Original Research

In recent years, hospitals have made considerable efforts to improve transitions of care, in part due to financial incentives from the Medicare Hospital Readmission Reduction Program (HRRP).¹ Initially focusing on three medical conditions, the HRRP has been associated with significant reductions in readmission rates.² Importantly, a small proportion of patients accounts for a very large proportion of hospital readmissions and hospital use.^3,4 Frequently hospitalized patients often have multiple chronic conditions and unique needs which may not be met by conventional approaches to healthcare delivery, including those influenced by the HRRP.^4-6 In light of this challenge, some hospitals have developed programs specifically focused on frequently hospitalized patients. A recent systematic review of these programs found relatively few studies of high quality, providing only limited insight in designing interventions to support this population.⁷ Moreover, no studies appear to have incorporated the patients’ perspectives into the design or adaptation of the model. Members of our research team developed and implemented the Complex High Admission Management Program (CHAMP) in January 2016 to address the needs of frequently hospitalized patients in our hospital. To enhance CHAMP and inform the design of programs serving similar populations in other health systems, we sought to identify factors associated with the onset and continuation of high hospital use. Our research question was, from the patients’ perspective, what factors contribute to patients’ becoming and continuing to be high users of hospital care.

METHODS

Setting, Study Design, and Participants

This qualitative study took place at Northwestern Memorial Hospital (NMH), an 894-bed urban academic hospital located in Chicago, Illinois. Between December 2016 and September 2017, we recruited adult patients admitted to the general medicine services. Eligible participants were identified with the assistance of a daily Northwestern Medicine Electronic Data Warehouse (EDW) search and included patients with two unplanned 30-day inpatient readmissions to NMH within the prior 12 months, in addition to one or more of the following criteria: (1) at least one readmission in the last six months; (2) a referral from one of the patient’s medical providers; or (3) at least three observation visits. We excluded patients whose preferred language was not English and those disoriented to person, place, or time. Considering NMH data showing that approximately one-third of high-utilizer patients have sickle cell disease, we used purposive sampling with the goal to compare findings within and between two groups of participants; those with and those without sickle cell disease. Our study was deemed exempt by the Northwestern University Institutional Review Board.

 

Participant Enrollment and Data Collection

We created an interview guide based on the research team’s experience with this population, a literature review, and our research question (See Appendix).^8,9 A research coordinator approached eligible participants during their hospital stay. The coordinator explained the study to eligible participants and obtained verbal consent for participation. The research coordinator then conducted one-on-one semi-structured interviews. Interviews were audio recorded for subsequent transcription and coding. Each interview lasted approximately 45 minutes. Participants were compensated with a $20 gift card for their time.

Analysis

Digital audio recordings from interviews were transcribed verbatim, deidentified, and analyzed using an iterative inductive team-based approach to coding.¹⁰ In our first cycle coding, all coders (KJO, SF, MMC, LO, KAC) independently reviewed and coded three transcripts using descriptive coding and subcoding to generate a preliminary codebook with code definitions.^10,11 Following the meetings to compare and compile our initial coding, each researcher then independently recoded the three transcripts with the developed codebook. The researchers met again to triangulate perspectives and reach a consensus on the final codebook. Using multiple coders is a standard process to control for subjective bias that one coder could bring to the coding process.¹² Following this meeting, the coders split into two teams of two (KJO, SF, and MMC, LO) to complete the coding of the remaining transcripts. Each team member independently coded the assigned transcripts and reconciled their codes with their counterpart; any discrepancies were resolved through discussion. Using this strategy, every transcript was coded by at least two team members. Our second coding cycle utilized pattern coding and involved identifying consistency both within and between transcripts; discovering associations between codes.^10,11,13 Constant comparison was used to compare responses among all participants, as well as between sickle-cell and nonsickle-cell participants.^13,14 Following team coding and reconciling, the analyses were presented to a broader research team for additional feedback and critique. All analyses were conducted using Dedoose version 8.0.35 (Los Angeles, California). Participant recruitment, interviews, and analysis of the transcripts continued until no new codes emerged and thematic saturation was achieved.

RESULTS

Participant Characteristics

Overall, we invited 34 patients to be interviewed; 26 consented and completed interviews (76.5%). Six (17.6%) patients declined participation, one (2.9%) was unable to complete the interview before hospital discharge, and one (2.9%) was excluded due to disorientation. Demographic characteristics of the 26 participants are shown in Table 1.

Four main themes emerged from our analysis. Table 2 summarizes these themes, subthemes, and provides representative quotes.

Major Medical Problem(s) are Universal, but High Hospital Use Varies in Onset

Not surprisingly, all participants described having at least one major medical problem. Some participants, such as those with genetic disorders, had experienced periods of high hospital use throughout their entire lifetime, while other participants experienced an onset of high hospital use as an adult after being previously healthy. Though most participants with genetic disorders had sickle cell anemia; one had a rare genetic disorder which caused chronic gastrointestinal symptoms. Participants typically described having a significant medical condition as well as other medical problems or complications from past surgery. Some participants described having a major medical problem which did not require frequent hospitalization until a complication or other medical problem arose, suggesting these new issues pushed them over a threshold beyond which self-management at home was less successful.

 

Course Fluctuates over Time and is Related to Psychological, Social, and Economic Factors

Participants identified psychological stress, social support, and financial constraints as factors which influence the course of their illness over time. Deaths in the family, breakups, and concerns about other family members were mentioned as specific forms of psychological stress and directly linked by participants to worsening of symptoms. Social support was present for most, but not all, participants, with no appreciable difference based on whether the participant had sickle cell disease. Social support was generally perceived as helpful, and several participants indicated a benefit to their own health when providing social support to others. Financial pressures also served as stressors and often impeded care due to lack of access to medications, other treatments, and housing.

Onset and Progression of Episodes Vary, but Generally Seem Uncontrollable

Regarding the onset of illness episodes, some participants described the sudden, unpredictable onset of symptoms, others described a more gradual onset which allowed them to attempt self-management. Regardless of the timing, episodes of illness were often perceived as spontaneous or triggered by factors outside of the participant’s control. Several participants, especially those with sickle cell disease, mentioned a relationship between their symptoms and the weather. Participants also noted the inconsistency in factors which may trigger an episode (ie, sometimes the factor exacerbated symptoms, while other times it did not). Participants also described having a high symptom burden with significant limitations in activities of daily living during episodes of illness. Pain was a very common component of symptoms regardless of whether or not the participant had sickle cell disease.

Individuals Seek Care after Self-Management Fails and Prefer to Avoid Hospitalization

Participants tried to control their symptoms with medications and typically sought care only when it was clear that this approach was not working, or they ran out of medications. This finding was consistent across both groups of participants (ie, those with and those without sickle cell disease). Many participants described very strong preferences not to come to the hospital; no participant described being in the hospital as a favorable or positive experience. Some participants mentioned that they had spent major holidays in the hospital and that they missed their family. No participant had a desire to come to the hospital.

DISCUSSION

In this study of frequently hospitalized patients, we found four major themes that illuminate patient perspectives about factors that contribute to high hospital use. While some of our findings corroborate those of previous studies, other emerging patterns were novel. Herein, we summarize key findings, provide context, and describe implications for the design of models of care for frequently hospitalized patients.

Similar to the findings of previous quantitative research, participants in our study described having a significant medical condition and typically had multiple medical conditions or complications.^4-6 Importantly, some participants described having a major medical problem which did not require frequent hospitalization until another medical problem or complication arose. This finding suggests that there may be an opportunity to identify patients with significant medical problems who are at elevated risk before the onset of high hospital use. Early identification of these high-risk patients could allow for the provision of additional support to prevent potential complications or address other factors which may contribute to the need for frequent hospitalization.

Participants in our study directly linked psychological stress to fluctuations in their course of illness. Previous research by Mautner and colleagues queried participants about childhood experiences and early life stressors and reported that early life instabilities and traumas were prevalent among patients with high levels of emergency and hospital-based healthcare utilization.¹⁵ Our participants identified more recent traumatic events (eg, the death of a loved one and breakups) when reflecting on factors contributing to illness exacerbations; early life trauma did not emerge as an identified contributor. Of note, unlike Mautner et al., we did not ask participants to reflect on childhood determinants of disease and illness specifically. Our findings suggest that psychological stress contributes to illness exacerbation, even for those patients without other significant psychiatric conditions (eg, depressive disorder, schizophrenia). Incorporating mental health professionals into programs for this patient population may improve health by teaching specific coping strategies, including cognitive-behavioral therapy for an acute stress disorder.^16,17

Social support was also a factor related to illness fluctuations over time. Notably, several participants indicated a benefit to their own health when providing social support to others, suggesting a role for peer support that may be reciprocally beneficial. This approach is supported by the literature. Williams and colleagues found that patients with sickle cell anemia experienced symptom improvement with peer support;¹⁸ while Johnson and colleagues recently reported a reduction in readmissions to acute care with the use of peer support for patients with severe mental illness.¹⁹

Financial constraints impeded care for some patients and served as a barrier to accessing medications, other treatments, and housing. Similar to the findings of prior quantitative research, our frequently hospitalized patients had a high proportion of patients with Medicaid and low proportion with private insurance, suggesting low socioeconomic status.^9,20 We did not formally collect data on income or economic status. Interestingly, prior qualitative studies have not identified financial constraints as a major theme, though this may be explained by differences in study populations and the overall objectives of the studies.^15,21 Importantly, the overwhelming majority of programs for frequently hospitalized patients identified in a recent systematic review included social workers.⁷ Our findings support the need to address financial constraints and the use of social workers in models of care for frequently hospitalized patients.

Many participants in our study felt that the factors contributing to exacerbations of illness were either inconsistent in their effect or out of their control. These findings have similarities to those from a qualitative study by Liu and colleagues in which they interviewed 20 “hospital-dependent” patients over 65 years of age.²¹ Though not explicitly focused on factors contributing to exacerbations, participants in their study felt that hospitalizations were generally inevitable. In our study, participants with sickle cell disease often identified changes in the weather as contributing to illness exacerbations. The relationship between weather and sickle cell disease remains incompletely understood, with an inconsistent association found in prior studies.²²

Participants in our study strongly desired to avoid hospitalization and typically sought hospital care when symptoms could not be controlled at home. This finding is in contrast to that from the study by Liu and colleagues where they found that hospital-dependent patients over 65 years had favorable perspectives of hospitalization because they felt safer and more secure in the hospital.²¹ Our participants were younger than those from the study by Liu and colleagues, had a high symptom burden, and may have been more concerned about control of those symptoms than the risk for clinical deterioration. Programs should aim to strengthen their support of patients’ self-management efforts early in the episode of illness and potentially offer home visits or a day hospital to avoid hospitalization. A recent systematic review found evidence that alternatives to inpatient care (eg, hospital-at-home) for low risk medical patients can achieve comparable outcomes at lower costs.²³ Similarly, some health systems have implemented day hospitals to treat low risk patients with uncomplicated sickle cell pain.^24,25

The heavy symptom burden experienced by participants in our study is notable. Pain was especially common. Programs may wish to partner with palliative care and addiction specialists to balance symptom relief with the simultaneous need to address comorbid substance and opioid use disorders when they are present.^4,9

Our study has several limitations. First, participants were recruited from the medicine service at a single academic hospital using criteria we developed to identify frequently hospitalized patients. Populations differ across hospitals and definitions of frequently hospitalized patients vary, limiting the generalizability of our findings. Second, we excluded patients whose preferred language was not English, as well as those disoriented to person, place, or time. It is possible that factors contributing to high hospital use differ for non-English speaking patients and those with cognitive deficits.

 

CONCLUSION

In this qualitative study, we identified factors associated with the onset and continuation of high hospital use. Emergent themes pointed to factors which influence patients’ onset of high hospital use, fluctuations in their illness over time, and triggers to seek care during an illness episode. These findings represent an important contribution to the literature because they allow patients’ perspectives to be incorporated into the design and adaptation of programs serving similar populations in other health systems. Programs that integrate patients’ perspectives into their design are likely to be better prepared to address patients’ needs and improve patient outcomes.

Acknowledgments

The authors thank the participants for their time and willingness to share their stories. The authors also thank Claire A. Knoten PhD and Erin Lambers PhD, former research team members who helped in the initial stages of the study.

Disclosures

The authors have nothing to disclose.

Funding

This project was funded by Northwestern Memorial Hospital and the Northwestern Medical Group.

 

In recent years, hospitals have made considerable efforts to improve transitions of care, in part due to financial incentives from the Medicare Hospital Readmission Reduction Program (HRRP).¹ Initially focusing on three medical conditions, the HRRP has been associated with significant reductions in readmission rates.² Importantly, a small proportion of patients accounts for a very large proportion of hospital readmissions and hospital use.^3,4 Frequently hospitalized patients often have multiple chronic conditions and unique needs which may not be met by conventional approaches to healthcare delivery, including those influenced by the HRRP.^4-6 In light of this challenge, some hospitals have developed programs specifically focused on frequently hospitalized patients. A recent systematic review of these programs found relatively few studies of high quality, providing only limited insight in designing interventions to support this population.⁷ Moreover, no studies appear to have incorporated the patients’ perspectives into the design or adaptation of the model. Members of our research team developed and implemented the Complex High Admission Management Program (CHAMP) in January 2016 to address the needs of frequently hospitalized patients in our hospital. To enhance CHAMP and inform the design of programs serving similar populations in other health systems, we sought to identify factors associated with the onset and continuation of high hospital use. Our research question was, from the patients’ perspective, what factors contribute to patients’ becoming and continuing to be high users of hospital care.

METHODS

Setting, Study Design, and Participants

This qualitative study took place at Northwestern Memorial Hospital (NMH), an 894-bed urban academic hospital located in Chicago, Illinois. Between December 2016 and September 2017, we recruited adult patients admitted to the general medicine services. Eligible participants were identified with the assistance of a daily Northwestern Medicine Electronic Data Warehouse (EDW) search and included patients with two unplanned 30-day inpatient readmissions to NMH within the prior 12 months, in addition to one or more of the following criteria: (1) at least one readmission in the last six months; (2) a referral from one of the patient’s medical providers; or (3) at least three observation visits. We excluded patients whose preferred language was not English and those disoriented to person, place, or time. Considering NMH data showing that approximately one-third of high-utilizer patients have sickle cell disease, we used purposive sampling with the goal to compare findings within and between two groups of participants; those with and those without sickle cell disease. Our study was deemed exempt by the Northwestern University Institutional Review Board.

 

Participant Enrollment and Data Collection

We created an interview guide based on the research team’s experience with this population, a literature review, and our research question (See Appendix).^8,9 A research coordinator approached eligible participants during their hospital stay. The coordinator explained the study to eligible participants and obtained verbal consent for participation. The research coordinator then conducted one-on-one semi-structured interviews. Interviews were audio recorded for subsequent transcription and coding. Each interview lasted approximately 45 minutes. Participants were compensated with a $20 gift card for their time.

Analysis

Digital audio recordings from interviews were transcribed verbatim, deidentified, and analyzed using an iterative inductive team-based approach to coding.¹⁰ In our first cycle coding, all coders (KJO, SF, MMC, LO, KAC) independently reviewed and coded three transcripts using descriptive coding and subcoding to generate a preliminary codebook with code definitions.^10,11 Following the meetings to compare and compile our initial coding, each researcher then independently recoded the three transcripts with the developed codebook. The researchers met again to triangulate perspectives and reach a consensus on the final codebook. Using multiple coders is a standard process to control for subjective bias that one coder could bring to the coding process.¹² Following this meeting, the coders split into two teams of two (KJO, SF, and MMC, LO) to complete the coding of the remaining transcripts. Each team member independently coded the assigned transcripts and reconciled their codes with their counterpart; any discrepancies were resolved through discussion. Using this strategy, every transcript was coded by at least two team members. Our second coding cycle utilized pattern coding and involved identifying consistency both within and between transcripts; discovering associations between codes.^10,11,13 Constant comparison was used to compare responses among all participants, as well as between sickle-cell and nonsickle-cell participants.^13,14 Following team coding and reconciling, the analyses were presented to a broader research team for additional feedback and critique. All analyses were conducted using Dedoose version 8.0.35 (Los Angeles, California). Participant recruitment, interviews, and analysis of the transcripts continued until no new codes emerged and thematic saturation was achieved.

RESULTS

Participant Characteristics

Overall, we invited 34 patients to be interviewed; 26 consented and completed interviews (76.5%). Six (17.6%) patients declined participation, one (2.9%) was unable to complete the interview before hospital discharge, and one (2.9%) was excluded due to disorientation. Demographic characteristics of the 26 participants are shown in Table 1.

Four main themes emerged from our analysis. Table 2 summarizes these themes, subthemes, and provides representative quotes.

Major Medical Problem(s) are Universal, but High Hospital Use Varies in Onset

Not surprisingly, all participants described having at least one major medical problem. Some participants, such as those with genetic disorders, had experienced periods of high hospital use throughout their entire lifetime, while other participants experienced an onset of high hospital use as an adult after being previously healthy. Though most participants with genetic disorders had sickle cell anemia; one had a rare genetic disorder which caused chronic gastrointestinal symptoms. Participants typically described having a significant medical condition as well as other medical problems or complications from past surgery. Some participants described having a major medical problem which did not require frequent hospitalization until a complication or other medical problem arose, suggesting these new issues pushed them over a threshold beyond which self-management at home was less successful.

 

Course Fluctuates over Time and is Related to Psychological, Social, and Economic Factors

Participants identified psychological stress, social support, and financial constraints as factors which influence the course of their illness over time. Deaths in the family, breakups, and concerns about other family members were mentioned as specific forms of psychological stress and directly linked by participants to worsening of symptoms. Social support was present for most, but not all, participants, with no appreciable difference based on whether the participant had sickle cell disease. Social support was generally perceived as helpful, and several participants indicated a benefit to their own health when providing social support to others. Financial pressures also served as stressors and often impeded care due to lack of access to medications, other treatments, and housing.

Onset and Progression of Episodes Vary, but Generally Seem Uncontrollable

Regarding the onset of illness episodes, some participants described the sudden, unpredictable onset of symptoms, others described a more gradual onset which allowed them to attempt self-management. Regardless of the timing, episodes of illness were often perceived as spontaneous or triggered by factors outside of the participant’s control. Several participants, especially those with sickle cell disease, mentioned a relationship between their symptoms and the weather. Participants also noted the inconsistency in factors which may trigger an episode (ie, sometimes the factor exacerbated symptoms, while other times it did not). Participants also described having a high symptom burden with significant limitations in activities of daily living during episodes of illness. Pain was a very common component of symptoms regardless of whether or not the participant had sickle cell disease.

Individuals Seek Care after Self-Management Fails and Prefer to Avoid Hospitalization

Participants tried to control their symptoms with medications and typically sought care only when it was clear that this approach was not working, or they ran out of medications. This finding was consistent across both groups of participants (ie, those with and those without sickle cell disease). Many participants described very strong preferences not to come to the hospital; no participant described being in the hospital as a favorable or positive experience. Some participants mentioned that they had spent major holidays in the hospital and that they missed their family. No participant had a desire to come to the hospital.

DISCUSSION

In this study of frequently hospitalized patients, we found four major themes that illuminate patient perspectives about factors that contribute to high hospital use. While some of our findings corroborate those of previous studies, other emerging patterns were novel. Herein, we summarize key findings, provide context, and describe implications for the design of models of care for frequently hospitalized patients.

Similar to the findings of previous quantitative research, participants in our study described having a significant medical condition and typically had multiple medical conditions or complications.^4-6 Importantly, some participants described having a major medical problem which did not require frequent hospitalization until another medical problem or complication arose. This finding suggests that there may be an opportunity to identify patients with significant medical problems who are at elevated risk before the onset of high hospital use. Early identification of these high-risk patients could allow for the provision of additional support to prevent potential complications or address other factors which may contribute to the need for frequent hospitalization.

Participants in our study directly linked psychological stress to fluctuations in their course of illness. Previous research by Mautner and colleagues queried participants about childhood experiences and early life stressors and reported that early life instabilities and traumas were prevalent among patients with high levels of emergency and hospital-based healthcare utilization.¹⁵ Our participants identified more recent traumatic events (eg, the death of a loved one and breakups) when reflecting on factors contributing to illness exacerbations; early life trauma did not emerge as an identified contributor. Of note, unlike Mautner et al., we did not ask participants to reflect on childhood determinants of disease and illness specifically. Our findings suggest that psychological stress contributes to illness exacerbation, even for those patients without other significant psychiatric conditions (eg, depressive disorder, schizophrenia). Incorporating mental health professionals into programs for this patient population may improve health by teaching specific coping strategies, including cognitive-behavioral therapy for an acute stress disorder.^16,17

Social support was also a factor related to illness fluctuations over time. Notably, several participants indicated a benefit to their own health when providing social support to others, suggesting a role for peer support that may be reciprocally beneficial. This approach is supported by the literature. Williams and colleagues found that patients with sickle cell anemia experienced symptom improvement with peer support;¹⁸ while Johnson and colleagues recently reported a reduction in readmissions to acute care with the use of peer support for patients with severe mental illness.¹⁹

Financial constraints impeded care for some patients and served as a barrier to accessing medications, other treatments, and housing. Similar to the findings of prior quantitative research, our frequently hospitalized patients had a high proportion of patients with Medicaid and low proportion with private insurance, suggesting low socioeconomic status.^9,20 We did not formally collect data on income or economic status. Interestingly, prior qualitative studies have not identified financial constraints as a major theme, though this may be explained by differences in study populations and the overall objectives of the studies.^15,21 Importantly, the overwhelming majority of programs for frequently hospitalized patients identified in a recent systematic review included social workers.⁷ Our findings support the need to address financial constraints and the use of social workers in models of care for frequently hospitalized patients.

Many participants in our study felt that the factors contributing to exacerbations of illness were either inconsistent in their effect or out of their control. These findings have similarities to those from a qualitative study by Liu and colleagues in which they interviewed 20 “hospital-dependent” patients over 65 years of age.²¹ Though not explicitly focused on factors contributing to exacerbations, participants in their study felt that hospitalizations were generally inevitable. In our study, participants with sickle cell disease often identified changes in the weather as contributing to illness exacerbations. The relationship between weather and sickle cell disease remains incompletely understood, with an inconsistent association found in prior studies.²²

Participants in our study strongly desired to avoid hospitalization and typically sought hospital care when symptoms could not be controlled at home. This finding is in contrast to that from the study by Liu and colleagues where they found that hospital-dependent patients over 65 years had favorable perspectives of hospitalization because they felt safer and more secure in the hospital.²¹ Our participants were younger than those from the study by Liu and colleagues, had a high symptom burden, and may have been more concerned about control of those symptoms than the risk for clinical deterioration. Programs should aim to strengthen their support of patients’ self-management efforts early in the episode of illness and potentially offer home visits or a day hospital to avoid hospitalization. A recent systematic review found evidence that alternatives to inpatient care (eg, hospital-at-home) for low risk medical patients can achieve comparable outcomes at lower costs.²³ Similarly, some health systems have implemented day hospitals to treat low risk patients with uncomplicated sickle cell pain.^24,25

The heavy symptom burden experienced by participants in our study is notable. Pain was especially common. Programs may wish to partner with palliative care and addiction specialists to balance symptom relief with the simultaneous need to address comorbid substance and opioid use disorders when they are present.^4,9

Our study has several limitations. First, participants were recruited from the medicine service at a single academic hospital using criteria we developed to identify frequently hospitalized patients. Populations differ across hospitals and definitions of frequently hospitalized patients vary, limiting the generalizability of our findings. Second, we excluded patients whose preferred language was not English, as well as those disoriented to person, place, or time. It is possible that factors contributing to high hospital use differ for non-English speaking patients and those with cognitive deficits.

 

CONCLUSION

In this qualitative study, we identified factors associated with the onset and continuation of high hospital use. Emergent themes pointed to factors which influence patients’ onset of high hospital use, fluctuations in their illness over time, and triggers to seek care during an illness episode. These findings represent an important contribution to the literature because they allow patients’ perspectives to be incorporated into the design and adaptation of programs serving similar populations in other health systems. Programs that integrate patients’ perspectives into their design are likely to be better prepared to address patients’ needs and improve patient outcomes.

Acknowledgments

The authors thank the participants for their time and willingness to share their stories. The authors also thank Claire A. Knoten PhD and Erin Lambers PhD, former research team members who helped in the initial stages of the study.

Disclosures

The authors have nothing to disclose.

Funding

This project was funded by Northwestern Memorial Hospital and the Northwestern Medical Group.

 

In recent years, hospitals have made considerable efforts to improve transitions of care, in part due to financial incentives from the Medicare Hospital Readmission Reduction Program (HRRP).¹ Initially focusing on three medical conditions, the HRRP has been associated with significant reductions in readmission rates.² Importantly, a small proportion of patients accounts for a very large proportion of hospital readmissions and hospital use.^3,4 Frequently hospitalized patients often have multiple chronic conditions and unique needs which may not be met by conventional approaches to healthcare delivery, including those influenced by the HRRP.^4-6 In light of this challenge, some hospitals have developed programs specifically focused on frequently hospitalized patients. A recent systematic review of these programs found relatively few studies of high quality, providing only limited insight in designing interventions to support this population.⁷ Moreover, no studies appear to have incorporated the patients’ perspectives into the design or adaptation of the model. Members of our research team developed and implemented the Complex High Admission Management Program (CHAMP) in January 2016 to address the needs of frequently hospitalized patients in our hospital. To enhance CHAMP and inform the design of programs serving similar populations in other health systems, we sought to identify factors associated with the onset and continuation of high hospital use. Our research question was, from the patients’ perspective, what factors contribute to patients’ becoming and continuing to be high users of hospital care.

METHODS

Setting, Study Design, and Participants

This qualitative study took place at Northwestern Memorial Hospital (NMH), an 894-bed urban academic hospital located in Chicago, Illinois. Between December 2016 and September 2017, we recruited adult patients admitted to the general medicine services. Eligible participants were identified with the assistance of a daily Northwestern Medicine Electronic Data Warehouse (EDW) search and included patients with two unplanned 30-day inpatient readmissions to NMH within the prior 12 months, in addition to one or more of the following criteria: (1) at least one readmission in the last six months; (2) a referral from one of the patient’s medical providers; or (3) at least three observation visits. We excluded patients whose preferred language was not English and those disoriented to person, place, or time. Considering NMH data showing that approximately one-third of high-utilizer patients have sickle cell disease, we used purposive sampling with the goal to compare findings within and between two groups of participants; those with and those without sickle cell disease. Our study was deemed exempt by the Northwestern University Institutional Review Board.

 

Participant Enrollment and Data Collection

We created an interview guide based on the research team’s experience with this population, a literature review, and our research question (See Appendix).^8,9 A research coordinator approached eligible participants during their hospital stay. The coordinator explained the study to eligible participants and obtained verbal consent for participation. The research coordinator then conducted one-on-one semi-structured interviews. Interviews were audio recorded for subsequent transcription and coding. Each interview lasted approximately 45 minutes. Participants were compensated with a $20 gift card for their time.

Analysis

Digital audio recordings from interviews were transcribed verbatim, deidentified, and analyzed using an iterative inductive team-based approach to coding.¹⁰ In our first cycle coding, all coders (KJO, SF, MMC, LO, KAC) independently reviewed and coded three transcripts using descriptive coding and subcoding to generate a preliminary codebook with code definitions.^10,11 Following the meetings to compare and compile our initial coding, each researcher then independently recoded the three transcripts with the developed codebook. The researchers met again to triangulate perspectives and reach a consensus on the final codebook. Using multiple coders is a standard process to control for subjective bias that one coder could bring to the coding process.¹² Following this meeting, the coders split into two teams of two (KJO, SF, and MMC, LO) to complete the coding of the remaining transcripts. Each team member independently coded the assigned transcripts and reconciled their codes with their counterpart; any discrepancies were resolved through discussion. Using this strategy, every transcript was coded by at least two team members. Our second coding cycle utilized pattern coding and involved identifying consistency both within and between transcripts; discovering associations between codes.^10,11,13 Constant comparison was used to compare responses among all participants, as well as between sickle-cell and nonsickle-cell participants.^13,14 Following team coding and reconciling, the analyses were presented to a broader research team for additional feedback and critique. All analyses were conducted using Dedoose version 8.0.35 (Los Angeles, California). Participant recruitment, interviews, and analysis of the transcripts continued until no new codes emerged and thematic saturation was achieved.

