in the form of decision aides can help patients develop more reasonable expectations and reduce utilization of certain discretionary procedures such as elective surgeries and prostate‐specific antigen testing.[29]

We did not specifically address clinical outcomes in our analysis because the clinical outcomes on which CMS currently adjusts VBP reimbursement focus on 30‐day mortality for specific diagnoses, nosocomial infections, and iatrogenic events.[30] Our data include only returned surveys from living patients, and it is likely that 30‐day mortality was similar throughout all subsets of patients. Additionally, the nosocomial and iatrogenic outcome measures used by CMS are sufficiently rare on the general floors and are unlikely to have significantly influenced our results.[31]

Our study has several strengths. Nearly all medical and surgical patient surveys returned during the study period were included, and therefore our calculations are likely to accurately reflect the Summary Star Rating that would have been assigned for the period. Second, the large sample size helps attenuate potential differences in commonly used outcome metrics. Third, by adjusting for a variety of demographic and clinical variables, we were able to decrease the likelihood of unidentified confounders.

Notably, we identified 38 (0.4%) surveys returned for patients under 18 years of age at admission. These surveys were included in our analysis because, to the best of our knowledge, they would have existed in the pool of surveys CMS could have used to assign a Summary Star Rating.

Our study also has limitations. First, geographically diverse data are needed to ensure generalizability. Second, we used the Charlson‐Deyo Comorbidity Index to describe the degree of illness for each patient. This index represents a patient's total illness burden but may not describe the relative severity of the patient's current illness relative to another patient. Third, we selected variables we felt were most likely to be associated with patient experience, but unidentified confounding remains possible. Fourth, attendings caring for ICU patients fall within the Division of Critical Care/Pulmonary Medicine. Therefore, we may have inadvertently placed patients into the ICU cohort who received a pulmonary/critical care consult on the general floors. Fifth, our data describe associations only for patients who returned surveys. Although there may be inherent biases in patients who return surveys, HCAHPS survey responses are used by CMS to determine a hospital's overall satisfaction score.

CONCLUSION

For patients who return HCAHPS surveys, resource utilization may be positively associated with a hospital's Summary Star Rating. These data suggest that hospitals with higher per‐patient expenditures may receive higher Summary Star Ratings, which could result in hospitals with higher per‐patient resource utilization appearing more attractive to healthcare consumers. Future studies should attempt to confirm our findings at other institutions and to determine causative factors.

Acknowledgements

The authors thank Jason Machan, PhD (Department of Orthopedics and Surgery, Warren Alpert Medical School, Brown University, Providence, Rhode Island) for his help with study design, and Ms. Brenda Foster (data analyst, University of Rochester Medical Center, Rochester, NY) for her help with data collection.

Disclosures: Nothing to report.

Imatinib mesylate (IM) represents the first-line treatment of chronic myeloid leukemia (CML) and gastrointestinal stromal tumors (GISTs). Its pharmacological activity is related to a specific action on several tyrosine kinases in different tumors, including Bcr-Abl in CML, c-Kit (CD117) in GIST, and platelet-derived growth factor receptor in dermatofibrosarcoma protuberans.^1,2

Imatinib mesylate has been shown to improve progression-free survival and overall survival²; however, it also has several side effects. Among the adverse effects (AEs), less than 10% are nonhematologic, such as nausea, vomiting, diarrhea, muscle cramps, and cutaneous reactions.^3,4

We followed patients who were treated with IM for 5 years to identify AEs of therapy.

Methods

The aim of this prospective study was to identify and collect data regarding IM cutaneous side effects so that clinicians can detect AEs early and differentiate them from AEs caused by other medications. All patients were subjected to a median of 5 years’ follow-up. We included all the patients treated with IM and excluded patients who had a history of eczematous dermatitis, psoriasis, renal impairment, or dyshidrosis palmoplantar. Before starting IM, all patients presented for a dermatologic visit. They were subsequently evaluated every 3 months.

The incidence rate was defined as the ratio of patients with cutaneous side effects and the total patients treated with IM. Furthermore, we calculated the ratio between each class of patient with a specific cutaneous manifestation and the entire cohort of patients with cutaneous side effects related to IM.

When necessary, microbiological, serological, and histopathological analyses were performed.

Results

In 60 months, we followed 220 patients treated with IM. Among them, 55 (25%) developed cutaneous side effects (35 males; 20 females). The incidence rate of the patients with cutaneous side effects was 1:4. The median age of the entire cohort was 52.5 years. Fifty patients were being treated for CML and 5 for GISTs. All patients received IM at a dosage of 400 mg daily.

The following skin diseases were observed in patients treated with IM (Table): 19 patients with maculopapular rash with pruritus (no maculopapular rash without pruritus was detected), 7 patients with eczematous dermatitis such as stasis dermatitis and seborrheic dermatitis, 6 patients with onychodystrophy melanonychia (Figure 1), 5 patients with psoriasis, 5 patients with skin cancers including basal cell carcinoma (BCC)(Figure 2), 3 patients with periorbital edema (Figure 3), 3 patients with mycosis, 3 patients with dermatofibromas, 2 patients with dyshidrosis palmoplantar, 1 patient with pityriasis rosea–like eruption (Figure 4), and 1 patient with actinic keratoses on the face. No hypopigmentation or hyperpigmentation, excluding the individual case of melanonychia, was observed.

All cutaneous diseases reported in this study appeared after IM therapy (median, 3.8 months). The median time to onset for each cutaneous disorder is reported in the Table. During the first dermatologic visit before starting IM therapy, none of the patients showed any of these cutaneous diseases.

The adverse cutaneous reactions were treated with appropriate drugs. Generally, eczematous dermatitis was treated using topical steroids, emollients, and oral antihistamines. In patients with maculopapular rash with pruritus, oral corticosteroids (eg, betamethasone 3 mg daily or prednisolone 1 mg/kg) in association with antihistamine was necessary. Psoriasis was completely improved with topical betamethasone 0.5 mg and calcipotriol 50 µg. Skin cancers were treated with surgical excision with histologic examination. All treatments are outlined in the Table.

Imatinib mesylate therapy was suspended in 2 patients with maculopapular rash with moderate to severe pruritus; however, despite the temporary suspension of the drug and the appropriate therapies (corticosteroids and antihistamines), cutaneous side effects reappeared 7 to 10 days after therapy resumed. Therefore, the treatment was permanently suspended in these 2 cases and IM was replaced with erlotinib, a second-generation Bcr-Abl tyrosine kinase inhibitor.

Comment

The introduction of IM for the treatment of GIST and CML has changed the history of these diseases. The drug typically is well tolerated and few patients have reported severe AEs. Mild skin reactions are relatively frequent, ranging from 7% to 21% of patients treated.³ In our case, the percentage was relatively higher (25%), likely because of close monitoring of patients, with an increase in the incidence rate.

Imatinib mesylate cutaneous reactions are dose dependent.⁴ Indeed, in all our cases, the cutaneous reactions arose with an IM dosage of 400 mg daily, which is compatible with the definition of dose-independent cutaneous AEs.

 

The most common cutaneous AEs reported in the literature were swelling/edema and maculopapular rash. Swelling is the most common AE described during therapy with IM with an incidence of 63% to 84%.⁵ Swelling often involves the periorbital area and occurs approximately 6 weeks after starting IM. Although its pathogenesis is uncertain, it has been shown that IM blocks the platelet-derived growth factor receptor expressed on blood vessels that regulates the transportation transcapillary. The inhibition of this receptor can lead to increased pore pressure, resulting in edema and erythema. Maculopapular eruptions (50% of cases) often affect the trunk and the limbs and are accompanied by pruritus. Commonly, these rashes arise after 9 weeks of IM therapy. These eruptions are self-limiting and only topical emollients and steroids are required, without any change in IM schedule. To treat maculopapular eruptions with pruritus, oral steroids and antihistamines may be helpful, without suspending IM treatment. When grade 2 or 3 pruriginous maculopapular eruptions arise, the suspension of IM combined with steroids and antihistamines may be necessary. When the readministration of IM is required, it is mandatory to start IM at a lower dose (50–100 mg/d), administering prednisolone 0.5 to 1.0 mg/kg daily. Then, the steroid gradually can be tapered.⁶ Critical cutaneous AEs that are resistant to supportive measures warrant suspension of IM therapy. However, the incidence of this event is small (<1% of all patients).⁷

Regarding severe cutaneous AEs from IM therapy, Hsiao et al⁸ reported the case of Stevens-Johnson syndrome. In this case, IM was immediately stopped and systemic steroids were started. Rarely, erythroderma (grade 4 toxicity) can develop for which a prompt and perpetual suspension of IM is necessary and supportive care therapy with oral and topical steroids is recommended.⁹

Hyperpigmentation induced by IM, mostly in patients with Fitzpatrick skin types V to VI and with a general prevalence of 16% to 40% in treated patients, often is related to a mutation of c-Kit or other kinases that are activated rather than inhibited by the drug, resulting in overstimulation of melanogenesis.¹⁰ The prevalence of Fitzpatrick skin types I to III determined the absence of pigmentation changes in our cohort, excluding melanonychia. Hyperpigmentation was observed in the skin as well as the appendages such as nails, resulting in melanonychia (Figure 1). However, Brazzelli et al¹¹ reported hypopigmentation in 5 white patients treated with IM; furthermore, they found a direct correlation between hypopigmentation and development of skin cancers in these patients. The susceptibility to develop skin cancers may persist, even without a clear manifestation of hypopigmentation, as reported in the current analysis. We documented BCC in 5 patients, 1 patient developed actinic keratoses, and 3 patients developed dermatofibromas. However, these neoplasms probably were not provoked by IM. On the contrary, we did not note squamous cell carcinoma, which was reported by Baskaynak et al¹² in 2 CML patients treated with IM.

