Peter J. Kaboli, MD, MS

Nurse practitioners (NPs) and physician assistants (PAs) provide healthcare in numerous environments internationally and in the United States.[1, 2] However, their role in the inpatient medicine setting is not well described.[2] In the United States, there are more than 157,000 NPs and 85,000 PAs with projected increases.[3, 4] Although both professions provide direct medical care, there are key differences.[1, 3, 4, 5] NPs typically complete a master's or doctoral degree with advanced clinical training beyond nursing. PAs complete at least 2 years of college courses similar to premedical school requirements. PA programs use a medical school‐based curriculum and train for about 2 years before awarding a master's degree. NPs are regulated through state nursing boards, whereas PAs are regulated through state licensing or medical boards. NPs and PAs have different, yet overlapping scopes of practice. A key difference is that PAs can only practice collaborating with a physician.[5, 6] Overall, both have been shown to provide healthcare that is similar in quality to physicians in specific primary care and surgical settings.[2]

NPs and PAs, often referred to as advanced practice providers (APPs), are employed primarily in outpatient clinic settings providing direct patient care. Most APP studies have focused on the outpatient setting, despite nearly a third of US healthcare expenditure for hospital care.[2, 7] Little is known about APP involvement, specific roles, or impact on outcomes in inpatient medicine settings where they are often referred to as NP or PA hospitalists.[2, 8, 9, 10]

The Veterans Health Administration (VHA) is 1 of the largest employers of APPs, with 3.6% of all NPs and 2.1% of all PAs reported to practice in the VHA.[11, 12, 13] As the largest fully integrated healthcare system in the US, the VHA had 8.8 million veterans enrolled and 703,500 inpatient admissions in 2012.[14] Although this makes the VHA an ideal environment to study the role of APPs, few studies have done so.[13, 15, 16, 17, 18, 19] Although studies have compared NPs and PAs to physicians, very little is known about how NPs differ from PAs when practicing in the same environment.

Our objective was to describe the scope of practice, defined as activities that an individual healthcare practitioner is licensed to perform, of NPs and PAs in the inpatient medicine setting and in the VHA. A secondary objective was to explore important outcomes that could potentially be affected by the presence of NPs and PAs on inpatient medicine.

METHODS

The Organizational Factors and Inpatient Medical Care Quality and Efficiency (OFIM) study provides a basis for this study with detail published elsewhere.[20] The OFIM study was conducted between 2010 and 2011 to evaluate quality of care in VHA inpatient medicine surveying chiefs of medicine (COM), inpatient medicine nurse managers (NM), attending physicians, and extant VHA survey data. The COM is the senior attending physician in charge of departments of medicine that include most medical subspecialties within the VHA medical centers. We used the subset of questions specific to NPs and PAs from the COM and NM surveys. Both COMs and NMs answered identical questions for NPs and PAs in 2 separate sections to avoid overlap of responses. NM survey responses were only used for the coordination of care regression model. Surveys were conducted by e‐mail with up to 4 reminders and a subsequent paper mailing. The inpatient medicine service included adult general internal medicine, medical subspecialties, and critical care. The study was approved by the institutional review boards of the VA Boston Healthcare System, the University of Iowa, and the Iowa City VA Healthcare System.

RESULTS

Of 123 inpatient medicine services that we surveyed, we included responses from the COMs of 118 services (response rate 95.2%); 5 responses were incomplete. Across 123 inpatient medicine services, we surveyed 264 nurse managers and received 198 responses (75.0%) from 114 inpatient medicine services. In the only model using NM responsesthe care coordination model104 inpatient medicine services had responses from both COM and NM surveys.

Of 118 VHA inpatient medicine services, 56 (47.5%) had APPs, of which 27 (48.2%) had NPs only, 15 (26.8%) had PAs only, and 14 (25.0%) had both NPs and PAs. FTEEs for NPs ranged from 0.5 to 7 (mean=2.22) and for PAs from 1 to 9 (mean=2.23) on the inpatient medicine service per hospital.

There were no significant differences on use of NPs and PAs by teaching affiliation, urban or rural setting, and geography. A significant difference was observed based on bed size (F[3,109]=5.13, P<0.001); facilities with both NPs and PAs had, on average, a larger number of inpatient beds (mean=79.0, standard deviation [SD]=32.3) compared to those without NPs or PAs (mean=50.1, SD=29.4) or with PAs only (mean=44.2, SD=20.5) using Tukey post hoc analysis.

The most common staffing model used staff (attending) physicians only working directly with APPs (N=29, 24.6%). Next most common was an academic model with staff physicians, housestaff, and APPs working together in teams (N=16, 13.4%). For performance evaluations, COMs contributed for both NPs (60.2%) and PAs (56.4%); in fewer cases, COMs completed evaluations of NPs (12.9%) and of PAs (29.0%) without input from other service managers (P=0.02).

Table 1 shows the differences reported by COMs between NPs and PAs scope of practice. Overall, 58.9% of NPs and 65.4% of PAs functioned somewhat or completely autonomously; 23.1% of NPs and 30.8% of PAs worked in a role closer to a ward assistant (eg, work directly with a physician, cowriting orders, and making care decisions with physician oversight). Tasks frequently performed by the majority of NPs and PAs included writing orders (87.9%), coordinating discharge plans (86.7%), communicating with consultants (83.1%), performing history and physicals (82.5%), writing daily progress notes (80.7%), communicating with primary care providers (73.5%), and working directly with hospitalists (72.8%). Less common tasks included serving on committees (46.4%), championing quality improvement activities (40.6%), and research (2.9%). There were no statistically significant differences between tasks, except for a higher proportion of services reporting PAs rather than NPs performing procedures (50.0% vs 22.0%, P=0.02) and teaching nonphysicians (50.0% vs 24.4%, P=0.04).

Table 2 reports location of practice in the hospital and workload. There were no significant differences in locations where NPs and PAs provided care. Overall, 81.9% of APPs worked in inpatient wards, 23.1% in step‐down units, 18.6% in intensive care units, 13.8% in skilled care units, and 4.9% in other locations. In addition, 97.4% of NPs and 89.3% of PAs worked weekdays, whereas only 7.9% of NPs and 17.9% of PAs worked nights. More PAs than NPs worked federal holidays (32.1% vs 7.9%, P=0.02) and weekends (32.1% vs 13.2%, P=0.08). Most NPs and PAs handled a caseload of 4 to 10 patients with a mean of 6.5, with no difference between the 2. The minority, 27.0% of NPs and 23.1% of PAs, were not assigned specific patients.

In multivariable adjusted analyses evaluating the association between patient satisfaction and use of APPs (Table 3), no significant differences were observed for patients' rating of the hospital (F[3,95]=0.19; P=0.90) or willingness to recommend the hospital (F[3,95]=0.54; P=0.65). Similarly, no significant differences were observed based on use of APPs for nurse overall job satisfaction (F[3,101]=1.85; P=0.14) or collegial relations with physicians (F[3,101]=0.96; P=0.41).

COM ratings of overall inpatient coordination were also nonsignificant (F[3, 100]=2.01; P=0.12), but their ratings of coordination were higher in facilities with NPs only than in those without either NPs or PAs (=1.63, P=0.08). Nurse manager ratings of overall inpatient coordination were not associated with APP use (F[3,91]=1.24; P=0.30), but were marginally lower with facilities using only PAs (=1.48; P=0.06). Nurse manager ratings of discharge coordination showed a significant effect for APP use (F[3,90]=3.30; P=0.02) with facilities having NPs only significantly higher than places without either NPs or PAs (=1.84, P=0.04).

DISCUSSION

Little evidence exists regarding the role of APPs in the inpatient medicine setting,[2] and important deficit concerns in medical knowledge, technical skills, and clinical experience have been raised.[27, 28] These concerns have called into question the appropriateness of involving APPs in the care of medical inpatients with extensive differential diagnoses and complex care requirements.[27, 28] In spite of these concerns, we found widespread use of APPs with almost half of the VHA inpatient medicine services reporting use, which stands in contrast to prior research.[9, 10, 22, 29, 30, 31, 32, 33, 34, 35] APPs practice in a variety of acute and subacute inpatient medicine settings including academic, community, rural, and urban settings without many discernable differences. The spectrum of activities performed by APPs in the VHA is similar to those reported in these inpatient medicine studies, although their scope of practice appears to be much broader than in these few small single academic center studies.[10, 22, 29, 30, 31, 32, 33, 34, 35, 36] For example, only 11% of hospitalist PAs did procedures in a 2006 Society of Hospital Medicine survey, whereas 50% did in our study.[36]

Interestingly, we found that VHA NPs and PAs perform very similar tasks with similar caseloads despite differences in their background, training, regulation, reimbursement, and the longstanding observation that nurse practitioners are not physician assistants.[1, 3, 4, 5] These findings may reflect that APP scope can be more extensive in the VHA. For example, PAs in the VHA practice under federal jurisdiction and can bypass state legislation of scope of practice.[13] It also may reflect ongoing expansion of the role of APPs in the healthcare system since prior studies.[33, 36]

We did, however, note a few significant differences in NP and PA scope. PAs are twice as likely to perform procedures as NPs in inpatient medicine. It is unclear why PAs may do more procedures, as acute care NPs also are commonly taught and perform similar procedures.[33] We also found that PAs teach nonphysician students twice as often as NPs. This may reflect the deep commitment shown by the VHA to PA education dating back to the 1960s.[13] Finally, we found that PAs were significantly more likely to work weekends and federal holidays, a finding that may have implications for inpatient medicine services hiring APPs. Although not statistically significant, PAs, in general, performed more clinically oriented tasks like history and physicals and more often worked directly with hospitalists.