RESULTS

Participant Characteristics

Overall, we invited 34 patients to be interviewed; 26 consented and completed interviews (76.5%). Six (17.6%) patients declined participation, one (2.9%) was unable to complete the interview before hospital discharge, and one (2.9%) was excluded due to disorientation. Demographic characteristics of the 26 participants are shown in Table 1.

Four main themes emerged from our analysis. Table 2 summarizes these themes, subthemes, and provides representative quotes.

Major Medical Problem(s) are Universal, but High Hospital Use Varies in Onset

Not surprisingly, all participants described having at least one major medical problem. Some participants, such as those with genetic disorders, had experienced periods of high hospital use throughout their entire lifetime, while other participants experienced an onset of high hospital use as an adult after being previously healthy. Though most participants with genetic disorders had sickle cell anemia; one had a rare genetic disorder which caused chronic gastrointestinal symptoms. Participants typically described having a significant medical condition as well as other medical problems or complications from past surgery. Some participants described having a major medical problem which did not require frequent hospitalization until a complication or other medical problem arose, suggesting these new issues pushed them over a threshold beyond which self-management at home was less successful.

 

Course Fluctuates over Time and is Related to Psychological, Social, and Economic Factors

Participants identified psychological stress, social support, and financial constraints as factors which influence the course of their illness over time. Deaths in the family, breakups, and concerns about other family members were mentioned as specific forms of psychological stress and directly linked by participants to worsening of symptoms. Social support was present for most, but not all, participants, with no appreciable difference based on whether the participant had sickle cell disease. Social support was generally perceived as helpful, and several participants indicated a benefit to their own health when providing social support to others. Financial pressures also served as stressors and often impeded care due to lack of access to medications, other treatments, and housing.

Onset and Progression of Episodes Vary, but Generally Seem Uncontrollable

Regarding the onset of illness episodes, some participants described the sudden, unpredictable onset of symptoms, others described a more gradual onset which allowed them to attempt self-management. Regardless of the timing, episodes of illness were often perceived as spontaneous or triggered by factors outside of the participant’s control. Several participants, especially those with sickle cell disease, mentioned a relationship between their symptoms and the weather. Participants also noted the inconsistency in factors which may trigger an episode (ie, sometimes the factor exacerbated symptoms, while other times it did not). Participants also described having a high symptom burden with significant limitations in activities of daily living during episodes of illness. Pain was a very common component of symptoms regardless of whether or not the participant had sickle cell disease.

Individuals Seek Care after Self-Management Fails and Prefer to Avoid Hospitalization

Participants tried to control their symptoms with medications and typically sought care only when it was clear that this approach was not working, or they ran out of medications. This finding was consistent across both groups of participants (ie, those with and those without sickle cell disease). Many participants described very strong preferences not to come to the hospital; no participant described being in the hospital as a favorable or positive experience. Some participants mentioned that they had spent major holidays in the hospital and that they missed their family. No participant had a desire to come to the hospital.

DISCUSSION

In this study of frequently hospitalized patients, we found four major themes that illuminate patient perspectives about factors that contribute to high hospital use. While some of our findings corroborate those of previous studies, other emerging patterns were novel. Herein, we summarize key findings, provide context, and describe implications for the design of models of care for frequently hospitalized patients.

Similar to the findings of previous quantitative research, participants in our study described having a significant medical condition and typically had multiple medical conditions or complications.^4-6 Importantly, some participants described having a major medical problem which did not require frequent hospitalization until another medical problem or complication arose. This finding suggests that there may be an opportunity to identify patients with significant medical problems who are at elevated risk before the onset of high hospital use. Early identification of these high-risk patients could allow for the provision of additional support to prevent potential complications or address other factors which may contribute to the need for frequent hospitalization.

Participants in our study directly linked psychological stress to fluctuations in their course of illness. Previous research by Mautner and colleagues queried participants about childhood experiences and early life stressors and reported that early life instabilities and traumas were prevalent among patients with high levels of emergency and hospital-based healthcare utilization.¹⁵ Our participants identified more recent traumatic events (eg, the death of a loved one and breakups) when reflecting on factors contributing to illness exacerbations; early life trauma did not emerge as an identified contributor. Of note, unlike Mautner et al., we did not ask participants to reflect on childhood determinants of disease and illness specifically. Our findings suggest that psychological stress contributes to illness exacerbation, even for those patients without other significant psychiatric conditions (eg, depressive disorder, schizophrenia). Incorporating mental health professionals into programs for this patient population may improve health by teaching specific coping strategies, including cognitive-behavioral therapy for an acute stress disorder.^16,17

Social support was also a factor related to illness fluctuations over time. Notably, several participants indicated a benefit to their own health when providing social support to others, suggesting a role for peer support that may be reciprocally beneficial. This approach is supported by the literature. Williams and colleagues found that patients with sickle cell anemia experienced symptom improvement with peer support;¹⁸ while Johnson and colleagues recently reported a reduction in readmissions to acute care with the use of peer support for patients with severe mental illness.¹⁹

Financial constraints impeded care for some patients and served as a barrier to accessing medications, other treatments, and housing. Similar to the findings of prior quantitative research, our frequently hospitalized patients had a high proportion of patients with Medicaid and low proportion with private insurance, suggesting low socioeconomic status.^9,20 We did not formally collect data on income or economic status. Interestingly, prior qualitative studies have not identified financial constraints as a major theme, though this may be explained by differences in study populations and the overall objectives of the studies.^15,21 Importantly, the overwhelming majority of programs for frequently hospitalized patients identified in a recent systematic review included social workers.⁷ Our findings support the need to address financial constraints and the use of social workers in models of care for frequently hospitalized patients.

Many participants in our study felt that the factors contributing to exacerbations of illness were either inconsistent in their effect or out of their control. These findings have similarities to those from a qualitative study by Liu and colleagues in which they interviewed 20 “hospital-dependent” patients over 65 years of age.²¹ Though not explicitly focused on factors contributing to exacerbations, participants in their study felt that hospitalizations were generally inevitable. In our study, participants with sickle cell disease often identified changes in the weather as contributing to illness exacerbations. The relationship between weather and sickle cell disease remains incompletely understood, with an inconsistent association found in prior studies.²²

Participants in our study strongly desired to avoid hospitalization and typically sought hospital care when symptoms could not be controlled at home. This finding is in contrast to that from the study by Liu and colleagues where they found that hospital-dependent patients over 65 years had favorable perspectives of hospitalization because they felt safer and more secure in the hospital.²¹ Our participants were younger than those from the study by Liu and colleagues, had a high symptom burden, and may have been more concerned about control of those symptoms than the risk for clinical deterioration. Programs should aim to strengthen their support of patients’ self-management efforts early in the episode of illness and potentially offer home visits or a day hospital to avoid hospitalization. A recent systematic review found evidence that alternatives to inpatient care (eg, hospital-at-home) for low risk medical patients can achieve comparable outcomes at lower costs.²³ Similarly, some health systems have implemented day hospitals to treat low risk patients with uncomplicated sickle cell pain.^24,25

The heavy symptom burden experienced by participants in our study is notable. Pain was especially common. Programs may wish to partner with palliative care and addiction specialists to balance symptom relief with the simultaneous need to address comorbid substance and opioid use disorders when they are present.^4,9

Our study has several limitations. First, participants were recruited from the medicine service at a single academic hospital using criteria we developed to identify frequently hospitalized patients. Populations differ across hospitals and definitions of frequently hospitalized patients vary, limiting the generalizability of our findings. Second, we excluded patients whose preferred language was not English, as well as those disoriented to person, place, or time. It is possible that factors contributing to high hospital use differ for non-English speaking patients and those with cognitive deficits.

 

CONCLUSION

In this qualitative study, we identified factors associated with the onset and continuation of high hospital use. Emergent themes pointed to factors which influence patients’ onset of high hospital use, fluctuations in their illness over time, and triggers to seek care during an illness episode. These findings represent an important contribution to the literature because they allow patients’ perspectives to be incorporated into the design and adaptation of programs serving similar populations in other health systems. Programs that integrate patients’ perspectives into their design are likely to be better prepared to address patients’ needs and improve patient outcomes.

Acknowledgments

The authors thank the participants for their time and willingness to share their stories. The authors also thank Claire A. Knoten PhD and Erin Lambers PhD, former research team members who helped in the initial stages of the study.

Disclosures

The authors have nothing to disclose.

Funding

This project was funded by Northwestern Memorial Hospital and the Northwestern Medical Group.

 

There is an increasing utilization of advanced practice providers (APPs) in the delivery of healthcare in the United States.^1,2 As of 2016, there were 157, 025 nurse practitioners (NPs) and 102,084 physician assistants (PAs) with a projected growth rate of 6.8% and 4.3%, respectively, which exceeds the physician growth rate of 1.1%.² This increased growth rate has been attributed to the expectation that APPs can enhance the quality of physician care, relieve physician shortages, and reduce service costs, as APPs are less expensive to hire than physicians.^3,4 Hospital medicine is the fastest growing medical field in the United States, and approximately 83% of hospitalist groups around the country utilize APPs; however, the demand for hospitalists continues to exceed the supply, and this has led to increased utilization of APPs in hospital medicine.^5-10

APPs receive very limited inpatient training and there is wide variation in their clinical abilities after graduation.¹¹ This is an issue that has become exacerbated in recent years by a change in the training process for PAs. Before 2005, PA programs were typically two to three years long and required the same prerequisite courses as medical schools.¹¹ PA students completed more than 2,000 hours of clinical rotations and then had to pass the Physician Assistant National Certifying Exam before they could practice.¹² Traditionally, PA programs typically attracted students with prior healthcare experience.¹¹ In 2005, PA programs began transitioning from bachelor’s degrees to requiring a master’s level degree for completion of the programs. This has shifted the demographics of the students matriculating to younger students with little-to-no prior healthcare experience; moreover, these fresh graduates lack exposure to hospital medicine.¹¹

NPs usually gain clinical experience working as registered nurses (RNs) for two or more years prior to entry into the NP program. NP programs for baccalaureate-prepared RNs vary in length from two to three years.² There is an acute care focus for NPs in training; however, there is no standardized training or licensure to ensure that hospital medicine competencies are met.^13-15 Some studies have shown that a lack of structured support has been found to affect NP role transition negatively during the first year of practice,¹⁶ and graduating NPs have indicated that they needed more out of their clinical education in terms of content, clinical experience, and competency testing.¹⁷

Hiring new APP graduates as hospitalists requires a longer and more rigorous onboarding process. On‐the‐job training in hospital medicine for new APP graduates can take as long as six to 12 months in order for them to acquire the basic skill set necessary to adequately manage hospitalized patients.¹⁵ This extended onboarding is costly because the APPs are receiving a full hospitalist salary, yet they are not functioning at full capacity. Ideally, there should be an intermediary training step between graduation and employment as hospitalist APPs. Studies have shown that APPs are interested in formal postgraduate hospital medicine training, even if it means having a lower stipend during the first year after graduating from their NP or PA program.^9,15,18

The growing need for hospitalists, driven by residency work-hour reform, increased age and complexity of patients, and the need to improve the quality of inpatient care while simultaneously reducing waste, has contributed to the increasing utilization of and need for highly qualified APPs in hospital medicine.^11,19,20 We established a fellowship to train APPs. The goal of this study was to determine if an APP fellowship is a cost-effective pipeline for filling vacancies within a hospitalist program.

 

METHODS

Design and Setting

Johns Hopkins Bayview Medical Center (JHBMC) is a 440 bed hospital in Baltimore Maryland. The hospitalist group was started in 1996 with one physician seeing approximately 500 discharges a year. Over the last 20 years, the group has grown and is now its own division with 57 providers, including 42 physicians, 11 APPs, and four APP fellows. The hospitalist division manages ~7,000 discharges a year, which corresponds to approximately 60% of admissions to general medicine. Hospitalist APPs help staff general medicine by working alongside doctors and admitting patients during the day and night. The APPs also staff the pulmonary step down unit with a pulmonary attending and the chemical dependency unit with an internal medicine addiction specialist.

The growth of the division of hospital medicine at JHBMC is a result of increasing volumes and reduced residency duty hours. The increasing full time equivalents (FTEs) resulted in a need for APPs; however, vacancies went unfilled for an average of 35 weeks due to the time it took to post open positions, interview applicants, and hire applicants through the credentialing process. Further, it took as long as 22 to 34 weeks for a new hire to work independently. The APP vacancies and onboarding resulted in increased costs to the division incurred by physician moonlighting to cover open shifts. The hourly physician moonlighting rate at JHBMC is $150. All costs were calculated on the basis of a 40-hour work week. We performed a pre- and postanalysis of outcomes of interest between January 2009 and June 2018. This study was exempt from institutional review board review.

Intervention

In 2014, a one year APP clinical fellowship in hospital medicine was started. The fellows evaluate and manage patients working one-on-one with an experienced hospitalist faculty member. The program consists of 80% clinical experience in the inpatient setting and 20% didactic instruction (Table 1). Up to four fellows are accepted each year and are eligible for hire after training if vacancies exist. The program is cost neutral and was financed by downsizing, through attrition, two physician FTEs. Four APP fellows’ salaries are the equivalent of two entry-level hospitalist physicians’ salaries at JHBMC. The annual salary for an APP fellow is $69,000.

Downsizing by two physician FTEs meant that one less doctor was scheduled every day. The patient load previously seen by that one doctor (10 patients) was absorbed by the MD–APP fellow dyads. Paired with a fellow, each physician sees a higher cap of 13 patients, and it takes six weeks for the fellows to ramp-up to this patient load. When the fellow first starts, the team sees 10 patients. Every two weeks, the pair’s census increases by one patient to the cap of 13. Collectively, the four APP fellow–MD dyads make it possible for four physicians to see an additional 12 patients. The two extra patients absorbed by the service per day results in a net increase in capacity of up to 730 patient encounters a year.

 

Outcomes and Analysis

Our main outcomes of interest were duration of onboarding and cost incurred by the division to (1) staff the service during a vacancy and (2) onboard new hires. Secondary outcomes included duration of vacancy and total time spent with the group. We collected basic demographic data on participants, including, age, gender, and race. Demographics and outcomes of interest were compared pre- (2009-2013) and post- (2014-2018) initiation of the APP clinical fellowship using the chi-square test, the t-test for normally distributed data, and the Wilcoxon rank-sum for nonnormally distributed data, as appropriate. The normality of the data distribution was tested using the Shapiro-Wilk W test. Two-tailed P values less than .05 were considered to be statistically significant. Results were analyzed using Stata/MP version 13.0 (StataCorp Inc, College Station, Texas).

RESULTS

Twelve fellows have been recruited, and of these, 10 have graduated. Two chose to leave the program prior to completion. Of the 10 fellows that have graduated, six have been hired into our group, one was hired within our facility, and three were hired as hospitalists at other institutions. The median time from APP school graduation to hire was also not different between the two groups (10.5 vs 3.9 months, P = .069). In addition, the total time that the new APP hires spent with the group was nonstatistically significantly different between the two periods (17.9 vs 18.3 months, P = .735). Both the mean duration of onboarding and the cost to the division were significantly reduced after implementation of the program (25.4 vs 11.0 weeks, P = .017 and $361,714 vs $66,000, P = .004; Table 2).

The yearly cost of an APP vacancy and onboarding is incurred by doctor moonlighting costs (at the rate of $150 per hour) to cover open shifts. The mean duration of vacancies and onboarding each year was 34.9 and 25.4 weeks, respectively, before the fellowship. The yearly cost of onboarding, after the establishment of the fellowship, is a maximum of $66,000, derived from physician moonlighting to cover the six-week ramp-up at the very beginning of the fellowship and the five weeks of orientation to the pulmonary and chemical dependency units after the fellowship (Table 3).

DISCUSSION

Our APP clinical fellowship in hospital medicine at JHBMC has produced several benefits. First, the fellowship has become a pipeline for filling APP vacancies within our division. We have been able to hire for four consecutive years from the fellowship. Second, the ready availability of high-functioning and efficient APP hospitalists has cut down on the onboarding time for our new APP hires. Many new APP graduates lack confidence in caring for complex hospitalized patients. Following our 12-month clinical fellowship, our matriculated fellows are able to practice at the top of their license immediately and confidently. Third, the reduced vacancy and shortened onboarding periods have reduced costs to the division. Fourth, the fellowship has created additional teaching avenues for the faculty. The medicine units at JHBMC are comprised of hospitalist and internal medicine residency services. The hospitalists spend the majority of their clinical time in direct patient care; however, they rotate on the residency service for two weeks out of the year. The majority of physicians welcome the chance to teach more, and partnering with an APP fellow provides that opportunity.

 

As we have developed and grown this program, the one great challenge has been what to do with graduating fellows when we cannot hire them. Fortunately, the market for highly qualified, well trained APPs is strong, and every one of the fellows that we could not hire within our group has been able to find a position either within our facility or outside our institution. To facilitate this process, program directors and recruiters are invited to meet with the fellows toward the end of their fellowship to share employment opportunities with them.

Our study has limitations. First, had the $276,000 from the attrition of two physicians been used to hire nonfellow APPs under the old model, then the costs of the two models would have been similar, but this was simply not possible because the positions could not be filled. Second, this is a single-site experience, and our findings may not be generalizable, particularly those pertaining to remuneration. Third, our study was underpowered to detect small but important differences in characteristics of APPs, especially time from graduation to hire, before and after the implementation of our fellowship. Further research comparing various programs both in structure and outcomes—such as fellows’ readiness for practice, costs, duration of vacancies, and provider satisfaction—are an important next step.

We have developed a pool of applicants within our division to fill vacancies left by turnover from senior NPs and PAs. This program has reduced costs and improved the joy of practice for both doctors and APPs. As the need for highly qualified NPs and PAs in hospital medicine continues to grow, we may see more APP fellowships in hospital medicine in the United States.

Acknowledgments

The authors thank the advanced practice providers who have helped us grow and refine our fellowship.

Disclosures

The authors have nothing to disclose

There is an increasing utilization of advanced practice providers (APPs) in the delivery of healthcare in the United States.^1,2 As of 2016, there were 157, 025 nurse practitioners (NPs) and 102,084 physician assistants (PAs) with a projected growth rate of 6.8% and 4.3%, respectively, which exceeds the physician growth rate of 1.1%.² This increased growth rate has been attributed to the expectation that APPs can enhance the quality of physician care, relieve physician shortages, and reduce service costs, as APPs are less expensive to hire than physicians.^3,4 Hospital medicine is the fastest growing medical field in the United States, and approximately 83% of hospitalist groups around the country utilize APPs; however, the demand for hospitalists continues to exceed the supply, and this has led to increased utilization of APPs in hospital medicine.^5-10

APPs receive very limited inpatient training and there is wide variation in their clinical abilities after graduation.¹¹ This is an issue that has become exacerbated in recent years by a change in the training process for PAs. Before 2005, PA programs were typically two to three years long and required the same prerequisite courses as medical schools.¹¹ PA students completed more than 2,000 hours of clinical rotations and then had to pass the Physician Assistant National Certifying Exam before they could practice.¹² Traditionally, PA programs typically attracted students with prior healthcare experience.¹¹ In 2005, PA programs began transitioning from bachelor’s degrees to requiring a master’s level degree for completion of the programs. This has shifted the demographics of the students matriculating to younger students with little-to-no prior healthcare experience; moreover, these fresh graduates lack exposure to hospital medicine.¹¹

NPs usually gain clinical experience working as registered nurses (RNs) for two or more years prior to entry into the NP program. NP programs for baccalaureate-prepared RNs vary in length from two to three years.² There is an acute care focus for NPs in training; however, there is no standardized training or licensure to ensure that hospital medicine competencies are met.^13-15 Some studies have shown that a lack of structured support has been found to affect NP role transition negatively during the first year of practice,¹⁶ and graduating NPs have indicated that they needed more out of their clinical education in terms of content, clinical experience, and competency testing.¹⁷

Hiring new APP graduates as hospitalists requires a longer and more rigorous onboarding process. On‐the‐job training in hospital medicine for new APP graduates can take as long as six to 12 months in order for them to acquire the basic skill set necessary to adequately manage hospitalized patients.¹⁵ This extended onboarding is costly because the APPs are receiving a full hospitalist salary, yet they are not functioning at full capacity. Ideally, there should be an intermediary training step between graduation and employment as hospitalist APPs. Studies have shown that APPs are interested in formal postgraduate hospital medicine training, even if it means having a lower stipend during the first year after graduating from their NP or PA program.^9,15,18

The growing need for hospitalists, driven by residency work-hour reform, increased age and complexity of patients, and the need to improve the quality of inpatient care while simultaneously reducing waste, has contributed to the increasing utilization of and need for highly qualified APPs in hospital medicine.^11,19,20 We established a fellowship to train APPs. The goal of this study was to determine if an APP fellowship is a cost-effective pipeline for filling vacancies within a hospitalist program.

 

METHODS

Design and Setting

Johns Hopkins Bayview Medical Center (JHBMC) is a 440 bed hospital in Baltimore Maryland. The hospitalist group was started in 1996 with one physician seeing approximately 500 discharges a year. Over the last 20 years, the group has grown and is now its own division with 57 providers, including 42 physicians, 11 APPs, and four APP fellows. The hospitalist division manages ~7,000 discharges a year, which corresponds to approximately 60% of admissions to general medicine. Hospitalist APPs help staff general medicine by working alongside doctors and admitting patients during the day and night. The APPs also staff the pulmonary step down unit with a pulmonary attending and the chemical dependency unit with an internal medicine addiction specialist.

The growth of the division of hospital medicine at JHBMC is a result of increasing volumes and reduced residency duty hours. The increasing full time equivalents (FTEs) resulted in a need for APPs; however, vacancies went unfilled for an average of 35 weeks due to the time it took to post open positions, interview applicants, and hire applicants through the credentialing process. Further, it took as long as 22 to 34 weeks for a new hire to work independently. The APP vacancies and onboarding resulted in increased costs to the division incurred by physician moonlighting to cover open shifts. The hourly physician moonlighting rate at JHBMC is $150. All costs were calculated on the basis of a 40-hour work week. We performed a pre- and postanalysis of outcomes of interest between January 2009 and June 2018. This study was exempt from institutional review board review.

Intervention

In 2014, a one year APP clinical fellowship in hospital medicine was started. The fellows evaluate and manage patients working one-on-one with an experienced hospitalist faculty member. The program consists of 80% clinical experience in the inpatient setting and 20% didactic instruction (Table 1). Up to four fellows are accepted each year and are eligible for hire after training if vacancies exist. The program is cost neutral and was financed by downsizing, through attrition, two physician FTEs. Four APP fellows’ salaries are the equivalent of two entry-level hospitalist physicians’ salaries at JHBMC. The annual salary for an APP fellow is $69,000.

Downsizing by two physician FTEs meant that one less doctor was scheduled every day. The patient load previously seen by that one doctor (10 patients) was absorbed by the MD–APP fellow dyads. Paired with a fellow, each physician sees a higher cap of 13 patients, and it takes six weeks for the fellows to ramp-up to this patient load. When the fellow first starts, the team sees 10 patients. Every two weeks, the pair’s census increases by one patient to the cap of 13. Collectively, the four APP fellow–MD dyads make it possible for four physicians to see an additional 12 patients. The two extra patients absorbed by the service per day results in a net increase in capacity of up to 730 patient encounters a year.

 

Outcomes and Analysis

Our main outcomes of interest were duration of onboarding and cost incurred by the division to (1) staff the service during a vacancy and (2) onboard new hires. Secondary outcomes included duration of vacancy and total time spent with the group. We collected basic demographic data on participants, including, age, gender, and race. Demographics and outcomes of interest were compared pre- (2009-2013) and post- (2014-2018) initiation of the APP clinical fellowship using the chi-square test, the t-test for normally distributed data, and the Wilcoxon rank-sum for nonnormally distributed data, as appropriate. The normality of the data distribution was tested using the Shapiro-Wilk W test. Two-tailed P values less than .05 were considered to be statistically significant. Results were analyzed using Stata/MP version 13.0 (StataCorp Inc, College Station, Texas).

RESULTS

Twelve fellows have been recruited, and of these, 10 have graduated. Two chose to leave the program prior to completion. Of the 10 fellows that have graduated, six have been hired into our group, one was hired within our facility, and three were hired as hospitalists at other institutions. The median time from APP school graduation to hire was also not different between the two groups (10.5 vs 3.9 months, P = .069). In addition, the total time that the new APP hires spent with the group was nonstatistically significantly different between the two periods (17.9 vs 18.3 months, P = .735). Both the mean duration of onboarding and the cost to the division were significantly reduced after implementation of the program (25.4 vs 11.0 weeks, P = .017 and $361,714 vs $66,000, P = .004; Table 2).

The yearly cost of an APP vacancy and onboarding is incurred by doctor moonlighting costs (at the rate of $150 per hour) to cover open shifts. The mean duration of vacancies and onboarding each year was 34.9 and 25.4 weeks, respectively, before the fellowship. The yearly cost of onboarding, after the establishment of the fellowship, is a maximum of $66,000, derived from physician moonlighting to cover the six-week ramp-up at the very beginning of the fellowship and the five weeks of orientation to the pulmonary and chemical dependency units after the fellowship (Table 3).

DISCUSSION

Our APP clinical fellowship in hospital medicine at JHBMC has produced several benefits. First, the fellowship has become a pipeline for filling APP vacancies within our division. We have been able to hire for four consecutive years from the fellowship. Second, the ready availability of high-functioning and efficient APP hospitalists has cut down on the onboarding time for our new APP hires. Many new APP graduates lack confidence in caring for complex hospitalized patients. Following our 12-month clinical fellowship, our matriculated fellows are able to practice at the top of their license immediately and confidently. Third, the reduced vacancy and shortened onboarding periods have reduced costs to the division. Fourth, the fellowship has created additional teaching avenues for the faculty. The medicine units at JHBMC are comprised of hospitalist and internal medicine residency services. The hospitalists spend the majority of their clinical time in direct patient care; however, they rotate on the residency service for two weeks out of the year. The majority of physicians welcome the chance to teach more, and partnering with an APP fellow provides that opportunity.

 

As we have developed and grown this program, the one great challenge has been what to do with graduating fellows when we cannot hire them. Fortunately, the market for highly qualified, well trained APPs is strong, and every one of the fellows that we could not hire within our group has been able to find a position either within our facility or outside our institution. To facilitate this process, program directors and recruiters are invited to meet with the fellows toward the end of their fellowship to share employment opportunities with them.

Our study has limitations. First, had the $276,000 from the attrition of two physicians been used to hire nonfellow APPs under the old model, then the costs of the two models would have been similar, but this was simply not possible because the positions could not be filled. Second, this is a single-site experience, and our findings may not be generalizable, particularly those pertaining to remuneration. Third, our study was underpowered to detect small but important differences in characteristics of APPs, especially time from graduation to hire, before and after the implementation of our fellowship. Further research comparing various programs both in structure and outcomes—such as fellows’ readiness for practice, costs, duration of vacancies, and provider satisfaction—are an important next step.