The administration of IM can be associated with exacerbation of psoriasis. Paradoxically, in genetically predisposed individuals, tumor necrosis factor α (TNF-α) antagonists, such as IM, seem to induce psoriasis, producing IFN-α rather than TNF-α and increasing inflammation.¹³ In fact, some research shows induction of psoriasis by anti–TNF-α drugs.^14-16 Two cases of IM associated with psoriasis have been reported, and both cases represented an exacerbation of previously diagnosed psoriasis.^13,17 On the contrary, in our analysis we reported 5 cases of psoriasis vulgaris induced by IM administration. Our patients developed cutaneous psoriatic lesions approximately 1.7 months after the start of IM therapy.

The pityriasis rosea–like eruption (Figure 4) presented as nonpruritic, erythematous, scaly patches on the trunk and extremities, and arose 3.6 months after the start of treatment. This particular cutaneous AE is rare. In 3 case reports, the IM dosage also was 400 mg daily.^18-20 The pathophysiology of this rare skin reaction stems from the pharmacological effect of IM rather than a hypersensitivity reaction.¹⁸

Deininger et al⁷ reported that patients with a high basophil count (>20%) rarely show urticarial eruptions after IM due to histamine release from basophils. Premedication with an antihistamine was helpful and the urticarial eruption resolved after normalization in basophil count.⁷

Given the importance of IM for patients who have limited therapeutic alternatives for their disease and the ability to safely treat the cutaneous AEs, as demonstrated in our analysis, the suspension of IM for dermatological complications is necessary only in rare cases, as shown by the low number of patients (n=2) who had to discontinue therapy. The cutaneous AEs should be diagnosed and treated early with less impact on chemotherapy treatments. The administration of IM should involve a coordinated effort among oncologists and dermatologists to prevent important complications.

Imatinib mesylate (IM) represents the first-line treatment of chronic myeloid leukemia (CML) and gastrointestinal stromal tumors (GISTs). Its pharmacological activity is related to a specific action on several tyrosine kinases in different tumors, including Bcr-Abl in CML, c-Kit (CD117) in GIST, and platelet-derived growth factor receptor in dermatofibrosarcoma protuberans.^1,2

Imatinib mesylate has been shown to improve progression-free survival and overall survival²; however, it also has several side effects. Among the adverse effects (AEs), less than 10% are nonhematologic, such as nausea, vomiting, diarrhea, muscle cramps, and cutaneous reactions.^3,4

We followed patients who were treated with IM for 5 years to identify AEs of therapy.

Methods

The aim of this prospective study was to identify and collect data regarding IM cutaneous side effects so that clinicians can detect AEs early and differentiate them from AEs caused by other medications. All patients were subjected to a median of 5 years’ follow-up. We included all the patients treated with IM and excluded patients who had a history of eczematous dermatitis, psoriasis, renal impairment, or dyshidrosis palmoplantar. Before starting IM, all patients presented for a dermatologic visit. They were subsequently evaluated every 3 months.

The incidence rate was defined as the ratio of patients with cutaneous side effects and the total patients treated with IM. Furthermore, we calculated the ratio between each class of patient with a specific cutaneous manifestation and the entire cohort of patients with cutaneous side effects related to IM.

When necessary, microbiological, serological, and histopathological analyses were performed.

Results

In 60 months, we followed 220 patients treated with IM. Among them, 55 (25%) developed cutaneous side effects (35 males; 20 females). The incidence rate of the patients with cutaneous side effects was 1:4. The median age of the entire cohort was 52.5 years. Fifty patients were being treated for CML and 5 for GISTs. All patients received IM at a dosage of 400 mg daily.

The following skin diseases were observed in patients treated with IM (Table): 19 patients with maculopapular rash with pruritus (no maculopapular rash without pruritus was detected), 7 patients with eczematous dermatitis such as stasis dermatitis and seborrheic dermatitis, 6 patients with onychodystrophy melanonychia (Figure 1), 5 patients with psoriasis, 5 patients with skin cancers including basal cell carcinoma (BCC)(Figure 2), 3 patients with periorbital edema (Figure 3), 3 patients with mycosis, 3 patients with dermatofibromas, 2 patients with dyshidrosis palmoplantar, 1 patient with pityriasis rosea–like eruption (Figure 4), and 1 patient with actinic keratoses on the face. No hypopigmentation or hyperpigmentation, excluding the individual case of melanonychia, was observed.

All cutaneous diseases reported in this study appeared after IM therapy (median, 3.8 months). The median time to onset for each cutaneous disorder is reported in the Table. During the first dermatologic visit before starting IM therapy, none of the patients showed any of these cutaneous diseases.

The adverse cutaneous reactions were treated with appropriate drugs. Generally, eczematous dermatitis was treated using topical steroids, emollients, and oral antihistamines. In patients with maculopapular rash with pruritus, oral corticosteroids (eg, betamethasone 3 mg daily or prednisolone 1 mg/kg) in association with antihistamine was necessary. Psoriasis was completely improved with topical betamethasone 0.5 mg and calcipotriol 50 µg. Skin cancers were treated with surgical excision with histologic examination. All treatments are outlined in the Table.

Imatinib mesylate therapy was suspended in 2 patients with maculopapular rash with moderate to severe pruritus; however, despite the temporary suspension of the drug and the appropriate therapies (corticosteroids and antihistamines), cutaneous side effects reappeared 7 to 10 days after therapy resumed. Therefore, the treatment was permanently suspended in these 2 cases and IM was replaced with erlotinib, a second-generation Bcr-Abl tyrosine kinase inhibitor.

Comment

The introduction of IM for the treatment of GIST and CML has changed the history of these diseases. The drug typically is well tolerated and few patients have reported severe AEs. Mild skin reactions are relatively frequent, ranging from 7% to 21% of patients treated.³ In our case, the percentage was relatively higher (25%), likely because of close monitoring of patients, with an increase in the incidence rate.

Imatinib mesylate cutaneous reactions are dose dependent.⁴ Indeed, in all our cases, the cutaneous reactions arose with an IM dosage of 400 mg daily, which is compatible with the definition of dose-independent cutaneous AEs.

 

The most common cutaneous AEs reported in the literature were swelling/edema and maculopapular rash. Swelling is the most common AE described during therapy with IM with an incidence of 63% to 84%.⁵ Swelling often involves the periorbital area and occurs approximately 6 weeks after starting IM. Although its pathogenesis is uncertain, it has been shown that IM blocks the platelet-derived growth factor receptor expressed on blood vessels that regulates the transportation transcapillary. The inhibition of this receptor can lead to increased pore pressure, resulting in edema and erythema. Maculopapular eruptions (50% of cases) often affect the trunk and the limbs and are accompanied by pruritus. Commonly, these rashes arise after 9 weeks of IM therapy. These eruptions are self-limiting and only topical emollients and steroids are required, without any change in IM schedule. To treat maculopapular eruptions with pruritus, oral steroids and antihistamines may be helpful, without suspending IM treatment. When grade 2 or 3 pruriginous maculopapular eruptions arise, the suspension of IM combined with steroids and antihistamines may be necessary. When the readministration of IM is required, it is mandatory to start IM at a lower dose (50–100 mg/d), administering prednisolone 0.5 to 1.0 mg/kg daily. Then, the steroid gradually can be tapered.⁶ Critical cutaneous AEs that are resistant to supportive measures warrant suspension of IM therapy. However, the incidence of this event is small (<1% of all patients).⁷

Regarding severe cutaneous AEs from IM therapy, Hsiao et al⁸ reported the case of Stevens-Johnson syndrome. In this case, IM was immediately stopped and systemic steroids were started. Rarely, erythroderma (grade 4 toxicity) can develop for which a prompt and perpetual suspension of IM is necessary and supportive care therapy with oral and topical steroids is recommended.⁹

Hyperpigmentation induced by IM, mostly in patients with Fitzpatrick skin types V to VI and with a general prevalence of 16% to 40% in treated patients, often is related to a mutation of c-Kit or other kinases that are activated rather than inhibited by the drug, resulting in overstimulation of melanogenesis.¹⁰ The prevalence of Fitzpatrick skin types I to III determined the absence of pigmentation changes in our cohort, excluding melanonychia. Hyperpigmentation was observed in the skin as well as the appendages such as nails, resulting in melanonychia (Figure 1). However, Brazzelli et al¹¹ reported hypopigmentation in 5 white patients treated with IM; furthermore, they found a direct correlation between hypopigmentation and development of skin cancers in these patients. The susceptibility to develop skin cancers may persist, even without a clear manifestation of hypopigmentation, as reported in the current analysis. We documented BCC in 5 patients, 1 patient developed actinic keratoses, and 3 patients developed dermatofibromas. However, these neoplasms probably were not provoked by IM. On the contrary, we did not note squamous cell carcinoma, which was reported by Baskaynak et al¹² in 2 CML patients treated with IM.