We found no difference in patient satisfaction or nurse satisfaction related to the presence of APPs, consistent with prior studies, where higher levels of satisfaction with APPs are observed in primary care but not hospital settings.[2, 10] However, it is surprising that no differences were observed for nurse satisfaction. NPs traditionally have a nursing focus, which might foster better relationships with nurses.[22] Expecting changes in either patient or nurse satisfaction with just the addition of APPs in the inpatient medicine setting without addressing other factors may be unrealistic. Patient satisfaction is a complex amalgam of various factors including patient expectations, sociodemographics, emotional and physical state, quality of care, and physician communication.[24] Similarly, nurse satisfaction depends on many factors including job stress, nursephysician collaboration, autonomy, staffing, and support.[37]

Finally, we found higher perception of both overall coordination of inpatient care and discharge coordination on services with NPs. A primary reason stated by medical centers to hire APPs is to improve continuity of care.[9] Prior research has shown better communication and collaboration between nurses, physicians, and NPs on inpatient medicine services.[21] NPs may feel that coordination of care is a major focus for their profession and may spend more time than physicians on care coordination activities.[38] Moreover, their background in both nursing and medicine may better lend itself to coordinating care between disciplines.[39] However, we were surprised to find that services with PAs had lower ratings of overall coordination by nurse managers given that care coordination also is a core competency of PA practice and a primary reason for medical centers to employ them.[9] The lack of a nursing background for PAs and potentially less overall medical experience than NPs possibly may contribute to this finding. However, our study does not suggest a direct explanation for this finding, and we had no measure of prior clinical experience, and thus it should be an area for further research.

There are a number of limitations to our study. First, findings from the VHA may not be generalizable to other healthcare systems.[39] However, VHA inpatient medicine services are, in general, structured similarly to non‐VHA settings and are often affiliated with academic medical centers. Further, this is the largest study to our knowledge to look at the specific roles and perceptions of care provided by both NPs and PAs in inpatient medicine. Second, we did not measure other outcomes of care that may be affected by the use of APPs, such as clinical outcomes, process of care measures, or cost‐effectiveness, some of which have been shown in small studies to be impacted by APPs in inpatient medicine.[10, 22, 29, 30, 31, 32, 33, 34, 35] Third, we are unable to attribute causality to our findings and may not have accounted for all the differences between services. Ideally, a randomized controlled trial of APPs in inpatient medicine would be helpful to address these concerns, but no such trials have been conducted. Finally, we did not survey APPs directly, but surveyed the chiefs of their service instead. The chiefs, however, are directly responsible for the scope of practice of all providers on their service and were directly involved in performance evaluations of most of these practitioners.

In conclusion, we found that NPs and PAs, functioning as APP hospitalists are more widely used and have a broader scope of practice on inpatient medicine than previously known or appreciated, at least in the VHA. In spite of their different backgrounds, training, regulations, and reimbursements, they appear to have a similar scope of practice with few differences in roles or perceived impact. Their impact on inpatient healthcare should be a subject of future research. In the meantime, inpatient medicine services should factor these findings into their decision making as they rapidly expand the use of APPs to provide better care to their patients and to address challenges in healthcare reform.[3, 27, 28, 40]

Nurse practitioners (NPs) and physician assistants (PAs) provide healthcare in numerous environments internationally and in the United States.[1, 2] However, their role in the inpatient medicine setting is not well described.[2] In the United States, there are more than 157,000 NPs and 85,000 PAs with projected increases.[3, 4] Although both professions provide direct medical care, there are key differences.[1, 3, 4, 5] NPs typically complete a master's or doctoral degree with advanced clinical training beyond nursing. PAs complete at least 2 years of college courses similar to premedical school requirements. PA programs use a medical school‐based curriculum and train for about 2 years before awarding a master's degree. NPs are regulated through state nursing boards, whereas PAs are regulated through state licensing or medical boards. NPs and PAs have different, yet overlapping scopes of practice. A key difference is that PAs can only practice collaborating with a physician.[5, 6] Overall, both have been shown to provide healthcare that is similar in quality to physicians in specific primary care and surgical settings.[2]

NPs and PAs, often referred to as advanced practice providers (APPs), are employed primarily in outpatient clinic settings providing direct patient care. Most APP studies have focused on the outpatient setting, despite nearly a third of US healthcare expenditure for hospital care.[2, 7] Little is known about APP involvement, specific roles, or impact on outcomes in inpatient medicine settings where they are often referred to as NP or PA hospitalists.[2, 8, 9, 10]

The Veterans Health Administration (VHA) is 1 of the largest employers of APPs, with 3.6% of all NPs and 2.1% of all PAs reported to practice in the VHA.[11, 12, 13] As the largest fully integrated healthcare system in the US, the VHA had 8.8 million veterans enrolled and 703,500 inpatient admissions in 2012.[14] Although this makes the VHA an ideal environment to study the role of APPs, few studies have done so.[13, 15, 16, 17, 18, 19] Although studies have compared NPs and PAs to physicians, very little is known about how NPs differ from PAs when practicing in the same environment.

Our objective was to describe the scope of practice, defined as activities that an individual healthcare practitioner is licensed to perform, of NPs and PAs in the inpatient medicine setting and in the VHA. A secondary objective was to explore important outcomes that could potentially be affected by the presence of NPs and PAs on inpatient medicine.

METHODS

The Organizational Factors and Inpatient Medical Care Quality and Efficiency (OFIM) study provides a basis for this study with detail published elsewhere.[20] The OFIM study was conducted between 2010 and 2011 to evaluate quality of care in VHA inpatient medicine surveying chiefs of medicine (COM), inpatient medicine nurse managers (NM), attending physicians, and extant VHA survey data. The COM is the senior attending physician in charge of departments of medicine that include most medical subspecialties within the VHA medical centers. We used the subset of questions specific to NPs and PAs from the COM and NM surveys. Both COMs and NMs answered identical questions for NPs and PAs in 2 separate sections to avoid overlap of responses. NM survey responses were only used for the coordination of care regression model. Surveys were conducted by e‐mail with up to 4 reminders and a subsequent paper mailing. The inpatient medicine service included adult general internal medicine, medical subspecialties, and critical care. The study was approved by the institutional review boards of the VA Boston Healthcare System, the University of Iowa, and the Iowa City VA Healthcare System.

RESULTS

Of 123 inpatient medicine services that we surveyed, we included responses from the COMs of 118 services (response rate 95.2%); 5 responses were incomplete. Across 123 inpatient medicine services, we surveyed 264 nurse managers and received 198 responses (75.0%) from 114 inpatient medicine services. In the only model using NM responsesthe care coordination model104 inpatient medicine services had responses from both COM and NM surveys.

Of 118 VHA inpatient medicine services, 56 (47.5%) had APPs, of which 27 (48.2%) had NPs only, 15 (26.8%) had PAs only, and 14 (25.0%) had both NPs and PAs. FTEEs for NPs ranged from 0.5 to 7 (mean=2.22) and for PAs from 1 to 9 (mean=2.23) on the inpatient medicine service per hospital.

There were no significant differences on use of NPs and PAs by teaching affiliation, urban or rural setting, and geography. A significant difference was observed based on bed size (F[3,109]=5.13, P<0.001); facilities with both NPs and PAs had, on average, a larger number of inpatient beds (mean=79.0, standard deviation [SD]=32.3) compared to those without NPs or PAs (mean=50.1, SD=29.4) or with PAs only (mean=44.2, SD=20.5) using Tukey post hoc analysis.

The most common staffing model used staff (attending) physicians only working directly with APPs (N=29, 24.6%). Next most common was an academic model with staff physicians, housestaff, and APPs working together in teams (N=16, 13.4%). For performance evaluations, COMs contributed for both NPs (60.2%) and PAs (56.4%); in fewer cases, COMs completed evaluations of NPs (12.9%) and of PAs (29.0%) without input from other service managers (P=0.02).

Table 1 shows the differences reported by COMs between NPs and PAs scope of practice. Overall, 58.9% of NPs and 65.4% of PAs functioned somewhat or completely autonomously; 23.1% of NPs and 30.8% of PAs worked in a role closer to a ward assistant (eg, work directly with a physician, cowriting orders, and making care decisions with physician oversight). Tasks frequently performed by the majority of NPs and PAs included writing orders (87.9%), coordinating discharge plans (86.7%), communicating with consultants (83.1%), performing history and physicals (82.5%), writing daily progress notes (80.7%), communicating with primary care providers (73.5%), and working directly with hospitalists (72.8%). Less common tasks included serving on committees (46.4%), championing quality improvement activities (40.6%), and research (2.9%). There were no statistically significant differences between tasks, except for a higher proportion of services reporting PAs rather than NPs performing procedures (50.0% vs 22.0%, P=0.02) and teaching nonphysicians (50.0% vs 24.4%, P=0.04).

Table 2 reports location of practice in the hospital and workload. There were no significant differences in locations where NPs and PAs provided care. Overall, 81.9% of APPs worked in inpatient wards, 23.1% in step‐down units, 18.6% in intensive care units, 13.8% in skilled care units, and 4.9% in other locations. In addition, 97.4% of NPs and 89.3% of PAs worked weekdays, whereas only 7.9% of NPs and 17.9% of PAs worked nights. More PAs than NPs worked federal holidays (32.1% vs 7.9%, P=0.02) and weekends (32.1% vs 13.2%, P=0.08). Most NPs and PAs handled a caseload of 4 to 10 patients with a mean of 6.5, with no difference between the 2. The minority, 27.0% of NPs and 23.1% of PAs, were not assigned specific patients.