We have developed a pool of applicants within our division to fill vacancies left by turnover from senior NPs and PAs. This program has reduced costs and improved the joy of practice for both doctors and APPs. As the need for highly qualified NPs and PAs in hospital medicine continues to grow, we may see more APP fellowships in hospital medicine in the United States.

Acknowledgments

The authors thank the advanced practice providers who have helped us grow and refine our fellowship.

Disclosures

The authors have nothing to disclose

There is an increasing utilization of advanced practice providers (APPs) in the delivery of healthcare in the United States.^1,2 As of 2016, there were 157, 025 nurse practitioners (NPs) and 102,084 physician assistants (PAs) with a projected growth rate of 6.8% and 4.3%, respectively, which exceeds the physician growth rate of 1.1%.² This increased growth rate has been attributed to the expectation that APPs can enhance the quality of physician care, relieve physician shortages, and reduce service costs, as APPs are less expensive to hire than physicians.^3,4 Hospital medicine is the fastest growing medical field in the United States, and approximately 83% of hospitalist groups around the country utilize APPs; however, the demand for hospitalists continues to exceed the supply, and this has led to increased utilization of APPs in hospital medicine.^5-10

APPs receive very limited inpatient training and there is wide variation in their clinical abilities after graduation.¹¹ This is an issue that has become exacerbated in recent years by a change in the training process for PAs. Before 2005, PA programs were typically two to three years long and required the same prerequisite courses as medical schools.¹¹ PA students completed more than 2,000 hours of clinical rotations and then had to pass the Physician Assistant National Certifying Exam before they could practice.¹² Traditionally, PA programs typically attracted students with prior healthcare experience.¹¹ In 2005, PA programs began transitioning from bachelor’s degrees to requiring a master’s level degree for completion of the programs. This has shifted the demographics of the students matriculating to younger students with little-to-no prior healthcare experience; moreover, these fresh graduates lack exposure to hospital medicine.¹¹

NPs usually gain clinical experience working as registered nurses (RNs) for two or more years prior to entry into the NP program. NP programs for baccalaureate-prepared RNs vary in length from two to three years.² There is an acute care focus for NPs in training; however, there is no standardized training or licensure to ensure that hospital medicine competencies are met.^13-15 Some studies have shown that a lack of structured support has been found to affect NP role transition negatively during the first year of practice,¹⁶ and graduating NPs have indicated that they needed more out of their clinical education in terms of content, clinical experience, and competency testing.¹⁷

Hiring new APP graduates as hospitalists requires a longer and more rigorous onboarding process. On‐the‐job training in hospital medicine for new APP graduates can take as long as six to 12 months in order for them to acquire the basic skill set necessary to adequately manage hospitalized patients.¹⁵ This extended onboarding is costly because the APPs are receiving a full hospitalist salary, yet they are not functioning at full capacity. Ideally, there should be an intermediary training step between graduation and employment as hospitalist APPs. Studies have shown that APPs are interested in formal postgraduate hospital medicine training, even if it means having a lower stipend during the first year after graduating from their NP or PA program.^9,15,18

The growing need for hospitalists, driven by residency work-hour reform, increased age and complexity of patients, and the need to improve the quality of inpatient care while simultaneously reducing waste, has contributed to the increasing utilization of and need for highly qualified APPs in hospital medicine.^11,19,20 We established a fellowship to train APPs. The goal of this study was to determine if an APP fellowship is a cost-effective pipeline for filling vacancies within a hospitalist program.

 

METHODS

Design and Setting

Johns Hopkins Bayview Medical Center (JHBMC) is a 440 bed hospital in Baltimore Maryland. The hospitalist group was started in 1996 with one physician seeing approximately 500 discharges a year. Over the last 20 years, the group has grown and is now its own division with 57 providers, including 42 physicians, 11 APPs, and four APP fellows. The hospitalist division manages ~7,000 discharges a year, which corresponds to approximately 60% of admissions to general medicine. Hospitalist APPs help staff general medicine by working alongside doctors and admitting patients during the day and night. The APPs also staff the pulmonary step down unit with a pulmonary attending and the chemical dependency unit with an internal medicine addiction specialist.

The growth of the division of hospital medicine at JHBMC is a result of increasing volumes and reduced residency duty hours. The increasing full time equivalents (FTEs) resulted in a need for APPs; however, vacancies went unfilled for an average of 35 weeks due to the time it took to post open positions, interview applicants, and hire applicants through the credentialing process. Further, it took as long as 22 to 34 weeks for a new hire to work independently. The APP vacancies and onboarding resulted in increased costs to the division incurred by physician moonlighting to cover open shifts. The hourly physician moonlighting rate at JHBMC is $150. All costs were calculated on the basis of a 40-hour work week. We performed a pre- and postanalysis of outcomes of interest between January 2009 and June 2018. This study was exempt from institutional review board review.

Intervention

In 2014, a one year APP clinical fellowship in hospital medicine was started. The fellows evaluate and manage patients working one-on-one with an experienced hospitalist faculty member. The program consists of 80% clinical experience in the inpatient setting and 20% didactic instruction (Table 1). Up to four fellows are accepted each year and are eligible for hire after training if vacancies exist. The program is cost neutral and was financed by downsizing, through attrition, two physician FTEs. Four APP fellows’ salaries are the equivalent of two entry-level hospitalist physicians’ salaries at JHBMC. The annual salary for an APP fellow is $69,000.

Downsizing by two physician FTEs meant that one less doctor was scheduled every day. The patient load previously seen by that one doctor (10 patients) was absorbed by the MD–APP fellow dyads. Paired with a fellow, each physician sees a higher cap of 13 patients, and it takes six weeks for the fellows to ramp-up to this patient load. When the fellow first starts, the team sees 10 patients. Every two weeks, the pair’s census increases by one patient to the cap of 13. Collectively, the four APP fellow–MD dyads make it possible for four physicians to see an additional 12 patients. The two extra patients absorbed by the service per day results in a net increase in capacity of up to 730 patient encounters a year.

 

Outcomes and Analysis

Our main outcomes of interest were duration of onboarding and cost incurred by the division to (1) staff the service during a vacancy and (2) onboard new hires. Secondary outcomes included duration of vacancy and total time spent with the group. We collected basic demographic data on participants, including, age, gender, and race. Demographics and outcomes of interest were compared pre- (2009-2013) and post- (2014-2018) initiation of the APP clinical fellowship using the chi-square test, the t-test for normally distributed data, and the Wilcoxon rank-sum for nonnormally distributed data, as appropriate. The normality of the data distribution was tested using the Shapiro-Wilk W test. Two-tailed P values less than .05 were considered to be statistically significant. Results were analyzed using Stata/MP version 13.0 (StataCorp Inc, College Station, Texas).

RESULTS

Twelve fellows have been recruited, and of these, 10 have graduated. Two chose to leave the program prior to completion. Of the 10 fellows that have graduated, six have been hired into our group, one was hired within our facility, and three were hired as hospitalists at other institutions. The median time from APP school graduation to hire was also not different between the two groups (10.5 vs 3.9 months, P = .069). In addition, the total time that the new APP hires spent with the group was nonstatistically significantly different between the two periods (17.9 vs 18.3 months, P = .735). Both the mean duration of onboarding and the cost to the division were significantly reduced after implementation of the program (25.4 vs 11.0 weeks, P = .017 and $361,714 vs $66,000, P = .004; Table 2).

The yearly cost of an APP vacancy and onboarding is incurred by doctor moonlighting costs (at the rate of $150 per hour) to cover open shifts. The mean duration of vacancies and onboarding each year was 34.9 and 25.4 weeks, respectively, before the fellowship. The yearly cost of onboarding, after the establishment of the fellowship, is a maximum of $66,000, derived from physician moonlighting to cover the six-week ramp-up at the very beginning of the fellowship and the five weeks of orientation to the pulmonary and chemical dependency units after the fellowship (Table 3).

DISCUSSION

Our APP clinical fellowship in hospital medicine at JHBMC has produced several benefits. First, the fellowship has become a pipeline for filling APP vacancies within our division. We have been able to hire for four consecutive years from the fellowship. Second, the ready availability of high-functioning and efficient APP hospitalists has cut down on the onboarding time for our new APP hires. Many new APP graduates lack confidence in caring for complex hospitalized patients. Following our 12-month clinical fellowship, our matriculated fellows are able to practice at the top of their license immediately and confidently. Third, the reduced vacancy and shortened onboarding periods have reduced costs to the division. Fourth, the fellowship has created additional teaching avenues for the faculty. The medicine units at JHBMC are comprised of hospitalist and internal medicine residency services. The hospitalists spend the majority of their clinical time in direct patient care; however, they rotate on the residency service for two weeks out of the year. The majority of physicians welcome the chance to teach more, and partnering with an APP fellow provides that opportunity.

 

As we have developed and grown this program, the one great challenge has been what to do with graduating fellows when we cannot hire them. Fortunately, the market for highly qualified, well trained APPs is strong, and every one of the fellows that we could not hire within our group has been able to find a position either within our facility or outside our institution. To facilitate this process, program directors and recruiters are invited to meet with the fellows toward the end of their fellowship to share employment opportunities with them.

Our study has limitations. First, had the $276,000 from the attrition of two physicians been used to hire nonfellow APPs under the old model, then the costs of the two models would have been similar, but this was simply not possible because the positions could not be filled. Second, this is a single-site experience, and our findings may not be generalizable, particularly those pertaining to remuneration. Third, our study was underpowered to detect small but important differences in characteristics of APPs, especially time from graduation to hire, before and after the implementation of our fellowship. Further research comparing various programs both in structure and outcomes—such as fellows’ readiness for practice, costs, duration of vacancies, and provider satisfaction—are an important next step.

We have developed a pool of applicants within our division to fill vacancies left by turnover from senior NPs and PAs. This program has reduced costs and improved the joy of practice for both doctors and APPs. As the need for highly qualified NPs and PAs in hospital medicine continues to grow, we may see more APP fellowships in hospital medicine in the United States.

Acknowledgments

The authors thank the advanced practice providers who have helped us grow and refine our fellowship.

Disclosures

The authors have nothing to disclose

Necrotizing infection of the extremity is a rare but potentially lethal diagnosis with a mortality rate in the range of 17% to 35%.^1-4 The plastic surgery service at the Malcom Randall Veterans Affairs Medical Center (MRVAMC) treats all hand emergencies, including upper extremity infection, in the North Florida/South Georgia Veterans Heath System. There has been a well-coordinated emergency hand care system in place for several years that includes specialty templates on the electronic health record, pre-existing urgent clinic appointments, and single service surgical specialty care.⁵ This facilitates a fluid line of communication between primary care, emergency department (ED) providers, and surgical specialties. The objective of the study was to evaluate our identification, treatment, and outcome of these serious infections.

Methods

The MRVAMC Institutional Review Board approved a retrospective review of necrotizing infection of the upper extremity treated at the facility by the plastic surgery service. Surgical cases over a 9-year period (June 5, 2008-June 5, 2017) were identified by CPT (current procedural technology) codes for amputation and/or debridement of the upper extremity. The charts were reviewed for evidence of necrotizing infection by clinical description or pathology report. The patients’ age, sex, etiology, comorbidities from their problem list, vitals, and laboratory results were recorded upon arrival at the hospital. Time from presentation to surgery, treatment, and outcomes were recorded.

Results

Ten patients were treated for necrotizing infection of the upper extremity over a 9-year period; all were men with an average age of 64 years. Etiologies included nail biting, “bug bites,” crush injuries, burns, suspected IV drug use, and unknown. Nine of 10 patients had diabetes mellitus (DM). Most did not show evidence of hemodynamic instability on hospital arrival (Table). One patient was hypotensive with a mean arterial blood pressure < 65 mm Hg, 2 had heart rates > 100 beats/min, 1 patient had a temperature > 38° C, and 7 had elevated white blood cell (WBC) counts ranging from 11 to 24 k/cmm. Two undiagnosed patients with DM (patients 1 and 8) expressed no complaints of pain and presented with blood glucose > 450 mg/dL with hemoglobin A_1c levels > 12%.

Infectious disease and critical care services were involved in the treatment of several cases when requested. A computed tomography (CT) scan was used in 2 of the patients (patients 1 and 4) to assist in the diagnosis (Figure 1). 

The patient with the largest debridement (patient 4) had a CT that was not suspicious for necrotizing infection the day prior to emergent surgery. Patient 3 was found to have a subclavian stenosis on CT angiography early in the postoperative course and was treated with a carotid to subclavian bypass by the vascular service.

Seven patients out of 10 were treated with surgery within 24 hours on hospital arrival. The severity of the pathology was not initially recognized in 2 of the patients earlier in the review. A third patient resisted surgical treatment until the second hospital day. Four patients had from 1 to 3 digital amputations, 2 patients had wrist disarticulations, and 1 had a distal forearm amputation. 

The proximal amputations were patients with DM who went to the operating room within 24 hours of admission. Cultures grew a wide range of microorganisms, including methicillin-resistant Staphylococcus aureus (MRSA), methicillin-susceptible Staphylococcus aureus (MSSA), β-hemolytic Streptococcus, Streptococcus viridans, Klebsiella pneumoniae, and Prevotella.

Antibiotics were managed by critical care, hospitalist, or infectious disease services and adjusted once final cultures were returned (Table). 

The patients all had a minimum of 2 procedures (range 2-5), including debridement and closure (Figures 2A and 2B and 3A and 3B). There were no perioperative deaths.

 

Discussion

Necrotizing infection of the upper extremity is a rare pathology with a substantial risk of amputation and mortality that requires a high index of suspicion and expeditious referral to a hand surgeon. It is well accepted that the key to survival is prompt surgical debridement of all necrotic tissue, ideally within 24 hours of hospital arrival.^2-4,6 Death is usually secondary to sepsis.³ The classic presentation of pain out of proportion to exam, hypotension, erythema, skin necrosis, elevated WBC count, and fever may not be present and can delay diagnosis.^1-4,6

DM is the most common comorbidity, and reviews have found the disease occurs more often in males, both which are consistent with our study.^1-3 Diabetic infections have been found to be more likely to present as necrotizing infection than are nondiabetic infections and be at a higher risk for amputation.⁷ The patients with the wrist disarticulations and forearm amputation had DM. A minor trauma can be a portal for infection, which can be monomicrobial or polymicrobial.^1,4 Once the diagnosis is suspected, prompt resuscitation, surgical debridement, IV antibiotics, and early intensive care are lifesaving. Hyperbaric oxygen is not available at MRVAMC and was not pursued as a transfer request due to its controversial benefit.⁶

There were no perioperative 30-day mortalities over a 9-year period in patients identified as having necrotizing infection of the upper extremity. This is attributed to an aggressive and well-coordinated, multisystem approach involving emergency, surgical, anesthesia, intensive care, and infectious disease services.

The hand trauma triage system in place at MRVAMC was started in 2008 and presented at the 38th Annual VA Surgeons Meeting in New Haven, Connecticut. The process starts at the level of the ED, urgent care or primary care provider and facilitates rapid access to subspecialty care by reducing unnecessary phone calls and appointment wait times.

All hand emergencies are covered by the plastic surgery service rather than the traditional split coverage between orthopedics and plastic surgery. This provides consistency and continuity for the patients and staff. The electronic health record consult template gives specific instructions to contact the on-call plastic surgeon. The resident/fellow gets called if patient is in-house, and faculty is called if the patient is outside the main hospital. The requesting provider gets instructions on treatment and follow-up. Clinic profiles have appointments reserved for urgent consults during the first hour so that patients can be sent to pre-anesthesia clinic or hand therapy, depending on the diagnosis. This triage system increased our hand trauma volume by a multiple of 6 between 2008 and 2012 but cut the appointment wait time > 1 week by half, as a percentage of consults, and did not significantly increase after-hour use of the operating room. The number of faculty and trainees stayed the same.

We did find that speed to diagnosis for necrotizing infection is an area that can be improved on with a higher clinical suspicion. There is a learning curve to the diagnosis and treatment, which can be prolonged when the index cases do not present themselves often and the patients do not appear in distress. This argues for consistency in hand-specific trauma coverage. The patients were most often initially seen by the resident and examined by a faculty member within hours. There were 4 different plastic surgery faculty involved in these cases, and they all included resident participation before, during, and after surgery. Debridement consists of wide local excision to bleeding tissue. Author review of the operative notes found the numbers of trips to the operating room for debridement can be reduced as the surgeon becomes more confident in the diagnosis and management, resulting in less “whittling” and a more definitive debridement, resulting in a faster recovery.

The LRINEC (Laboratory Risk Indicator for Necrotizing Fasciitis) is a tool that helps to distinguish necrotizing infection from other forms of soft tissue infection by using a point system for laboratory values that include C-reactive protein (CRP), white blood count, hemoglobin, sodium, creatinine, and glucose values.⁸ We do not routinely request CRP results, but 1 of the 2 patients (patient 9) who had the full complement of laboratory tests would have met high-risk criteria. The diagnostic accuracy of this tool has been questioned⁹; however, the authors welcome any method that can rapidly and noninvasively assist in getting the patient proper attention.

The patients were not seen for long-term follow-up, but some did return to the main hospital or clinic for other pathology and were pleased to show off their grip strength after a 3-ray amputation (patient 1) and aesthetics after upper arm and forearm debridement and skin graft reconstruction (patient 4, Figure 4).

A single-ray amputation can be expected to result in a loss of grip and pinch strength, about 43.3% and 33.6%, respectively; however, given the alternative of further loss of life or limb, this was considered a reasonable trade-off.¹⁰ One wrist disarticulation and the forearm amputation were seen by amputee clinic for prosthetic fitting many months after the amputations once the wounds were healed and edema had subsided.

 

Conclusion

A well-coordinated multidisciplinary effort was the key to successful identification and treatment of this serious life- and limb-threatening infection at our institution. We did identify room for improvement in making an earlier diagnosis and performing a more aggressive first debridement.

Acknowledgments
This project is the result of work supported with resources and use of facilities at the Malcom Randall VA Medical Center in Gainesville, Florida.

Necrotizing infection of the extremity is a rare but potentially lethal diagnosis with a mortality rate in the range of 17% to 35%.^1-4 The plastic surgery service at the Malcom Randall Veterans Affairs Medical Center (MRVAMC) treats all hand emergencies, including upper extremity infection, in the North Florida/South Georgia Veterans Heath System. There has been a well-coordinated emergency hand care system in place for several years that includes specialty templates on the electronic health record, pre-existing urgent clinic appointments, and single service surgical specialty care.⁵ This facilitates a fluid line of communication between primary care, emergency department (ED) providers, and surgical specialties. The objective of the study was to evaluate our identification, treatment, and outcome of these serious infections.

Methods

The MRVAMC Institutional Review Board approved a retrospective review of necrotizing infection of the upper extremity treated at the facility by the plastic surgery service. Surgical cases over a 9-year period (June 5, 2008-June 5, 2017) were identified by CPT (current procedural technology) codes for amputation and/or debridement of the upper extremity. The charts were reviewed for evidence of necrotizing infection by clinical description or pathology report. The patients’ age, sex, etiology, comorbidities from their problem list, vitals, and laboratory results were recorded upon arrival at the hospital. Time from presentation to surgery, treatment, and outcomes were recorded.

Results

Ten patients were treated for necrotizing infection of the upper extremity over a 9-year period; all were men with an average age of 64 years. Etiologies included nail biting, “bug bites,” crush injuries, burns, suspected IV drug use, and unknown. Nine of 10 patients had diabetes mellitus (DM). Most did not show evidence of hemodynamic instability on hospital arrival (Table). One patient was hypotensive with a mean arterial blood pressure < 65 mm Hg, 2 had heart rates > 100 beats/min, 1 patient had a temperature > 38° C, and 7 had elevated white blood cell (WBC) counts ranging from 11 to 24 k/cmm. Two undiagnosed patients with DM (patients 1 and 8) expressed no complaints of pain and presented with blood glucose > 450 mg/dL with hemoglobin A_1c levels > 12%.

Infectious disease and critical care services were involved in the treatment of several cases when requested. A computed tomography (CT) scan was used in 2 of the patients (patients 1 and 4) to assist in the diagnosis (Figure 1). 

The patient with the largest debridement (patient 4) had a CT that was not suspicious for necrotizing infection the day prior to emergent surgery. Patient 3 was found to have a subclavian stenosis on CT angiography early in the postoperative course and was treated with a carotid to subclavian bypass by the vascular service.

Seven patients out of 10 were treated with surgery within 24 hours on hospital arrival. The severity of the pathology was not initially recognized in 2 of the patients earlier in the review. A third patient resisted surgical treatment until the second hospital day. Four patients had from 1 to 3 digital amputations, 2 patients had wrist disarticulations, and 1 had a distal forearm amputation. 

The proximal amputations were patients with DM who went to the operating room within 24 hours of admission. Cultures grew a wide range of microorganisms, including methicillin-resistant Staphylococcus aureus (MRSA), methicillin-susceptible Staphylococcus aureus (MSSA), β-hemolytic Streptococcus, Streptococcus viridans, Klebsiella pneumoniae, and Prevotella.

Antibiotics were managed by critical care, hospitalist, or infectious disease services and adjusted once final cultures were returned (Table). 

The patients all had a minimum of 2 procedures (range 2-5), including debridement and closure (Figures 2A and 2B and 3A and 3B). There were no perioperative deaths.

 

Discussion

Necrotizing infection of the upper extremity is a rare pathology with a substantial risk of amputation and mortality that requires a high index of suspicion and expeditious referral to a hand surgeon. It is well accepted that the key to survival is prompt surgical debridement of all necrotic tissue, ideally within 24 hours of hospital arrival.^2-4,6 Death is usually secondary to sepsis.³ The classic presentation of pain out of proportion to exam, hypotension, erythema, skin necrosis, elevated WBC count, and fever may not be present and can delay diagnosis.^1-4,6

DM is the most common comorbidity, and reviews have found the disease occurs more often in males, both which are consistent with our study.^1-3 Diabetic infections have been found to be more likely to present as necrotizing infection than are nondiabetic infections and be at a higher risk for amputation.⁷ The patients with the wrist disarticulations and forearm amputation had DM. A minor trauma can be a portal for infection, which can be monomicrobial or polymicrobial.^1,4 Once the diagnosis is suspected, prompt resuscitation, surgical debridement, IV antibiotics, and early intensive care are lifesaving. Hyperbaric oxygen is not available at MRVAMC and was not pursued as a transfer request due to its controversial benefit.⁶

There were no perioperative 30-day mortalities over a 9-year period in patients identified as having necrotizing infection of the upper extremity. This is attributed to an aggressive and well-coordinated, multisystem approach involving emergency, surgical, anesthesia, intensive care, and infectious disease services.

The hand trauma triage system in place at MRVAMC was started in 2008 and presented at the 38th Annual VA Surgeons Meeting in New Haven, Connecticut. The process starts at the level of the ED, urgent care or primary care provider and facilitates rapid access to subspecialty care by reducing unnecessary phone calls and appointment wait times.

All hand emergencies are covered by the plastic surgery service rather than the traditional split coverage between orthopedics and plastic surgery. This provides consistency and continuity for the patients and staff. The electronic health record consult template gives specific instructions to contact the on-call plastic surgeon. The resident/fellow gets called if patient is in-house, and faculty is called if the patient is outside the main hospital. The requesting provider gets instructions on treatment and follow-up. Clinic profiles have appointments reserved for urgent consults during the first hour so that patients can be sent to pre-anesthesia clinic or hand therapy, depending on the diagnosis. This triage system increased our hand trauma volume by a multiple of 6 between 2008 and 2012 but cut the appointment wait time > 1 week by half, as a percentage of consults, and did not significantly increase after-hour use of the operating room. The number of faculty and trainees stayed the same.

We did find that speed to diagnosis for necrotizing infection is an area that can be improved on with a higher clinical suspicion. There is a learning curve to the diagnosis and treatment, which can be prolonged when the index cases do not present themselves often and the patients do not appear in distress. This argues for consistency in hand-specific trauma coverage. The patients were most often initially seen by the resident and examined by a faculty member within hours. There were 4 different plastic surgery faculty involved in these cases, and they all included resident participation before, during, and after surgery. Debridement consists of wide local excision to bleeding tissue. Author review of the operative notes found the numbers of trips to the operating room for debridement can be reduced as the surgeon becomes more confident in the diagnosis and management, resulting in less “whittling” and a more definitive debridement, resulting in a faster recovery.

The LRINEC (Laboratory Risk Indicator for Necrotizing Fasciitis) is a tool that helps to distinguish necrotizing infection from other forms of soft tissue infection by using a point system for laboratory values that include C-reactive protein (CRP), white blood count, hemoglobin, sodium, creatinine, and glucose values.⁸ We do not routinely request CRP results, but 1 of the 2 patients (patient 9) who had the full complement of laboratory tests would have met high-risk criteria. The diagnostic accuracy of this tool has been questioned⁹; however, the authors welcome any method that can rapidly and noninvasively assist in getting the patient proper attention.

The patients were not seen for long-term follow-up, but some did return to the main hospital or clinic for other pathology and were pleased to show off their grip strength after a 3-ray amputation (patient 1) and aesthetics after upper arm and forearm debridement and skin graft reconstruction (patient 4, Figure 4).

A single-ray amputation can be expected to result in a loss of grip and pinch strength, about 43.3% and 33.6%, respectively; however, given the alternative of further loss of life or limb, this was considered a reasonable trade-off.¹⁰ One wrist disarticulation and the forearm amputation were seen by amputee clinic for prosthetic fitting many months after the amputations once the wounds were healed and edema had subsided.

 

Conclusion

A well-coordinated multidisciplinary effort was the key to successful identification and treatment of this serious life- and limb-threatening infection at our institution. We did identify room for improvement in making an earlier diagnosis and performing a more aggressive first debridement.

Acknowledgments
This project is the result of work supported with resources and use of facilities at the Malcom Randall VA Medical Center in Gainesville, Florida.

Necrotizing infection of the extremity is a rare but potentially lethal diagnosis with a mortality rate in the range of 17% to 35%.^1-4 The plastic surgery service at the Malcom Randall Veterans Affairs Medical Center (MRVAMC) treats all hand emergencies, including upper extremity infection, in the North Florida/South Georgia Veterans Heath System. There has been a well-coordinated emergency hand care system in place for several years that includes specialty templates on the electronic health record, pre-existing urgent clinic appointments, and single service surgical specialty care.⁵ This facilitates a fluid line of communication between primary care, emergency department (ED) providers, and surgical specialties. The objective of the study was to evaluate our identification, treatment, and outcome of these serious infections.

Methods

The MRVAMC Institutional Review Board approved a retrospective review of necrotizing infection of the upper extremity treated at the facility by the plastic surgery service. Surgical cases over a 9-year period (June 5, 2008-June 5, 2017) were identified by CPT (current procedural technology) codes for amputation and/or debridement of the upper extremity. The charts were reviewed for evidence of necrotizing infection by clinical description or pathology report. The patients’ age, sex, etiology, comorbidities from their problem list, vitals, and laboratory results were recorded upon arrival at the hospital. Time from presentation to surgery, treatment, and outcomes were recorded.

Results

Ten patients were treated for necrotizing infection of the upper extremity over a 9-year period; all were men with an average age of 64 years. Etiologies included nail biting, “bug bites,” crush injuries, burns, suspected IV drug use, and unknown. Nine of 10 patients had diabetes mellitus (DM). Most did not show evidence of hemodynamic instability on hospital arrival (Table). One patient was hypotensive with a mean arterial blood pressure < 65 mm Hg, 2 had heart rates > 100 beats/min, 1 patient had a temperature > 38° C, and 7 had elevated white blood cell (WBC) counts ranging from 11 to 24 k/cmm. Two undiagnosed patients with DM (patients 1 and 8) expressed no complaints of pain and presented with blood glucose > 450 mg/dL with hemoglobin A_1c levels > 12%.