The administration of IM can be associated with exacerbation of psoriasis. Paradoxically, in genetically predisposed individuals, tumor necrosis factor α (TNF-α) antagonists, such as IM, seem to induce psoriasis, producing IFN-α rather than TNF-α and increasing inflammation.¹³ In fact, some research shows induction of psoriasis by anti–TNF-α drugs.^14-16 Two cases of IM associated with psoriasis have been reported, and both cases represented an exacerbation of previously diagnosed psoriasis.^13,17 On the contrary, in our analysis we reported 5 cases of psoriasis vulgaris induced by IM administration. Our patients developed cutaneous psoriatic lesions approximately 1.7 months after the start of IM therapy.

The pityriasis rosea–like eruption (Figure 4) presented as nonpruritic, erythematous, scaly patches on the trunk and extremities, and arose 3.6 months after the start of treatment. This particular cutaneous AE is rare. In 3 case reports, the IM dosage also was 400 mg daily.^18-20 The pathophysiology of this rare skin reaction stems from the pharmacological effect of IM rather than a hypersensitivity reaction.¹⁸

Deininger et al⁷ reported that patients with a high basophil count (>20%) rarely show urticarial eruptions after IM due to histamine release from basophils. Premedication with an antihistamine was helpful and the urticarial eruption resolved after normalization in basophil count.⁷

Given the importance of IM for patients who have limited therapeutic alternatives for their disease and the ability to safely treat the cutaneous AEs, as demonstrated in our analysis, the suspension of IM for dermatological complications is necessary only in rare cases, as shown by the low number of patients (n=2) who had to discontinue therapy. The cutaneous AEs should be diagnosed and treated early with less impact on chemotherapy treatments. The administration of IM should involve a coordinated effort among oncologists and dermatologists to prevent important complications.

Imatinib mesylate (IM) represents the first-line treatment of chronic myeloid leukemia (CML) and gastrointestinal stromal tumors (GISTs). Its pharmacological activity is related to a specific action on several tyrosine kinases in different tumors, including Bcr-Abl in CML, c-Kit (CD117) in GIST, and platelet-derived growth factor receptor in dermatofibrosarcoma protuberans.^1,2

Imatinib mesylate has been shown to improve progression-free survival and overall survival²; however, it also has several side effects. Among the adverse effects (AEs), less than 10% are nonhematologic, such as nausea, vomiting, diarrhea, muscle cramps, and cutaneous reactions.^3,4

We followed patients who were treated with IM for 5 years to identify AEs of therapy.

Methods

The aim of this prospective study was to identify and collect data regarding IM cutaneous side effects so that clinicians can detect AEs early and differentiate them from AEs caused by other medications. All patients were subjected to a median of 5 years’ follow-up. We included all the patients treated with IM and excluded patients who had a history of eczematous dermatitis, psoriasis, renal impairment, or dyshidrosis palmoplantar. Before starting IM, all patients presented for a dermatologic visit. They were subsequently evaluated every 3 months.

The incidence rate was defined as the ratio of patients with cutaneous side effects and the total patients treated with IM. Furthermore, we calculated the ratio between each class of patient with a specific cutaneous manifestation and the entire cohort of patients with cutaneous side effects related to IM.

When necessary, microbiological, serological, and histopathological analyses were performed.

Results

In 60 months, we followed 220 patients treated with IM. Among them, 55 (25%) developed cutaneous side effects (35 males; 20 females). The incidence rate of the patients with cutaneous side effects was 1:4. The median age of the entire cohort was 52.5 years. Fifty patients were being treated for CML and 5 for GISTs. All patients received IM at a dosage of 400 mg daily.

The following skin diseases were observed in patients treated with IM (Table): 19 patients with maculopapular rash with pruritus (no maculopapular rash without pruritus was detected), 7 patients with eczematous dermatitis such as stasis dermatitis and seborrheic dermatitis, 6 patients with onychodystrophy melanonychia (Figure 1), 5 patients with psoriasis, 5 patients with skin cancers including basal cell carcinoma (BCC)(Figure 2), 3 patients with periorbital edema (Figure 3), 3 patients with mycosis, 3 patients with dermatofibromas, 2 patients with dyshidrosis palmoplantar, 1 patient with pityriasis rosea–like eruption (Figure 4), and 1 patient with actinic keratoses on the face. No hypopigmentation or hyperpigmentation, excluding the individual case of melanonychia, was observed.

All cutaneous diseases reported in this study appeared after IM therapy (median, 3.8 months). The median time to onset for each cutaneous disorder is reported in the Table. During the first dermatologic visit before starting IM therapy, none of the patients showed any of these cutaneous diseases.

The adverse cutaneous reactions were treated with appropriate drugs. Generally, eczematous dermatitis was treated using topical steroids, emollients, and oral antihistamines. In patients with maculopapular rash with pruritus, oral corticosteroids (eg, betamethasone 3 mg daily or prednisolone 1 mg/kg) in association with antihistamine was necessary. Psoriasis was completely improved with topical betamethasone 0.5 mg and calcipotriol 50 µg. Skin cancers were treated with surgical excision with histologic examination. All treatments are outlined in the Table.

Imatinib mesylate therapy was suspended in 2 patients with maculopapular rash with moderate to severe pruritus; however, despite the temporary suspension of the drug and the appropriate therapies (corticosteroids and antihistamines), cutaneous side effects reappeared 7 to 10 days after therapy resumed. Therefore, the treatment was permanently suspended in these 2 cases and IM was replaced with erlotinib, a second-generation Bcr-Abl tyrosine kinase inhibitor.

Comment

The introduction of IM for the treatment of GIST and CML has changed the history of these diseases. The drug typically is well tolerated and few patients have reported severe AEs. Mild skin reactions are relatively frequent, ranging from 7% to 21% of patients treated.³ In our case, the percentage was relatively higher (25%), likely because of close monitoring of patients, with an increase in the incidence rate.

Imatinib mesylate cutaneous reactions are dose dependent.⁴ Indeed, in all our cases, the cutaneous reactions arose with an IM dosage of 400 mg daily, which is compatible with the definition of dose-independent cutaneous AEs.

 

The most common cutaneous AEs reported in the literature were swelling/edema and maculopapular rash. Swelling is the most common AE described during therapy with IM with an incidence of 63% to 84%.⁵ Swelling often involves the periorbital area and occurs approximately 6 weeks after starting IM. Although its pathogenesis is uncertain, it has been shown that IM blocks the platelet-derived growth factor receptor expressed on blood vessels that regulates the transportation transcapillary. The inhibition of this receptor can lead to increased pore pressure, resulting in edema and erythema. Maculopapular eruptions (50% of cases) often affect the trunk and the limbs and are accompanied by pruritus. Commonly, these rashes arise after 9 weeks of IM therapy. These eruptions are self-limiting and only topical emollients and steroids are required, without any change in IM schedule. To treat maculopapular eruptions with pruritus, oral steroids and antihistamines may be helpful, without suspending IM treatment. When grade 2 or 3 pruriginous maculopapular eruptions arise, the suspension of IM combined with steroids and antihistamines may be necessary. When the readministration of IM is required, it is mandatory to start IM at a lower dose (50–100 mg/d), administering prednisolone 0.5 to 1.0 mg/kg daily. Then, the steroid gradually can be tapered.⁶ Critical cutaneous AEs that are resistant to supportive measures warrant suspension of IM therapy. However, the incidence of this event is small (<1% of all patients).⁷

Regarding severe cutaneous AEs from IM therapy, Hsiao et al⁸ reported the case of Stevens-Johnson syndrome. In this case, IM was immediately stopped and systemic steroids were started. Rarely, erythroderma (grade 4 toxicity) can develop for which a prompt and perpetual suspension of IM is necessary and supportive care therapy with oral and topical steroids is recommended.⁹

Hyperpigmentation induced by IM, mostly in patients with Fitzpatrick skin types V to VI and with a general prevalence of 16% to 40% in treated patients, often is related to a mutation of c-Kit or other kinases that are activated rather than inhibited by the drug, resulting in overstimulation of melanogenesis.¹⁰ The prevalence of Fitzpatrick skin types I to III determined the absence of pigmentation changes in our cohort, excluding melanonychia. Hyperpigmentation was observed in the skin as well as the appendages such as nails, resulting in melanonychia (Figure 1). However, Brazzelli et al¹¹ reported hypopigmentation in 5 white patients treated with IM; furthermore, they found a direct correlation between hypopigmentation and development of skin cancers in these patients. The susceptibility to develop skin cancers may persist, even without a clear manifestation of hypopigmentation, as reported in the current analysis. We documented BCC in 5 patients, 1 patient developed actinic keratoses, and 3 patients developed dermatofibromas. However, these neoplasms probably were not provoked by IM. On the contrary, we did not note squamous cell carcinoma, which was reported by Baskaynak et al¹² in 2 CML patients treated with IM.