In multivariable adjusted analyses evaluating the association between patient satisfaction and use of APPs (Table 3), no significant differences were observed for patients' rating of the hospital (F[3,95]=0.19; P=0.90) or willingness to recommend the hospital (F[3,95]=0.54; P=0.65). Similarly, no significant differences were observed based on use of APPs for nurse overall job satisfaction (F[3,101]=1.85; P=0.14) or collegial relations with physicians (F[3,101]=0.96; P=0.41).

COM ratings of overall inpatient coordination were also nonsignificant (F[3, 100]=2.01; P=0.12), but their ratings of coordination were higher in facilities with NPs only than in those without either NPs or PAs (=1.63, P=0.08). Nurse manager ratings of overall inpatient coordination were not associated with APP use (F[3,91]=1.24; P=0.30), but were marginally lower with facilities using only PAs (=1.48; P=0.06). Nurse manager ratings of discharge coordination showed a significant effect for APP use (F[3,90]=3.30; P=0.02) with facilities having NPs only significantly higher than places without either NPs or PAs (=1.84, P=0.04).

DISCUSSION

Little evidence exists regarding the role of APPs in the inpatient medicine setting,[2] and important deficit concerns in medical knowledge, technical skills, and clinical experience have been raised.[27, 28] These concerns have called into question the appropriateness of involving APPs in the care of medical inpatients with extensive differential diagnoses and complex care requirements.[27, 28] In spite of these concerns, we found widespread use of APPs with almost half of the VHA inpatient medicine services reporting use, which stands in contrast to prior research.[9, 10, 22, 29, 30, 31, 32, 33, 34, 35] APPs practice in a variety of acute and subacute inpatient medicine settings including academic, community, rural, and urban settings without many discernable differences. The spectrum of activities performed by APPs in the VHA is similar to those reported in these inpatient medicine studies, although their scope of practice appears to be much broader than in these few small single academic center studies.[10, 22, 29, 30, 31, 32, 33, 34, 35, 36] For example, only 11% of hospitalist PAs did procedures in a 2006 Society of Hospital Medicine survey, whereas 50% did in our study.[36]

Interestingly, we found that VHA NPs and PAs perform very similar tasks with similar caseloads despite differences in their background, training, regulation, reimbursement, and the longstanding observation that nurse practitioners are not physician assistants.[1, 3, 4, 5] These findings may reflect that APP scope can be more extensive in the VHA. For example, PAs in the VHA practice under federal jurisdiction and can bypass state legislation of scope of practice.[13] It also may reflect ongoing expansion of the role of APPs in the healthcare system since prior studies.[33, 36]

We did, however, note a few significant differences in NP and PA scope. PAs are twice as likely to perform procedures as NPs in inpatient medicine. It is unclear why PAs may do more procedures, as acute care NPs also are commonly taught and perform similar procedures.[33] We also found that PAs teach nonphysician students twice as often as NPs. This may reflect the deep commitment shown by the VHA to PA education dating back to the 1960s.[13] Finally, we found that PAs were significantly more likely to work weekends and federal holidays, a finding that may have implications for inpatient medicine services hiring APPs. Although not statistically significant, PAs, in general, performed more clinically oriented tasks like history and physicals and more often worked directly with hospitalists.

We found no difference in patient satisfaction or nurse satisfaction related to the presence of APPs, consistent with prior studies, where higher levels of satisfaction with APPs are observed in primary care but not hospital settings.[2, 10] However, it is surprising that no differences were observed for nurse satisfaction. NPs traditionally have a nursing focus, which might foster better relationships with nurses.[22] Expecting changes in either patient or nurse satisfaction with just the addition of APPs in the inpatient medicine setting without addressing other factors may be unrealistic. Patient satisfaction is a complex amalgam of various factors including patient expectations, sociodemographics, emotional and physical state, quality of care, and physician communication.[24] Similarly, nurse satisfaction depends on many factors including job stress, nursephysician collaboration, autonomy, staffing, and support.[37]

Finally, we found higher perception of both overall coordination of inpatient care and discharge coordination on services with NPs. A primary reason stated by medical centers to hire APPs is to improve continuity of care.[9] Prior research has shown better communication and collaboration between nurses, physicians, and NPs on inpatient medicine services.[21] NPs may feel that coordination of care is a major focus for their profession and may spend more time than physicians on care coordination activities.[38] Moreover, their background in both nursing and medicine may better lend itself to coordinating care between disciplines.[39] However, we were surprised to find that services with PAs had lower ratings of overall coordination by nurse managers given that care coordination also is a core competency of PA practice and a primary reason for medical centers to employ them.[9] The lack of a nursing background for PAs and potentially less overall medical experience than NPs possibly may contribute to this finding. However, our study does not suggest a direct explanation for this finding, and we had no measure of prior clinical experience, and thus it should be an area for further research.

There are a number of limitations to our study. First, findings from the VHA may not be generalizable to other healthcare systems.[39] However, VHA inpatient medicine services are, in general, structured similarly to non‐VHA settings and are often affiliated with academic medical centers. Further, this is the largest study to our knowledge to look at the specific roles and perceptions of care provided by both NPs and PAs in inpatient medicine. Second, we did not measure other outcomes of care that may be affected by the use of APPs, such as clinical outcomes, process of care measures, or cost‐effectiveness, some of which have been shown in small studies to be impacted by APPs in inpatient medicine.[10, 22, 29, 30, 31, 32, 33, 34, 35] Third, we are unable to attribute causality to our findings and may not have accounted for all the differences between services. Ideally, a randomized controlled trial of APPs in inpatient medicine would be helpful to address these concerns, but no such trials have been conducted. Finally, we did not survey APPs directly, but surveyed the chiefs of their service instead. The chiefs, however, are directly responsible for the scope of practice of all providers on their service and were directly involved in performance evaluations of most of these practitioners.

In conclusion, we found that NPs and PAs, functioning as APP hospitalists are more widely used and have a broader scope of practice on inpatient medicine than previously known or appreciated, at least in the VHA. In spite of their different backgrounds, training, regulations, and reimbursements, they appear to have a similar scope of practice with few differences in roles or perceived impact. Their impact on inpatient healthcare should be a subject of future research. In the meantime, inpatient medicine services should factor these findings into their decision making as they rapidly expand the use of APPs to provide better care to their patients and to address challenges in healthcare reform.[3, 27, 28, 40]

Nurse practitioners (NPs) and physician assistants (PAs) provide healthcare in numerous environments internationally and in the United States.[1, 2] However, their role in the inpatient medicine setting is not well described.[2] In the United States, there are more than 157,000 NPs and 85,000 PAs with projected increases.[3, 4] Although both professions provide direct medical care, there are key differences.[1, 3, 4, 5] NPs typically complete a master's or doctoral degree with advanced clinical training beyond nursing. PAs complete at least 2 years of college courses similar to premedical school requirements. PA programs use a medical school‐based curriculum and train for about 2 years before awarding a master's degree. NPs are regulated through state nursing boards, whereas PAs are regulated through state licensing or medical boards. NPs and PAs have different, yet overlapping scopes of practice. A key difference is that PAs can only practice collaborating with a physician.[5, 6] Overall, both have been shown to provide healthcare that is similar in quality to physicians in specific primary care and surgical settings.[2]

NPs and PAs, often referred to as advanced practice providers (APPs), are employed primarily in outpatient clinic settings providing direct patient care. Most APP studies have focused on the outpatient setting, despite nearly a third of US healthcare expenditure for hospital care.[2, 7] Little is known about APP involvement, specific roles, or impact on outcomes in inpatient medicine settings where they are often referred to as NP or PA hospitalists.[2, 8, 9, 10]

The Veterans Health Administration (VHA) is 1 of the largest employers of APPs, with 3.6% of all NPs and 2.1% of all PAs reported to practice in the VHA.[11, 12, 13] As the largest fully integrated healthcare system in the US, the VHA had 8.8 million veterans enrolled and 703,500 inpatient admissions in 2012.[14] Although this makes the VHA an ideal environment to study the role of APPs, few studies have done so.[13, 15, 16, 17, 18, 19] Although studies have compared NPs and PAs to physicians, very little is known about how NPs differ from PAs when practicing in the same environment.

Our objective was to describe the scope of practice, defined as activities that an individual healthcare practitioner is licensed to perform, of NPs and PAs in the inpatient medicine setting and in the VHA. A secondary objective was to explore important outcomes that could potentially be affected by the presence of NPs and PAs on inpatient medicine.

METHODS

The Organizational Factors and Inpatient Medical Care Quality and Efficiency (OFIM) study provides a basis for this study with detail published elsewhere.[20] The OFIM study was conducted between 2010 and 2011 to evaluate quality of care in VHA inpatient medicine surveying chiefs of medicine (COM), inpatient medicine nurse managers (NM), attending physicians, and extant VHA survey data. The COM is the senior attending physician in charge of departments of medicine that include most medical subspecialties within the VHA medical centers. We used the subset of questions specific to NPs and PAs from the COM and NM surveys. Both COMs and NMs answered identical questions for NPs and PAs in 2 separate sections to avoid overlap of responses. NM survey responses were only used for the coordination of care regression model. Surveys were conducted by e‐mail with up to 4 reminders and a subsequent paper mailing. The inpatient medicine service included adult general internal medicine, medical subspecialties, and critical care. The study was approved by the institutional review boards of the VA Boston Healthcare System, the University of Iowa, and the Iowa City VA Healthcare System.

RESULTS

Of 123 inpatient medicine services that we surveyed, we included responses from the COMs of 118 services (response rate 95.2%); 5 responses were incomplete. Across 123 inpatient medicine services, we surveyed 264 nurse managers and received 198 responses (75.0%) from 114 inpatient medicine services. In the only model using NM responsesthe care coordination model104 inpatient medicine services had responses from both COM and NM surveys.