Infectious disease and critical care services were involved in the treatment of several cases when requested. A computed tomography (CT) scan was used in 2 of the patients (patients 1 and 4) to assist in the diagnosis (Figure 1). 

The patient with the largest debridement (patient 4) had a CT that was not suspicious for necrotizing infection the day prior to emergent surgery. Patient 3 was found to have a subclavian stenosis on CT angiography early in the postoperative course and was treated with a carotid to subclavian bypass by the vascular service.

Seven patients out of 10 were treated with surgery within 24 hours on hospital arrival. The severity of the pathology was not initially recognized in 2 of the patients earlier in the review. A third patient resisted surgical treatment until the second hospital day. Four patients had from 1 to 3 digital amputations, 2 patients had wrist disarticulations, and 1 had a distal forearm amputation. 

The proximal amputations were patients with DM who went to the operating room within 24 hours of admission. Cultures grew a wide range of microorganisms, including methicillin-resistant Staphylococcus aureus (MRSA), methicillin-susceptible Staphylococcus aureus (MSSA), β-hemolytic Streptococcus, Streptococcus viridans, Klebsiella pneumoniae, and Prevotella.

Antibiotics were managed by critical care, hospitalist, or infectious disease services and adjusted once final cultures were returned (Table). 

The patients all had a minimum of 2 procedures (range 2-5), including debridement and closure (Figures 2A and 2B and 3A and 3B). There were no perioperative deaths.

 

Discussion

Necrotizing infection of the upper extremity is a rare pathology with a substantial risk of amputation and mortality that requires a high index of suspicion and expeditious referral to a hand surgeon. It is well accepted that the key to survival is prompt surgical debridement of all necrotic tissue, ideally within 24 hours of hospital arrival.^2-4,6 Death is usually secondary to sepsis.³ The classic presentation of pain out of proportion to exam, hypotension, erythema, skin necrosis, elevated WBC count, and fever may not be present and can delay diagnosis.^1-4,6

DM is the most common comorbidity, and reviews have found the disease occurs more often in males, both which are consistent with our study.^1-3 Diabetic infections have been found to be more likely to present as necrotizing infection than are nondiabetic infections and be at a higher risk for amputation.⁷ The patients with the wrist disarticulations and forearm amputation had DM. A minor trauma can be a portal for infection, which can be monomicrobial or polymicrobial.^1,4 Once the diagnosis is suspected, prompt resuscitation, surgical debridement, IV antibiotics, and early intensive care are lifesaving. Hyperbaric oxygen is not available at MRVAMC and was not pursued as a transfer request due to its controversial benefit.⁶

There were no perioperative 30-day mortalities over a 9-year period in patients identified as having necrotizing infection of the upper extremity. This is attributed to an aggressive and well-coordinated, multisystem approach involving emergency, surgical, anesthesia, intensive care, and infectious disease services.

The hand trauma triage system in place at MRVAMC was started in 2008 and presented at the 38th Annual VA Surgeons Meeting in New Haven, Connecticut. The process starts at the level of the ED, urgent care or primary care provider and facilitates rapid access to subspecialty care by reducing unnecessary phone calls and appointment wait times.

All hand emergencies are covered by the plastic surgery service rather than the traditional split coverage between orthopedics and plastic surgery. This provides consistency and continuity for the patients and staff. The electronic health record consult template gives specific instructions to contact the on-call plastic surgeon. The resident/fellow gets called if patient is in-house, and faculty is called if the patient is outside the main hospital. The requesting provider gets instructions on treatment and follow-up. Clinic profiles have appointments reserved for urgent consults during the first hour so that patients can be sent to pre-anesthesia clinic or hand therapy, depending on the diagnosis. This triage system increased our hand trauma volume by a multiple of 6 between 2008 and 2012 but cut the appointment wait time > 1 week by half, as a percentage of consults, and did not significantly increase after-hour use of the operating room. The number of faculty and trainees stayed the same.

We did find that speed to diagnosis for necrotizing infection is an area that can be improved on with a higher clinical suspicion. There is a learning curve to the diagnosis and treatment, which can be prolonged when the index cases do not present themselves often and the patients do not appear in distress. This argues for consistency in hand-specific trauma coverage. The patients were most often initially seen by the resident and examined by a faculty member within hours. There were 4 different plastic surgery faculty involved in these cases, and they all included resident participation before, during, and after surgery. Debridement consists of wide local excision to bleeding tissue. Author review of the operative notes found the numbers of trips to the operating room for debridement can be reduced as the surgeon becomes more confident in the diagnosis and management, resulting in less “whittling” and a more definitive debridement, resulting in a faster recovery.

The LRINEC (Laboratory Risk Indicator for Necrotizing Fasciitis) is a tool that helps to distinguish necrotizing infection from other forms of soft tissue infection by using a point system for laboratory values that include C-reactive protein (CRP), white blood count, hemoglobin, sodium, creatinine, and glucose values.⁸ We do not routinely request CRP results, but 1 of the 2 patients (patient 9) who had the full complement of laboratory tests would have met high-risk criteria. The diagnostic accuracy of this tool has been questioned⁹; however, the authors welcome any method that can rapidly and noninvasively assist in getting the patient proper attention.

The patients were not seen for long-term follow-up, but some did return to the main hospital or clinic for other pathology and were pleased to show off their grip strength after a 3-ray amputation (patient 1) and aesthetics after upper arm and forearm debridement and skin graft reconstruction (patient 4, Figure 4).

A single-ray amputation can be expected to result in a loss of grip and pinch strength, about 43.3% and 33.6%, respectively; however, given the alternative of further loss of life or limb, this was considered a reasonable trade-off.¹⁰ One wrist disarticulation and the forearm amputation were seen by amputee clinic for prosthetic fitting many months after the amputations once the wounds were healed and edema had subsided.

 

Conclusion

A well-coordinated multidisciplinary effort was the key to successful identification and treatment of this serious life- and limb-threatening infection at our institution. We did identify room for improvement in making an earlier diagnosis and performing a more aggressive first debridement.

Acknowledgments
This project is the result of work supported with resources and use of facilities at the Malcom Randall VA Medical Center in Gainesville, Florida.

Continuous positive airway pressure therapy (CPAP) is the first-line treatment for obstructive sleep apnea (OSA) recommended by the American College of Physicians and the American Academy of Sleep Medicine.^1,2 CPAP reduces the apnea hypopnea index (AHI), improves oxyhemoglobin desaturation, and reduces cortical arousals associated with apneic/hypopneic events.³ Despite being an effective treatment for OSA, a significant limitation of CPAP is treatment adherence. Factors associated with CPAP adherence include disease and patient characteristics, perceived self-efficacy, treatment titration procedure, device technology factors, adverse effects, and psychosocial factors.⁴

Recent studies suggest that insomnia and depression may be associated with OSA. According to a review by Luyster and colleagues, insomnia is present in 39% to 58% of patients with OSA.⁵ Since OSA may disturb sleep by the number of nightly awakenings, OSA may cause or worsen insomnia. Furthermore, insomnia may exacerbate sleep apnea thus impeding the effectiveness of sleep apnea treatment.

In some studies, the presence of insomnia symptoms prior to initiating CPAP treatment has been found to be associated with reduced CPAP adherence. For example, in 2010, Wickwire and colleagues found that there was a negative association with the average nightly minutes of CPAP use for those patients with OSA that reported symptoms of sleep maintenance insomnia.⁶ This was not found for those patients with OSA who reported symptoms of sleep onset insomnia or reported no insomnia at all. In another study by Pieh and colleagues, self-reported insomnia symptoms were predictive of CPAP adherence (defined as < 4 hours use/night) at a 6-month follow-up.⁷ However, results from a separate study indicated that insomnia was not associated with 6-month CPAP adherence.⁸

Depressive symptoms are commonly reported by patients with OSA, and higher rates of depressive symptomatology in patients with OSA have been observed in a number of prevalence studies when compared with the general population.^9,10 Between 15% and 56% of patients with OSA are diagnosed with a depressive disorder compared with 6.6% of the general population.¹¹ OSA may be causally related with depression or coexist as a separate disorder. Apnea severity has been shown to exacerbate depressive symptoms, and treatment with CPAP can improve depressive symptoms.^12,13 Unfortunately, depression has been found to reduce CPAP adherence. For example, Law and colleagues found that depression was independently associated with poorer adherence during home-based auto-PAP titration.¹⁴ Furthermore, in a study by Gurlanick and colleagues, depressive symptoms were independently associated with reduced CPAP adherence in surgical patients with OSA.¹⁵

To the best of our knowledge, the combined impact of both insomnia and depression on CPAP adherence has not been investigated. In military populations this may be especially important as CPAP adherence has been reported to be worse in military patients with posttraumatic stress disorder (PTSD) and other psychiatric disorders, and there are increasing rates of insomnia and OSA in the military.^16,17 We hypothesize that active-duty and retired military patients with self-reported insomnia and depression will have reduced short and long-term CPAP adherence.

 

Methods

This is a retrospective cohort study that reviewed charts of active-duty and retired military members diagnosed with OSA by the Sleep Medicine Clinic at Naval Medical Center San Diego in California using a home sleep test (HST). The HSTs were interpreted by board-certified physicians in sleep medicine. Prior to the HST, all patients completed a sleep questionnaire that included self-reports of daytime sleepiness, using the Epworth Sleepiness Scale (ESS), depression using the Center for Epidemiologic Studies Depression Scale (CES-D) and insomnia using the Insomnia Severity Index (ISI).

The study population included active-duty and veteran patients diagnosed with OSA who chose treatment with a CPAP and attended the sleep clinic’s OSA educational class, which discussed the diagnosis and treatment of OSA. Inclusion criteria were patients aged > 18 years and diagnosed with OSA at the Naval Medical Center San Diego sleep lab between June 2014 and June 2015.

The study population was stratified into 4 groups: (1) those with OSA but no self-reported depression or insomnia; (2) those with OSA and self-reported depression but no insomnia; (3) those with OSA and insomnia but no depression; and (4) those with OSA and self-reported depression and insomnia. Charts were excluded from the review if there were incomplete data or if the patient selected an alternative treatment for OSA, such as an oral appliance. A total of 120 charts were included in the final review. This study was approved by the Naval Medical Center San Diego Institutional Review Board.

Data Collection

Data collected included the individual’s age, sex, minimum oxygen saturation during sleep, body mass index (BMI), height, weight, ESS score at time of diagnosis, date of HST, and date of attendance at the clinic’s OSA group treatment class. Diagnosis of OSA was based on the patient’s ≥ 5 AHI. OSA severity was divided into mild (AHI 5-14), moderate (AHI 15-29), or severe (AHI ≥ 30). A patient with a CES-D score > 14 was considered to have clinically significant depression, and a patient with an ISI score of > 14 was considered to have clinically significant insomnia. ISI is a reliable and valid instrument to quantify perceived insomnia severity.¹⁸ The CES-D was used only as an indicator of symptoms relating to depression, not to clinically diagnose depression. It also has been used extensively to investigate levels of depression without a psychiatric diagnosis.¹⁹

Follow-up CPAP adherence was collected at 3- and 12-month intervals after the date of the patient’s OSA treatment group class and included AHI, median pressure setting, median days used, average time used per night, and percentage of days used for more than 4 hours for the previous 30 days. Data were obtained through Sleep Data and ResMed websites, which receive patient adherence data directly from the patient’s CPAP device. Patients were considered to be adherent with CPAP usage based on the Medicare definition: Use of the CPAP device > 4 hours per night for at least 70% of nights during a 30-day period). The 3-month time frame was used as a short interval because that is when patients are seen in the pulmonary clinic for their initial follow-up appointment. Patients are seen again at 12 months because durable medical equipment supplies must be reordered after 12 months, which requires a patient visit.

 

Statistical Analysis

Linear regression methods were used to characterize any potential relationships between the predictor variables and the target outcome variables associated with CPAP adherence at 3 and 12 months. Scatterplots were produced to assess whether linear structure was sufficient to characterize any detectable relationships, or whether there existed more complex, nonlinear relationships. The best-fitting linear regression line was examined in relation to the confidence bands of the corresponding LOESS line to determine whether a more complicated model structure was needed to capture the relationship.

Standard tests of assumptions required for these methods were also carried out: QQ plots of residuals to test for normality, the Durbin-Watson test for independence of residuals, and the nonconstant variance score test for heteroskedasticity (ie, Breusch-Pagan test). The results of these assumptions tests are reported only in cases in which the assumptions were revealed to be untenable. In cases in which suspicious outlying observations may have biased analyses, robust versions of the corresponding models were constructed. In no cases did the resulting conclusions change; only the results of the original analysis are reported. All analyses were carried out in R (R Foundation, r-project.org). Statistical significance was defined as P < .05.

Results

Our study population was predominately male (90%) with a median age of 41 years (range 22-65) and BMI of 29.8 (range 7.7-57.2)(Table 1). 

Subjects had a median ESS score of 13 (range 1-23), median ISI score of 14.3 (range 0-28), and a median CES-D score of 16 (range 0-42)(Tables 2 and 3). 

Most of the patients were on auto-CPAP (78%) and had mild OSA with an AHI of 11.1 (range 5.1-81.9). Median CPAP use at 3 months was 5 hours and 15 minutes, and the median CPAP use at 12 months was 6 hours and 3 minutes.

Predictors of CPAP Adherence

OSA severity, as measured by the AHI, was the only promising predictor of CPAP use at 3 months (b, 2.128; t₈₀, 2.854; P = .005; adjusted R², 0.081). The severity of self-reported daytime sleepiness prior to a diagnosis of OSA, as measured by the ESS, did not predict 3-month CPAP adherence (b, 0.688; t₇₇, 0.300; P = .765; adjusted R², -0.012). Self-reported depression as measured by the CES-D also did not predict CPAP use at 3 months (b, -0.078; t₈₀, -0.014; P = .941; adjusted R², -0.012). Similarly, self-reported insomnia, as measured by the ISI, did not predict 3-month CPAP adherence (b, 1.765; t₈₀, 0.939; P = .350; adjusted R², -0.001). Furthermore, a model that incorporated both depression and insomnia proved no better at accounting for variation in 3-month CPAP use (R², -0.012). Demographic variables, such as age, sex, or BMI did not predict 3-month CPAP adherence (all Ps > .20). Finally, median CPAP pressure approached statistical significance as a predictor of 3-month CPAP adherence (b, 9.493; t₆₆, 1.881; P = .064; adjusted R², 0.037) (Figure 1).

CPAP Use at 12 months

The results for CPAP use at 12 months mirrored the results for 3 months with one main exception: OSA severity, as measured by the AHI, did not predict CPAP use at 12 months (b, 1.158; t₅₂, 1.245; P = .219; adjusted R², 0.010). Neither adding a quadratic predictor nor log transforming the AHI values produced a better model (R², -0.0007 vs R², 0.0089, respectively). The severity of self-reported daytime sleepiness, as measured by the ESS, did not predict 12-month CPAP adherence (b, -2.201; t₅₀, -0.752; P = .456; adjusted R² = -0.0086). Self-reported depression as measured by the CES-D also did not predict CPAP use at 12 months (b, 0.034, t₅₂, 0.022; P = .983; adjusted R², -0.092). Self-reported insomnia, as measured by the ISI, also did not predict 12-month CPAP adherence (b, 1.765; t₈₀, 0.939; P = .350; adjusted R² = -0.001). Furthermore, a model that incorporated both depression and insomnia proved no better at accounting for variation in 12-month CPAP use, (R², -0.0298). 

Demographic variables, such as age, sex, or BMI failed to predict 12-month CPAP adherence (all Ps > .15). Finally, median CPAP pressure, in contrast to its promising value as a predictor of 3-month CPAP adherence, did not predict CPAP adherence at 12 months (b, -6.516; t₂₀, -1.021; P = .319; adjusted R² = 0.002) (Figure 2).

 

Discussion

Our study did not provide evidence that self-reported depressive and insomnia symptoms, as measured by the CES-D and ISI, can serve as useful predictors of short and long-term CPAP adherence in a sample of active-duty and retired military. OSA severity, as measured by the AHI, was the only promising predictor of CPAP adherence at 3 months.

Insomnia has been shown to improve with the use of CPAP. In a pilot study, Krakow and colleagues investigated the use of CPAP, oral appliances, or bilateral turbinectomy on patients with OSA and chronic insomnia.²⁰ Objective measures of insomnia improved with 1 night of CPAP titration. Björnsdóttir and colleagues evaluated the long-term effects of positive airway pressure (PAP) treatment on 705 adults with middle insomnia.²¹ They found after 2 years of PAP treatment combined with cognitive behavioral therapy for insomnia, patients had reduced symptoms of middle insomnia. It is possible that persistent insomnia is associated with more severe OSA which was not studied in our population.²²

As reported in other studies, it is possible that patients with depressive symptoms can improve with CPAP use, suggesting that depression and CPAP use are not totally unrelated. Edwards and colleagues studied the impact of CPAP on depressive symptoms in men and woman. They found that depressive symptoms are common in OSA and markedly improve with CPAP.²³ Bopparaju and colleagues found a high prevalence of anxiety and depression in patients with OSA but did not influence CPAP adherence.²⁴

The results of this study differ from some previous findings where depression was found to predict CPAP adherence.¹⁰ This may be due in part to differences in the type of instrument used to assess depression. Wells and colleagues found that baseline depressive symptoms did not correlate with CPAP adherence and that patients with greater CPAP adherence had improvement in OSA and depressive symptoms.²⁵ Furthermore, patients with residual OSA symptoms using CPAP had more depressive symptoms, suggesting that it is the improvement in OSA symptoms that may be correlated with the improvement in depressive symptoms. Although soldiers with PTSD may have reduced CPAP adherence, use of CPAP is associated with improvement in PTSD symptoms.^11,26

Limitations

This study had several limitations, including a small sample size. Study patients were also from a single institution, and the majority of patients had mild-to-moderate OSA. A multicenter prospective study with a larger sample size that included more severe patients with OSA may have shown different results. The participants in this study were limited to members from the active-duty and retired military population. The findings in this population may not be transferrable to the general public. Another study limitation was that the ISI and the CES-D were only administered prior to the initiation of CPAP. If the CES-D and ISI were administered at the 3- and 12-month follow-up visits, we could determine whether short and long-term CPAP improved these symptoms or whether there was no association between CPAP adherence with insomnia and depressive symptoms. Another limitation is that we did not have access to information about potential PTSD symptomatology, which has been associated with reduced CPAP adherence and is more common in a military and veteran population.¹¹

 

Conclusion

This study found little evidence that symptoms of depression and insomnia are useful predictors of CPAP adherence, in either short- or long-term use, in an active-duty and retired military sample. Although these were not found to be predictors of CPAP adherence, further research will be necessary to determine whether CPAP adherence improves symptoms of depression and insomnia in military and veteran populations. Apnea severity did predict CPAP adherence in the short term, but not for any length of time beyond 3 months. More research is needed to explore strategies to improve CPAP adherence in military populations.

Continuous positive airway pressure therapy (CPAP) is the first-line treatment for obstructive sleep apnea (OSA) recommended by the American College of Physicians and the American Academy of Sleep Medicine.^1,2 CPAP reduces the apnea hypopnea index (AHI), improves oxyhemoglobin desaturation, and reduces cortical arousals associated with apneic/hypopneic events.³ Despite being an effective treatment for OSA, a significant limitation of CPAP is treatment adherence. Factors associated with CPAP adherence include disease and patient characteristics, perceived self-efficacy, treatment titration procedure, device technology factors, adverse effects, and psychosocial factors.⁴

Recent studies suggest that insomnia and depression may be associated with OSA. According to a review by Luyster and colleagues, insomnia is present in 39% to 58% of patients with OSA.⁵ Since OSA may disturb sleep by the number of nightly awakenings, OSA may cause or worsen insomnia. Furthermore, insomnia may exacerbate sleep apnea thus impeding the effectiveness of sleep apnea treatment.

In some studies, the presence of insomnia symptoms prior to initiating CPAP treatment has been found to be associated with reduced CPAP adherence. For example, in 2010, Wickwire and colleagues found that there was a negative association with the average nightly minutes of CPAP use for those patients with OSA that reported symptoms of sleep maintenance insomnia.⁶ This was not found for those patients with OSA who reported symptoms of sleep onset insomnia or reported no insomnia at all. In another study by Pieh and colleagues, self-reported insomnia symptoms were predictive of CPAP adherence (defined as < 4 hours use/night) at a 6-month follow-up.⁷ However, results from a separate study indicated that insomnia was not associated with 6-month CPAP adherence.⁸

Depressive symptoms are commonly reported by patients with OSA, and higher rates of depressive symptomatology in patients with OSA have been observed in a number of prevalence studies when compared with the general population.^9,10 Between 15% and 56% of patients with OSA are diagnosed with a depressive disorder compared with 6.6% of the general population.¹¹ OSA may be causally related with depression or coexist as a separate disorder. Apnea severity has been shown to exacerbate depressive symptoms, and treatment with CPAP can improve depressive symptoms.^12,13 Unfortunately, depression has been found to reduce CPAP adherence. For example, Law and colleagues found that depression was independently associated with poorer adherence during home-based auto-PAP titration.¹⁴ Furthermore, in a study by Gurlanick and colleagues, depressive symptoms were independently associated with reduced CPAP adherence in surgical patients with OSA.¹⁵

To the best of our knowledge, the combined impact of both insomnia and depression on CPAP adherence has not been investigated. In military populations this may be especially important as CPAP adherence has been reported to be worse in military patients with posttraumatic stress disorder (PTSD) and other psychiatric disorders, and there are increasing rates of insomnia and OSA in the military.^16,17 We hypothesize that active-duty and retired military patients with self-reported insomnia and depression will have reduced short and long-term CPAP adherence.

 

Methods

This is a retrospective cohort study that reviewed charts of active-duty and retired military members diagnosed with OSA by the Sleep Medicine Clinic at Naval Medical Center San Diego in California using a home sleep test (HST). The HSTs were interpreted by board-certified physicians in sleep medicine. Prior to the HST, all patients completed a sleep questionnaire that included self-reports of daytime sleepiness, using the Epworth Sleepiness Scale (ESS), depression using the Center for Epidemiologic Studies Depression Scale (CES-D) and insomnia using the Insomnia Severity Index (ISI).

The study population included active-duty and veteran patients diagnosed with OSA who chose treatment with a CPAP and attended the sleep clinic’s OSA educational class, which discussed the diagnosis and treatment of OSA. Inclusion criteria were patients aged > 18 years and diagnosed with OSA at the Naval Medical Center San Diego sleep lab between June 2014 and June 2015.

The study population was stratified into 4 groups: (1) those with OSA but no self-reported depression or insomnia; (2) those with OSA and self-reported depression but no insomnia; (3) those with OSA and insomnia but no depression; and (4) those with OSA and self-reported depression and insomnia. Charts were excluded from the review if there were incomplete data or if the patient selected an alternative treatment for OSA, such as an oral appliance. A total of 120 charts were included in the final review. This study was approved by the Naval Medical Center San Diego Institutional Review Board.

Data Collection

Data collected included the individual’s age, sex, minimum oxygen saturation during sleep, body mass index (BMI), height, weight, ESS score at time of diagnosis, date of HST, and date of attendance at the clinic’s OSA group treatment class. Diagnosis of OSA was based on the patient’s ≥ 5 AHI. OSA severity was divided into mild (AHI 5-14), moderate (AHI 15-29), or severe (AHI ≥ 30). A patient with a CES-D score > 14 was considered to have clinically significant depression, and a patient with an ISI score of > 14 was considered to have clinically significant insomnia. ISI is a reliable and valid instrument to quantify perceived insomnia severity.¹⁸ The CES-D was used only as an indicator of symptoms relating to depression, not to clinically diagnose depression. It also has been used extensively to investigate levels of depression without a psychiatric diagnosis.¹⁹

Follow-up CPAP adherence was collected at 3- and 12-month intervals after the date of the patient’s OSA treatment group class and included AHI, median pressure setting, median days used, average time used per night, and percentage of days used for more than 4 hours for the previous 30 days. Data were obtained through Sleep Data and ResMed websites, which receive patient adherence data directly from the patient’s CPAP device. Patients were considered to be adherent with CPAP usage based on the Medicare definition: Use of the CPAP device > 4 hours per night for at least 70% of nights during a 30-day period). The 3-month time frame was used as a short interval because that is when patients are seen in the pulmonary clinic for their initial follow-up appointment. Patients are seen again at 12 months because durable medical equipment supplies must be reordered after 12 months, which requires a patient visit.

 

Statistical Analysis

Linear regression methods were used to characterize any potential relationships between the predictor variables and the target outcome variables associated with CPAP adherence at 3 and 12 months. Scatterplots were produced to assess whether linear structure was sufficient to characterize any detectable relationships, or whether there existed more complex, nonlinear relationships. The best-fitting linear regression line was examined in relation to the confidence bands of the corresponding LOESS line to determine whether a more complicated model structure was needed to capture the relationship.

Standard tests of assumptions required for these methods were also carried out: QQ plots of residuals to test for normality, the Durbin-Watson test for independence of residuals, and the nonconstant variance score test for heteroskedasticity (ie, Breusch-Pagan test). The results of these assumptions tests are reported only in cases in which the assumptions were revealed to be untenable. In cases in which suspicious outlying observations may have biased analyses, robust versions of the corresponding models were constructed. In no cases did the resulting conclusions change; only the results of the original analysis are reported. All analyses were carried out in R (R Foundation, r-project.org). Statistical significance was defined as P < .05.

Results

Our study population was predominately male (90%) with a median age of 41 years (range 22-65) and BMI of 29.8 (range 7.7-57.2)(Table 1). 

Subjects had a median ESS score of 13 (range 1-23), median ISI score of 14.3 (range 0-28), and a median CES-D score of 16 (range 0-42)(Tables 2 and 3). 

Most of the patients were on auto-CPAP (78%) and had mild OSA with an AHI of 11.1 (range 5.1-81.9). Median CPAP use at 3 months was 5 hours and 15 minutes, and the median CPAP use at 12 months was 6 hours and 3 minutes.

Predictors of CPAP Adherence

OSA severity, as measured by the AHI, was the only promising predictor of CPAP use at 3 months (b, 2.128; t₈₀, 2.854; P = .005; adjusted R², 0.081). The severity of self-reported daytime sleepiness prior to a diagnosis of OSA, as measured by the ESS, did not predict 3-month CPAP adherence (b, 0.688; t₇₇, 0.300; P = .765; adjusted R², -0.012). Self-reported depression as measured by the CES-D also did not predict CPAP use at 3 months (b, -0.078; t₈₀, -0.014; P = .941; adjusted R², -0.012). Similarly, self-reported insomnia, as measured by the ISI, did not predict 3-month CPAP adherence (b, 1.765; t₈₀, 0.939; P = .350; adjusted R², -0.001). Furthermore, a model that incorporated both depression and insomnia proved no better at accounting for variation in 3-month CPAP use (R², -0.012). Demographic variables, such as age, sex, or BMI did not predict 3-month CPAP adherence (all Ps > .20). Finally, median CPAP pressure approached statistical significance as a predictor of 3-month CPAP adherence (b, 9.493; t₆₆, 1.881; P = .064; adjusted R², 0.037) (Figure 1).

CPAP Use at 12 months

The results for CPAP use at 12 months mirrored the results for 3 months with one main exception: OSA severity, as measured by the AHI, did not predict CPAP use at 12 months (b, 1.158; t₅₂, 1.245; P = .219; adjusted R², 0.010). Neither adding a quadratic predictor nor log transforming the AHI values produced a better model (R², -0.0007 vs R², 0.0089, respectively). The severity of self-reported daytime sleepiness, as measured by the ESS, did not predict 12-month CPAP adherence (b, -2.201; t₅₀, -0.752; P = .456; adjusted R² = -0.0086). Self-reported depression as measured by the CES-D also did not predict CPAP use at 12 months (b, 0.034, t₅₂, 0.022; P = .983; adjusted R², -0.092). Self-reported insomnia, as measured by the ISI, also did not predict 12-month CPAP adherence (b, 1.765; t₈₀, 0.939; P = .350; adjusted R² = -0.001). Furthermore, a model that incorporated both depression and insomnia proved no better at accounting for variation in 12-month CPAP use, (R², -0.0298). 