The administration of IM can be associated with exacerbation of psoriasis. Paradoxically, in genetically predisposed individuals, tumor necrosis factor α (TNF-α) antagonists, such as IM, seem to induce psoriasis, producing IFN-α rather than TNF-α and increasing inflammation.¹³ In fact, some research shows induction of psoriasis by anti–TNF-α drugs.^14-16 Two cases of IM associated with psoriasis have been reported, and both cases represented an exacerbation of previously diagnosed psoriasis.^13,17 On the contrary, in our analysis we reported 5 cases of psoriasis vulgaris induced by IM administration. Our patients developed cutaneous psoriatic lesions approximately 1.7 months after the start of IM therapy.

The pityriasis rosea–like eruption (Figure 4) presented as nonpruritic, erythematous, scaly patches on the trunk and extremities, and arose 3.6 months after the start of treatment. This particular cutaneous AE is rare. In 3 case reports, the IM dosage also was 400 mg daily.^18-20 The pathophysiology of this rare skin reaction stems from the pharmacological effect of IM rather than a hypersensitivity reaction.¹⁸

Deininger et al⁷ reported that patients with a high basophil count (>20%) rarely show urticarial eruptions after IM due to histamine release from basophils. Premedication with an antihistamine was helpful and the urticarial eruption resolved after normalization in basophil count.⁷

Given the importance of IM for patients who have limited therapeutic alternatives for their disease and the ability to safely treat the cutaneous AEs, as demonstrated in our analysis, the suspension of IM for dermatological complications is necessary only in rare cases, as shown by the low number of patients (n=2) who had to discontinue therapy. The cutaneous AEs should be diagnosed and treated early with less impact on chemotherapy treatments. The administration of IM should involve a coordinated effort among oncologists and dermatologists to prevent important complications.

The distal or proximal femur with tumor endoprosthesis is commonly replaced after segmental resections for bone tumors, complex trauma, or revision arthroplasty. In conventional joint replacements, correct rotational alignment of the component is referenced off anatomical landmarks in the proximal or distal femur. After tumor resection, however, these landmarks are often not available for rotational orientation. There are no reports of studies validating a particular method of establishing rotation in these cases.

To establish a guide for rotational alignment of tumor endoprostheses, we set out to define the natural location of the linea aspera (LA) based on axial computed tomography (CT) scans. The LA is often the most outstanding visible bony landmark on a cross-section of the femur during surgery, and it would be helpful to know its normal orientation in relation to the true anteroposterior (AP) axis of the femur and to the femoral version. We wanted to answer these 5 questions:

1. Is the prominence of the LA easily identifiable on cross-section at different levels of the femoral shaft?

2. Does an axis passing through the LA correspond to the AP axis of the femur?

3. If not, is this axis offset internally or externally and by how much?

4. Is this offset constant at all levels of the femoral shaft?

5. How does the LA axis relate to the femoral neck axis at these levels?

The answers determine if the LA can be reliably used for rotational alignment of tumor endoprostheses.

Materials and Methods

After this study received Institutional Review Board approval, we retrospectively reviewed whole-body fluorine-18-deoxyglucose (FDG) positron emission tomography–computed tomography (PET-CT) studies performed in our hospital between 2003 and 2006 to identify those with full-length bilateral femur CT scans. These scans were available on the hospital’s computerized picture archiving system (General Electric). Patients could be included in the study as long as they were at least 18 years old at time of scan and did not have any pathology that deformed the femur, broke a cortex, or otherwise caused any gross asymmetry of the femur. Of the 72 patients with full-length femur CT scans, 3 were excluded: 1 with a congenital hip dysplasia, 1 with an old, malunited femoral fracture, and 1 who was 15 years old at time of scan.

Axial Slice Selection

For each patient, scout AP films were used to measure femoral shaft length from the top of the greater trochanter to the end of the lateral femoral condyle. The levels of the proximal third, midshaft, and distal third were then calculated based on this length. The LA was studied on the axial slices nearest these levels. Next, we scrolled through the scans to identify an axial slice that best showed the femoral neck axis. The literature on CT measurement of femoral anteversion is varied. Some articles describe a technique that uses 2 superimposed axial slices, and others describe a single axial slice.^1-3 We used 1 axial slice to draw the femoral neck axis because our computer software could not superimpose 2 images on 1 screen and because the CT scans were not made under specific protocols to measure anteversion but rather were part of a cancer staging work-up. Axial cuts were made at 5-mm intervals, and not all scans included a single slice capturing the head, neck, and greater trochanter. Therefore, we used a (previously described) method in which the femoral neck axis is drawn on a slice that most captured the femoral neck, usually toward its base.4 Last, in order to draw the posterior condyle (PC) axis, we selected an axial slice that showed the posterior-most aspects of the femoral condyles at the intercondylar notch.

 

Determining Anteroposterior and Posterior Condyle Axes of Femur

As we made all measurements for each femur off a single CT scan, we were able to use a straight horizontal line—drawn on-screen with a software tool—as a reference for measuring rotation. On a distal femur cut, the PC axis is drawn by connecting the posterior-most points of both condyles. The software calculates the angle formed—the PC angle (Figure 1). This angle, the degree to which the PC axis deviates from a straight horizontal line on-screen, can be used to account for gross rotation of the limb on comparison of images. The AP axis of the femur is the axis perpendicular to the PC axis. As such, the PC angle can also be used to determine degree of deviation of the AP axis from a straight vertical line on-screen. The AP axis was used when calculating the LA axis at the various levels of the femur (Figure 2).

 

Femoral Version

We used the software tool to draw the femoral neck axis. From the end of this line, a straight horizontal line is drawn on-screen (Figure 3). The software calculates the angle formed—the femoral neck axis angle. We assigned a positive value for a femoral head that pointed anteriorly on the image and a negative value for a head that pointed posteriorly. Adjusting for external rotation of the limb involved calculating the femoral version by subtracting the PC angle from the neck axis angle; adjusting for internal rotation involved adding these 2 angles.

Linea Aspera Morphology

After viewing the first 20 CT scans, we identified 3 types of LA morphology. Type I presents as a thickening on the posterior cortex with a sharp apex; type II presents as a flat-faced but distinct ridge of bone between the medial and lateral lips; and in type III there is no distinct cortical thickening with blunted medial and lateral lips; the latter is always more prominent.

Linea Aspera Axis Offset

From the most posterior point of the LA, a line drawn forward bisecting the femoral canal defined the LA axis. In type I morphology, the posterior-most point was the apex; in type II, the middle of flat posterior surface was used as the starting point; in type III, the lateral lip was used, as it was sharper than the medial lip. This line is again referenced with a straight horizontal line across the image. The PC angle is then added to account for limb rotation, and the result is the LA angle. As the AP axis is perpendicular to the PC axis, the LA angle is subtracted from 90°; the difference represents the amount of offset of the LA axis from the AP axis. By convention, we assigned this a positive value for an LA lateral to the midpoint of the femur and a negative value for an LA medial to the midpoint (Figure 4).

Linea Aspera Axis and Femoral Neck Axis

The angle between the LA axis and the PC axis was measured. The femoral version angle was subtracted from that angle to obtain the arc between the LA axis and the femoral neck axis.

Statistical Analyses

All analyses were performed with SAS 9.1 (SAS Institute). All tests were 2-sided and conducted at the .05 significance level. No adjustments were made for multiple testing. Statistical analysis was performed with nonparametric tests and without making assumptions about the distribution of the study population. Univariate analyses were performed to test for significant side-to-side differences in femoral length, femoral version angle, and LA torsion angles at each level. A multivariate analysis was performed to test for interactions between sex, side, and level. In all analyses, P < .05 was used as the cutoff value for statistical significance.

 

Results

Femoral lengths varied by side and sex. The left side was longer than the right by a mean of 1.3 mm (P = .008). With multivariate analysis taking into account sex and age (cumulated per decade), there was still a significant effect of side on femoral length. Sex also had a significant effect on femoral length, with females’ femurs shorter by 21.7 mm (standard error, 5.0 mm). Mean (SD) anteversion of the femoral neck was 7.9° (12.7°) on the left and 13.3° (13.0°) on the right; the difference between sides was significant (P < .001). In a multivariate analysis performed to identify potential predictors of femoral version, side still had a significant (P < .001) independent effect; sex and age did not have an effect.

LA morphology varied according to femoral shaft level (Table 1). The morphology was type I in 75% of patients at the distal femur and 74% of patients at the midshaft femur, while only 53% of patients had a type I morphology at the proximal femur. The proportion of type III morphology was larger in the proximal femur (41%) than in the other locations.