Of 118 VHA inpatient medicine services, 56 (47.5%) had APPs, of which 27 (48.2%) had NPs only, 15 (26.8%) had PAs only, and 14 (25.0%) had both NPs and PAs. FTEEs for NPs ranged from 0.5 to 7 (mean=2.22) and for PAs from 1 to 9 (mean=2.23) on the inpatient medicine service per hospital.

There were no significant differences on use of NPs and PAs by teaching affiliation, urban or rural setting, and geography. A significant difference was observed based on bed size (F[3,109]=5.13, P<0.001); facilities with both NPs and PAs had, on average, a larger number of inpatient beds (mean=79.0, standard deviation [SD]=32.3) compared to those without NPs or PAs (mean=50.1, SD=29.4) or with PAs only (mean=44.2, SD=20.5) using Tukey post hoc analysis.

The most common staffing model used staff (attending) physicians only working directly with APPs (N=29, 24.6%). Next most common was an academic model with staff physicians, housestaff, and APPs working together in teams (N=16, 13.4%). For performance evaluations, COMs contributed for both NPs (60.2%) and PAs (56.4%); in fewer cases, COMs completed evaluations of NPs (12.9%) and of PAs (29.0%) without input from other service managers (P=0.02).

Table 1 shows the differences reported by COMs between NPs and PAs scope of practice. Overall, 58.9% of NPs and 65.4% of PAs functioned somewhat or completely autonomously; 23.1% of NPs and 30.8% of PAs worked in a role closer to a ward assistant (eg, work directly with a physician, cowriting orders, and making care decisions with physician oversight). Tasks frequently performed by the majority of NPs and PAs included writing orders (87.9%), coordinating discharge plans (86.7%), communicating with consultants (83.1%), performing history and physicals (82.5%), writing daily progress notes (80.7%), communicating with primary care providers (73.5%), and working directly with hospitalists (72.8%). Less common tasks included serving on committees (46.4%), championing quality improvement activities (40.6%), and research (2.9%). There were no statistically significant differences between tasks, except for a higher proportion of services reporting PAs rather than NPs performing procedures (50.0% vs 22.0%, P=0.02) and teaching nonphysicians (50.0% vs 24.4%, P=0.04).

Table 2 reports location of practice in the hospital and workload. There were no significant differences in locations where NPs and PAs provided care. Overall, 81.9% of APPs worked in inpatient wards, 23.1% in step‐down units, 18.6% in intensive care units, 13.8% in skilled care units, and 4.9% in other locations. In addition, 97.4% of NPs and 89.3% of PAs worked weekdays, whereas only 7.9% of NPs and 17.9% of PAs worked nights. More PAs than NPs worked federal holidays (32.1% vs 7.9%, P=0.02) and weekends (32.1% vs 13.2%, P=0.08). Most NPs and PAs handled a caseload of 4 to 10 patients with a mean of 6.5, with no difference between the 2. The minority, 27.0% of NPs and 23.1% of PAs, were not assigned specific patients.

In multivariable adjusted analyses evaluating the association between patient satisfaction and use of APPs (Table 3), no significant differences were observed for patients' rating of the hospital (F[3,95]=0.19; P=0.90) or willingness to recommend the hospital (F[3,95]=0.54; P=0.65). Similarly, no significant differences were observed based on use of APPs for nurse overall job satisfaction (F[3,101]=1.85; P=0.14) or collegial relations with physicians (F[3,101]=0.96; P=0.41).

COM ratings of overall inpatient coordination were also nonsignificant (F[3, 100]=2.01; P=0.12), but their ratings of coordination were higher in facilities with NPs only than in those without either NPs or PAs (=1.63, P=0.08). Nurse manager ratings of overall inpatient coordination were not associated with APP use (F[3,91]=1.24; P=0.30), but were marginally lower with facilities using only PAs (=1.48; P=0.06). Nurse manager ratings of discharge coordination showed a significant effect for APP use (F[3,90]=3.30; P=0.02) with facilities having NPs only significantly higher than places without either NPs or PAs (=1.84, P=0.04).

DISCUSSION

Little evidence exists regarding the role of APPs in the inpatient medicine setting,[2] and important deficit concerns in medical knowledge, technical skills, and clinical experience have been raised.[27, 28] These concerns have called into question the appropriateness of involving APPs in the care of medical inpatients with extensive differential diagnoses and complex care requirements.[27, 28] In spite of these concerns, we found widespread use of APPs with almost half of the VHA inpatient medicine services reporting use, which stands in contrast to prior research.[9, 10, 22, 29, 30, 31, 32, 33, 34, 35] APPs practice in a variety of acute and subacute inpatient medicine settings including academic, community, rural, and urban settings without many discernable differences. The spectrum of activities performed by APPs in the VHA is similar to those reported in these inpatient medicine studies, although their scope of practice appears to be much broader than in these few small single academic center studies.[10, 22, 29, 30, 31, 32, 33, 34, 35, 36] For example, only 11% of hospitalist PAs did procedures in a 2006 Society of Hospital Medicine survey, whereas 50% did in our study.[36]

Interestingly, we found that VHA NPs and PAs perform very similar tasks with similar caseloads despite differences in their background, training, regulation, reimbursement, and the longstanding observation that nurse practitioners are not physician assistants.[1, 3, 4, 5] These findings may reflect that APP scope can be more extensive in the VHA. For example, PAs in the VHA practice under federal jurisdiction and can bypass state legislation of scope of practice.[13] It also may reflect ongoing expansion of the role of APPs in the healthcare system since prior studies.[33, 36]

We did, however, note a few significant differences in NP and PA scope. PAs are twice as likely to perform procedures as NPs in inpatient medicine. It is unclear why PAs may do more procedures, as acute care NPs also are commonly taught and perform similar procedures.[33] We also found that PAs teach nonphysician students twice as often as NPs. This may reflect the deep commitment shown by the VHA to PA education dating back to the 1960s.[13] Finally, we found that PAs were significantly more likely to work weekends and federal holidays, a finding that may have implications for inpatient medicine services hiring APPs. Although not statistically significant, PAs, in general, performed more clinically oriented tasks like history and physicals and more often worked directly with hospitalists.

We found no difference in patient satisfaction or nurse satisfaction related to the presence of APPs, consistent with prior studies, where higher levels of satisfaction with APPs are observed in primary care but not hospital settings.[2, 10] However, it is surprising that no differences were observed for nurse satisfaction. NPs traditionally have a nursing focus, which might foster better relationships with nurses.[22] Expecting changes in either patient or nurse satisfaction with just the addition of APPs in the inpatient medicine setting without addressing other factors may be unrealistic. Patient satisfaction is a complex amalgam of various factors including patient expectations, sociodemographics, emotional and physical state, quality of care, and physician communication.[24] Similarly, nurse satisfaction depends on many factors including job stress, nursephysician collaboration, autonomy, staffing, and support.[37]

Finally, we found higher perception of both overall coordination of inpatient care and discharge coordination on services with NPs. A primary reason stated by medical centers to hire APPs is to improve continuity of care.[9] Prior research has shown better communication and collaboration between nurses, physicians, and NPs on inpatient medicine services.[21] NPs may feel that coordination of care is a major focus for their profession and may spend more time than physicians on care coordination activities.[38] Moreover, their background in both nursing and medicine may better lend itself to coordinating care between disciplines.[39] However, we were surprised to find that services with PAs had lower ratings of overall coordination by nurse managers given that care coordination also is a core competency of PA practice and a primary reason for medical centers to employ them.[9] The lack of a nursing background for PAs and potentially less overall medical experience than NPs possibly may contribute to this finding. However, our study does not suggest a direct explanation for this finding, and we had no measure of prior clinical experience, and thus it should be an area for further research.

There are a number of limitations to our study. First, findings from the VHA may not be generalizable to other healthcare systems.[39] However, VHA inpatient medicine services are, in general, structured similarly to non‐VHA settings and are often affiliated with academic medical centers. Further, this is the largest study to our knowledge to look at the specific roles and perceptions of care provided by both NPs and PAs in inpatient medicine. Second, we did not measure other outcomes of care that may be affected by the use of APPs, such as clinical outcomes, process of care measures, or cost‐effectiveness, some of which have been shown in small studies to be impacted by APPs in inpatient medicine.[10, 22, 29, 30, 31, 32, 33, 34, 35] Third, we are unable to attribute causality to our findings and may not have accounted for all the differences between services. Ideally, a randomized controlled trial of APPs in inpatient medicine would be helpful to address these concerns, but no such trials have been conducted. Finally, we did not survey APPs directly, but surveyed the chiefs of their service instead. The chiefs, however, are directly responsible for the scope of practice of all providers on their service and were directly involved in performance evaluations of most of these practitioners.

In conclusion, we found that NPs and PAs, functioning as APP hospitalists are more widely used and have a broader scope of practice on inpatient medicine than previously known or appreciated, at least in the VHA. In spite of their different backgrounds, training, regulations, and reimbursements, they appear to have a similar scope of practice with few differences in roles or perceived impact. Their impact on inpatient healthcare should be a subject of future research. In the meantime, inpatient medicine services should factor these findings into their decision making as they rapidly expand the use of APPs to provide better care to their patients and to address challenges in healthcare reform.[3, 27, 28, 40]

Recent trends show a marked increase in outpatient use of chronic opioid therapy (COT) for chronic noncancer pain (CNCP)[1, 2] without decreases in reported CNCP,[3] raising concerns about the efficacy and risk‐to‐benefit ratio of opioids in this population.[4, 5, 6, 7, 8] Increasing rates of outpatient use likely are accompanied by increasing rates of opioid exposure among patients admitted to the hospital. To our knowledge there are no published data regarding the prevalence of COT during the months preceding hospitalization.