Demographic variables, such as age, sex, or BMI failed to predict 12-month CPAP adherence (all Ps > .15). Finally, median CPAP pressure, in contrast to its promising value as a predictor of 3-month CPAP adherence, did not predict CPAP adherence at 12 months (b, -6.516; t₂₀, -1.021; P = .319; adjusted R² = 0.002) (Figure 2).

 

Discussion

Our study did not provide evidence that self-reported depressive and insomnia symptoms, as measured by the CES-D and ISI, can serve as useful predictors of short and long-term CPAP adherence in a sample of active-duty and retired military. OSA severity, as measured by the AHI, was the only promising predictor of CPAP adherence at 3 months.

Insomnia has been shown to improve with the use of CPAP. In a pilot study, Krakow and colleagues investigated the use of CPAP, oral appliances, or bilateral turbinectomy on patients with OSA and chronic insomnia.²⁰ Objective measures of insomnia improved with 1 night of CPAP titration. Björnsdóttir and colleagues evaluated the long-term effects of positive airway pressure (PAP) treatment on 705 adults with middle insomnia.²¹ They found after 2 years of PAP treatment combined with cognitive behavioral therapy for insomnia, patients had reduced symptoms of middle insomnia. It is possible that persistent insomnia is associated with more severe OSA which was not studied in our population.²²

As reported in other studies, it is possible that patients with depressive symptoms can improve with CPAP use, suggesting that depression and CPAP use are not totally unrelated. Edwards and colleagues studied the impact of CPAP on depressive symptoms in men and woman. They found that depressive symptoms are common in OSA and markedly improve with CPAP.²³ Bopparaju and colleagues found a high prevalence of anxiety and depression in patients with OSA but did not influence CPAP adherence.²⁴

The results of this study differ from some previous findings where depression was found to predict CPAP adherence.¹⁰ This may be due in part to differences in the type of instrument used to assess depression. Wells and colleagues found that baseline depressive symptoms did not correlate with CPAP adherence and that patients with greater CPAP adherence had improvement in OSA and depressive symptoms.²⁵ Furthermore, patients with residual OSA symptoms using CPAP had more depressive symptoms, suggesting that it is the improvement in OSA symptoms that may be correlated with the improvement in depressive symptoms. Although soldiers with PTSD may have reduced CPAP adherence, use of CPAP is associated with improvement in PTSD symptoms.^11,26

Limitations

This study had several limitations, including a small sample size. Study patients were also from a single institution, and the majority of patients had mild-to-moderate OSA. A multicenter prospective study with a larger sample size that included more severe patients with OSA may have shown different results. The participants in this study were limited to members from the active-duty and retired military population. The findings in this population may not be transferrable to the general public. Another study limitation was that the ISI and the CES-D were only administered prior to the initiation of CPAP. If the CES-D and ISI were administered at the 3- and 12-month follow-up visits, we could determine whether short and long-term CPAP improved these symptoms or whether there was no association between CPAP adherence with insomnia and depressive symptoms. Another limitation is that we did not have access to information about potential PTSD symptomatology, which has been associated with reduced CPAP adherence and is more common in a military and veteran population.¹¹

 

Conclusion

This study found little evidence that symptoms of depression and insomnia are useful predictors of CPAP adherence, in either short- or long-term use, in an active-duty and retired military sample. Although these were not found to be predictors of CPAP adherence, further research will be necessary to determine whether CPAP adherence improves symptoms of depression and insomnia in military and veteran populations. Apnea severity did predict CPAP adherence in the short term, but not for any length of time beyond 3 months. More research is needed to explore strategies to improve CPAP adherence in military populations.

Continuous positive airway pressure therapy (CPAP) is the first-line treatment for obstructive sleep apnea (OSA) recommended by the American College of Physicians and the American Academy of Sleep Medicine.^1,2 CPAP reduces the apnea hypopnea index (AHI), improves oxyhemoglobin desaturation, and reduces cortical arousals associated with apneic/hypopneic events.³ Despite being an effective treatment for OSA, a significant limitation of CPAP is treatment adherence. Factors associated with CPAP adherence include disease and patient characteristics, perceived self-efficacy, treatment titration procedure, device technology factors, adverse effects, and psychosocial factors.⁴

Recent studies suggest that insomnia and depression may be associated with OSA. According to a review by Luyster and colleagues, insomnia is present in 39% to 58% of patients with OSA.⁵ Since OSA may disturb sleep by the number of nightly awakenings, OSA may cause or worsen insomnia. Furthermore, insomnia may exacerbate sleep apnea thus impeding the effectiveness of sleep apnea treatment.

In some studies, the presence of insomnia symptoms prior to initiating CPAP treatment has been found to be associated with reduced CPAP adherence. For example, in 2010, Wickwire and colleagues found that there was a negative association with the average nightly minutes of CPAP use for those patients with OSA that reported symptoms of sleep maintenance insomnia.⁶ This was not found for those patients with OSA who reported symptoms of sleep onset insomnia or reported no insomnia at all. In another study by Pieh and colleagues, self-reported insomnia symptoms were predictive of CPAP adherence (defined as < 4 hours use/night) at a 6-month follow-up.⁷ However, results from a separate study indicated that insomnia was not associated with 6-month CPAP adherence.⁸

Depressive symptoms are commonly reported by patients with OSA, and higher rates of depressive symptomatology in patients with OSA have been observed in a number of prevalence studies when compared with the general population.^9,10 Between 15% and 56% of patients with OSA are diagnosed with a depressive disorder compared with 6.6% of the general population.¹¹ OSA may be causally related with depression or coexist as a separate disorder. Apnea severity has been shown to exacerbate depressive symptoms, and treatment with CPAP can improve depressive symptoms.^12,13 Unfortunately, depression has been found to reduce CPAP adherence. For example, Law and colleagues found that depression was independently associated with poorer adherence during home-based auto-PAP titration.¹⁴ Furthermore, in a study by Gurlanick and colleagues, depressive symptoms were independently associated with reduced CPAP adherence in surgical patients with OSA.¹⁵

To the best of our knowledge, the combined impact of both insomnia and depression on CPAP adherence has not been investigated. In military populations this may be especially important as CPAP adherence has been reported to be worse in military patients with posttraumatic stress disorder (PTSD) and other psychiatric disorders, and there are increasing rates of insomnia and OSA in the military.^16,17 We hypothesize that active-duty and retired military patients with self-reported insomnia and depression will have reduced short and long-term CPAP adherence.

 

Methods

This is a retrospective cohort study that reviewed charts of active-duty and retired military members diagnosed with OSA by the Sleep Medicine Clinic at Naval Medical Center San Diego in California using a home sleep test (HST). The HSTs were interpreted by board-certified physicians in sleep medicine. Prior to the HST, all patients completed a sleep questionnaire that included self-reports of daytime sleepiness, using the Epworth Sleepiness Scale (ESS), depression using the Center for Epidemiologic Studies Depression Scale (CES-D) and insomnia using the Insomnia Severity Index (ISI).

The study population included active-duty and veteran patients diagnosed with OSA who chose treatment with a CPAP and attended the sleep clinic’s OSA educational class, which discussed the diagnosis and treatment of OSA. Inclusion criteria were patients aged > 18 years and diagnosed with OSA at the Naval Medical Center San Diego sleep lab between June 2014 and June 2015.

The study population was stratified into 4 groups: (1) those with OSA but no self-reported depression or insomnia; (2) those with OSA and self-reported depression but no insomnia; (3) those with OSA and insomnia but no depression; and (4) those with OSA and self-reported depression and insomnia. Charts were excluded from the review if there were incomplete data or if the patient selected an alternative treatment for OSA, such as an oral appliance. A total of 120 charts were included in the final review. This study was approved by the Naval Medical Center San Diego Institutional Review Board.

Data Collection

Data collected included the individual’s age, sex, minimum oxygen saturation during sleep, body mass index (BMI), height, weight, ESS score at time of diagnosis, date of HST, and date of attendance at the clinic’s OSA group treatment class. Diagnosis of OSA was based on the patient’s ≥ 5 AHI. OSA severity was divided into mild (AHI 5-14), moderate (AHI 15-29), or severe (AHI ≥ 30). A patient with a CES-D score > 14 was considered to have clinically significant depression, and a patient with an ISI score of > 14 was considered to have clinically significant insomnia. ISI is a reliable and valid instrument to quantify perceived insomnia severity.¹⁸ The CES-D was used only as an indicator of symptoms relating to depression, not to clinically diagnose depression. It also has been used extensively to investigate levels of depression without a psychiatric diagnosis.¹⁹

Follow-up CPAP adherence was collected at 3- and 12-month intervals after the date of the patient’s OSA treatment group class and included AHI, median pressure setting, median days used, average time used per night, and percentage of days used for more than 4 hours for the previous 30 days. Data were obtained through Sleep Data and ResMed websites, which receive patient adherence data directly from the patient’s CPAP device. Patients were considered to be adherent with CPAP usage based on the Medicare definition: Use of the CPAP device > 4 hours per night for at least 70% of nights during a 30-day period). The 3-month time frame was used as a short interval because that is when patients are seen in the pulmonary clinic for their initial follow-up appointment. Patients are seen again at 12 months because durable medical equipment supplies must be reordered after 12 months, which requires a patient visit.

 

Statistical Analysis

Linear regression methods were used to characterize any potential relationships between the predictor variables and the target outcome variables associated with CPAP adherence at 3 and 12 months. Scatterplots were produced to assess whether linear structure was sufficient to characterize any detectable relationships, or whether there existed more complex, nonlinear relationships. The best-fitting linear regression line was examined in relation to the confidence bands of the corresponding LOESS line to determine whether a more complicated model structure was needed to capture the relationship.

Standard tests of assumptions required for these methods were also carried out: QQ plots of residuals to test for normality, the Durbin-Watson test for independence of residuals, and the nonconstant variance score test for heteroskedasticity (ie, Breusch-Pagan test). The results of these assumptions tests are reported only in cases in which the assumptions were revealed to be untenable. In cases in which suspicious outlying observations may have biased analyses, robust versions of the corresponding models were constructed. In no cases did the resulting conclusions change; only the results of the original analysis are reported. All analyses were carried out in R (R Foundation, r-project.org). Statistical significance was defined as P < .05.

Results

Our study population was predominately male (90%) with a median age of 41 years (range 22-65) and BMI of 29.8 (range 7.7-57.2)(Table 1). 

Subjects had a median ESS score of 13 (range 1-23), median ISI score of 14.3 (range 0-28), and a median CES-D score of 16 (range 0-42)(Tables 2 and 3). 

Most of the patients were on auto-CPAP (78%) and had mild OSA with an AHI of 11.1 (range 5.1-81.9). Median CPAP use at 3 months was 5 hours and 15 minutes, and the median CPAP use at 12 months was 6 hours and 3 minutes.

Predictors of CPAP Adherence

OSA severity, as measured by the AHI, was the only promising predictor of CPAP use at 3 months (b, 2.128; t₈₀, 2.854; P = .005; adjusted R², 0.081). The severity of self-reported daytime sleepiness prior to a diagnosis of OSA, as measured by the ESS, did not predict 3-month CPAP adherence (b, 0.688; t₇₇, 0.300; P = .765; adjusted R², -0.012). Self-reported depression as measured by the CES-D also did not predict CPAP use at 3 months (b, -0.078; t₈₀, -0.014; P = .941; adjusted R², -0.012). Similarly, self-reported insomnia, as measured by the ISI, did not predict 3-month CPAP adherence (b, 1.765; t₈₀, 0.939; P = .350; adjusted R², -0.001). Furthermore, a model that incorporated both depression and insomnia proved no better at accounting for variation in 3-month CPAP use (R², -0.012). Demographic variables, such as age, sex, or BMI did not predict 3-month CPAP adherence (all Ps > .20). Finally, median CPAP pressure approached statistical significance as a predictor of 3-month CPAP adherence (b, 9.493; t₆₆, 1.881; P = .064; adjusted R², 0.037) (Figure 1).

CPAP Use at 12 months

The results for CPAP use at 12 months mirrored the results for 3 months with one main exception: OSA severity, as measured by the AHI, did not predict CPAP use at 12 months (b, 1.158; t₅₂, 1.245; P = .219; adjusted R², 0.010). Neither adding a quadratic predictor nor log transforming the AHI values produced a better model (R², -0.0007 vs R², 0.0089, respectively). The severity of self-reported daytime sleepiness, as measured by the ESS, did not predict 12-month CPAP adherence (b, -2.201; t₅₀, -0.752; P = .456; adjusted R² = -0.0086). Self-reported depression as measured by the CES-D also did not predict CPAP use at 12 months (b, 0.034, t₅₂, 0.022; P = .983; adjusted R², -0.092). Self-reported insomnia, as measured by the ISI, also did not predict 12-month CPAP adherence (b, 1.765; t₈₀, 0.939; P = .350; adjusted R² = -0.001). Furthermore, a model that incorporated both depression and insomnia proved no better at accounting for variation in 12-month CPAP use, (R², -0.0298). 

Demographic variables, such as age, sex, or BMI failed to predict 12-month CPAP adherence (all Ps > .15). Finally, median CPAP pressure, in contrast to its promising value as a predictor of 3-month CPAP adherence, did not predict CPAP adherence at 12 months (b, -6.516; t₂₀, -1.021; P = .319; adjusted R² = 0.002) (Figure 2).

 

Discussion

Our study did not provide evidence that self-reported depressive and insomnia symptoms, as measured by the CES-D and ISI, can serve as useful predictors of short and long-term CPAP adherence in a sample of active-duty and retired military. OSA severity, as measured by the AHI, was the only promising predictor of CPAP adherence at 3 months.

Insomnia has been shown to improve with the use of CPAP. In a pilot study, Krakow and colleagues investigated the use of CPAP, oral appliances, or bilateral turbinectomy on patients with OSA and chronic insomnia.²⁰ Objective measures of insomnia improved with 1 night of CPAP titration. Björnsdóttir and colleagues evaluated the long-term effects of positive airway pressure (PAP) treatment on 705 adults with middle insomnia.²¹ They found after 2 years of PAP treatment combined with cognitive behavioral therapy for insomnia, patients had reduced symptoms of middle insomnia. It is possible that persistent insomnia is associated with more severe OSA which was not studied in our population.²²

As reported in other studies, it is possible that patients with depressive symptoms can improve with CPAP use, suggesting that depression and CPAP use are not totally unrelated. Edwards and colleagues studied the impact of CPAP on depressive symptoms in men and woman. They found that depressive symptoms are common in OSA and markedly improve with CPAP.²³ Bopparaju and colleagues found a high prevalence of anxiety and depression in patients with OSA but did not influence CPAP adherence.²⁴

The results of this study differ from some previous findings where depression was found to predict CPAP adherence.¹⁰ This may be due in part to differences in the type of instrument used to assess depression. Wells and colleagues found that baseline depressive symptoms did not correlate with CPAP adherence and that patients with greater CPAP adherence had improvement in OSA and depressive symptoms.²⁵ Furthermore, patients with residual OSA symptoms using CPAP had more depressive symptoms, suggesting that it is the improvement in OSA symptoms that may be correlated with the improvement in depressive symptoms. Although soldiers with PTSD may have reduced CPAP adherence, use of CPAP is associated with improvement in PTSD symptoms.^11,26

Limitations

This study had several limitations, including a small sample size. Study patients were also from a single institution, and the majority of patients had mild-to-moderate OSA. A multicenter prospective study with a larger sample size that included more severe patients with OSA may have shown different results. The participants in this study were limited to members from the active-duty and retired military population. The findings in this population may not be transferrable to the general public. Another study limitation was that the ISI and the CES-D were only administered prior to the initiation of CPAP. If the CES-D and ISI were administered at the 3- and 12-month follow-up visits, we could determine whether short and long-term CPAP improved these symptoms or whether there was no association between CPAP adherence with insomnia and depressive symptoms. Another limitation is that we did not have access to information about potential PTSD symptomatology, which has been associated with reduced CPAP adherence and is more common in a military and veteran population.¹¹

 

Conclusion

This study found little evidence that symptoms of depression and insomnia are useful predictors of CPAP adherence, in either short- or long-term use, in an active-duty and retired military sample. Although these were not found to be predictors of CPAP adherence, further research will be necessary to determine whether CPAP adherence improves symptoms of depression and insomnia in military and veteran populations. Apnea severity did predict CPAP adherence in the short term, but not for any length of time beyond 3 months. More research is needed to explore strategies to improve CPAP adherence in military populations.

Historically, the Veterans Health Administration (VHA) has excelled at improving veterans’ access to health care and enhancing foundational services, such as prosthetics and other veteran-centric services, and this continues to be the VHA’s top priority.¹ Travel distance and time are often barriers to accessing health care for many veterans.^2-11 For veterans with disabilities who must overcome additional physical, cognitive, and emotional obstacles to access vital rehabilitation services, these geographic obstacles are magnified. Further compounding the challenge is that rehabilitation therapies frequently require multiple encounters. Telerehabilitation is a promising solution for veterans in need of rehabilitation to regain optimal functioning. This alternative mode of service delivery can help veterans overcome geographic access barriers by delivering health care directly to veterans in their homes or nearby community-based outpatient clinics.^12,13

A growing body of evidence supports telerehabilitation. In a 2017 systematic review and meta-analysis, Cottrell and colleagues reviewed and analyzed data from 13 studies that met their inclusion criteria; specifically, their meta-analytic sample comprised adults aged ≥ 18 years presenting with any diagnosed primary musculoskeletal condition; treatment interventions via a real-time telerehabilitation medium, trials that had a comparison group with the same condition; provided clinical outcomes data, and included published randomized and nonrandomized controlled trials.¹⁴ Based on their aggregated results, they concluded that real-time telerehabilitation was effective in improving physical function (standardized mean difference [SMD], 0.63; 95% CI, 0.92-2.33; I², 93%), and reducing pain (SMD, 0.66; 95% CI, −0.27- .60; I², 96%) in patients with any diagnosed primary musculoskeletal condition.¹⁴

Two other systematic reviews conducted by Pietrzak and colleagues and Agostini and colleagues also demonstrated the clinical effectiveness of telerehabilitation.^15,16 Clinical effectiveness was defined as changes in health, functional status, and satisfaction with the telerehabilitation services delivered. The studies examined in the review included those that provided online self-management and education in addition to exercise via teleconferencing in real time.

Pietrzak and colleagues found that Internet-based osteoarthritis self-management interventions significantly improved 4 of 6 health status measures reviewed (ie, pain, fatigue, activity limitation, health distress, disability, and self‐reported global health).¹⁵ User acceptance and satisfaction were high (≥ 70% satisfied) in all studies meeting the inclusion criteria.

Agostini and colleagues found that telerehabilitation was more effective than other modes of delivering rehabilitation to regain motor function in cardiac (SMD, 0.24; 95% CI, 0.04-0.43) and total knee arthroplasty (Timed Up and Go test: SMD, −5.17; 95% CI, −9.79- −0.55) patients.¹⁶ Some evidence from VHA and non-VHA studies also support the use of telerehabilitation to reduce health care costs,^17-19 improve treatment adherence,^12,20 and enhance patient physical, cognitive and mobility function, as well as patient satisfaction and health-related quality of life.^13,21-24

Since the first recorded use of telehealth in 1959, the application of technology to deliver health care, including rehabilitation services, has increased exponentially.¹⁴ In fiscal year (FY) 2017 alone, the VA provided > 2 million episodes of care for > 700,000 veterans using telehealth services.²⁵

Although the process for accessing telerehabilitation may vary throughout the VA, typically a few common factors make a veteran eligible for this mode of rehabilitation care delivery: Veterans must meet criteria for a specific program (eg, amputation, occupational therapy, and physical therapy) and receive VA care from a VA medical facility or clinic that offers telehealth services. Care providers must believe that the veteran would benefit from telerehabilitation (eg, limited mobility and long-distance travel to the facility) and that they would be able to receive an appropriate consult. The veteran must meet the following requirements: (1) willingness to consent to a visit via telehealth; (2) access to required equipment/e-mail; and (3) a caregiver to assist if they are unable to complete a visit independently.

In this article, we provide an overview of the growth of telerehabilitation in the VHA. Data are presented for specific telerehabilitation programs over time and by rurality.

 

Methods

The VHA Support Service Center works with VHA program offices and field users to provide field-focused business, clinical, and special topic reports. An online portal provides access to these customizable reports organized as data cubes, which represent data dimensions (ie, clinic type) and measures (ie, number of unique patients). For this study, we used the Connected Care, Telehealth, Call Centers Clinical Video Telehealth/Store and Forward Telehealth data cube clinical stop codes to identify the numbers of telerehabilitation veteran users and encounters across time. The following telerehabilitation clinic-stop codes were selected: 197 (polytrauma/traumatic brain injury [TBI]–individuals), 201 (Physical Medicine and Rehabilitation [PM&R] Service), 205 (physical therapy), 206 (occupational therapy), 211 (PM&R amputation clinic), 418 (amputation clinic), 214 (kinesiotherapy), and 240 (PM&R assistive technology clinic). Data for total unique patients served and the total number of encounters were extracted at the national level and by rurality from FY 2012 to FY 2017, providing the past 5 years of VHA telerehabilitation data.

It is important to note that in FY 2015, the VHA changed its definition of rurality to a rural-urban commuting areas (RUCA)-based system (www.ruralhealth.va.gov/rural-definition.asp). Prior to FY 2015, the VHA used the US Census Bureau (CB) urbanized area definitions. According to CB, an urbanized area contains a central city and surrounding area that totals > 50,000 in population. It also includes places outside of urbanized areas with populations > 2,500. Rural areas are defined as all other areas. VHA added a third category, highly rural, which is defined as areas that had < 7 people per square mile. In the RUCA system, each census tract defined by the CB is given a score. The VHA definitions are as follows:

• Urban (U)—census tracts with RUCA scores of 1.0 or 1.1. These tracts are determined by the CB as being in an urban core and having the majority of their workers commute within that same core (1.0). If 30% to 49% commute to an even larger urban core, then the code is 1.1;
• Rural (R)—all tracts not receiving scores in the urban or highly rural tiers; and
• Highly rural (H)—tracts with a RUCA score of 10.0. These are the most remote occupied land areas. Less than 10% of workers travel to CB-defined urbanized areas or urban clusters.

In addition, VHA recently added an “I” category to complement “U,” “R,” and “H.” The “I” value is assigned to veterans living on the US insular islands (ie, territories): Guam, American Samoa, Northern Marianas, and US Virgin Islands. For the analysis by rurality in this study, we excluded veterans living in the insular islands and those of unknown rurality (< 1.0% of patients and encounters). Further, because the numbers of highly rural veterans were relatively small (< 2% of patients and encounters), the rural and highly rural categories were combined and compared with urban-dwelling veterans.

Results

Overall, the workload for telerehabilitation nearly quadrupled over the 5-year period (Table 1 and Figure 1). 

In FY 2012, there were 4,397 unique individuals receiving telerehabilitation in the selected telerehabilitation clinics. By FY 2017, this number had grown to 16,319 veterans. 

Similar increases were seen for total encounters, growing from 6,643 in FY 2012 to 22,179 in FY 2017 (Figure 2). The rate of the increase for the number of unique patients seen and telerehabilitation encounter totals across years were higher from FY 2012 to FY 2015 than from FY 2015 to FY 2017.

 

Interesting trends were seen by clinic type. Some clinics increased substantially, whereas others showed only moderate increases, and in 1 case (PM&R Service), a decrease. For example, there is significant growth in the number of patients and encounters involving physical therapy through telerehabilitation. This telerehabilitation clinic increased its workload from 1,676 patients with 3,016 encounters in FY 2012 to 9,136 patients with 11,834 encounters in FY 2017, accounting for 62.6% of total growth in patients and 56.8% of total growth in encounters.

Other clinics showing substantial growth over time included occupational therapy and polytrauma/TBI-individual secondary evaluation. Kinesiotherapy telerehabilitation was almost nonexistent in the VHA during FY 2012, with only 23 patients having 23 encounters. By FY 2017, there were 563 patients with 624 kinesiotherapy telerehabilitation encounters, equating to staggering increases in 5 years: 2,348% for patients and 2,613% for encounters. Similarly, the Physical Medicine and Rehabilitation Assistive Technology clinics had very low numbers in FY 2012 (patients, 2; encounters, 3) and increased over time; albeit, at a slow rate.

Trends by Rurality

Trends by rural location of patients and encounters must be interpreted with caution because of the changing rural definition between FY 2014 and FY 2015 (Tables 2 and 3; Figures 3 and 4). 

Nevertheless, the number of veterans seen and encounters performed via telerehabilitation increased in both urban and rural settings during the time under investigation. 

Under both the legacy and RUCA definitions of rural, the percentage increase was greater for rural veterans than that for urban veterans.

The increased total number of patients seen between FY 2012 and FY 2014 (old definition) was 225% for rural veterans vs 134% for urban veterans. Between FY 2015 and FY 2017 (new definition), the increase was lower for both groups (rural, 13.4%; urban, 7.3%), but rural veterans still increased at a higher rate than did urban dwellers.

Discussion

Our primary aim was to provide data on the growth of telerehabilitation in the VHA over the past 5 years. Our secondary aim was to examine growth in the use of telerehabilitation by rurality. Specifically, we provided an overview of telerehabilitation growth in terms of unique patients and overall encounters in the VHA by rurality from FY 2012 to FY 2014 and FY 2015 to FY 2017 using the following programs: Polytrauma/TBI, PM&R Service, physical therapy, occupational therapy, PM&R amputation clinic, amputation clinic, kinesiotherapy, and PM&R assistive technology clinic. Our findings demonstrated a noteworthy increase in telerehabilitation encounters and unique patients over time for these programs. These findings were consistent with the overall trend of continued growth and expansion of telehealth within the VHA.

Our findings reveal an upward trend in the total number of rural encounters and rural unique patients despite the change in the VA’s definition of rurality in FY 2015. To our knowledge, urban and rural use of telerehabilitation has not been examined previously. Under both definitions of rurality, encounters and unique patients show an important increase over time, and by year-end 2017, more than half of all patients and encounters were attributed to rural patients (53.7% and 53.9%, respectively). Indeed, the upward trend may have been more pronounced if the rural definition had not changed in FY 2015. Our early VHA stroke patients study on the difference between rural-urban patients and taxonomies showed that the RUCA definition was more likely to reduce the number of rural patients by 8.5% than the early definition used by the VHA.²⁶

It is notable that although the use of tele-delivery of rehabilitation has continually increased, the rate of this increase was steeper from FY 2012 to FY 2014 than FY 2015 to FY 2017. For the programs under consideration in this study, the total number of rural patients/encounters increased throughout the observed periods. However, urban patients and encounters increased through FY 2016 and experienced a slight decrease in FY 2017.

The appearance of a slower rate of increase may be due to a rapid initial rate of increase through early adopters and “crossing the diffusion chasm,” a well-documented process of slower diffusion between the time of invention to penetration that often characterizes the spread of successful telehealth innovations.²⁷ Integrating technology into care delivery innovation requires the integration of technical, clinical, and administrative processes and can take time to scale successfully.²⁸

With an emphasis on increasing access to rehabilitation services, the VHA can expect to see a continuing increase in both the number and the percentage of telerehabilitation rural patients and encounters. The VHA has several telerehabilitation initiatives underway through the VHA’s Physical Medicine and Rehabilitation Telerehabilitation Enterprise Wide Initiative (TREWI) and Rural Veterans Telerehabilitation Initiative. These projects demonstrate the feasibility of this delivery approach and facilitate integration of this modality in clinical workflows. However, to sustain these efforts, facilities will need more infrastructure and personnel resources dedicated to the delivery of services.