The LA axis of the femur did not correspond exactly to the AP axis at all femoral levels. At the distal femur, mean (SD) lateral offset of the LA axis was 5.5° (7.5°) on the left and 8.3° (8.9°) on the right. At the midshaft, mean (SD) medial offset of the LA axis was 3.1° (8.4°) on the left and 1.2° (7.9°) on the right. At the proximal femur, mean (SD) lateral offset of the LA axis was 5.4° (9.2°) on the left and 6.2° (8.3°) on the right. The side-to-side differences were statistically significant for the distal femur and midshaft but not the proximal femur. Table 2 lists the 95% confidence intervals for the mean values. As the range of differences was small (0.7°-2.8°), and the differences may not be clinically detected on gross inspection during surgery, we pooled both sides’ values to arrive at a single mean for each level. The LA axis was offset a mean (SD) of 6.9° (8.3°) laterally at the distal femur, 2.2° (8.2°) medially at the midshaft, and 5.8° (8.6°) laterally at the proximal femur. Figure 5 shows the frequency of distribution of LA axis offset.Offset of the LA axis from the AP axis of the femur was significantly (P < .001) different for each femoral level, even when a multivariate analysis was performed to determine the effect of sex, age, or side. Age and sex had no significant effect on mean offset of LA axis from AP axis.

We compared the mean arc between femoral neck axis and LA axis after referencing both off the PC axis. At the distal femur, mean (SD) arc between these 2 axes was 76.6° (13.1°) on the left and 68.3° (13.6°) on the right (mean difference, 8.3°); at the midshaft, mean (SD) arc was 85.2° (13.5°) on the left and 77.9° (13.1°) on the right (mean difference, 7.4°); at the proximal femur, mean (SD) arc was 76.7° (11.9°) on the left and 70.5° (12.8°) on the right (mean difference, 6.2°). The side-to-side differences were statistically significant (P < .001) for all locations.

In multivariate analysis, sex and age did not have an effect on mean arc between the 2 axes. Side and femoral level, however, had a significant effect (P < .001).

Discussion

In total hip arthroplasty, the goal is to restore femoral anteversion, usually referenced to the remaining femoral neck segment.³ In total knee arthroplasty (TKA), proper rotation preserves normal patellofemoral tracking.⁵ Various landmarks are used, such as the PCs or the epicondyles. After tumor resections, these landmarks are often lost.⁶ However, there are no reports of studies validating a particular method of achieving proper rotational orientation of tumor endoprostheses, though several methods are being used. One method involves inserting 2 drill bits before osteotomy—one proximal to the intended level of resection on the anterior femur, and the other on the anterior tibial shaft. The straight line formed can establish a plane of rotation (and length), which the surgeon must aim to restore when the components are placed. This method is useful for distal femur resections but not proximal femur resections. Another method, based on the LA’s anatomical position on the posterior aspect of the femur,⁴ uses the prominence of the LA to align the prosthesis. With this method, the LA is assumed to be directly posterior (6 o’clock) on the femur. However, this assumption has not been confirmed by any study. A third method, described by Heck and Carnesale,⁵ involves marking the anterior aspect of the femur after resection and aligning the components to it. The authors cautioned against using the LA as a landmark, saying that its course is highly variable.

The LA is a narrow, elevated length of bone, with medial and lateral lips, that serves as an attachment site for muscles in the posterior thigh. Proximally, the LA presents with lateral, medial, and intermediate lips. In the midshaft, it is often elevated by an underlying bony ridge or pilaster complex. Distally, it diverges into 2 ridges that form the triangular popliteal surface.^1,7 For the LA to be a reliable landmark, first it must be clearly identifiable on viewing a femoral cross-section. The LA that presents with type I or II morphology is distinctly identifiable, and an axis from its apex and bisecting the canal can easily be constructed. In our study, the LA presented with type I or II morphology in 82% of distal femoral sections and 99% of midshaft femoral sections. Therefore, the LA is a conspicuous landmark at these levels. In the proximal femur, 59% had type I or II morphology. Type III morphology could be identified on cross-sections by the persisting prominence of the lateral lip. However, it may be difficult to appreciate the LA with this morphology at surgery.

Once the LA is identified, its normal cross-sectional position must be defined. One way to do this is to establish the relationship of its axis (LA axis) to the true AP axis. Based on mean values, the LA axis is laterally offset 7º at the distal third of the femur, medially offset 2º at the midshaft, and laterally offset 6º at the proximal third. Therefore, for ideal placement with the LA used for orientation, the component must be internally rotated 7º relative to the LA for femoral resection at the distal third, externally rotated 2º for resection at the midshaft, and internally rotated 6º for resection at the proximal third. Studies have demonstrated that joint contact forces and mechanical alignment of the lower limb can be altered with as little as 5º of femoral malrotation.^8,9 Although such a small degree of malrotation is often asymptomatic, it can have long-term effects on soft-tissue tension and patellar tracking.^10,11 Rotating-platform mobile-bearing TKA designs can compensate for femoral malrotation, but they may have little to no effect on patellar tracking.¹² Therefore, we think aligning the components as near as possible to their natural orientation can prove beneficial in long-term patient management.

Another way of defining the normal cross-sectional position of the LA is to relate it to the femoral neck axis. We measured the difference between these 2 axes. Mean differences were 72º (distal femur), 81.5º (midshaft), and 73.5º (proximal third). Mean arc differences at all levels were larger on the left side—a reflection of the femoral neck being less anteverted on that side in our measurements. Standard deviations were smaller for measurements of LA axis offset from AP axis (range, 7.5°-9.2°) than for measurements of arc between LA axis and femoral neck axis (range, 11.9°-13.6°). This finding indicates there is less variation in the former method, making it preferable for defining the cross-sectional position of the LA.

It has been said that the course of the LA is variable, and our data provide confirmation. The LA does not lie directly posterior (6 o’clock), and it does not trace a straight longitudinal course along the posterior femur, as demonstrated by the different LA axis offsets at 3 levels. However, we may still use it as a landmark if we remain aware how much the LA is offset from the AP axis at each femoral level. Figures 6A-6D, which show CT scans of a patient who underwent distal femoral resection and replacement with an endoprosthesis, illustrate how the LA axis was measured before surgery and how proper prosthesis placement was confirmed after surgery.

In hip arthroplasty, restoration of normal femoral version is the reference for endoprosthetic placement. The literature on “normal” femoral anteversion varies with the method used. In a review of studies on CT-measured adult femoral version, reported values ranged from 6.3° to 40°.² Mean femoral version in our study ranged from 8° to 13°. Orthopedics textbooks generally put the value at 10° to 15º, and this seems to be the range that surgeons target.⁶ However, we found a statistically significant mean side-to-side difference of 5.4°. This finding is possibly explained by our large sample—it was larger than the samples used in other studies of CT-measured femoral version. Other studies have found mean side-to-side differences of up to 4.0º.⁵ Another explanation for our finding is that the studies may differ methodologically. The studies that established values for femoral anteversion were based on CT protocols—thinner slices (1-5 mm), use of foot holders to standardize limb rotation, use of 2 axial cuts in proximal femur to establish femoral neck axis^2,13—designed specifically for this measurement. As the CT scans reviewed in our study are not designed for this purpose, errors in femoral version measurement may have been introduced, which may also explain why there is larger variation in measurements of the arc between the LA axis and the femoral neck axis.

Conclusion

The LA does not lie directly on the posterior surface of the femur. It deviates 6.9° laterally at the distal femur, 2.2° medially at the midshaft, and 6.9° laterally at the proximal third. As the LA is an easily identifiable structure on cross-sections of the femoral shaft at the midshaft and distal third of the femur, it may be useful as a rotational landmark for resections at these levels if these deviations are considered during tumor endoprosthetic replacements.

The distal or proximal femur with tumor endoprosthesis is commonly replaced after segmental resections for bone tumors, complex trauma, or revision arthroplasty. In conventional joint replacements, correct rotational alignment of the component is referenced off anatomical landmarks in the proximal or distal femur. After tumor resection, however, these landmarks are often not available for rotational orientation. There are no reports of studies validating a particular method of establishing rotation in these cases.

To establish a guide for rotational alignment of tumor endoprostheses, we set out to define the natural location of the linea aspera (LA) based on axial computed tomography (CT) scans. The LA is often the most outstanding visible bony landmark on a cross-section of the femur during surgery, and it would be helpful to know its normal orientation in relation to the true anteroposterior (AP) axis of the femur and to the femoral version. We wanted to answer these 5 questions:

1. Is the prominence of the LA easily identifiable on cross-section at different levels of the femoral shaft?

2. Does an axis passing through the LA correspond to the AP axis of the femur?

3. If not, is this axis offset internally or externally and by how much?

4. Is this offset constant at all levels of the femoral shaft?

5. How does the LA axis relate to the femoral neck axis at these levels?

The answers determine if the LA can be reliably used for rotational alignment of tumor endoprostheses.

Materials and Methods

After this study received Institutional Review Board approval, we retrospectively reviewed whole-body fluorine-18-deoxyglucose (FDG) positron emission tomography–computed tomography (PET-CT) studies performed in our hospital between 2003 and 2006 to identify those with full-length bilateral femur CT scans. These scans were available on the hospital’s computerized picture archiving system (General Electric). Patients could be included in the study as long as they were at least 18 years old at time of scan and did not have any pathology that deformed the femur, broke a cortex, or otherwise caused any gross asymmetry of the femur. Of the 72 patients with full-length femur CT scans, 3 were excluded: 1 with a congenital hip dysplasia, 1 with an old, malunited femoral fracture, and 1 who was 15 years old at time of scan.