Opioid use has been linked to increased emergency room utilization[9, 10] and emergency hospitalization,[11] but associations between opioid use and inpatient metrics (eg, mortality, readmission) have not been explored. Furthermore, lack of knowledge about the prevalence of opioid use prior to hospitalization may impede efforts to improve inpatient pain management and satisfaction with care. Although there is reason to expect that strategies to safely and effectively treat acute pain during the inpatient stay differ between opioid‐nave patients and opioid‐exposed patients, evidence regarding treatment strategies is limited.[12, 13, 14] Opioid pain medications are associated with hospital adverse events, with both prior opioid exposure and lack of opioid use as proposed risk factors.[15] A better understanding of the prevalence and characteristics of hospitalized COT patients is fundamental to future work to achieve safer and more effective inpatient pain management.

The primary purpose of this study was to determine the prevalence of prior COT among hospitalized medical patients. Additionally, we aimed to characterize inpatients with occasional and chronic opioid therapy prior to admission in comparison to opioid‐nave inpatients, as differences between these groups may suggest directions for further investigation into the distinct needs or challenges of hospitalized opioid‐exposed patients.

METHODS

We used inpatient and outpatient administrative data from the Department of Veterans Affairs (VA) Healthcare System. The primary data source to identify acute medical admissions was the VA Patient Treatment File, a national administrative database of all inpatient admissions, including patient demographic characteristics, primary and secondary diagnoses (using International Classification of Diseases, 9th Revision, Clinical Modification [ICD‐9‐CM], codes), and hospitalization characteristics. Outpatient pharmacy data were from the VA Pharmacy Prescription Data Files. The VA Vital Status Files provided dates of death.

We identified all first acute medical admissions to 129 VA hospitals during fiscal years (FYs) 2009 to 2011 (October 2009September 2011). We defined first admissions as the initial medical hospitalization occurring following a minimum 365‐day hospitalization‐free period. Patients were required to demonstrate pharmacy use by receipt of any outpatient medication from the VA on 2 separate occasions within 270 days preceding the first admission, to avoid misclassification of patients who routinely obtained medications only from a non‐VA provider. Patients admitted from extended care facilities were excluded.

We grouped patients by opioid‐use status based on outpatient prescription records: (1) no opioid use, defined as no opioid prescriptions in the 6 months prior to hospitalization; (2) occasional opioid use, defined as patients who received any opioid prescription during the 6 months prior but did not meet definition of chronic use; and (3) chronic opioid therapy, defined as 90 or more days' supply of opioids received within 6 months preceding hospitalization. We did not specify continuous prescribing. Opioids included in the definition were codeine, dihydrocodeine, fentanyl (mucosal and topical), hydrocodone, hydromorphone, meperidine, methadone, morphine, oxycodone, oxymorphone, pentazocine, propoxyphene, tapentadol, and tramadol.[16, 17]

We compared groups by demographic variables including age, sex, race, income, rural vs urban residence (determined from Rural‐Urban Commuting Area codes), region based on hospital location; overall comorbidity using the Charlson Comorbidity Index (CCI);[18] and 10 selected conditions to characterize comorbidity (see Supporting Information, Appendix A, in the online version of this article). These 10 conditions were chosen based on probable associations with chronic opioid use or high prevalence among hospitalized veterans.[9, 19, 20]

We used a CNCP definition based on ICD‐9‐CM codes.[9] This definition did not include episodic conditions such as migraine[2] or a measure of pain intensity.[21] All conditions were determined from diagnoses coded during any encounter in the year prior to hospitalization, exclusive of the first (ie, index) admission. We also determined the frequency of palliative care use, defined as presence of ICD‐9‐CM code V667 during index hospitalization or within the past year. Patients with palliative care use (n=3070) were excluded from further analyses.

We compared opioid use groups by baseline characteristics using the [2] statistic to determine if the distribution was nonrandom. We used analysis of variance to compare hospital length of stay between groups. We used the [2] statistic to compare rates of 4 outcomes of interest: intensive care unit (ICU) admission during the index hospitalization, discharge disposition other than home, 30‐day readmission rate, and in‐hospital or 30‐day mortality.

To assess the association between opioid‐use status and the 4 outcomes of interest, we constructed 2 multivariable regression models; the first was adjusted only for admission diagnosis using the Clinical Classification Software (CCS),[22] and the second was adjusted for demographics, CCI, and the 10 selected comorbidities in addition to admission diagnosis.

The authors had full access to and take full responsibility for the integrity of the data. All analyses were conducted using SAS statistical software version 9.2 (SAS Institute, Cary, NC). The study was approved by the University of Iowa institutional review board and the Iowa City VA Health Care System Research and Development Committee.

RESULTS

DISCUSSION

This observational study is, to our knowledge, the first to report prevalence of and characteristics associated with prior opioid use among hospitalized medical patients. The prevalence of any opioid use and of COT was substantially higher in this hospitalized cohort than reported in outpatient settings. The prevalence of any opioid use during 1 year (FY 2009) among all veterans with VA primary care use was 26.1%.[23] A study of incident prescribing rates among veterans with new diagnoses of noncancer‐related pain demonstrated 11% received an opioid prescription within 1 year.[24] Using a definition of 90 consecutive prescription days to define COT, Dobscha et al.[25] found that 5% of veterans with persistent elevated pain intensity and no previous opioid prescriptions subsequently received COT within 12 months. The high prevalence we found likely reflects cumulative effects of incident use as well as an increased symptom burden in a population defined by need for medical hospitalization.

Although a veteran population may not be generalizable to a nonveteran setting, we do note prior studies reporting prevalence of any opioid use in outpatient cohorts (in 2000 and 2005) of between 18% and 30%, with higher rates among women and patients over 65 years of age.[1, 2]

Our work was purposefully inclusive of cancer patients so that we might assess the degree to which cancer diagnoses accounted for prior opioid use in hospitalized patients. Surprisingly, the rate of COT for patients with cancer was lower than that for patients with CNCP, perhaps reflecting that a cancer condition defined in administrative data may not constitute a pain‐causing disease.

Recognition of the prevalence of opioid therapy is important as we work to understand and improve safety, satisfaction, utilization, and long‐term health outcomes associated with hospitalization. Our finding that over half of medical inpatients have preexisting CNCP diagnoses, and a not entirely overlapping proportion has prior opioid exposure, implies a need for future work to refine expectations and strategies for inpatient management, potentially tailored to prior opioid use and presence of CNCP.

A recent Joint Commission sentinel event alert[26] highlights opioid adverse events in the hospital and identifies both lack of previous opioid therapy and prior opioid therapy as factors increasing risk. ICU admission during the hospital stay may reflect adverse events such as opioid‐induced respiratory depression; in our study, patients with no opioid use prior to admission were more likely to have an ICU stay, although the effect was small. One might speculate that clinicians, accustomed to treating pain in opioid‐exposed patients, are using inappropriately large starting dosages of narcotics for inpatients without first assessing prior opioid exposure. Another possible explanation is that patients on COT are admitted to the hospital with less severe illness, potentially reflecting functional, social, or access limitations that compromise ability to manage illness in the outpatient setting. More detailed comparison of illness severity is beyond the scope of the present work.

Patient satisfaction with pain management is reflected in 2 of the Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) questions, and is publically reported.[27] HCAHPS results also figure in the formula for the Centers for Medicare and Medicaid Services value‐based purchasing.[28] Preadmission pain is predictive of postoperative pain[29, 30] and may shape patient expectations; how preadmission opioid use modulates nonsurgical pain and satisfaction with management in the medical inpatient remains to be studied. The high prevalence of prior COT underscores the importance of understanding characteristics of patients on COT, and potential differences and disparities in pain management, when designing interventions to augment patient satisfaction with pain management.

Although the age distribution and patterns of comorbidities differed between the opioid‐use groups, opioid therapy remained a small but significant predictor of hospital readmission; this association was independent of CNCP diagnosis. Functional outcomes are recognized as important measures of efficacy of outpatient pain management strategies,[31] with some evidence that opioids are associated with worse functioning.[32, 33] Functional limitations, as well as inadequately or inappropriately treated pain, may drive both admissions and readmissions. Alternately, COT may be a marker for unmeasured factors that increase a patient's risk of returning to the hospital. Further work is needed to elucidate the relationship between COT and healthcare utilization associated with the inpatient stay.

Our finding that patients on COT have an increased mortality risk is concerning, given the rapid expansion in use of these medications. Although pain is increasingly prevalent toward end of life,[34] we did not observe an association between either CNCP (data not shown) or occasional opioid use and mortality. COT may complicate chronic disease through adverse drug effects including respiratory depression, apnea, or endocrine or immune alteration. Complex chronically ill patients with conditions such as COPD, HF, or diabetes may be particularly susceptible to these effects. Incident use of morphine is associated with increased mortality in acute coronary syndrome and HF[35, 36]: we are not aware of any work describing the relationship between prior opioid use and incident use during hospitalization in medical patients.

CONCLUSION

Nearly 1 in 4 hospitalized veterans has current or recent COT at the time of hospital admission for nonsurgical conditions; nearly half have been prescribed any opioids. Practitioners designing interventions to improve pain management in the inpatient setting should account for prior opioid use. Patients who are on COT prior to hospitalization differ in age and comorbidities from their counterparts who are not on COT. Further elucidation of differences between opioid‐use groups may help providers address care needs during the transition to posthospitalization care. CNCP diagnoses and chronic opioid exposure are different entities and cannot serve as proxies in administrative data. Additional work on utilization and outcomes in specific patient populations may improve our understanding of the long‐term health effects of chronic opioid therapy.