In an ongoing evaluation of the TREWI, several factors seem to influence the uptake of the VHA Office of Rural Health TREWI programs. These factors are the presence or absence of a local site champion; the quality of hospital leadership support; the quality of past relationships between telerehabilitation sending sites and receiving sites; barriers to getting a telehealth service agreement in place; the availability of space; administrative know-how on setting up clinics appropriately; time involved to bring on staff; contracting issues; equipment availability and installation; cultural issues in embracing technologic innovation; training burden; hassle factors; and limited funds. Although early adopters may be able to negotiate and push through many of the barriers associated with the diffusion of telerehabilitation, the numerous barriers may slow its larger systemwide diffusion.

Telerehabilitation is a promising mode to deliver care to rural veterans who otherwise may not have access to this type of specialty care. Therefore, the identification of elements that foster telerehabilitation growth in future investigations can assist policy makers and key stakeholders in optimally leveraging program resources for maximal productivity. Future studies investigating the drivers of increases in telerehabilitation growth by rurality are warranted. Furthermore, more research is needed to examine telerehabilitation growth quality of care outcomes (eg, patient and provider satisfaction) to ensure that care is not only timely and accessible, but of high quality.

 

Conclusion

Disparities between rural and urban veterans compel a mode of expanding delivery of care. The VHA has embraced the use of telehealth modalities to extend its reach of rehabilitation services to veterans with disability and rehabilitation needs. Growth in telerehabilitation rural patient encounters increases access to rehabilitative care, reduces patient and caregiver travel burden, and helps ensure treatment adherence. Telerehabilitation utilization (unique patients and total encounters) is growing more rapidly for rural veterans than for their urban counterparts. Overall, telerehabilitation is filling a gap for rural veterans, as well as veterans in general with challenges in accessibility to health care. In order to make full use of the telerehabilitation services across its health care system, VA health care facilities may need to expand their effort in telerehabilitation dissemination and education among providers and veterans, particularly among providers who are less familiar with telerehabilitation services and among veterans who live in rural or highly rural areas and need special rehabilitation care.

Historically, the Veterans Health Administration (VHA) has excelled at improving veterans’ access to health care and enhancing foundational services, such as prosthetics and other veteran-centric services, and this continues to be the VHA’s top priority.¹ Travel distance and time are often barriers to accessing health care for many veterans.^2-11 For veterans with disabilities who must overcome additional physical, cognitive, and emotional obstacles to access vital rehabilitation services, these geographic obstacles are magnified. Further compounding the challenge is that rehabilitation therapies frequently require multiple encounters. Telerehabilitation is a promising solution for veterans in need of rehabilitation to regain optimal functioning. This alternative mode of service delivery can help veterans overcome geographic access barriers by delivering health care directly to veterans in their homes or nearby community-based outpatient clinics.^12,13

A growing body of evidence supports telerehabilitation. In a 2017 systematic review and meta-analysis, Cottrell and colleagues reviewed and analyzed data from 13 studies that met their inclusion criteria; specifically, their meta-analytic sample comprised adults aged ≥ 18 years presenting with any diagnosed primary musculoskeletal condition; treatment interventions via a real-time telerehabilitation medium, trials that had a comparison group with the same condition; provided clinical outcomes data, and included published randomized and nonrandomized controlled trials.¹⁴ Based on their aggregated results, they concluded that real-time telerehabilitation was effective in improving physical function (standardized mean difference [SMD], 0.63; 95% CI, 0.92-2.33; I², 93%), and reducing pain (SMD, 0.66; 95% CI, −0.27- .60; I², 96%) in patients with any diagnosed primary musculoskeletal condition.¹⁴

Two other systematic reviews conducted by Pietrzak and colleagues and Agostini and colleagues also demonstrated the clinical effectiveness of telerehabilitation.^15,16 Clinical effectiveness was defined as changes in health, functional status, and satisfaction with the telerehabilitation services delivered. The studies examined in the review included those that provided online self-management and education in addition to exercise via teleconferencing in real time.

Pietrzak and colleagues found that Internet-based osteoarthritis self-management interventions significantly improved 4 of 6 health status measures reviewed (ie, pain, fatigue, activity limitation, health distress, disability, and self‐reported global health).¹⁵ User acceptance and satisfaction were high (≥ 70% satisfied) in all studies meeting the inclusion criteria.

Agostini and colleagues found that telerehabilitation was more effective than other modes of delivering rehabilitation to regain motor function in cardiac (SMD, 0.24; 95% CI, 0.04-0.43) and total knee arthroplasty (Timed Up and Go test: SMD, −5.17; 95% CI, −9.79- −0.55) patients.¹⁶ Some evidence from VHA and non-VHA studies also support the use of telerehabilitation to reduce health care costs,^17-19 improve treatment adherence,^12,20 and enhance patient physical, cognitive and mobility function, as well as patient satisfaction and health-related quality of life.^13,21-24

Since the first recorded use of telehealth in 1959, the application of technology to deliver health care, including rehabilitation services, has increased exponentially.¹⁴ In fiscal year (FY) 2017 alone, the VA provided > 2 million episodes of care for > 700,000 veterans using telehealth services.²⁵

Although the process for accessing telerehabilitation may vary throughout the VA, typically a few common factors make a veteran eligible for this mode of rehabilitation care delivery: Veterans must meet criteria for a specific program (eg, amputation, occupational therapy, and physical therapy) and receive VA care from a VA medical facility or clinic that offers telehealth services. Care providers must believe that the veteran would benefit from telerehabilitation (eg, limited mobility and long-distance travel to the facility) and that they would be able to receive an appropriate consult. The veteran must meet the following requirements: (1) willingness to consent to a visit via telehealth; (2) access to required equipment/e-mail; and (3) a caregiver to assist if they are unable to complete a visit independently.

In this article, we provide an overview of the growth of telerehabilitation in the VHA. Data are presented for specific telerehabilitation programs over time and by rurality.

 

Methods

The VHA Support Service Center works with VHA program offices and field users to provide field-focused business, clinical, and special topic reports. An online portal provides access to these customizable reports organized as data cubes, which represent data dimensions (ie, clinic type) and measures (ie, number of unique patients). For this study, we used the Connected Care, Telehealth, Call Centers Clinical Video Telehealth/Store and Forward Telehealth data cube clinical stop codes to identify the numbers of telerehabilitation veteran users and encounters across time. The following telerehabilitation clinic-stop codes were selected: 197 (polytrauma/traumatic brain injury [TBI]–individuals), 201 (Physical Medicine and Rehabilitation [PM&R] Service), 205 (physical therapy), 206 (occupational therapy), 211 (PM&R amputation clinic), 418 (amputation clinic), 214 (kinesiotherapy), and 240 (PM&R assistive technology clinic). Data for total unique patients served and the total number of encounters were extracted at the national level and by rurality from FY 2012 to FY 2017, providing the past 5 years of VHA telerehabilitation data.

It is important to note that in FY 2015, the VHA changed its definition of rurality to a rural-urban commuting areas (RUCA)-based system (www.ruralhealth.va.gov/rural-definition.asp). Prior to FY 2015, the VHA used the US Census Bureau (CB) urbanized area definitions. According to CB, an urbanized area contains a central city and surrounding area that totals > 50,000 in population. It also includes places outside of urbanized areas with populations > 2,500. Rural areas are defined as all other areas. VHA added a third category, highly rural, which is defined as areas that had < 7 people per square mile. In the RUCA system, each census tract defined by the CB is given a score. The VHA definitions are as follows:

• Urban (U)—census tracts with RUCA scores of 1.0 or 1.1. These tracts are determined by the CB as being in an urban core and having the majority of their workers commute within that same core (1.0). If 30% to 49% commute to an even larger urban core, then the code is 1.1;
• Rural (R)—all tracts not receiving scores in the urban or highly rural tiers; and
• Highly rural (H)—tracts with a RUCA score of 10.0. These are the most remote occupied land areas. Less than 10% of workers travel to CB-defined urbanized areas or urban clusters.

In addition, VHA recently added an “I” category to complement “U,” “R,” and “H.” The “I” value is assigned to veterans living on the US insular islands (ie, territories): Guam, American Samoa, Northern Marianas, and US Virgin Islands. For the analysis by rurality in this study, we excluded veterans living in the insular islands and those of unknown rurality (< 1.0% of patients and encounters). Further, because the numbers of highly rural veterans were relatively small (< 2% of patients and encounters), the rural and highly rural categories were combined and compared with urban-dwelling veterans.

Results

Overall, the workload for telerehabilitation nearly quadrupled over the 5-year period (Table 1 and Figure 1). 

In FY 2012, there were 4,397 unique individuals receiving telerehabilitation in the selected telerehabilitation clinics. By FY 2017, this number had grown to 16,319 veterans. 

Similar increases were seen for total encounters, growing from 6,643 in FY 2012 to 22,179 in FY 2017 (Figure 2). The rate of the increase for the number of unique patients seen and telerehabilitation encounter totals across years were higher from FY 2012 to FY 2015 than from FY 2015 to FY 2017.

 

Interesting trends were seen by clinic type. Some clinics increased substantially, whereas others showed only moderate increases, and in 1 case (PM&R Service), a decrease. For example, there is significant growth in the number of patients and encounters involving physical therapy through telerehabilitation. This telerehabilitation clinic increased its workload from 1,676 patients with 3,016 encounters in FY 2012 to 9,136 patients with 11,834 encounters in FY 2017, accounting for 62.6% of total growth in patients and 56.8% of total growth in encounters.

Other clinics showing substantial growth over time included occupational therapy and polytrauma/TBI-individual secondary evaluation. Kinesiotherapy telerehabilitation was almost nonexistent in the VHA during FY 2012, with only 23 patients having 23 encounters. By FY 2017, there were 563 patients with 624 kinesiotherapy telerehabilitation encounters, equating to staggering increases in 5 years: 2,348% for patients and 2,613% for encounters. Similarly, the Physical Medicine and Rehabilitation Assistive Technology clinics had very low numbers in FY 2012 (patients, 2; encounters, 3) and increased over time; albeit, at a slow rate.

Trends by Rurality

Trends by rural location of patients and encounters must be interpreted with caution because of the changing rural definition between FY 2014 and FY 2015 (Tables 2 and 3; Figures 3 and 4). 

Nevertheless, the number of veterans seen and encounters performed via telerehabilitation increased in both urban and rural settings during the time under investigation. 

Under both the legacy and RUCA definitions of rural, the percentage increase was greater for rural veterans than that for urban veterans.

The increased total number of patients seen between FY 2012 and FY 2014 (old definition) was 225% for rural veterans vs 134% for urban veterans. Between FY 2015 and FY 2017 (new definition), the increase was lower for both groups (rural, 13.4%; urban, 7.3%), but rural veterans still increased at a higher rate than did urban dwellers.

Discussion

Our primary aim was to provide data on the growth of telerehabilitation in the VHA over the past 5 years. Our secondary aim was to examine growth in the use of telerehabilitation by rurality. Specifically, we provided an overview of telerehabilitation growth in terms of unique patients and overall encounters in the VHA by rurality from FY 2012 to FY 2014 and FY 2015 to FY 2017 using the following programs: Polytrauma/TBI, PM&R Service, physical therapy, occupational therapy, PM&R amputation clinic, amputation clinic, kinesiotherapy, and PM&R assistive technology clinic. Our findings demonstrated a noteworthy increase in telerehabilitation encounters and unique patients over time for these programs. These findings were consistent with the overall trend of continued growth and expansion of telehealth within the VHA.

Our findings reveal an upward trend in the total number of rural encounters and rural unique patients despite the change in the VA’s definition of rurality in FY 2015. To our knowledge, urban and rural use of telerehabilitation has not been examined previously. Under both definitions of rurality, encounters and unique patients show an important increase over time, and by year-end 2017, more than half of all patients and encounters were attributed to rural patients (53.7% and 53.9%, respectively). Indeed, the upward trend may have been more pronounced if the rural definition had not changed in FY 2015. Our early VHA stroke patients study on the difference between rural-urban patients and taxonomies showed that the RUCA definition was more likely to reduce the number of rural patients by 8.5% than the early definition used by the VHA.²⁶

It is notable that although the use of tele-delivery of rehabilitation has continually increased, the rate of this increase was steeper from FY 2012 to FY 2014 than FY 2015 to FY 2017. For the programs under consideration in this study, the total number of rural patients/encounters increased throughout the observed periods. However, urban patients and encounters increased through FY 2016 and experienced a slight decrease in FY 2017.

The appearance of a slower rate of increase may be due to a rapid initial rate of increase through early adopters and “crossing the diffusion chasm,” a well-documented process of slower diffusion between the time of invention to penetration that often characterizes the spread of successful telehealth innovations.²⁷ Integrating technology into care delivery innovation requires the integration of technical, clinical, and administrative processes and can take time to scale successfully.²⁸

With an emphasis on increasing access to rehabilitation services, the VHA can expect to see a continuing increase in both the number and the percentage of telerehabilitation rural patients and encounters. The VHA has several telerehabilitation initiatives underway through the VHA’s Physical Medicine and Rehabilitation Telerehabilitation Enterprise Wide Initiative (TREWI) and Rural Veterans Telerehabilitation Initiative. These projects demonstrate the feasibility of this delivery approach and facilitate integration of this modality in clinical workflows. However, to sustain these efforts, facilities will need more infrastructure and personnel resources dedicated to the delivery of services.

In an ongoing evaluation of the TREWI, several factors seem to influence the uptake of the VHA Office of Rural Health TREWI programs. These factors are the presence or absence of a local site champion; the quality of hospital leadership support; the quality of past relationships between telerehabilitation sending sites and receiving sites; barriers to getting a telehealth service agreement in place; the availability of space; administrative know-how on setting up clinics appropriately; time involved to bring on staff; contracting issues; equipment availability and installation; cultural issues in embracing technologic innovation; training burden; hassle factors; and limited funds. Although early adopters may be able to negotiate and push through many of the barriers associated with the diffusion of telerehabilitation, the numerous barriers may slow its larger systemwide diffusion.

Telerehabilitation is a promising mode to deliver care to rural veterans who otherwise may not have access to this type of specialty care. Therefore, the identification of elements that foster telerehabilitation growth in future investigations can assist policy makers and key stakeholders in optimally leveraging program resources for maximal productivity. Future studies investigating the drivers of increases in telerehabilitation growth by rurality are warranted. Furthermore, more research is needed to examine telerehabilitation growth quality of care outcomes (eg, patient and provider satisfaction) to ensure that care is not only timely and accessible, but of high quality.

 

Conclusion

Disparities between rural and urban veterans compel a mode of expanding delivery of care. The VHA has embraced the use of telehealth modalities to extend its reach of rehabilitation services to veterans with disability and rehabilitation needs. Growth in telerehabilitation rural patient encounters increases access to rehabilitative care, reduces patient and caregiver travel burden, and helps ensure treatment adherence. Telerehabilitation utilization (unique patients and total encounters) is growing more rapidly for rural veterans than for their urban counterparts. Overall, telerehabilitation is filling a gap for rural veterans, as well as veterans in general with challenges in accessibility to health care. In order to make full use of the telerehabilitation services across its health care system, VA health care facilities may need to expand their effort in telerehabilitation dissemination and education among providers and veterans, particularly among providers who are less familiar with telerehabilitation services and among veterans who live in rural or highly rural areas and need special rehabilitation care.

Historically, the Veterans Health Administration (VHA) has excelled at improving veterans’ access to health care and enhancing foundational services, such as prosthetics and other veteran-centric services, and this continues to be the VHA’s top priority.¹ Travel distance and time are often barriers to accessing health care for many veterans.^2-11 For veterans with disabilities who must overcome additional physical, cognitive, and emotional obstacles to access vital rehabilitation services, these geographic obstacles are magnified. Further compounding the challenge is that rehabilitation therapies frequently require multiple encounters. Telerehabilitation is a promising solution for veterans in need of rehabilitation to regain optimal functioning. This alternative mode of service delivery can help veterans overcome geographic access barriers by delivering health care directly to veterans in their homes or nearby community-based outpatient clinics.^12,13

A growing body of evidence supports telerehabilitation. In a 2017 systematic review and meta-analysis, Cottrell and colleagues reviewed and analyzed data from 13 studies that met their inclusion criteria; specifically, their meta-analytic sample comprised adults aged ≥ 18 years presenting with any diagnosed primary musculoskeletal condition; treatment interventions via a real-time telerehabilitation medium, trials that had a comparison group with the same condition; provided clinical outcomes data, and included published randomized and nonrandomized controlled trials.¹⁴ Based on their aggregated results, they concluded that real-time telerehabilitation was effective in improving physical function (standardized mean difference [SMD], 0.63; 95% CI, 0.92-2.33; I², 93%), and reducing pain (SMD, 0.66; 95% CI, −0.27- .60; I², 96%) in patients with any diagnosed primary musculoskeletal condition.¹⁴

Two other systematic reviews conducted by Pietrzak and colleagues and Agostini and colleagues also demonstrated the clinical effectiveness of telerehabilitation.^15,16 Clinical effectiveness was defined as changes in health, functional status, and satisfaction with the telerehabilitation services delivered. The studies examined in the review included those that provided online self-management and education in addition to exercise via teleconferencing in real time.

Pietrzak and colleagues found that Internet-based osteoarthritis self-management interventions significantly improved 4 of 6 health status measures reviewed (ie, pain, fatigue, activity limitation, health distress, disability, and self‐reported global health).¹⁵ User acceptance and satisfaction were high (≥ 70% satisfied) in all studies meeting the inclusion criteria.

Agostini and colleagues found that telerehabilitation was more effective than other modes of delivering rehabilitation to regain motor function in cardiac (SMD, 0.24; 95% CI, 0.04-0.43) and total knee arthroplasty (Timed Up and Go test: SMD, −5.17; 95% CI, −9.79- −0.55) patients.¹⁶ Some evidence from VHA and non-VHA studies also support the use of telerehabilitation to reduce health care costs,^17-19 improve treatment adherence,^12,20 and enhance patient physical, cognitive and mobility function, as well as patient satisfaction and health-related quality of life.^13,21-24

Since the first recorded use of telehealth in 1959, the application of technology to deliver health care, including rehabilitation services, has increased exponentially.¹⁴ In fiscal year (FY) 2017 alone, the VA provided > 2 million episodes of care for > 700,000 veterans using telehealth services.²⁵

Although the process for accessing telerehabilitation may vary throughout the VA, typically a few common factors make a veteran eligible for this mode of rehabilitation care delivery: Veterans must meet criteria for a specific program (eg, amputation, occupational therapy, and physical therapy) and receive VA care from a VA medical facility or clinic that offers telehealth services. Care providers must believe that the veteran would benefit from telerehabilitation (eg, limited mobility and long-distance travel to the facility) and that they would be able to receive an appropriate consult. The veteran must meet the following requirements: (1) willingness to consent to a visit via telehealth; (2) access to required equipment/e-mail; and (3) a caregiver to assist if they are unable to complete a visit independently.

In this article, we provide an overview of the growth of telerehabilitation in the VHA. Data are presented for specific telerehabilitation programs over time and by rurality.

 

Methods

The VHA Support Service Center works with VHA program offices and field users to provide field-focused business, clinical, and special topic reports. An online portal provides access to these customizable reports organized as data cubes, which represent data dimensions (ie, clinic type) and measures (ie, number of unique patients). For this study, we used the Connected Care, Telehealth, Call Centers Clinical Video Telehealth/Store and Forward Telehealth data cube clinical stop codes to identify the numbers of telerehabilitation veteran users and encounters across time. The following telerehabilitation clinic-stop codes were selected: 197 (polytrauma/traumatic brain injury [TBI]–individuals), 201 (Physical Medicine and Rehabilitation [PM&R] Service), 205 (physical therapy), 206 (occupational therapy), 211 (PM&R amputation clinic), 418 (amputation clinic), 214 (kinesiotherapy), and 240 (PM&R assistive technology clinic). Data for total unique patients served and the total number of encounters were extracted at the national level and by rurality from FY 2012 to FY 2017, providing the past 5 years of VHA telerehabilitation data.

It is important to note that in FY 2015, the VHA changed its definition of rurality to a rural-urban commuting areas (RUCA)-based system (www.ruralhealth.va.gov/rural-definition.asp). Prior to FY 2015, the VHA used the US Census Bureau (CB) urbanized area definitions. According to CB, an urbanized area contains a central city and surrounding area that totals > 50,000 in population. It also includes places outside of urbanized areas with populations > 2,500. Rural areas are defined as all other areas. VHA added a third category, highly rural, which is defined as areas that had < 7 people per square mile. In the RUCA system, each census tract defined by the CB is given a score. The VHA definitions are as follows:

• Urban (U)—census tracts with RUCA scores of 1.0 or 1.1. These tracts are determined by the CB as being in an urban core and having the majority of their workers commute within that same core (1.0). If 30% to 49% commute to an even larger urban core, then the code is 1.1;
• Rural (R)—all tracts not receiving scores in the urban or highly rural tiers; and
• Highly rural (H)—tracts with a RUCA score of 10.0. These are the most remote occupied land areas. Less than 10% of workers travel to CB-defined urbanized areas or urban clusters.

In addition, VHA recently added an “I” category to complement “U,” “R,” and “H.” The “I” value is assigned to veterans living on the US insular islands (ie, territories): Guam, American Samoa, Northern Marianas, and US Virgin Islands. For the analysis by rurality in this study, we excluded veterans living in the insular islands and those of unknown rurality (< 1.0% of patients and encounters). Further, because the numbers of highly rural veterans were relatively small (< 2% of patients and encounters), the rural and highly rural categories were combined and compared with urban-dwelling veterans.

Results

Overall, the workload for telerehabilitation nearly quadrupled over the 5-year period (Table 1 and Figure 1). 

In FY 2012, there were 4,397 unique individuals receiving telerehabilitation in the selected telerehabilitation clinics. By FY 2017, this number had grown to 16,319 veterans. 

Similar increases were seen for total encounters, growing from 6,643 in FY 2012 to 22,179 in FY 2017 (Figure 2). The rate of the increase for the number of unique patients seen and telerehabilitation encounter totals across years were higher from FY 2012 to FY 2015 than from FY 2015 to FY 2017.

 

Interesting trends were seen by clinic type. Some clinics increased substantially, whereas others showed only moderate increases, and in 1 case (PM&R Service), a decrease. For example, there is significant growth in the number of patients and encounters involving physical therapy through telerehabilitation. This telerehabilitation clinic increased its workload from 1,676 patients with 3,016 encounters in FY 2012 to 9,136 patients with 11,834 encounters in FY 2017, accounting for 62.6% of total growth in patients and 56.8% of total growth in encounters.

Other clinics showing substantial growth over time included occupational therapy and polytrauma/TBI-individual secondary evaluation. Kinesiotherapy telerehabilitation was almost nonexistent in the VHA during FY 2012, with only 23 patients having 23 encounters. By FY 2017, there were 563 patients with 624 kinesiotherapy telerehabilitation encounters, equating to staggering increases in 5 years: 2,348% for patients and 2,613% for encounters. Similarly, the Physical Medicine and Rehabilitation Assistive Technology clinics had very low numbers in FY 2012 (patients, 2; encounters, 3) and increased over time; albeit, at a slow rate.

Trends by Rurality

Trends by rural location of patients and encounters must be interpreted with caution because of the changing rural definition between FY 2014 and FY 2015 (Tables 2 and 3; Figures 3 and 4). 

Nevertheless, the number of veterans seen and encounters performed via telerehabilitation increased in both urban and rural settings during the time under investigation. 

Under both the legacy and RUCA definitions of rural, the percentage increase was greater for rural veterans than that for urban veterans.

The increased total number of patients seen between FY 2012 and FY 2014 (old definition) was 225% for rural veterans vs 134% for urban veterans. Between FY 2015 and FY 2017 (new definition), the increase was lower for both groups (rural, 13.4%; urban, 7.3%), but rural veterans still increased at a higher rate than did urban dwellers.

Discussion

Our primary aim was to provide data on the growth of telerehabilitation in the VHA over the past 5 years. Our secondary aim was to examine growth in the use of telerehabilitation by rurality. Specifically, we provided an overview of telerehabilitation growth in terms of unique patients and overall encounters in the VHA by rurality from FY 2012 to FY 2014 and FY 2015 to FY 2017 using the following programs: Polytrauma/TBI, PM&R Service, physical therapy, occupational therapy, PM&R amputation clinic, amputation clinic, kinesiotherapy, and PM&R assistive technology clinic. Our findings demonstrated a noteworthy increase in telerehabilitation encounters and unique patients over time for these programs. These findings were consistent with the overall trend of continued growth and expansion of telehealth within the VHA.

Our findings reveal an upward trend in the total number of rural encounters and rural unique patients despite the change in the VA’s definition of rurality in FY 2015. To our knowledge, urban and rural use of telerehabilitation has not been examined previously. Under both definitions of rurality, encounters and unique patients show an important increase over time, and by year-end 2017, more than half of all patients and encounters were attributed to rural patients (53.7% and 53.9%, respectively). Indeed, the upward trend may have been more pronounced if the rural definition had not changed in FY 2015. Our early VHA stroke patients study on the difference between rural-urban patients and taxonomies showed that the RUCA definition was more likely to reduce the number of rural patients by 8.5% than the early definition used by the VHA.²⁶

It is notable that although the use of tele-delivery of rehabilitation has continually increased, the rate of this increase was steeper from FY 2012 to FY 2014 than FY 2015 to FY 2017. For the programs under consideration in this study, the total number of rural patients/encounters increased throughout the observed periods. However, urban patients and encounters increased through FY 2016 and experienced a slight decrease in FY 2017.

The appearance of a slower rate of increase may be due to a rapid initial rate of increase through early adopters and “crossing the diffusion chasm,” a well-documented process of slower diffusion between the time of invention to penetration that often characterizes the spread of successful telehealth innovations.²⁷ Integrating technology into care delivery innovation requires the integration of technical, clinical, and administrative processes and can take time to scale successfully.²⁸

With an emphasis on increasing access to rehabilitation services, the VHA can expect to see a continuing increase in both the number and the percentage of telerehabilitation rural patients and encounters. The VHA has several telerehabilitation initiatives underway through the VHA’s Physical Medicine and Rehabilitation Telerehabilitation Enterprise Wide Initiative (TREWI) and Rural Veterans Telerehabilitation Initiative. These projects demonstrate the feasibility of this delivery approach and facilitate integration of this modality in clinical workflows. However, to sustain these efforts, facilities will need more infrastructure and personnel resources dedicated to the delivery of services.

In an ongoing evaluation of the TREWI, several factors seem to influence the uptake of the VHA Office of Rural Health TREWI programs. These factors are the presence or absence of a local site champion; the quality of hospital leadership support; the quality of past relationships between telerehabilitation sending sites and receiving sites; barriers to getting a telehealth service agreement in place; the availability of space; administrative know-how on setting up clinics appropriately; time involved to bring on staff; contracting issues; equipment availability and installation; cultural issues in embracing technologic innovation; training burden; hassle factors; and limited funds. Although early adopters may be able to negotiate and push through many of the barriers associated with the diffusion of telerehabilitation, the numerous barriers may slow its larger systemwide diffusion.

Telerehabilitation is a promising mode to deliver care to rural veterans who otherwise may not have access to this type of specialty care. Therefore, the identification of elements that foster telerehabilitation growth in future investigations can assist policy makers and key stakeholders in optimally leveraging program resources for maximal productivity. Future studies investigating the drivers of increases in telerehabilitation growth by rurality are warranted. Furthermore, more research is needed to examine telerehabilitation growth quality of care outcomes (eg, patient and provider satisfaction) to ensure that care is not only timely and accessible, but of high quality.