Axial Slice Selection

For each patient, scout AP films were used to measure femoral shaft length from the top of the greater trochanter to the end of the lateral femoral condyle. The levels of the proximal third, midshaft, and distal third were then calculated based on this length. The LA was studied on the axial slices nearest these levels. Next, we scrolled through the scans to identify an axial slice that best showed the femoral neck axis. The literature on CT measurement of femoral anteversion is varied. Some articles describe a technique that uses 2 superimposed axial slices, and others describe a single axial slice.^1-3 We used 1 axial slice to draw the femoral neck axis because our computer software could not superimpose 2 images on 1 screen and because the CT scans were not made under specific protocols to measure anteversion but rather were part of a cancer staging work-up. Axial cuts were made at 5-mm intervals, and not all scans included a single slice capturing the head, neck, and greater trochanter. Therefore, we used a (previously described) method in which the femoral neck axis is drawn on a slice that most captured the femoral neck, usually toward its base.4 Last, in order to draw the posterior condyle (PC) axis, we selected an axial slice that showed the posterior-most aspects of the femoral condyles at the intercondylar notch.

 

Determining Anteroposterior and Posterior Condyle Axes of Femur

As we made all measurements for each femur off a single CT scan, we were able to use a straight horizontal line—drawn on-screen with a software tool—as a reference for measuring rotation. On a distal femur cut, the PC axis is drawn by connecting the posterior-most points of both condyles. The software calculates the angle formed—the PC angle (Figure 1). This angle, the degree to which the PC axis deviates from a straight horizontal line on-screen, can be used to account for gross rotation of the limb on comparison of images. The AP axis of the femur is the axis perpendicular to the PC axis. As such, the PC angle can also be used to determine degree of deviation of the AP axis from a straight vertical line on-screen. The AP axis was used when calculating the LA axis at the various levels of the femur (Figure 2).

 

Femoral Version

We used the software tool to draw the femoral neck axis. From the end of this line, a straight horizontal line is drawn on-screen (Figure 3). The software calculates the angle formed—the femoral neck axis angle. We assigned a positive value for a femoral head that pointed anteriorly on the image and a negative value for a head that pointed posteriorly. Adjusting for external rotation of the limb involved calculating the femoral version by subtracting the PC angle from the neck axis angle; adjusting for internal rotation involved adding these 2 angles.

Linea Aspera Morphology

After viewing the first 20 CT scans, we identified 3 types of LA morphology. Type I presents as a thickening on the posterior cortex with a sharp apex; type II presents as a flat-faced but distinct ridge of bone between the medial and lateral lips; and in type III there is no distinct cortical thickening with blunted medial and lateral lips; the latter is always more prominent.

Linea Aspera Axis Offset

From the most posterior point of the LA, a line drawn forward bisecting the femoral canal defined the LA axis. In type I morphology, the posterior-most point was the apex; in type II, the middle of flat posterior surface was used as the starting point; in type III, the lateral lip was used, as it was sharper than the medial lip. This line is again referenced with a straight horizontal line across the image. The PC angle is then added to account for limb rotation, and the result is the LA angle. As the AP axis is perpendicular to the PC axis, the LA angle is subtracted from 90°; the difference represents the amount of offset of the LA axis from the AP axis. By convention, we assigned this a positive value for an LA lateral to the midpoint of the femur and a negative value for an LA medial to the midpoint (Figure 4).

Linea Aspera Axis and Femoral Neck Axis

The angle between the LA axis and the PC axis was measured. The femoral version angle was subtracted from that angle to obtain the arc between the LA axis and the femoral neck axis.

Statistical Analyses

All analyses were performed with SAS 9.1 (SAS Institute). All tests were 2-sided and conducted at the .05 significance level. No adjustments were made for multiple testing. Statistical analysis was performed with nonparametric tests and without making assumptions about the distribution of the study population. Univariate analyses were performed to test for significant side-to-side differences in femoral length, femoral version angle, and LA torsion angles at each level. A multivariate analysis was performed to test for interactions between sex, side, and level. In all analyses, P < .05 was used as the cutoff value for statistical significance.

 

Results

Femoral lengths varied by side and sex. The left side was longer than the right by a mean of 1.3 mm (P = .008). With multivariate analysis taking into account sex and age (cumulated per decade), there was still a significant effect of side on femoral length. Sex also had a significant effect on femoral length, with females’ femurs shorter by 21.7 mm (standard error, 5.0 mm). Mean (SD) anteversion of the femoral neck was 7.9° (12.7°) on the left and 13.3° (13.0°) on the right; the difference between sides was significant (P < .001). In a multivariate analysis performed to identify potential predictors of femoral version, side still had a significant (P < .001) independent effect; sex and age did not have an effect.

LA morphology varied according to femoral shaft level (Table 1). The morphology was type I in 75% of patients at the distal femur and 74% of patients at the midshaft femur, while only 53% of patients had a type I morphology at the proximal femur. The proportion of type III morphology was larger in the proximal femur (41%) than in the other locations.

The LA axis of the femur did not correspond exactly to the AP axis at all femoral levels. At the distal femur, mean (SD) lateral offset of the LA axis was 5.5° (7.5°) on the left and 8.3° (8.9°) on the right. At the midshaft, mean (SD) medial offset of the LA axis was 3.1° (8.4°) on the left and 1.2° (7.9°) on the right. At the proximal femur, mean (SD) lateral offset of the LA axis was 5.4° (9.2°) on the left and 6.2° (8.3°) on the right. The side-to-side differences were statistically significant for the distal femur and midshaft but not the proximal femur. Table 2 lists the 95% confidence intervals for the mean values. As the range of differences was small (0.7°-2.8°), and the differences may not be clinically detected on gross inspection during surgery, we pooled both sides’ values to arrive at a single mean for each level. The LA axis was offset a mean (SD) of 6.9° (8.3°) laterally at the distal femur, 2.2° (8.2°) medially at the midshaft, and 5.8° (8.6°) laterally at the proximal femur. Figure 5 shows the frequency of distribution of LA axis offset.Offset of the LA axis from the AP axis of the femur was significantly (P < .001) different for each femoral level, even when a multivariate analysis was performed to determine the effect of sex, age, or side. Age and sex had no significant effect on mean offset of LA axis from AP axis.

We compared the mean arc between femoral neck axis and LA axis after referencing both off the PC axis. At the distal femur, mean (SD) arc between these 2 axes was 76.6° (13.1°) on the left and 68.3° (13.6°) on the right (mean difference, 8.3°); at the midshaft, mean (SD) arc was 85.2° (13.5°) on the left and 77.9° (13.1°) on the right (mean difference, 7.4°); at the proximal femur, mean (SD) arc was 76.7° (11.9°) on the left and 70.5° (12.8°) on the right (mean difference, 6.2°). The side-to-side differences were statistically significant (P < .001) for all locations.

In multivariate analysis, sex and age did not have an effect on mean arc between the 2 axes. Side and femoral level, however, had a significant effect (P < .001).

Discussion

In total hip arthroplasty, the goal is to restore femoral anteversion, usually referenced to the remaining femoral neck segment.³ In total knee arthroplasty (TKA), proper rotation preserves normal patellofemoral tracking.⁵ Various landmarks are used, such as the PCs or the epicondyles. After tumor resections, these landmarks are often lost.⁶ However, there are no reports of studies validating a particular method of achieving proper rotational orientation of tumor endoprostheses, though several methods are being used. One method involves inserting 2 drill bits before osteotomy—one proximal to the intended level of resection on the anterior femur, and the other on the anterior tibial shaft. The straight line formed can establish a plane of rotation (and length), which the surgeon must aim to restore when the components are placed. This method is useful for distal femur resections but not proximal femur resections. Another method, based on the LA’s anatomical position on the posterior aspect of the femur,⁴ uses the prominence of the LA to align the prosthesis. With this method, the LA is assumed to be directly posterior (6 o’clock) on the femur. However, this assumption has not been confirmed by any study. A third method, described by Heck and Carnesale,⁵ involves marking the anterior aspect of the femur after resection and aligning the components to it. The authors cautioned against using the LA as a landmark, saying that its course is highly variable.

The LA is a narrow, elevated length of bone, with medial and lateral lips, that serves as an attachment site for muscles in the posterior thigh. Proximally, the LA presents with lateral, medial, and intermediate lips. In the midshaft, it is often elevated by an underlying bony ridge or pilaster complex. Distally, it diverges into 2 ridges that form the triangular popliteal surface.^1,7 For the LA to be a reliable landmark, first it must be clearly identifiable on viewing a femoral cross-section. The LA that presents with type I or II morphology is distinctly identifiable, and an axis from its apex and bisecting the canal can easily be constructed. In our study, the LA presented with type I or II morphology in 82% of distal femoral sections and 99% of midshaft femoral sections. Therefore, the LA is a conspicuous landmark at these levels. In the proximal femur, 59% had type I or II morphology. Type III morphology could be identified on cross-sections by the persisting prominence of the lateral lip. However, it may be difficult to appreciate the LA with this morphology at surgery.