Disclosures: Dr. Mosher is supported by the Veterans Administration (VA) Quality Scholars Fellowship, Office of Academic Affiliations, Department of Veterans Affairs. Dr. Cram is supported by a K24 award from NIAMS (AR062133) at the National Institutes of Health. The preliminary results of this article were presented at the Society of General Internal Medicine Annual Meeting in Denver, Colordao, April 2013. The views expressed in this article are those of the authors and do not necessarily represent the views of the Department of Veterans Affairs. Data are available to researchers with VA accreditation, the statistical code and the protocol are available to interested readers by contacting Dr. Mosher. The authors report no conflict of interest in regard to this study.

Recent trends show a marked increase in outpatient use of chronic opioid therapy (COT) for chronic noncancer pain (CNCP)[1, 2] without decreases in reported CNCP,[3] raising concerns about the efficacy and risk‐to‐benefit ratio of opioids in this population.[4, 5, 6, 7, 8] Increasing rates of outpatient use likely are accompanied by increasing rates of opioid exposure among patients admitted to the hospital. To our knowledge there are no published data regarding the prevalence of COT during the months preceding hospitalization.

Opioid use has been linked to increased emergency room utilization[9, 10] and emergency hospitalization,[11] but associations between opioid use and inpatient metrics (eg, mortality, readmission) have not been explored. Furthermore, lack of knowledge about the prevalence of opioid use prior to hospitalization may impede efforts to improve inpatient pain management and satisfaction with care. Although there is reason to expect that strategies to safely and effectively treat acute pain during the inpatient stay differ between opioid‐nave patients and opioid‐exposed patients, evidence regarding treatment strategies is limited.[12, 13, 14] Opioid pain medications are associated with hospital adverse events, with both prior opioid exposure and lack of opioid use as proposed risk factors.[15] A better understanding of the prevalence and characteristics of hospitalized COT patients is fundamental to future work to achieve safer and more effective inpatient pain management.

The primary purpose of this study was to determine the prevalence of prior COT among hospitalized medical patients. Additionally, we aimed to characterize inpatients with occasional and chronic opioid therapy prior to admission in comparison to opioid‐nave inpatients, as differences between these groups may suggest directions for further investigation into the distinct needs or challenges of hospitalized opioid‐exposed patients.

METHODS

We used inpatient and outpatient administrative data from the Department of Veterans Affairs (VA) Healthcare System. The primary data source to identify acute medical admissions was the VA Patient Treatment File, a national administrative database of all inpatient admissions, including patient demographic characteristics, primary and secondary diagnoses (using International Classification of Diseases, 9th Revision, Clinical Modification [ICD‐9‐CM], codes), and hospitalization characteristics. Outpatient pharmacy data were from the VA Pharmacy Prescription Data Files. The VA Vital Status Files provided dates of death.

We identified all first acute medical admissions to 129 VA hospitals during fiscal years (FYs) 2009 to 2011 (October 2009September 2011). We defined first admissions as the initial medical hospitalization occurring following a minimum 365‐day hospitalization‐free period. Patients were required to demonstrate pharmacy use by receipt of any outpatient medication from the VA on 2 separate occasions within 270 days preceding the first admission, to avoid misclassification of patients who routinely obtained medications only from a non‐VA provider. Patients admitted from extended care facilities were excluded.

We grouped patients by opioid‐use status based on outpatient prescription records: (1) no opioid use, defined as no opioid prescriptions in the 6 months prior to hospitalization; (2) occasional opioid use, defined as patients who received any opioid prescription during the 6 months prior but did not meet definition of chronic use; and (3) chronic opioid therapy, defined as 90 or more days' supply of opioids received within 6 months preceding hospitalization. We did not specify continuous prescribing. Opioids included in the definition were codeine, dihydrocodeine, fentanyl (mucosal and topical), hydrocodone, hydromorphone, meperidine, methadone, morphine, oxycodone, oxymorphone, pentazocine, propoxyphene, tapentadol, and tramadol.[16, 17]

We compared groups by demographic variables including age, sex, race, income, rural vs urban residence (determined from Rural‐Urban Commuting Area codes), region based on hospital location; overall comorbidity using the Charlson Comorbidity Index (CCI);[18] and 10 selected conditions to characterize comorbidity (see Supporting Information, Appendix A, in the online version of this article). These 10 conditions were chosen based on probable associations with chronic opioid use or high prevalence among hospitalized veterans.[9, 19, 20]

We used a CNCP definition based on ICD‐9‐CM codes.[9] This definition did not include episodic conditions such as migraine[2] or a measure of pain intensity.[21] All conditions were determined from diagnoses coded during any encounter in the year prior to hospitalization, exclusive of the first (ie, index) admission. We also determined the frequency of palliative care use, defined as presence of ICD‐9‐CM code V667 during index hospitalization or within the past year. Patients with palliative care use (n=3070) were excluded from further analyses.

We compared opioid use groups by baseline characteristics using the [2] statistic to determine if the distribution was nonrandom. We used analysis of variance to compare hospital length of stay between groups. We used the [2] statistic to compare rates of 4 outcomes of interest: intensive care unit (ICU) admission during the index hospitalization, discharge disposition other than home, 30‐day readmission rate, and in‐hospital or 30‐day mortality.

To assess the association between opioid‐use status and the 4 outcomes of interest, we constructed 2 multivariable regression models; the first was adjusted only for admission diagnosis using the Clinical Classification Software (CCS),[22] and the second was adjusted for demographics, CCI, and the 10 selected comorbidities in addition to admission diagnosis.

The authors had full access to and take full responsibility for the integrity of the data. All analyses were conducted using SAS statistical software version 9.2 (SAS Institute, Cary, NC). The study was approved by the University of Iowa institutional review board and the Iowa City VA Health Care System Research and Development Committee.

RESULTS

DISCUSSION

This observational study is, to our knowledge, the first to report prevalence of and characteristics associated with prior opioid use among hospitalized medical patients. The prevalence of any opioid use and of COT was substantially higher in this hospitalized cohort than reported in outpatient settings. The prevalence of any opioid use during 1 year (FY 2009) among all veterans with VA primary care use was 26.1%.[23] A study of incident prescribing rates among veterans with new diagnoses of noncancer‐related pain demonstrated 11% received an opioid prescription within 1 year.[24] Using a definition of 90 consecutive prescription days to define COT, Dobscha et al.[25] found that 5% of veterans with persistent elevated pain intensity and no previous opioid prescriptions subsequently received COT within 12 months. The high prevalence we found likely reflects cumulative effects of incident use as well as an increased symptom burden in a population defined by need for medical hospitalization.

Although a veteran population may not be generalizable to a nonveteran setting, we do note prior studies reporting prevalence of any opioid use in outpatient cohorts (in 2000 and 2005) of between 18% and 30%, with higher rates among women and patients over 65 years of age.[1, 2]

Our work was purposefully inclusive of cancer patients so that we might assess the degree to which cancer diagnoses accounted for prior opioid use in hospitalized patients. Surprisingly, the rate of COT for patients with cancer was lower than that for patients with CNCP, perhaps reflecting that a cancer condition defined in administrative data may not constitute a pain‐causing disease.

Recognition of the prevalence of opioid therapy is important as we work to understand and improve safety, satisfaction, utilization, and long‐term health outcomes associated with hospitalization. Our finding that over half of medical inpatients have preexisting CNCP diagnoses, and a not entirely overlapping proportion has prior opioid exposure, implies a need for future work to refine expectations and strategies for inpatient management, potentially tailored to prior opioid use and presence of CNCP.

A recent Joint Commission sentinel event alert[26] highlights opioid adverse events in the hospital and identifies both lack of previous opioid therapy and prior opioid therapy as factors increasing risk. ICU admission during the hospital stay may reflect adverse events such as opioid‐induced respiratory depression; in our study, patients with no opioid use prior to admission were more likely to have an ICU stay, although the effect was small. One might speculate that clinicians, accustomed to treating pain in opioid‐exposed patients, are using inappropriately large starting dosages of narcotics for inpatients without first assessing prior opioid exposure. Another possible explanation is that patients on COT are admitted to the hospital with less severe illness, potentially reflecting functional, social, or access limitations that compromise ability to manage illness in the outpatient setting. More detailed comparison of illness severity is beyond the scope of the present work.

Patient satisfaction with pain management is reflected in 2 of the Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) questions, and is publically reported.[27] HCAHPS results also figure in the formula for the Centers for Medicare and Medicaid Services value‐based purchasing.[28] Preadmission pain is predictive of postoperative pain[29, 30] and may shape patient expectations; how preadmission opioid use modulates nonsurgical pain and satisfaction with management in the medical inpatient remains to be studied. The high prevalence of prior COT underscores the importance of understanding characteristics of patients on COT, and potential differences and disparities in pain management, when designing interventions to augment patient satisfaction with pain management.

Although the age distribution and patterns of comorbidities differed between the opioid‐use groups, opioid therapy remained a small but significant predictor of hospital readmission; this association was independent of CNCP diagnosis. Functional outcomes are recognized as important measures of efficacy of outpatient pain management strategies,[31] with some evidence that opioids are associated with worse functioning.[32, 33] Functional limitations, as well as inadequately or inappropriately treated pain, may drive both admissions and readmissions. Alternately, COT may be a marker for unmeasured factors that increase a patient's risk of returning to the hospital. Further work is needed to elucidate the relationship between COT and healthcare utilization associated with the inpatient stay.

Our finding that patients on COT have an increased mortality risk is concerning, given the rapid expansion in use of these medications. Although pain is increasingly prevalent toward end of life,[34] we did not observe an association between either CNCP (data not shown) or occasional opioid use and mortality. COT may complicate chronic disease through adverse drug effects including respiratory depression, apnea, or endocrine or immune alteration. Complex chronically ill patients with conditions such as COPD, HF, or diabetes may be particularly susceptible to these effects. Incident use of morphine is associated with increased mortality in acute coronary syndrome and HF[35, 36]: we are not aware of any work describing the relationship between prior opioid use and incident use during hospitalization in medical patients.