 

Conclusion

Disparities between rural and urban veterans compel a mode of expanding delivery of care. The VHA has embraced the use of telehealth modalities to extend its reach of rehabilitation services to veterans with disability and rehabilitation needs. Growth in telerehabilitation rural patient encounters increases access to rehabilitative care, reduces patient and caregiver travel burden, and helps ensure treatment adherence. Telerehabilitation utilization (unique patients and total encounters) is growing more rapidly for rural veterans than for their urban counterparts. Overall, telerehabilitation is filling a gap for rural veterans, as well as veterans in general with challenges in accessibility to health care. In order to make full use of the telerehabilitation services across its health care system, VA health care facilities may need to expand their effort in telerehabilitation dissemination and education among providers and veterans, particularly among providers who are less familiar with telerehabilitation services and among veterans who live in rural or highly rural areas and need special rehabilitation care.

Knee osteoarthritis (OA) affects almost 9.3 million adults in the US and accounts for $27 billion in annual health care expenses.^1,2 Due to the increasing cost of health care and an aging population, there has been renewed interest in establishing criteria for nonarthroplasty treatment of knee OA.

In 2013, using the RAND/UCLA Appropriateness method, the American Academy of Orthopaedic Surgeons (AAOS) developed an appropriate use criteria (AUC) for nonarthroplasty management of primary OA of the knee, based on orthopaedic literature and expert opinion.³ Interventions such as activity modification, weight loss, prescribed physical therapy, nonsteroidal anti-inflammatory drugs, tramadol, prescribed oral or transcutaneous opioids, acetaminophen, intra-articular corticosteroids, hinged or unloading knee braces, arthroscopic partial menisectomy or loose body removal, and realignment osteotomy were assessed. An algorithm was developed for 576 patients scenarios that incorporated patient-specific, prognostic/predictor variables to assign designations of “appropriate,” “may be appropriate,” or “rarely appropriate,” to treatment interventions.^4,5 An online version of the algorithm (orthoguidelines.org) is available for physicians and surgeons to judge appropriateness of nonarthroplasty treatments; however, it is not intended to mandate candidacy for treatment or intervention.

Clinical evaluation of the AAOS AUC is necessary to determine how treatment recommendations correlate with current practice. A recent examination of the AAOS Appropriateness System for Surgical Management of Knee OA found that prognostic/predictor variables, such as patient age, OA severity, and pattern of knee OA involvement were more heavily weighted when determining arthroplasty appropriateness than was pain severity or functional loss.⁶ Furthermore, non-AAOS AUC prognostic/predictor variables, such as race and gender, have been linked to disparities in utilization of knee OA interventions.^7-9 Such disparities can be costly not just from a patient perceptive, but also employer and societal perspectives.¹⁰

The Department of Veterans Affairs (VA) health care system represents a model of equal-access-to care system in the US that is ideal for examination of issues about health care utilization and any disparities within the AAOS AUC model and has previously been used to assess utilization of total knee arthroplasty.⁹ The aim of this study was to characterize utilization of the AAOS AUC for nonarthroplasty treatment of knee OA in a VA patient population. We asked the following questions: (1) What variables are predictive of receiving a greater number of AAOS AUC evaluated nonarthroplasty treatments? (2) What variables are predictive of receiving “rarely appropriate” AAOS AUC evaluated nonarthroplasty treatment? (3) What factors are predictive of duration of nonarthroplasty care until total knee arthroplasty (TKA)?

Methods

The institutional review board at the Louis Stokes Cleveland VA Medical Center in Ohio approved a retrospective chart review of nonarthroplasty treatments utilized by patients presenting to its orthopaedic section who subsequently underwent knee arthroplasty between 2013 and 2016. Eligibility criteria included patients aged ≥ 30 years with a diagnosis of unilateral or bilateral primary knee OA. Patients with posttraumatic OA, inflammatory arthritis, and a history of infectious arthritis or Charcot arthropathy of the knee were excluded. Patients with a body mass index (BMI) > 40 or a hemoglobin A_1c > 8.0 at presentation were excluded as nonarthroplasty care was the recommended course of treatment above these thresholds.

 

Data collected included race, gender, duration of nonarthroplasty treatment, BMI, and Kellgren-Lawrence classification of knee OA at time of presentation for symptomatic knee OA.¹¹ All AAOS AUC-evaluated nonarthroplasty treatments utilized prior to arthroplasty intervention also were recorded (Table 1). 

Indications and classifications for each subject were entered into the AAOS AUC online algorithm, and every AAOS AUC evaluated treatment utilized was assigned a rating of appropriate, may be appropriate, or rarely appropriate, based on the algorithm results for that clinical scenario (Table 2). 

Information regarding anti-inflammatory, analgesic, and prescribed oral or transcutaneous opioid use for chronic knee pain during the period of nonoperative management of knee OA prior to TKA was obtained by review of medication lists and reconciliation with orthopaedic consultation notes in the electronic health record. Peri-operative anti-inflammatory, analgesic, and prescribed oral or transcutaneous opioid use did not constitute an AUC intervention.

Statistical Analysis

Statistical analysis was completed with GraphPad Software Prism 7.0a (La Jolla, CA) and Mathworks MatLab R2016b software (Natick, MA). Univariate analysis with Student t tests with Welch corrections in the setting of unequal variance, Mann-Whitney nonparametric tests, and Fisher exact test were generated in the appropriate setting. Multivariable analyses also were conducted. For continuous outcomes, stepwise multiple linear regression was used to generate predictive models; for binary outcomes, binomial logistic regression was used.

Factors analyzed in regression modeling for the total number of AAOS AUC evaluated nonarthroplasty treatments utilized and the likelihood of receiving a rarely appropriate treatment included gender, race, function-limiting pain, range of motion (ROM), ligamentous instability, arthritis pattern, limb alignment, mechanical symptoms, BMI, age, and Kellgren-Lawrence grade. Factors analyzed in timing of TKA included the above variables plus the total number of AUC interventions, whether the patient received an inappropriate intervention, and average appropriateness of the interventions received. Residual analysis with Cook’s distance was used to identify outliers in regression. Observations with Cook’s distance > 3 times the mean Cook’s distance were identified as potential outliers, and models were adjusted accordingly. All statistical analyses were 2-tailed. Statistical significance was set to P ≤ .05 for all outputs.

Results

In the study, 97.8% of participants identified as male, and the mean age was 62.8 years (Table 3). 

The study group was predominantly white (70.3%). All participants had a diagnosis of primary OA. The majority of patients were aged 51 to 70 years (68.1%) and presented with pain occurring following short-distance ambulation (79.1%) but without mechanical symptoms (80.2%). On examination, the majority of patients were found to have full knee ROM (53.8%), no ligamentous instability (97.8%), and normal limb alignment (60.4%). Radiographically, patients most often had multicompartmental disease (69.2%) with evidence of severe joint-space narrowing (63.7%), resulting in a plurality of patients having a Kellgren-Lawrence arthritis grade of 3 (46.2%) (Table 4).

Appropriate Use Criteria Interventions

Patients received a mean of 5.2 AAOS AUC evaluated interventions before undergoing arthroplasty management at a mean of 32.3 months (range 2-181 months) from initial presentation. The majority of these interventions were classified as either appropriate or may be appropriate, according to the AUC definitions (95.1%). Self-management and physical therapy programs were widely utilized (100% and 90.1%, respectively), with all use of these interventions classified as appropriate.

 

Hinged or unloader knee braces were utilized in about half the study patients; this intervention was classified as rarely appropriate in 4.4% of these patients. Medical therapy was also widely used, with all use of NSAIDs, acetaminophen, and tramadol classified as appropriate or may be appropriate. Oral or transcutaneous opioid medications were prescribed in 14.3% of patients, with 92.3% of this use classified as rarely appropriate. Although the opioid medication prescribing provider was not specifically evaluated, there were no instances in which the orthopaedic service provided an oral or transcutaneous opioid prescriptions. Procedural interventions, with the exception of corticosteroid injections, were uncommon; no patient received realignment osteotomy, and only 12.1% of patients underwent arthroscopy. The use of arthroscopy was deemed rarely appropriate in 72.7% of these cases.

Factors Associated With AAOS AUC Intervention Use

There was no difference in the number of AAOS AUC evaluated interventions received based on BMI (mean [SD] BMI < 35, 5.2 [1.0] vs BMI ≥ 35, 5.3 [1.1], P = .49), age (mean [SD] aged < 60 years, 5.4 [1.0] vs aged ≥ 60 years, 5.1 [1.2], P = .23), or Kellgren-Lawrence arthritic grade (mean [SD] grade ≤ 2, 5.5 [1.0] vs grade > 2, 5.1 [1.1], P = .06). These variables also were not associated with receiving a rarely appropriate intervention (mean [SD] BMI < 35, 0.27 [0.5] vs BMI > 35, 0.2 [0.4], P = .81; aged > 60 years, 0.3 [0.5] vs aged < 60 years, 0.2 [0.4], P = .26; Kellgren-Lawrence grade < 2, 0.4 [0.6] vs grade > 2, 0.2 [0.4], P = .1).

Regression modeling to predict total number of AAOS AUC evaluated interventions received produced a significant model (R² = 0.111, P = .006). The presence of ligamentous instability (β coefficient, -1.61) and the absence of mechanical symptoms (β coefficient, -0.67) were negative predictors of number of AUC interventions received. Variance inflation factors were 1.014 and 1.012, respectively. Likewise, regression modeling to identify factors predictive of receiving a rarely appropriate intervention also produced a significant model (pseudo R²= 0.06, P = .025), with lower Kellgren-Lawrence grade the only significant predictor of receiving a rarely appropriate intervention (odds ratio [OR] 0.54; 95% CI, 0.42 -0.72, per unit increase).

Timing from presentation to arthroplasty intervention was also evaluated. Age was a negative predictor (β coefficient -1.61), while positive predictors were reduced ROM (β coefficient 15.72) and having more AUC interventions (β coefficient 7.31) (model R²= 0.29, P = < .001). Age was the most significant predictor. Variance inflations factors were 1.02, 1.01, and 1.03, respectively. Receiving a rarely appropriate intervention was not associated with TKA timing.

Discussion

This single-center retrospective study examined the utilization of AAOS AUC-evaluated nonarthroplasty interventions for symptomatic knee OA prior to TKA. The aims of this study were to validate the AAOS AUC in a clinical setting and identify predictors of AAOS AUC utilization. In particular, this study focused on the number of interventions utilized prior to knee arthroplasty, whether interventions receiving a designation of rarely appropriate were used, and the duration of nonarthroplasty treatment.

 

Patients with knee instability used fewer total AAOS AUC evaluated interventions prior to TKA. Subjective instability has been reported as high as 27% in patients with OA and has been associated with fear of falling, poor balance confidence, activity limitations, and lower Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) physical function scores.¹² However, it has not been found to correlate with knee laxity.¹³ Nevertheless, significant functional impairment with the risk of falling may reduce the number of nonarthroplasty interventions attempted. On the other hand, the presence of mechanical symptoms resulted in greater utilization of nonarthroplasty interventions. This is likely due to the greater utilization of arthroscopic partial menisectomy or loose body removal in this group of patients. Despite its inclusion as an AAOS AUC evaluated intervention, arthroscopy remains a contentious treatment for symptomatic knee pain in the setting of OA.^14,15

For every unit decrease in Kellgren-Lawrence OA grade, patients were 54% more likely to receive a rarely appropriate intervention prior to knee arthroplasty. This is supported by the recent literature examining the AAOS AUC for surgical management of knee OA. Riddle and colleagues developed a classification tree to determine the contributions of various prognostic variables in final classifications of the 864 clinical vignettes used to develop the appropriateness algorithm and found that OA severity was strongly favored, with only 4 of the 432 vignettes with severe knee OA judged as rarely appropriate for surgical intervention.⁶

Our findings, too, may be explained by an AAOS AUC system that too heavily weighs radiographic severity of knee OA, resulting in more frequent rarely appropriate interventions in patients with less severe arthritis, including nonarthroplasty treatments. It is likely that rarely appropriate interventions were attempted in this subset of our study cohort based on patient’s subjective symptoms and functional status, both of which have been shown to be discordant with radiographic severity of knee OA.¹⁶

Oral or transcutaneous prescribed opioid medications were the most frequent intervention that received a rarely appropriate designation. Patients with preoperative opioid use undergoing TKA have been shown to have a greater risk for postoperative complications and longer hospital stay, particularly those patients aged < 75 years. Younger age, use of more interventions, and decreased knee ROM at presentation were predictive of longer duration of nonarthroplasty treatment. The use of more AAOS AUC evaluated interventions in these patients suggests that the AAOS AUC model may effectively be used to manage symptomatic OA, increasing the time from presentation to knee arthroplasty.

Interestingly, the use of rarely appropriate interventions did not affect TKA timing, as would be expected in a clinically effective nonarthroplasty treatment model. The reasons for rarely appropriate nonsurgical interventions are complex and require further investigation. One possible explanation is that decreased ROM was a marker for mechanical symptoms that necessitated additional intervention in the form of knee arthroscopy, delaying time to TKA.

Limitations

There are several limitations of this study. First, the small sample size (N = 90) requires acknowledgment; however, this limitation reflects the difficulty in following patients for years prior to an operative intervention. Second, the study population consists of veterans using the VA system and may not be reflective of the general population, differing with respect to gender, racial, and socioeconomic factors. Nevertheless, studies examining TKA utilization found, aside from racial and ethnic variability, patient gender and age do not affect arthroplasty utilization rate in the VA system.¹⁷

 

Additional limitations stem from the retrospective nature of this study. While the Computerized Patient Record System and centralized care of the VA system allows for review of all physical therapy consultations, orthotic consultations, and medications within the VA system, any treatments and intervention delivered by non-VA providers were not captured. Furthermore, the ability to assess for confounding variables limiting the prescription of certain medications, such as chronic kidney disease with NSAIDs or liver disease with acetaminophen, was limited by our study design.

Although our study suffers from selection bias with respect to examination of nonarthroplasty treatment in patients who have ultimately undergone TKA, we feel that this subset of patients with symptomatic knee OA represents the majority of patients evaluated for knee OA by orthopaedic surgeons in the clinic setting. It should be noted that although realignment osteotomies were sometimes indicated as appropriate by AAOS AUC model in our study population, this intervention was never performed due to patient and surgeon preference. Additionally, although it is not an AAOS AUC evaluated intervention, viscosupplementation was sporadically used during the study period; however, it is now off formulary at the investigation institution.

Conclusion

Our study suggests that patients without knee instability use more nonarthroplasty treatments over a longer period before TKA, and those patients with less severe knee OA are at risk of receiving an intervention judged to be rarely appropriate by the AAOS AUC. Such interventions do not affect timing of TKA. Nonarthroplasty care should be individualized to patients’ needs, and the decision to proceed with arthroplasty should be considered only after exhausting appropriate conservative measures. We recommend that providers use the AAOS AUC, especially when treating younger patients with less severe knee OA, particularly if considering opiate therapy or knee arthroscopy.

Acknowledgments
The authors would like to acknowledge Patrick Getty, MD, for his surgical care of some of the study patients. This material is the result of work supported with resources and the use of facilities at the Louis Stokes Cleveland VA Medical Center in Ohio.

Knee osteoarthritis (OA) affects almost 9.3 million adults in the US and accounts for $27 billion in annual health care expenses.^1,2 Due to the increasing cost of health care and an aging population, there has been renewed interest in establishing criteria for nonarthroplasty treatment of knee OA.

In 2013, using the RAND/UCLA Appropriateness method, the American Academy of Orthopaedic Surgeons (AAOS) developed an appropriate use criteria (AUC) for nonarthroplasty management of primary OA of the knee, based on orthopaedic literature and expert opinion.³ Interventions such as activity modification, weight loss, prescribed physical therapy, nonsteroidal anti-inflammatory drugs, tramadol, prescribed oral or transcutaneous opioids, acetaminophen, intra-articular corticosteroids, hinged or unloading knee braces, arthroscopic partial menisectomy or loose body removal, and realignment osteotomy were assessed. An algorithm was developed for 576 patients scenarios that incorporated patient-specific, prognostic/predictor variables to assign designations of “appropriate,” “may be appropriate,” or “rarely appropriate,” to treatment interventions.^4,5 An online version of the algorithm (orthoguidelines.org) is available for physicians and surgeons to judge appropriateness of nonarthroplasty treatments; however, it is not intended to mandate candidacy for treatment or intervention.

Clinical evaluation of the AAOS AUC is necessary to determine how treatment recommendations correlate with current practice. A recent examination of the AAOS Appropriateness System for Surgical Management of Knee OA found that prognostic/predictor variables, such as patient age, OA severity, and pattern of knee OA involvement were more heavily weighted when determining arthroplasty appropriateness than was pain severity or functional loss.⁶ Furthermore, non-AAOS AUC prognostic/predictor variables, such as race and gender, have been linked to disparities in utilization of knee OA interventions.^7-9 Such disparities can be costly not just from a patient perceptive, but also employer and societal perspectives.¹⁰

The Department of Veterans Affairs (VA) health care system represents a model of equal-access-to care system in the US that is ideal for examination of issues about health care utilization and any disparities within the AAOS AUC model and has previously been used to assess utilization of total knee arthroplasty.⁹ The aim of this study was to characterize utilization of the AAOS AUC for nonarthroplasty treatment of knee OA in a VA patient population. We asked the following questions: (1) What variables are predictive of receiving a greater number of AAOS AUC evaluated nonarthroplasty treatments? (2) What variables are predictive of receiving “rarely appropriate” AAOS AUC evaluated nonarthroplasty treatment? (3) What factors are predictive of duration of nonarthroplasty care until total knee arthroplasty (TKA)?

Methods

The institutional review board at the Louis Stokes Cleveland VA Medical Center in Ohio approved a retrospective chart review of nonarthroplasty treatments utilized by patients presenting to its orthopaedic section who subsequently underwent knee arthroplasty between 2013 and 2016. Eligibility criteria included patients aged ≥ 30 years with a diagnosis of unilateral or bilateral primary knee OA. Patients with posttraumatic OA, inflammatory arthritis, and a history of infectious arthritis or Charcot arthropathy of the knee were excluded. Patients with a body mass index (BMI) > 40 or a hemoglobin A_1c > 8.0 at presentation were excluded as nonarthroplasty care was the recommended course of treatment above these thresholds.

 

Data collected included race, gender, duration of nonarthroplasty treatment, BMI, and Kellgren-Lawrence classification of knee OA at time of presentation for symptomatic knee OA.¹¹ All AAOS AUC-evaluated nonarthroplasty treatments utilized prior to arthroplasty intervention also were recorded (Table 1). 

Indications and classifications for each subject were entered into the AAOS AUC online algorithm, and every AAOS AUC evaluated treatment utilized was assigned a rating of appropriate, may be appropriate, or rarely appropriate, based on the algorithm results for that clinical scenario (Table 2). 

Information regarding anti-inflammatory, analgesic, and prescribed oral or transcutaneous opioid use for chronic knee pain during the period of nonoperative management of knee OA prior to TKA was obtained by review of medication lists and reconciliation with orthopaedic consultation notes in the electronic health record. Peri-operative anti-inflammatory, analgesic, and prescribed oral or transcutaneous opioid use did not constitute an AUC intervention.

Statistical Analysis

Statistical analysis was completed with GraphPad Software Prism 7.0a (La Jolla, CA) and Mathworks MatLab R2016b software (Natick, MA). Univariate analysis with Student t tests with Welch corrections in the setting of unequal variance, Mann-Whitney nonparametric tests, and Fisher exact test were generated in the appropriate setting. Multivariable analyses also were conducted. For continuous outcomes, stepwise multiple linear regression was used to generate predictive models; for binary outcomes, binomial logistic regression was used.

Factors analyzed in regression modeling for the total number of AAOS AUC evaluated nonarthroplasty treatments utilized and the likelihood of receiving a rarely appropriate treatment included gender, race, function-limiting pain, range of motion (ROM), ligamentous instability, arthritis pattern, limb alignment, mechanical symptoms, BMI, age, and Kellgren-Lawrence grade. Factors analyzed in timing of TKA included the above variables plus the total number of AUC interventions, whether the patient received an inappropriate intervention, and average appropriateness of the interventions received. Residual analysis with Cook’s distance was used to identify outliers in regression. Observations with Cook’s distance > 3 times the mean Cook’s distance were identified as potential outliers, and models were adjusted accordingly. All statistical analyses were 2-tailed. Statistical significance was set to P ≤ .05 for all outputs.

Results

In the study, 97.8% of participants identified as male, and the mean age was 62.8 years (Table 3). 

The study group was predominantly white (70.3%). All participants had a diagnosis of primary OA. The majority of patients were aged 51 to 70 years (68.1%) and presented with pain occurring following short-distance ambulation (79.1%) but without mechanical symptoms (80.2%). On examination, the majority of patients were found to have full knee ROM (53.8%), no ligamentous instability (97.8%), and normal limb alignment (60.4%). Radiographically, patients most often had multicompartmental disease (69.2%) with evidence of severe joint-space narrowing (63.7%), resulting in a plurality of patients having a Kellgren-Lawrence arthritis grade of 3 (46.2%) (Table 4).

Appropriate Use Criteria Interventions

Patients received a mean of 5.2 AAOS AUC evaluated interventions before undergoing arthroplasty management at a mean of 32.3 months (range 2-181 months) from initial presentation. The majority of these interventions were classified as either appropriate or may be appropriate, according to the AUC definitions (95.1%). Self-management and physical therapy programs were widely utilized (100% and 90.1%, respectively), with all use of these interventions classified as appropriate.

 

Hinged or unloader knee braces were utilized in about half the study patients; this intervention was classified as rarely appropriate in 4.4% of these patients. Medical therapy was also widely used, with all use of NSAIDs, acetaminophen, and tramadol classified as appropriate or may be appropriate. Oral or transcutaneous opioid medications were prescribed in 14.3% of patients, with 92.3% of this use classified as rarely appropriate. Although the opioid medication prescribing provider was not specifically evaluated, there were no instances in which the orthopaedic service provided an oral or transcutaneous opioid prescriptions. Procedural interventions, with the exception of corticosteroid injections, were uncommon; no patient received realignment osteotomy, and only 12.1% of patients underwent arthroscopy. The use of arthroscopy was deemed rarely appropriate in 72.7% of these cases.

Factors Associated With AAOS AUC Intervention Use

There was no difference in the number of AAOS AUC evaluated interventions received based on BMI (mean [SD] BMI < 35, 5.2 [1.0] vs BMI ≥ 35, 5.3 [1.1], P = .49), age (mean [SD] aged < 60 years, 5.4 [1.0] vs aged ≥ 60 years, 5.1 [1.2], P = .23), or Kellgren-Lawrence arthritic grade (mean [SD] grade ≤ 2, 5.5 [1.0] vs grade > 2, 5.1 [1.1], P = .06). These variables also were not associated with receiving a rarely appropriate intervention (mean [SD] BMI < 35, 0.27 [0.5] vs BMI > 35, 0.2 [0.4], P = .81; aged > 60 years, 0.3 [0.5] vs aged < 60 years, 0.2 [0.4], P = .26; Kellgren-Lawrence grade < 2, 0.4 [0.6] vs grade > 2, 0.2 [0.4], P = .1).

Regression modeling to predict total number of AAOS AUC evaluated interventions received produced a significant model (R² = 0.111, P = .006). The presence of ligamentous instability (β coefficient, -1.61) and the absence of mechanical symptoms (β coefficient, -0.67) were negative predictors of number of AUC interventions received. Variance inflation factors were 1.014 and 1.012, respectively. Likewise, regression modeling to identify factors predictive of receiving a rarely appropriate intervention also produced a significant model (pseudo R²= 0.06, P = .025), with lower Kellgren-Lawrence grade the only significant predictor of receiving a rarely appropriate intervention (odds ratio [OR] 0.54; 95% CI, 0.42 -0.72, per unit increase).

Timing from presentation to arthroplasty intervention was also evaluated. Age was a negative predictor (β coefficient -1.61), while positive predictors were reduced ROM (β coefficient 15.72) and having more AUC interventions (β coefficient 7.31) (model R²= 0.29, P = < .001). Age was the most significant predictor. Variance inflations factors were 1.02, 1.01, and 1.03, respectively. Receiving a rarely appropriate intervention was not associated with TKA timing.

Discussion

This single-center retrospective study examined the utilization of AAOS AUC-evaluated nonarthroplasty interventions for symptomatic knee OA prior to TKA. The aims of this study were to validate the AAOS AUC in a clinical setting and identify predictors of AAOS AUC utilization. In particular, this study focused on the number of interventions utilized prior to knee arthroplasty, whether interventions receiving a designation of rarely appropriate were used, and the duration of nonarthroplasty treatment.

 

Patients with knee instability used fewer total AAOS AUC evaluated interventions prior to TKA. Subjective instability has been reported as high as 27% in patients with OA and has been associated with fear of falling, poor balance confidence, activity limitations, and lower Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) physical function scores.¹² However, it has not been found to correlate with knee laxity.¹³ Nevertheless, significant functional impairment with the risk of falling may reduce the number of nonarthroplasty interventions attempted. On the other hand, the presence of mechanical symptoms resulted in greater utilization of nonarthroplasty interventions. This is likely due to the greater utilization of arthroscopic partial menisectomy or loose body removal in this group of patients. Despite its inclusion as an AAOS AUC evaluated intervention, arthroscopy remains a contentious treatment for symptomatic knee pain in the setting of OA.^14,15

For every unit decrease in Kellgren-Lawrence OA grade, patients were 54% more likely to receive a rarely appropriate intervention prior to knee arthroplasty. This is supported by the recent literature examining the AAOS AUC for surgical management of knee OA. Riddle and colleagues developed a classification tree to determine the contributions of various prognostic variables in final classifications of the 864 clinical vignettes used to develop the appropriateness algorithm and found that OA severity was strongly favored, with only 4 of the 432 vignettes with severe knee OA judged as rarely appropriate for surgical intervention.⁶

Our findings, too, may be explained by an AAOS AUC system that too heavily weighs radiographic severity of knee OA, resulting in more frequent rarely appropriate interventions in patients with less severe arthritis, including nonarthroplasty treatments. It is likely that rarely appropriate interventions were attempted in this subset of our study cohort based on patient’s subjective symptoms and functional status, both of which have been shown to be discordant with radiographic severity of knee OA.¹⁶

Oral or transcutaneous prescribed opioid medications were the most frequent intervention that received a rarely appropriate designation. Patients with preoperative opioid use undergoing TKA have been shown to have a greater risk for postoperative complications and longer hospital stay, particularly those patients aged < 75 years. Younger age, use of more interventions, and decreased knee ROM at presentation were predictive of longer duration of nonarthroplasty treatment. The use of more AAOS AUC evaluated interventions in these patients suggests that the AAOS AUC model may effectively be used to manage symptomatic OA, increasing the time from presentation to knee arthroplasty.

Interestingly, the use of rarely appropriate interventions did not affect TKA timing, as would be expected in a clinically effective nonarthroplasty treatment model. The reasons for rarely appropriate nonsurgical interventions are complex and require further investigation. One possible explanation is that decreased ROM was a marker for mechanical symptoms that necessitated additional intervention in the form of knee arthroscopy, delaying time to TKA.