Once the LA is identified, its normal cross-sectional position must be defined. One way to do this is to establish the relationship of its axis (LA axis) to the true AP axis. Based on mean values, the LA axis is laterally offset 7º at the distal third of the femur, medially offset 2º at the midshaft, and laterally offset 6º at the proximal third. Therefore, for ideal placement with the LA used for orientation, the component must be internally rotated 7º relative to the LA for femoral resection at the distal third, externally rotated 2º for resection at the midshaft, and internally rotated 6º for resection at the proximal third. Studies have demonstrated that joint contact forces and mechanical alignment of the lower limb can be altered with as little as 5º of femoral malrotation.^8,9 Although such a small degree of malrotation is often asymptomatic, it can have long-term effects on soft-tissue tension and patellar tracking.^10,11 Rotating-platform mobile-bearing TKA designs can compensate for femoral malrotation, but they may have little to no effect on patellar tracking.¹² Therefore, we think aligning the components as near as possible to their natural orientation can prove beneficial in long-term patient management.

Another way of defining the normal cross-sectional position of the LA is to relate it to the femoral neck axis. We measured the difference between these 2 axes. Mean differences were 72º (distal femur), 81.5º (midshaft), and 73.5º (proximal third). Mean arc differences at all levels were larger on the left side—a reflection of the femoral neck being less anteverted on that side in our measurements. Standard deviations were smaller for measurements of LA axis offset from AP axis (range, 7.5°-9.2°) than for measurements of arc between LA axis and femoral neck axis (range, 11.9°-13.6°). This finding indicates there is less variation in the former method, making it preferable for defining the cross-sectional position of the LA.

It has been said that the course of the LA is variable, and our data provide confirmation. The LA does not lie directly posterior (6 o’clock), and it does not trace a straight longitudinal course along the posterior femur, as demonstrated by the different LA axis offsets at 3 levels. However, we may still use it as a landmark if we remain aware how much the LA is offset from the AP axis at each femoral level. Figures 6A-6D, which show CT scans of a patient who underwent distal femoral resection and replacement with an endoprosthesis, illustrate how the LA axis was measured before surgery and how proper prosthesis placement was confirmed after surgery.

In hip arthroplasty, restoration of normal femoral version is the reference for endoprosthetic placement. The literature on “normal” femoral anteversion varies with the method used. In a review of studies on CT-measured adult femoral version, reported values ranged from 6.3° to 40°.² Mean femoral version in our study ranged from 8° to 13°. Orthopedics textbooks generally put the value at 10° to 15º, and this seems to be the range that surgeons target.⁶ However, we found a statistically significant mean side-to-side difference of 5.4°. This finding is possibly explained by our large sample—it was larger than the samples used in other studies of CT-measured femoral version. Other studies have found mean side-to-side differences of up to 4.0º.⁵ Another explanation for our finding is that the studies may differ methodologically. The studies that established values for femoral anteversion were based on CT protocols—thinner slices (1-5 mm), use of foot holders to standardize limb rotation, use of 2 axial cuts in proximal femur to establish femoral neck axis^2,13—designed specifically for this measurement. As the CT scans reviewed in our study are not designed for this purpose, errors in femoral version measurement may have been introduced, which may also explain why there is larger variation in measurements of the arc between the LA axis and the femoral neck axis.

Conclusion

The LA does not lie directly on the posterior surface of the femur. It deviates 6.9° laterally at the distal femur, 2.2° medially at the midshaft, and 6.9° laterally at the proximal third. As the LA is an easily identifiable structure on cross-sections of the femoral shaft at the midshaft and distal third of the femur, it may be useful as a rotational landmark for resections at these levels if these deviations are considered during tumor endoprosthetic replacements.

The distal or proximal femur with tumor endoprosthesis is commonly replaced after segmental resections for bone tumors, complex trauma, or revision arthroplasty. In conventional joint replacements, correct rotational alignment of the component is referenced off anatomical landmarks in the proximal or distal femur. After tumor resection, however, these landmarks are often not available for rotational orientation. There are no reports of studies validating a particular method of establishing rotation in these cases.

To establish a guide for rotational alignment of tumor endoprostheses, we set out to define the natural location of the linea aspera (LA) based on axial computed tomography (CT) scans. The LA is often the most outstanding visible bony landmark on a cross-section of the femur during surgery, and it would be helpful to know its normal orientation in relation to the true anteroposterior (AP) axis of the femur and to the femoral version. We wanted to answer these 5 questions:

1. Is the prominence of the LA easily identifiable on cross-section at different levels of the femoral shaft?

2. Does an axis passing through the LA correspond to the AP axis of the femur?

3. If not, is this axis offset internally or externally and by how much?

4. Is this offset constant at all levels of the femoral shaft?

5. How does the LA axis relate to the femoral neck axis at these levels?

The answers determine if the LA can be reliably used for rotational alignment of tumor endoprostheses.

Materials and Methods

After this study received Institutional Review Board approval, we retrospectively reviewed whole-body fluorine-18-deoxyglucose (FDG) positron emission tomography–computed tomography (PET-CT) studies performed in our hospital between 2003 and 2006 to identify those with full-length bilateral femur CT scans. These scans were available on the hospital’s computerized picture archiving system (General Electric). Patients could be included in the study as long as they were at least 18 years old at time of scan and did not have any pathology that deformed the femur, broke a cortex, or otherwise caused any gross asymmetry of the femur. Of the 72 patients with full-length femur CT scans, 3 were excluded: 1 with a congenital hip dysplasia, 1 with an old, malunited femoral fracture, and 1 who was 15 years old at time of scan.

Axial Slice Selection

For each patient, scout AP films were used to measure femoral shaft length from the top of the greater trochanter to the end of the lateral femoral condyle. The levels of the proximal third, midshaft, and distal third were then calculated based on this length. The LA was studied on the axial slices nearest these levels. Next, we scrolled through the scans to identify an axial slice that best showed the femoral neck axis. The literature on CT measurement of femoral anteversion is varied. Some articles describe a technique that uses 2 superimposed axial slices, and others describe a single axial slice.^1-3 We used 1 axial slice to draw the femoral neck axis because our computer software could not superimpose 2 images on 1 screen and because the CT scans were not made under specific protocols to measure anteversion but rather were part of a cancer staging work-up. Axial cuts were made at 5-mm intervals, and not all scans included a single slice capturing the head, neck, and greater trochanter. Therefore, we used a (previously described) method in which the femoral neck axis is drawn on a slice that most captured the femoral neck, usually toward its base.4 Last, in order to draw the posterior condyle (PC) axis, we selected an axial slice that showed the posterior-most aspects of the femoral condyles at the intercondylar notch.

 

Determining Anteroposterior and Posterior Condyle Axes of Femur

As we made all measurements for each femur off a single CT scan, we were able to use a straight horizontal line—drawn on-screen with a software tool—as a reference for measuring rotation. On a distal femur cut, the PC axis is drawn by connecting the posterior-most points of both condyles. The software calculates the angle formed—the PC angle (Figure 1). This angle, the degree to which the PC axis deviates from a straight horizontal line on-screen, can be used to account for gross rotation of the limb on comparison of images. The AP axis of the femur is the axis perpendicular to the PC axis. As such, the PC angle can also be used to determine degree of deviation of the AP axis from a straight vertical line on-screen. The AP axis was used when calculating the LA axis at the various levels of the femur (Figure 2).

 

Femoral Version

We used the software tool to draw the femoral neck axis. From the end of this line, a straight horizontal line is drawn on-screen (Figure 3). The software calculates the angle formed—the femoral neck axis angle. We assigned a positive value for a femoral head that pointed anteriorly on the image and a negative value for a head that pointed posteriorly. Adjusting for external rotation of the limb involved calculating the femoral version by subtracting the PC angle from the neck axis angle; adjusting for internal rotation involved adding these 2 angles.

Linea Aspera Morphology

After viewing the first 20 CT scans, we identified 3 types of LA morphology. Type I presents as a thickening on the posterior cortex with a sharp apex; type II presents as a flat-faced but distinct ridge of bone between the medial and lateral lips; and in type III there is no distinct cortical thickening with blunted medial and lateral lips; the latter is always more prominent.

Linea Aspera Axis Offset

From the most posterior point of the LA, a line drawn forward bisecting the femoral canal defined the LA axis. In type I morphology, the posterior-most point was the apex; in type II, the middle of flat posterior surface was used as the starting point; in type III, the lateral lip was used, as it was sharper than the medial lip. This line is again referenced with a straight horizontal line across the image. The PC angle is then added to account for limb rotation, and the result is the LA angle. As the AP axis is perpendicular to the PC axis, the LA angle is subtracted from 90°; the difference represents the amount of offset of the LA axis from the AP axis. By convention, we assigned this a positive value for an LA lateral to the midpoint of the femur and a negative value for an LA medial to the midpoint (Figure 4).

Linea Aspera Axis and Femoral Neck Axis

The angle between the LA axis and the PC axis was measured. The femoral version angle was subtracted from that angle to obtain the arc between the LA axis and the femoral neck axis.