CONCLUSION

Nearly 1 in 4 hospitalized veterans has current or recent COT at the time of hospital admission for nonsurgical conditions; nearly half have been prescribed any opioids. Practitioners designing interventions to improve pain management in the inpatient setting should account for prior opioid use. Patients who are on COT prior to hospitalization differ in age and comorbidities from their counterparts who are not on COT. Further elucidation of differences between opioid‐use groups may help providers address care needs during the transition to posthospitalization care. CNCP diagnoses and chronic opioid exposure are different entities and cannot serve as proxies in administrative data. Additional work on utilization and outcomes in specific patient populations may improve our understanding of the long‐term health effects of chronic opioid therapy.

Disclosures: Dr. Mosher is supported by the Veterans Administration (VA) Quality Scholars Fellowship, Office of Academic Affiliations, Department of Veterans Affairs. Dr. Cram is supported by a K24 award from NIAMS (AR062133) at the National Institutes of Health. The preliminary results of this article were presented at the Society of General Internal Medicine Annual Meeting in Denver, Colordao, April 2013. The views expressed in this article are those of the authors and do not necessarily represent the views of the Department of Veterans Affairs. Data are available to researchers with VA accreditation, the statistical code and the protocol are available to interested readers by contacting Dr. Mosher. The authors report no conflict of interest in regard to this study.

Recent trends show a marked increase in outpatient use of chronic opioid therapy (COT) for chronic noncancer pain (CNCP)[1, 2] without decreases in reported CNCP,[3] raising concerns about the efficacy and risk‐to‐benefit ratio of opioids in this population.[4, 5, 6, 7, 8] Increasing rates of outpatient use likely are accompanied by increasing rates of opioid exposure among patients admitted to the hospital. To our knowledge there are no published data regarding the prevalence of COT during the months preceding hospitalization.

Opioid use has been linked to increased emergency room utilization[9, 10] and emergency hospitalization,[11] but associations between opioid use and inpatient metrics (eg, mortality, readmission) have not been explored. Furthermore, lack of knowledge about the prevalence of opioid use prior to hospitalization may impede efforts to improve inpatient pain management and satisfaction with care. Although there is reason to expect that strategies to safely and effectively treat acute pain during the inpatient stay differ between opioid‐nave patients and opioid‐exposed patients, evidence regarding treatment strategies is limited.[12, 13, 14] Opioid pain medications are associated with hospital adverse events, with both prior opioid exposure and lack of opioid use as proposed risk factors.[15] A better understanding of the prevalence and characteristics of hospitalized COT patients is fundamental to future work to achieve safer and more effective inpatient pain management.

The primary purpose of this study was to determine the prevalence of prior COT among hospitalized medical patients. Additionally, we aimed to characterize inpatients with occasional and chronic opioid therapy prior to admission in comparison to opioid‐nave inpatients, as differences between these groups may suggest directions for further investigation into the distinct needs or challenges of hospitalized opioid‐exposed patients.

METHODS

We used inpatient and outpatient administrative data from the Department of Veterans Affairs (VA) Healthcare System. The primary data source to identify acute medical admissions was the VA Patient Treatment File, a national administrative database of all inpatient admissions, including patient demographic characteristics, primary and secondary diagnoses (using International Classification of Diseases, 9th Revision, Clinical Modification [ICD‐9‐CM], codes), and hospitalization characteristics. Outpatient pharmacy data were from the VA Pharmacy Prescription Data Files. The VA Vital Status Files provided dates of death.

We identified all first acute medical admissions to 129 VA hospitals during fiscal years (FYs) 2009 to 2011 (October 2009September 2011). We defined first admissions as the initial medical hospitalization occurring following a minimum 365‐day hospitalization‐free period. Patients were required to demonstrate pharmacy use by receipt of any outpatient medication from the VA on 2 separate occasions within 270 days preceding the first admission, to avoid misclassification of patients who routinely obtained medications only from a non‐VA provider. Patients admitted from extended care facilities were excluded.

We grouped patients by opioid‐use status based on outpatient prescription records: (1) no opioid use, defined as no opioid prescriptions in the 6 months prior to hospitalization; (2) occasional opioid use, defined as patients who received any opioid prescription during the 6 months prior but did not meet definition of chronic use; and (3) chronic opioid therapy, defined as 90 or more days' supply of opioids received within 6 months preceding hospitalization. We did not specify continuous prescribing. Opioids included in the definition were codeine, dihydrocodeine, fentanyl (mucosal and topical), hydrocodone, hydromorphone, meperidine, methadone, morphine, oxycodone, oxymorphone, pentazocine, propoxyphene, tapentadol, and tramadol.[16, 17]

We compared groups by demographic variables including age, sex, race, income, rural vs urban residence (determined from Rural‐Urban Commuting Area codes), region based on hospital location; overall comorbidity using the Charlson Comorbidity Index (CCI);[18] and 10 selected conditions to characterize comorbidity (see Supporting Information, Appendix A, in the online version of this article). These 10 conditions were chosen based on probable associations with chronic opioid use or high prevalence among hospitalized veterans.[9, 19, 20]

We used a CNCP definition based on ICD‐9‐CM codes.[9] This definition did not include episodic conditions such as migraine[2] or a measure of pain intensity.[21] All conditions were determined from diagnoses coded during any encounter in the year prior to hospitalization, exclusive of the first (ie, index) admission. We also determined the frequency of palliative care use, defined as presence of ICD‐9‐CM code V667 during index hospitalization or within the past year. Patients with palliative care use (n=3070) were excluded from further analyses.

We compared opioid use groups by baseline characteristics using the [2] statistic to determine if the distribution was nonrandom. We used analysis of variance to compare hospital length of stay between groups. We used the [2] statistic to compare rates of 4 outcomes of interest: intensive care unit (ICU) admission during the index hospitalization, discharge disposition other than home, 30‐day readmission rate, and in‐hospital or 30‐day mortality.

To assess the association between opioid‐use status and the 4 outcomes of interest, we constructed 2 multivariable regression models; the first was adjusted only for admission diagnosis using the Clinical Classification Software (CCS),[22] and the second was adjusted for demographics, CCI, and the 10 selected comorbidities in addition to admission diagnosis.

The authors had full access to and take full responsibility for the integrity of the data. All analyses were conducted using SAS statistical software version 9.2 (SAS Institute, Cary, NC). The study was approved by the University of Iowa institutional review board and the Iowa City VA Health Care System Research and Development Committee.

RESULTS

DISCUSSION

This observational study is, to our knowledge, the first to report prevalence of and characteristics associated with prior opioid use among hospitalized medical patients. The prevalence of any opioid use and of COT was substantially higher in this hospitalized cohort than reported in outpatient settings. The prevalence of any opioid use during 1 year (FY 2009) among all veterans with VA primary care use was 26.1%.[23] A study of incident prescribing rates among veterans with new diagnoses of noncancer‐related pain demonstrated 11% received an opioid prescription within 1 year.[24] Using a definition of 90 consecutive prescription days to define COT, Dobscha et al.[25] found that 5% of veterans with persistent elevated pain intensity and no previous opioid prescriptions subsequently received COT within 12 months. The high prevalence we found likely reflects cumulative effects of incident use as well as an increased symptom burden in a population defined by need for medical hospitalization.

Although a veteran population may not be generalizable to a nonveteran setting, we do note prior studies reporting prevalence of any opioid use in outpatient cohorts (in 2000 and 2005) of between 18% and 30%, with higher rates among women and patients over 65 years of age.[1, 2]

Our work was purposefully inclusive of cancer patients so that we might assess the degree to which cancer diagnoses accounted for prior opioid use in hospitalized patients. Surprisingly, the rate of COT for patients with cancer was lower than that for patients with CNCP, perhaps reflecting that a cancer condition defined in administrative data may not constitute a pain‐causing disease.

Recognition of the prevalence of opioid therapy is important as we work to understand and improve safety, satisfaction, utilization, and long‐term health outcomes associated with hospitalization. Our finding that over half of medical inpatients have preexisting CNCP diagnoses, and a not entirely overlapping proportion has prior opioid exposure, implies a need for future work to refine expectations and strategies for inpatient management, potentially tailored to prior opioid use and presence of CNCP.

A recent Joint Commission sentinel event alert[26] highlights opioid adverse events in the hospital and identifies both lack of previous opioid therapy and prior opioid therapy as factors increasing risk. ICU admission during the hospital stay may reflect adverse events such as opioid‐induced respiratory depression; in our study, patients with no opioid use prior to admission were more likely to have an ICU stay, although the effect was small. One might speculate that clinicians, accustomed to treating pain in opioid‐exposed patients, are using inappropriately large starting dosages of narcotics for inpatients without first assessing prior opioid exposure. Another possible explanation is that patients on COT are admitted to the hospital with less severe illness, potentially reflecting functional, social, or access limitations that compromise ability to manage illness in the outpatient setting. More detailed comparison of illness severity is beyond the scope of the present work.

Patient satisfaction with pain management is reflected in 2 of the Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) questions, and is publically reported.[27] HCAHPS results also figure in the formula for the Centers for Medicare and Medicaid Services value‐based purchasing.[28] Preadmission pain is predictive of postoperative pain[29, 30] and may shape patient expectations; how preadmission opioid use modulates nonsurgical pain and satisfaction with management in the medical inpatient remains to be studied. The high prevalence of prior COT underscores the importance of understanding characteristics of patients on COT, and potential differences and disparities in pain management, when designing interventions to augment patient satisfaction with pain management.