Limitations

There are several limitations of this study. First, the small sample size (N = 90) requires acknowledgment; however, this limitation reflects the difficulty in following patients for years prior to an operative intervention. Second, the study population consists of veterans using the VA system and may not be reflective of the general population, differing with respect to gender, racial, and socioeconomic factors. Nevertheless, studies examining TKA utilization found, aside from racial and ethnic variability, patient gender and age do not affect arthroplasty utilization rate in the VA system.¹⁷

 

Additional limitations stem from the retrospective nature of this study. While the Computerized Patient Record System and centralized care of the VA system allows for review of all physical therapy consultations, orthotic consultations, and medications within the VA system, any treatments and intervention delivered by non-VA providers were not captured. Furthermore, the ability to assess for confounding variables limiting the prescription of certain medications, such as chronic kidney disease with NSAIDs or liver disease with acetaminophen, was limited by our study design.

Although our study suffers from selection bias with respect to examination of nonarthroplasty treatment in patients who have ultimately undergone TKA, we feel that this subset of patients with symptomatic knee OA represents the majority of patients evaluated for knee OA by orthopaedic surgeons in the clinic setting. It should be noted that although realignment osteotomies were sometimes indicated as appropriate by AAOS AUC model in our study population, this intervention was never performed due to patient and surgeon preference. Additionally, although it is not an AAOS AUC evaluated intervention, viscosupplementation was sporadically used during the study period; however, it is now off formulary at the investigation institution.

Conclusion

Our study suggests that patients without knee instability use more nonarthroplasty treatments over a longer period before TKA, and those patients with less severe knee OA are at risk of receiving an intervention judged to be rarely appropriate by the AAOS AUC. Such interventions do not affect timing of TKA. Nonarthroplasty care should be individualized to patients’ needs, and the decision to proceed with arthroplasty should be considered only after exhausting appropriate conservative measures. We recommend that providers use the AAOS AUC, especially when treating younger patients with less severe knee OA, particularly if considering opiate therapy or knee arthroscopy.

Acknowledgments
The authors would like to acknowledge Patrick Getty, MD, for his surgical care of some of the study patients. This material is the result of work supported with resources and the use of facilities at the Louis Stokes Cleveland VA Medical Center in Ohio.

Knee osteoarthritis (OA) affects almost 9.3 million adults in the US and accounts for $27 billion in annual health care expenses.^1,2 Due to the increasing cost of health care and an aging population, there has been renewed interest in establishing criteria for nonarthroplasty treatment of knee OA.

In 2013, using the RAND/UCLA Appropriateness method, the American Academy of Orthopaedic Surgeons (AAOS) developed an appropriate use criteria (AUC) for nonarthroplasty management of primary OA of the knee, based on orthopaedic literature and expert opinion.³ Interventions such as activity modification, weight loss, prescribed physical therapy, nonsteroidal anti-inflammatory drugs, tramadol, prescribed oral or transcutaneous opioids, acetaminophen, intra-articular corticosteroids, hinged or unloading knee braces, arthroscopic partial menisectomy or loose body removal, and realignment osteotomy were assessed. An algorithm was developed for 576 patients scenarios that incorporated patient-specific, prognostic/predictor variables to assign designations of “appropriate,” “may be appropriate,” or “rarely appropriate,” to treatment interventions.^4,5 An online version of the algorithm (orthoguidelines.org) is available for physicians and surgeons to judge appropriateness of nonarthroplasty treatments; however, it is not intended to mandate candidacy for treatment or intervention.

Clinical evaluation of the AAOS AUC is necessary to determine how treatment recommendations correlate with current practice. A recent examination of the AAOS Appropriateness System for Surgical Management of Knee OA found that prognostic/predictor variables, such as patient age, OA severity, and pattern of knee OA involvement were more heavily weighted when determining arthroplasty appropriateness than was pain severity or functional loss.⁶ Furthermore, non-AAOS AUC prognostic/predictor variables, such as race and gender, have been linked to disparities in utilization of knee OA interventions.^7-9 Such disparities can be costly not just from a patient perceptive, but also employer and societal perspectives.¹⁰

The Department of Veterans Affairs (VA) health care system represents a model of equal-access-to care system in the US that is ideal for examination of issues about health care utilization and any disparities within the AAOS AUC model and has previously been used to assess utilization of total knee arthroplasty.⁹ The aim of this study was to characterize utilization of the AAOS AUC for nonarthroplasty treatment of knee OA in a VA patient population. We asked the following questions: (1) What variables are predictive of receiving a greater number of AAOS AUC evaluated nonarthroplasty treatments? (2) What variables are predictive of receiving “rarely appropriate” AAOS AUC evaluated nonarthroplasty treatment? (3) What factors are predictive of duration of nonarthroplasty care until total knee arthroplasty (TKA)?

Methods

The institutional review board at the Louis Stokes Cleveland VA Medical Center in Ohio approved a retrospective chart review of nonarthroplasty treatments utilized by patients presenting to its orthopaedic section who subsequently underwent knee arthroplasty between 2013 and 2016. Eligibility criteria included patients aged ≥ 30 years with a diagnosis of unilateral or bilateral primary knee OA. Patients with posttraumatic OA, inflammatory arthritis, and a history of infectious arthritis or Charcot arthropathy of the knee were excluded. Patients with a body mass index (BMI) > 40 or a hemoglobin A_1c > 8.0 at presentation were excluded as nonarthroplasty care was the recommended course of treatment above these thresholds.

 

Data collected included race, gender, duration of nonarthroplasty treatment, BMI, and Kellgren-Lawrence classification of knee OA at time of presentation for symptomatic knee OA.¹¹ All AAOS AUC-evaluated nonarthroplasty treatments utilized prior to arthroplasty intervention also were recorded (Table 1). 

Indications and classifications for each subject were entered into the AAOS AUC online algorithm, and every AAOS AUC evaluated treatment utilized was assigned a rating of appropriate, may be appropriate, or rarely appropriate, based on the algorithm results for that clinical scenario (Table 2). 

Information regarding anti-inflammatory, analgesic, and prescribed oral or transcutaneous opioid use for chronic knee pain during the period of nonoperative management of knee OA prior to TKA was obtained by review of medication lists and reconciliation with orthopaedic consultation notes in the electronic health record. Peri-operative anti-inflammatory, analgesic, and prescribed oral or transcutaneous opioid use did not constitute an AUC intervention.

Statistical Analysis

Statistical analysis was completed with GraphPad Software Prism 7.0a (La Jolla, CA) and Mathworks MatLab R2016b software (Natick, MA). Univariate analysis with Student t tests with Welch corrections in the setting of unequal variance, Mann-Whitney nonparametric tests, and Fisher exact test were generated in the appropriate setting. Multivariable analyses also were conducted. For continuous outcomes, stepwise multiple linear regression was used to generate predictive models; for binary outcomes, binomial logistic regression was used.

Factors analyzed in regression modeling for the total number of AAOS AUC evaluated nonarthroplasty treatments utilized and the likelihood of receiving a rarely appropriate treatment included gender, race, function-limiting pain, range of motion (ROM), ligamentous instability, arthritis pattern, limb alignment, mechanical symptoms, BMI, age, and Kellgren-Lawrence grade. Factors analyzed in timing of TKA included the above variables plus the total number of AUC interventions, whether the patient received an inappropriate intervention, and average appropriateness of the interventions received. Residual analysis with Cook’s distance was used to identify outliers in regression. Observations with Cook’s distance > 3 times the mean Cook’s distance were identified as potential outliers, and models were adjusted accordingly. All statistical analyses were 2-tailed. Statistical significance was set to P ≤ .05 for all outputs.

Results

In the study, 97.8% of participants identified as male, and the mean age was 62.8 years (Table 3). 

The study group was predominantly white (70.3%). All participants had a diagnosis of primary OA. The majority of patients were aged 51 to 70 years (68.1%) and presented with pain occurring following short-distance ambulation (79.1%) but without mechanical symptoms (80.2%). On examination, the majority of patients were found to have full knee ROM (53.8%), no ligamentous instability (97.8%), and normal limb alignment (60.4%). Radiographically, patients most often had multicompartmental disease (69.2%) with evidence of severe joint-space narrowing (63.7%), resulting in a plurality of patients having a Kellgren-Lawrence arthritis grade of 3 (46.2%) (Table 4).

Appropriate Use Criteria Interventions

Patients received a mean of 5.2 AAOS AUC evaluated interventions before undergoing arthroplasty management at a mean of 32.3 months (range 2-181 months) from initial presentation. The majority of these interventions were classified as either appropriate or may be appropriate, according to the AUC definitions (95.1%). Self-management and physical therapy programs were widely utilized (100% and 90.1%, respectively), with all use of these interventions classified as appropriate.

 

Hinged or unloader knee braces were utilized in about half the study patients; this intervention was classified as rarely appropriate in 4.4% of these patients. Medical therapy was also widely used, with all use of NSAIDs, acetaminophen, and tramadol classified as appropriate or may be appropriate. Oral or transcutaneous opioid medications were prescribed in 14.3% of patients, with 92.3% of this use classified as rarely appropriate. Although the opioid medication prescribing provider was not specifically evaluated, there were no instances in which the orthopaedic service provided an oral or transcutaneous opioid prescriptions. Procedural interventions, with the exception of corticosteroid injections, were uncommon; no patient received realignment osteotomy, and only 12.1% of patients underwent arthroscopy. The use of arthroscopy was deemed rarely appropriate in 72.7% of these cases.

Factors Associated With AAOS AUC Intervention Use

There was no difference in the number of AAOS AUC evaluated interventions received based on BMI (mean [SD] BMI < 35, 5.2 [1.0] vs BMI ≥ 35, 5.3 [1.1], P = .49), age (mean [SD] aged < 60 years, 5.4 [1.0] vs aged ≥ 60 years, 5.1 [1.2], P = .23), or Kellgren-Lawrence arthritic grade (mean [SD] grade ≤ 2, 5.5 [1.0] vs grade > 2, 5.1 [1.1], P = .06). These variables also were not associated with receiving a rarely appropriate intervention (mean [SD] BMI < 35, 0.27 [0.5] vs BMI > 35, 0.2 [0.4], P = .81; aged > 60 years, 0.3 [0.5] vs aged < 60 years, 0.2 [0.4], P = .26; Kellgren-Lawrence grade < 2, 0.4 [0.6] vs grade > 2, 0.2 [0.4], P = .1).

Regression modeling to predict total number of AAOS AUC evaluated interventions received produced a significant model (R² = 0.111, P = .006). The presence of ligamentous instability (β coefficient, -1.61) and the absence of mechanical symptoms (β coefficient, -0.67) were negative predictors of number of AUC interventions received. Variance inflation factors were 1.014 and 1.012, respectively. Likewise, regression modeling to identify factors predictive of receiving a rarely appropriate intervention also produced a significant model (pseudo R²= 0.06, P = .025), with lower Kellgren-Lawrence grade the only significant predictor of receiving a rarely appropriate intervention (odds ratio [OR] 0.54; 95% CI, 0.42 -0.72, per unit increase).

Timing from presentation to arthroplasty intervention was also evaluated. Age was a negative predictor (β coefficient -1.61), while positive predictors were reduced ROM (β coefficient 15.72) and having more AUC interventions (β coefficient 7.31) (model R²= 0.29, P = < .001). Age was the most significant predictor. Variance inflations factors were 1.02, 1.01, and 1.03, respectively. Receiving a rarely appropriate intervention was not associated with TKA timing.

Discussion

This single-center retrospective study examined the utilization of AAOS AUC-evaluated nonarthroplasty interventions for symptomatic knee OA prior to TKA. The aims of this study were to validate the AAOS AUC in a clinical setting and identify predictors of AAOS AUC utilization. In particular, this study focused on the number of interventions utilized prior to knee arthroplasty, whether interventions receiving a designation of rarely appropriate were used, and the duration of nonarthroplasty treatment.

 

Patients with knee instability used fewer total AAOS AUC evaluated interventions prior to TKA. Subjective instability has been reported as high as 27% in patients with OA and has been associated with fear of falling, poor balance confidence, activity limitations, and lower Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) physical function scores.¹² However, it has not been found to correlate with knee laxity.¹³ Nevertheless, significant functional impairment with the risk of falling may reduce the number of nonarthroplasty interventions attempted. On the other hand, the presence of mechanical symptoms resulted in greater utilization of nonarthroplasty interventions. This is likely due to the greater utilization of arthroscopic partial menisectomy or loose body removal in this group of patients. Despite its inclusion as an AAOS AUC evaluated intervention, arthroscopy remains a contentious treatment for symptomatic knee pain in the setting of OA.^14,15

For every unit decrease in Kellgren-Lawrence OA grade, patients were 54% more likely to receive a rarely appropriate intervention prior to knee arthroplasty. This is supported by the recent literature examining the AAOS AUC for surgical management of knee OA. Riddle and colleagues developed a classification tree to determine the contributions of various prognostic variables in final classifications of the 864 clinical vignettes used to develop the appropriateness algorithm and found that OA severity was strongly favored, with only 4 of the 432 vignettes with severe knee OA judged as rarely appropriate for surgical intervention.⁶

Our findings, too, may be explained by an AAOS AUC system that too heavily weighs radiographic severity of knee OA, resulting in more frequent rarely appropriate interventions in patients with less severe arthritis, including nonarthroplasty treatments. It is likely that rarely appropriate interventions were attempted in this subset of our study cohort based on patient’s subjective symptoms and functional status, both of which have been shown to be discordant with radiographic severity of knee OA.¹⁶

Oral or transcutaneous prescribed opioid medications were the most frequent intervention that received a rarely appropriate designation. Patients with preoperative opioid use undergoing TKA have been shown to have a greater risk for postoperative complications and longer hospital stay, particularly those patients aged < 75 years. Younger age, use of more interventions, and decreased knee ROM at presentation were predictive of longer duration of nonarthroplasty treatment. The use of more AAOS AUC evaluated interventions in these patients suggests that the AAOS AUC model may effectively be used to manage symptomatic OA, increasing the time from presentation to knee arthroplasty.

Interestingly, the use of rarely appropriate interventions did not affect TKA timing, as would be expected in a clinically effective nonarthroplasty treatment model. The reasons for rarely appropriate nonsurgical interventions are complex and require further investigation. One possible explanation is that decreased ROM was a marker for mechanical symptoms that necessitated additional intervention in the form of knee arthroscopy, delaying time to TKA.

Limitations

There are several limitations of this study. First, the small sample size (N = 90) requires acknowledgment; however, this limitation reflects the difficulty in following patients for years prior to an operative intervention. Second, the study population consists of veterans using the VA system and may not be reflective of the general population, differing with respect to gender, racial, and socioeconomic factors. Nevertheless, studies examining TKA utilization found, aside from racial and ethnic variability, patient gender and age do not affect arthroplasty utilization rate in the VA system.¹⁷

 

Additional limitations stem from the retrospective nature of this study. While the Computerized Patient Record System and centralized care of the VA system allows for review of all physical therapy consultations, orthotic consultations, and medications within the VA system, any treatments and intervention delivered by non-VA providers were not captured. Furthermore, the ability to assess for confounding variables limiting the prescription of certain medications, such as chronic kidney disease with NSAIDs or liver disease with acetaminophen, was limited by our study design.

Although our study suffers from selection bias with respect to examination of nonarthroplasty treatment in patients who have ultimately undergone TKA, we feel that this subset of patients with symptomatic knee OA represents the majority of patients evaluated for knee OA by orthopaedic surgeons in the clinic setting. It should be noted that although realignment osteotomies were sometimes indicated as appropriate by AAOS AUC model in our study population, this intervention was never performed due to patient and surgeon preference. Additionally, although it is not an AAOS AUC evaluated intervention, viscosupplementation was sporadically used during the study period; however, it is now off formulary at the investigation institution.

Conclusion

Our study suggests that patients without knee instability use more nonarthroplasty treatments over a longer period before TKA, and those patients with less severe knee OA are at risk of receiving an intervention judged to be rarely appropriate by the AAOS AUC. Such interventions do not affect timing of TKA. Nonarthroplasty care should be individualized to patients’ needs, and the decision to proceed with arthroplasty should be considered only after exhausting appropriate conservative measures. We recommend that providers use the AAOS AUC, especially when treating younger patients with less severe knee OA, particularly if considering opiate therapy or knee arthroscopy.

Acknowledgments
The authors would like to acknowledge Patrick Getty, MD, for his surgical care of some of the study patients. This material is the result of work supported with resources and the use of facilities at the Louis Stokes Cleveland VA Medical Center in Ohio.

Psoriasis is a chronic skin condition affecting approximately 2% to 3% of the population.^1,2 Beyond cutaneous manifestations, psoriasis is a systemic inflammatory state that is associated with an increased risk for cardiovascular disease, including obesity,^3,4 type 2 diabetes mellitus,^5,6 hypertension,⁵ dyslipidemia,^3,7 metabolic syndrome,⁷ atherosclerosis,⁸ peripheral vascular disease,⁹ coronary artery calcification,¹⁰ myocardial infarction,^11-13 stroke,^9,14 and cardiac death.^15,16

Psoriasis also has been associated with inflammatory bowel disease (IBD), possibly because of similar autoimmune mechanisms in the pathogenesis of both diseases.^17,18 However, there is no literature regarding the risk for acute gastrointestinal pathologies such as appendicitis, cholecystitis, or diverticulitis in patients with psoriasis.

The primary objective of this study was to examine if patients with psoriasis are at increased risk for appendicitis, cholecystitis, or diverticulitis compared to the general population. The secondary objective was to determine if patients with severe psoriasis (ie, patients treated with phototherapy or systemic therapy) are at a higher risk for these conditions compared to patients with mild psoriasis.

Methods

Patients and Tools
A descriptive, population-based cohort study design with controls from a matched cohort was used to ascertain the effect of psoriasis status on patients’ risk for appendicitis, cholecystitis, or diverticulitis. Our cohort was selected using administrative data from Kaiser Permanente Southern California (KPSC) during the study period (January 1, 2004, through December 31, 2016).

Kaiser Permanente Southern California is a large integrated health maintenance organization that includes approximately 4 million patients as of December 31, 2016, and includes roughly 20% of the region’s population. The geographic area served extends from Bakersfield in the lower California Central Valley to San Diego on the border with Mexico. Membership demographics, socioeconomic status, and ethnicity composition are representative of California.

Patients were included if they had a diagnosis of psoriasis (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] code 696.1; International Classification of Diseases, Tenth Revision, Clinical Modification [ICD-10-CM] codes L40.0, L40.4, L40.8, or L40.9) for at least 3 visits between January 1, 2004, and December 31, 2016. Patients were not excluded if they also had a diagnosis of psoriatic arthritis (ICD-9-CM code 696.0; ICD-10-CM code L40.5x). Patients also must have been continuously enrolled for at least 1 year before and 1 year after the index date, which was defined as the date of the third psoriasis diagnosis.

Each patient with psoriasis was assigned to 1 of 2 cohorts: (1) severe psoriasis: patients who received UVB phototherapy, psoralen plus UVA phototherapy, methotrexate, acitretin, cyclosporine, apremilast, etanercept, adalimumab, infliximab, ustekinumab, efalizumab, alefacept, secukinumab, or ixekizumab during the study period; and (2) mild psoriasis: patients who had a diagnosis of psoriasis who did not receive one of these therapies during the study period.

Patients were excluded if they had a history of appendicitis, cholecystitis, or diverticulitis at any time before the index date. Only patients older than 18 years were included.

Patients with psoriasis were frequency matched (1:5) with healthy patients, also from the KPSC network. Individuals were matched by age, sex, and ethnicity.

Statistical Analysis
Baseline characteristics were described with means and SD for continuous variables as well as percentages for categorical variables. Chi-square tests for categorical variables and the Mann-Whitney U Test for continuous variables were used to compare the patients’ characteristics by psoriasis status. Cox proportional hazards regression models were used to examine the risk for appendicitis, cholecystitis, or diverticulitis among patients with and without psoriasis and among patients with mild and severe psoriasis. Proportionality assumption was validated using Pearson product moment correlation between the scaled Schoenfeld residuals and log transformed time for each covariate.

Results were presented as crude (unadjusted) hazard ratios (HRs) and adjusted HRs, where confounding factors (ie, age, sex, ethnicity, body mass index [BMI], alcohol use, smoking status, income, education, and membership length) were adjusted. All tests were performed with SAS EG 5.1 and R software. P<.05 was considered statistically significant. Results are reported with the 95% confidence interval (CI), when appropriate.

 

Results

A total of 1,690,214 KPSC patients were eligible for the study; 10,307 (0.6%) met diagnostic and inclusion criteria for the psoriasis cohort. Patients with psoriasis had a significantly higher mean BMI (29.9 vs 28.7; P<.0001) as well as higher mean rates of alcohol use (56% vs 53%; P<.0001) and smoking (47% vs 38%; P<.01) compared to controls. Psoriasis patients had a shorter average duration of membership within the Kaiser network (P=.0001) compared to controls.

A total of 7416 patients met criteria for mild psoriasis and 2891 patients met criteria for severe psoriasis (eTable). Patients with severe psoriasis were significantly younger and had significantly higher mean BMI compared to patients with mild psoriasis (P<.0001 and P=.0001, respectively). No significant difference in rates of alcohol or tobacco use was detected among patients with mild and severe psoriasis.

Appendicitis
The prevalence of appendicitis was not significantly different between patients with and without psoriasis or between patients with mild and severe psoriasis, though the incidence rate was slightly higher among patients with psoriasis (0.80 per 1000 patient-years compared to 0.62 per 1000 patient-years among patients without psoriasis)(Table 1). However, there was not a significant difference in risk for appendicitis between healthy patients, patients with severe psoriasis, and patients with mild psoriasis after adjusting for potential confounding factors (Table 2). Interestingly, patients with severe psoriasis who had a diagnosis of appendicitis had a significantly shorter time to diagnosis of appendicitis compared to patients with mild psoriasis (7.4 years vs 8.1 years; P<.0001).

Cholecystitis
Psoriasis patients also did not have an increased prevalence of cholecystitis compared to healthy patients. However, patients with severe psoriasis had a significantly higher prevalence of cholecystitis compared to patients with mild psoriasis (P=.0038). Overall, patients with psoriasis had a slightly higher incidence rate (1.72 per 1000 patient-years) compared to healthy patients (1.46 per 1000 patient-years). Moreover, the time to diagnosis of cholecystitis was significantly shorter for patients with severe psoriasis than for patients with mild psoriasis (7.4 years vs 8.1 years; P<.0001). Mild psoriasis was associated with a significantly increased risk (HR, 1.33; 95% CI, 1.09-1.63; P<.01) for cholecystitis compared to individuals without psoriasis in both the crude and adjusted models (Table 2). There was no difference between mild psoriasis patients and severe psoriasis patients in risk for cholecystitis.

Diverticulitis
Patients with psoriasis had a significantly greater prevalence of diverticulitis compared to the control cohort (5.1% vs 4.2%; P<.0001). There was no difference in prevalence between the severe psoriasis group and the mild psoriasis group (P=.96), but the time to diagnosis of diverticulitis was shorter in the severe psoriasis group than in the mild psoriasis group (7.2 years vs 7.9 years; P<.0001). Psoriasis patients had an incidence rate of diverticulitis of 6.61 per 1000 patient-years compared to 5.38 per 1000 patient-years in the control group. Psoriasis conferred a higher risk for diverticulitis in both the crude and adjusted models (HR, 1.23; 95% CI, 1.11-1.35 [P<.001] and HR, 1.16; 95% CI, 1.05-1.29; [P<.01], respectively)(Table 3); however, when stratified by disease severity, only patients with severe psoriasis were found to be at higher risk (HR, 1.26; 95% CI, 1.15-1.61; P<.001 for the adjusted model).

 

Comment

The objective of this study was to examine the background risks for specific gastrointestinal pathologies in a large cohort of patients with psoriasis compared to the general population. After adjusting for measured confounders, patients with severe psoriasis had a significantly higher risk of diverticulitis compared to the general population. Although more patients with severe psoriasis developed appendicitis or cholecystitis, the difference was not significant.

The pathogenesis of diverticulosis and diverticulitis has been thought to be related to increased intracolonic pressure and decreased dietary fiber intake, leading to formation of diverticula in the colon.¹⁹ Our study did not correct for differences in diet between the 2 groups, making it a possible confounding variable. Studies evaluating dietary habits of psoriatic patients have found that adult males with psoriasis might consume less fiber compared to healthy patients,²⁰ and psoriasis patients also might consume less whole-grain fiber.²¹ Furthermore, fiber deficiency also might affect gut flora, causing low-grade chronic inflammation,¹⁸ which also has been supported by response to anti-inflammatory medications such as mesalazine.²² Given the autoimmune association between psoriasis and IBD, it is possible that psoriasis also might create an environment of chronic inflammation in the gut, predisposing patients with psoriasis to diverticulitis. However, further research is needed to better evaluate this possibility.

Our study also does not address any potential effects on outcomes of specific treatments for psoriasis. Brandl et al²³ found that patients on immunosuppressive therapy for autoimmune diseases had longer hospital and intensive care unit stays, higher rates of emergency operations, and higher mortality while hospitalized. Because our results suggest that patients with severe psoriasis, who are therefore more likely to require treatment with an immunomodulator, are at higher risk for diverticulitis, these patients also might be at risk for poorer outcomes.

There is no literature evaluating the relationship between psoriasis and appendicitis. Our study found a slightly lower incidence rate compared to the national trend (9.38 per 10,000 patient-years in the United States in 2008) in both healthy patients and psoriasis patients.²⁴ Of note, this statistic includes children, whereas our study did not, which might in part account for the lower rate. However, Cheluvappa et al²⁵ hypothesized a relationship between appendicitis and subsequent appendectomy at a young age and protection against IBD. They also found that the mechanism for protection involves downregulation of the helper T cell (T_H17) pathway,²⁵ which also has been found to play a role in psoriasis pathogenesis.^26,27 Although our results suggest that the risk for appendicitis is not increased for patients with psoriasis, further research might be able to determine if appendicitis and subsequent appendectomy also can offer protection against development of psoriasis.

We found that patients with severe psoriasis had a higher incidence rate of cholecystitis compared to patients with mild psoriasis. Egeberg et al²⁸ found an increased risk for cholelithiasis among patients with psoriasis, which may contribute to a higher rate of cholecystitis. Although both acute and chronic cholecystitis were incorporated in this study, a Russian study found that chronic cholecystitis may be a predictor of progression of psoriasis.²⁹ Moreover, patients with severe psoriasis had a shorter duration to diagnosis of cholecystitis than patients with mild psoriasis. It is possible that patients with severe psoriasis are in a state of greater chronic inflammation than those with mild psoriasis, and therefore, when combined with other risk factors for cholecystitis, may progress to disease more quickly. Alternatively, this finding could be treatment related, as there have been reported cases of cholecystitis related to etanercept use in patients treated for psoriasis and juvenile polyarticular rheumatoid arthritis.^30,31 The relationship is not yet well defined, however, and further research is necessary to evaluate this association.

Study Strengths
Key strengths of this study include the large sample size and diversity of the patient population. Kaiser Permanente Southern California membership generally is representative of the broader community, making our results fairly generalizable to populations with health insurance. Use of a matched control cohort allows the results to be more specific to the disease of interest, and the population-based design minimizes bias.

Study Limitations
This study has several limitations. Although the cohorts were categorized based on type of treatment received, exact therapies were not specified. As a retrospective study, it is difficult to control for potential confounding variables that are not included in the electronic medical record. The results of this study also demonstrated significantly shorter durations to diagnosis of all 3 conditions, indicating that surveillance bias may be present.

Conclusion

Patients with psoriasis may be at an increased risk for diverticulitis compared to patients without psoriasis, which could be due to the chronic inflammatory state induced by psoriasis. Therefore, it may be beneficial for clinicians to evaluate psoriasis patients for other risk factors for diverticulitis and subsequently provide counseling to these patients to minimize their risk for diverticulitis. Psoriasis patients do not appear to be at an increased risk for appendicitis or cholecystitis compared to controls; however, further research is needed for confirmation.