Statistical Analyses

All analyses were performed with SAS 9.1 (SAS Institute). All tests were 2-sided and conducted at the .05 significance level. No adjustments were made for multiple testing. Statistical analysis was performed with nonparametric tests and without making assumptions about the distribution of the study population. Univariate analyses were performed to test for significant side-to-side differences in femoral length, femoral version angle, and LA torsion angles at each level. A multivariate analysis was performed to test for interactions between sex, side, and level. In all analyses, P < .05 was used as the cutoff value for statistical significance.

 

Results

Femoral lengths varied by side and sex. The left side was longer than the right by a mean of 1.3 mm (P = .008). With multivariate analysis taking into account sex and age (cumulated per decade), there was still a significant effect of side on femoral length. Sex also had a significant effect on femoral length, with females’ femurs shorter by 21.7 mm (standard error, 5.0 mm). Mean (SD) anteversion of the femoral neck was 7.9° (12.7°) on the left and 13.3° (13.0°) on the right; the difference between sides was significant (P < .001). In a multivariate analysis performed to identify potential predictors of femoral version, side still had a significant (P < .001) independent effect; sex and age did not have an effect.

LA morphology varied according to femoral shaft level (Table 1). The morphology was type I in 75% of patients at the distal femur and 74% of patients at the midshaft femur, while only 53% of patients had a type I morphology at the proximal femur. The proportion of type III morphology was larger in the proximal femur (41%) than in the other locations.

The LA axis of the femur did not correspond exactly to the AP axis at all femoral levels. At the distal femur, mean (SD) lateral offset of the LA axis was 5.5° (7.5°) on the left and 8.3° (8.9°) on the right. At the midshaft, mean (SD) medial offset of the LA axis was 3.1° (8.4°) on the left and 1.2° (7.9°) on the right. At the proximal femur, mean (SD) lateral offset of the LA axis was 5.4° (9.2°) on the left and 6.2° (8.3°) on the right. The side-to-side differences were statistically significant for the distal femur and midshaft but not the proximal femur. Table 2 lists the 95% confidence intervals for the mean values. As the range of differences was small (0.7°-2.8°), and the differences may not be clinically detected on gross inspection during surgery, we pooled both sides’ values to arrive at a single mean for each level. The LA axis was offset a mean (SD) of 6.9° (8.3°) laterally at the distal femur, 2.2° (8.2°) medially at the midshaft, and 5.8° (8.6°) laterally at the proximal femur. Figure 5 shows the frequency of distribution of LA axis offset.Offset of the LA axis from the AP axis of the femur was significantly (P < .001) different for each femoral level, even when a multivariate analysis was performed to determine the effect of sex, age, or side. Age and sex had no significant effect on mean offset of LA axis from AP axis.

We compared the mean arc between femoral neck axis and LA axis after referencing both off the PC axis. At the distal femur, mean (SD) arc between these 2 axes was 76.6° (13.1°) on the left and 68.3° (13.6°) on the right (mean difference, 8.3°); at the midshaft, mean (SD) arc was 85.2° (13.5°) on the left and 77.9° (13.1°) on the right (mean difference, 7.4°); at the proximal femur, mean (SD) arc was 76.7° (11.9°) on the left and 70.5° (12.8°) on the right (mean difference, 6.2°). The side-to-side differences were statistically significant (P < .001) for all locations.

In multivariate analysis, sex and age did not have an effect on mean arc between the 2 axes. Side and femoral level, however, had a significant effect (P < .001).

Discussion

In total hip arthroplasty, the goal is to restore femoral anteversion, usually referenced to the remaining femoral neck segment.³ In total knee arthroplasty (TKA), proper rotation preserves normal patellofemoral tracking.⁵ Various landmarks are used, such as the PCs or the epicondyles. After tumor resections, these landmarks are often lost.⁶ However, there are no reports of studies validating a particular method of achieving proper rotational orientation of tumor endoprostheses, though several methods are being used. One method involves inserting 2 drill bits before osteotomy—one proximal to the intended level of resection on the anterior femur, and the other on the anterior tibial shaft. The straight line formed can establish a plane of rotation (and length), which the surgeon must aim to restore when the components are placed. This method is useful for distal femur resections but not proximal femur resections. Another method, based on the LA’s anatomical position on the posterior aspect of the femur,⁴ uses the prominence of the LA to align the prosthesis. With this method, the LA is assumed to be directly posterior (6 o’clock) on the femur. However, this assumption has not been confirmed by any study. A third method, described by Heck and Carnesale,⁵ involves marking the anterior aspect of the femur after resection and aligning the components to it. The authors cautioned against using the LA as a landmark, saying that its course is highly variable.

The LA is a narrow, elevated length of bone, with medial and lateral lips, that serves as an attachment site for muscles in the posterior thigh. Proximally, the LA presents with lateral, medial, and intermediate lips. In the midshaft, it is often elevated by an underlying bony ridge or pilaster complex. Distally, it diverges into 2 ridges that form the triangular popliteal surface.^1,7 For the LA to be a reliable landmark, first it must be clearly identifiable on viewing a femoral cross-section. The LA that presents with type I or II morphology is distinctly identifiable, and an axis from its apex and bisecting the canal can easily be constructed. In our study, the LA presented with type I or II morphology in 82% of distal femoral sections and 99% of midshaft femoral sections. Therefore, the LA is a conspicuous landmark at these levels. In the proximal femur, 59% had type I or II morphology. Type III morphology could be identified on cross-sections by the persisting prominence of the lateral lip. However, it may be difficult to appreciate the LA with this morphology at surgery.

Once the LA is identified, its normal cross-sectional position must be defined. One way to do this is to establish the relationship of its axis (LA axis) to the true AP axis. Based on mean values, the LA axis is laterally offset 7º at the distal third of the femur, medially offset 2º at the midshaft, and laterally offset 6º at the proximal third. Therefore, for ideal placement with the LA used for orientation, the component must be internally rotated 7º relative to the LA for femoral resection at the distal third, externally rotated 2º for resection at the midshaft, and internally rotated 6º for resection at the proximal third. Studies have demonstrated that joint contact forces and mechanical alignment of the lower limb can be altered with as little as 5º of femoral malrotation.^8,9 Although such a small degree of malrotation is often asymptomatic, it can have long-term effects on soft-tissue tension and patellar tracking.^10,11 Rotating-platform mobile-bearing TKA designs can compensate for femoral malrotation, but they may have little to no effect on patellar tracking.¹² Therefore, we think aligning the components as near as possible to their natural orientation can prove beneficial in long-term patient management.

Another way of defining the normal cross-sectional position of the LA is to relate it to the femoral neck axis. We measured the difference between these 2 axes. Mean differences were 72º (distal femur), 81.5º (midshaft), and 73.5º (proximal third). Mean arc differences at all levels were larger on the left side—a reflection of the femoral neck being less anteverted on that side in our measurements. Standard deviations were smaller for measurements of LA axis offset from AP axis (range, 7.5°-9.2°) than for measurements of arc between LA axis and femoral neck axis (range, 11.9°-13.6°). This finding indicates there is less variation in the former method, making it preferable for defining the cross-sectional position of the LA.

It has been said that the course of the LA is variable, and our data provide confirmation. The LA does not lie directly posterior (6 o’clock), and it does not trace a straight longitudinal course along the posterior femur, as demonstrated by the different LA axis offsets at 3 levels. However, we may still use it as a landmark if we remain aware how much the LA is offset from the AP axis at each femoral level. Figures 6A-6D, which show CT scans of a patient who underwent distal femoral resection and replacement with an endoprosthesis, illustrate how the LA axis was measured before surgery and how proper prosthesis placement was confirmed after surgery.

In hip arthroplasty, restoration of normal femoral version is the reference for endoprosthetic placement. The literature on “normal” femoral anteversion varies with the method used. In a review of studies on CT-measured adult femoral version, reported values ranged from 6.3° to 40°.² Mean femoral version in our study ranged from 8° to 13°. Orthopedics textbooks generally put the value at 10° to 15º, and this seems to be the range that surgeons target.⁶ However, we found a statistically significant mean side-to-side difference of 5.4°. This finding is possibly explained by our large sample—it was larger than the samples used in other studies of CT-measured femoral version. Other studies have found mean side-to-side differences of up to 4.0º.⁵ Another explanation for our finding is that the studies may differ methodologically. The studies that established values for femoral anteversion were based on CT protocols—thinner slices (1-5 mm), use of foot holders to standardize limb rotation, use of 2 axial cuts in proximal femur to establish femoral neck axis^2,13—designed specifically for this measurement. As the CT scans reviewed in our study are not designed for this purpose, errors in femoral version measurement may have been introduced, which may also explain why there is larger variation in measurements of the arc between the LA axis and the femoral neck axis.

Conclusion

The LA does not lie directly on the posterior surface of the femur. It deviates 6.9° laterally at the distal femur, 2.2° medially at the midshaft, and 6.9° laterally at the proximal third. As the LA is an easily identifiable structure on cross-sections of the femoral shaft at the midshaft and distal third of the femur, it may be useful as a rotational landmark for resections at these levels if these deviations are considered during tumor endoprosthetic replacements.