Although the age distribution and patterns of comorbidities differed between the opioid‐use groups, opioid therapy remained a small but significant predictor of hospital readmission; this association was independent of CNCP diagnosis. Functional outcomes are recognized as important measures of efficacy of outpatient pain management strategies,[31] with some evidence that opioids are associated with worse functioning.[32, 33] Functional limitations, as well as inadequately or inappropriately treated pain, may drive both admissions and readmissions. Alternately, COT may be a marker for unmeasured factors that increase a patient's risk of returning to the hospital. Further work is needed to elucidate the relationship between COT and healthcare utilization associated with the inpatient stay.

Our finding that patients on COT have an increased mortality risk is concerning, given the rapid expansion in use of these medications. Although pain is increasingly prevalent toward end of life,[34] we did not observe an association between either CNCP (data not shown) or occasional opioid use and mortality. COT may complicate chronic disease through adverse drug effects including respiratory depression, apnea, or endocrine or immune alteration. Complex chronically ill patients with conditions such as COPD, HF, or diabetes may be particularly susceptible to these effects. Incident use of morphine is associated with increased mortality in acute coronary syndrome and HF[35, 36]: we are not aware of any work describing the relationship between prior opioid use and incident use during hospitalization in medical patients.

CONCLUSION

Nearly 1 in 4 hospitalized veterans has current or recent COT at the time of hospital admission for nonsurgical conditions; nearly half have been prescribed any opioids. Practitioners designing interventions to improve pain management in the inpatient setting should account for prior opioid use. Patients who are on COT prior to hospitalization differ in age and comorbidities from their counterparts who are not on COT. Further elucidation of differences between opioid‐use groups may help providers address care needs during the transition to posthospitalization care. CNCP diagnoses and chronic opioid exposure are different entities and cannot serve as proxies in administrative data. Additional work on utilization and outcomes in specific patient populations may improve our understanding of the long‐term health effects of chronic opioid therapy.

Disclosures: Dr. Mosher is supported by the Veterans Administration (VA) Quality Scholars Fellowship, Office of Academic Affiliations, Department of Veterans Affairs. Dr. Cram is supported by a K24 award from NIAMS (AR062133) at the National Institutes of Health. The preliminary results of this article were presented at the Society of General Internal Medicine Annual Meeting in Denver, Colordao, April 2013. The views expressed in this article are those of the authors and do not necessarily represent the views of the Department of Veterans Affairs. Data are available to researchers with VA accreditation, the statistical code and the protocol are available to interested readers by contacting Dr. Mosher. The authors report no conflict of interest in regard to this study.

Acute upper gastrointestinal hemorrhage (UGIH) is one of the most common hospital admissions for acute care. Estimates indicate that 300,000 patients (100‐150 cases per 100,000 adults) are admitted annually with an associated economic impact of $2.5 billion.15 The current standard management of UGIH requires hospital admission and esophagogastroduodenoscopy (EGD) by a gastroenterologist for diagnosis and/or treatment. This management strategy results in a high consumption of hospital resources and costs.

Simultaneously, hospitalists have dramatically changed the delivery of inpatient care in the United States and are recognized as a location‐driven subspecialty for the care of acute hospitalized patients, similar to emergency medicine. Currently there are 20,000 hospitalists, and more than one‐third of general medicine inpatients are cared for by hospitalists.6, 7

Previous studies have shown that hospitalist care offers better or comparable outcomes, with lower overall length of stay (LOS) and costs compared to traditional providers.810 However, most of these studies were performed in single institutions, had weak designs or little‐to‐no adjustment for severity of illness, or were limited to 7 specific diseases (pneumonia, congestive heart failure [CHF], chest pain, ischemic stroke, urinary tract infection, chronic obstructive lung disease [COPD], and acute myocardial infarction [AMI]).8

Furthermore, less is known about the effect of hospitalists on conditions that may be dependent upon specialist consultation for procedures and/or treatment plans. In this study, gastroenterologists performed diagnostic and/or therapeutic endoscopy work as consultants to the attending physicians in the management of acute inpatient UGIH.

To explore the effects of hospitalists on care of patients with acute UGIH, we examined data from the Multicenter Hospitalist (MCH) trial. The objectives of our study were to compare clinical outcomesin‐hospital mortality and complications (ie, recurrent bleeding, intensive care unit [ICU] transfer, decompensation, transfusion, reendoscopy, 30‐day readmission)and efficiency (LOS and costs) in hospitalized acute UGIH patients cared for by hospitalists and nonhospitalists in 6 academic centers in the United States during a 2‐year period.

Patients and Methods

Results

Discussion

This is the first study that has looked at the effect of hospitalists on clinical outcomes and efficiency in patients admitted for acute UGIH, a condition highly dependent upon another specialty for procedures and management. This is also one of only a few studies on UGIH that adjusted for severity of illness (Rockall score), comorbidity, performance of early endoscopypatient‐level confounders usually unaccounted for in prior research.

We show that hospitalists and nonhospitalists caring for acute UGIH patients had overall similar unadjusted outcomes; except for blood transfusion and 30‐day readmission rates. Unfortunately, due to the small number of events for readmissions, we were unable to perform adjusted analysis for readmission. Differences between hospitalists and nonhospitalists on blood transfusion rates were not substantiated on multivariable adjustments.

As for efficiency, univariable and multivariable analyses revealed that LOS was similar between provider types while costs were greater in UGIH patients attended by hospitalists.

Reductions in resource use, particularly costs, may be achieved by increasing throughput (eg, reducing LOS) or by decreasing service intensity (eg, using fewer ancillary services and specialty consultations).26 Specifically in acute UGIH, LOS is significantly affected by performance of early EGD.27, 28 In these studies, gastroenterologist‐led teams, compared to internists and surgeons, have easier access to endoscopy, thus reducing LOS and overall costs.27, 28

Similarly, prior studies have shown that the mechanism by which hospitalists lower costs is by decreasing LOS.810, 29 There are several hypotheses on how hospitalists affect LOS. Hospitalists, by being available all day, are thought to respond quickly to acute symptoms or new test results, are more efficient in navigating the complex hospital environment, or develop greater expertise as a result of added inpatient experience.8 On the downside, although the hospitalist model reduces overall LOS and costs, they also provide higher intensity of care as reflected by greater costs when broken down per hospital day.29 Thus, the cost differential we found may represent higher intensity of care by hospitalists in their management of acute UGIH, as higher intensity care without decreasing LOS can translate to higher costs.

In addition, patients with acute UGIH are unique in several respects. In contrast to diseases like heart failure, COPD, and pneumonia, in which the admitting provider has the option to request a subspecialist consultation, all patients with acute UGIH need a gastroenterologist to perform endoscopy as part of the management. These patients are usually admitted to general medicine wards, aggressively resuscitated with intravenous fluids, with a nonurgent gastroenterology consult or EGD performed on the next available schedule.

Aside from LOS being greatly affected by performance of early EGD and/or delay in consulting gastroenterology, sicker patients require longer hospitalization and drive LOS and healthcare costs up. It was therefore crucial that we accounted for severity of illness, comorbidity, and performance of early EGD in our regression models for LOS and costs. This approach allows us to acquire a more accurate estimate on the effects of hospitalist on LOS and costs in patients admitted with acute UGIH.

Our findings suggest that the academic hospitalist model of care may not have as great of an impact on hospital efficiency in certain patient groups that require nonurgent subspecialty consultations. Future studies should focus on elucidating these relationships.

Limitations

This study has several limitations. First, clinical data were abstracted at 6 sites by different abstractors so it is possible there were variations in how data were collected. To reduce variation, a standardized abstraction form with instructions was developed and the primary investigator (PI) was available for specific questions during the abstraction process. Second, only 5 out of the 6 sites used TSI accounting systems. Although similar, interhospital costs captured by TSI may vary among sites in terms of classifying direct and indirect costs, potentially resulting in misclassification bias in our cost estimates.17 We addressed these issues by including the hospital site variable in our regression models, regardless of its significance. Third, consent rates across sites vary from 70% to 85%. It is possible that patients who refused enrollment in the MCH trial are systematically different and may introduce bias in our analysis.

Furthermore, the study was designed as a natural experiment based on a rotational call cycle between hospitalist‐led and nonhospitalist‐led teams. It is possible that the order of patient assignment might not be completely naturally random as we intended. However, the study period was for 2 years and we expect the effect of order would have averaged out in time.

There are many hospitalist models of care. In terms of generalizability, the study pertains only to academic hospitalists and may not be applicable to hospitalists practicing in community hospitals. For example, the nonhospitalist comparison group is likely different in the community and academic settings. Community nonhospitalists (traditional practitioners) are usually internists covering both inpatient and outpatient responsibilities at the same time. In contrast, academic nonhospitalists are internists or subspecialists serving as ward attendings for a limited period (usually 1 month) with considerable variation in their nonattending responsibilities (eg, research, clinic, administration). Furthermore, academic nonhospitalist providers might be a self‐selected group by their willingness to serve as a ward attending, making them more hospitalist‐like. Changes and variability of inpatient attendings may also affect our findings when compared to prior work. Finally, it is also possible that having residents at academic medical centers may attenuate the effect of hospitalists more than in community‐based models.

Conclusions/Implications

Compared to nonhospitalists, academic hospitalist care of acute UGIH patients had similar overall clinical outcomes. However, our finding of similar LOS yet higher costs for patients cared for by hospitalists support 1 proposed mechanism in which hospitalists decrease healthcare costs: providing higher intensity of care per day of hospitalization. However, in academic hospitalist models, this higher intensity hypothesis should be revisited, especially in certain patient groups in which timing and involvement of subspecialists may influence discharge decisions, affecting LOS and overall costs.

Due to inherent limitations in this observational study, future studies should focus on verifying and elucidating these relationships further. Lastly, understanding which patient groups receive the greatest potential benefit from this model will help guide both organizational efforts and quality improvement strategies.