Samuel F. Hohmann, PhD

Interhospital transfers (IHTs) to academic medical centers (AMCs) or their affiliated hospitals may benefit patients who require unique specialty and procedural services. However, IHTs also introduce a potentially risky transition of care for patients suffering from complex or unstable medical problems.[1] Components of this risk include the dangers associated with transportation and the disrupted continuity of care that may lead to delays or errors in care.[2, 3] Furthermore, referring and accepting providers may face barriers to optimal handoffs including a lack of shared communication standards and difficulty accessing external medical records.[3, 4, 5] Although some authors have recommended the creation of formal guidelines for interhospital transfer processes for all patients to mitigate the risks of transfer, the available guidelines governing the IHT triage and communication process are limited to critically ill patients.[6]

A recent study of a diverse patient and hospital dataset demonstrated that interhospital transfer patients have a higher risk of mortality, increased length of stay (LOS), and increased risk of adverse events as compared with non‐transfer patients.[7] However, it is unknown if these findings persist in the population of patients transferred specifically to AMCs or their affiliated hospitals (the combination is hereafter referred to as academic health systems [AHSs]). AMCs provide a disproportionate share of IHT care for complex patients and have a vested interest in improving the outcomes of these transitions.[8] Prior single‐center studies of acute care adult medical patients accepted to AMCs have shown that IHT is associated with a longer LOS, increased in‐hospital mortality, and higher resource use.[9, 10] However, it is difficult to generalize from single‐center studies due to the variation in referral practices, geography, and network characteristics. Additionally, AMC referral systems, patient mix, and utilization of hospitalists have likely changed substantially in the nearly 2 decades since those reports were published.

Hospitalists and general internists often manage the transfer acceptance processes for internal medicine services at receiving hospitals, helping to triage and coordinate care for IHT patients. As a result, it is important for hospitalists to understand the characteristics and outcomes of the IHT population. In addition to informing the decision making around transfer for a given patient, such an understanding is the foundation for helping providers and institutions begin to systematically identify and mitigate peritransfer risks.

We conducted this large multicenter study to describe the characteristics and outcomes of a current, nationally representative IHT patient population discharged by hospitalists and general internists at AHSs. To identify unique features of the IHT population, we compared patients transferred from another hospital to an AHS to those admitted to the AHS directly from the AHS's emergency department (ED). Based on our anecdotal experiences and the prior single‐center study findings in adult medical populations,[9, 10] we hypothesized that the IHT population would be sicker, stay in the hospital and intensive care unit (ICU) longer, and have higher costs and in‐hospital mortality than ED patients. Although there may be fundamental differences between the 2 groups related to disease and patient condition, we hypothesized that outcome differences would persist even after adjusting for patient factors such as demographics, disease‐specific risk of mortality, and ICU utilization.

PATIENTS AND METHODS

We conducted a retrospective cohort study using data from the University HealthSystem Consortium (UHC) Clinical Database and Resource Manager (CDB/RM). UHC is an alliance of 120 academic medical centers and 300 of their affiliated hospitals for the purposes of collaboration on performance improvement. Each year, a subset of participating hospitals submits data on all of their inpatient discharges to the CDB/RM, which totals approximately 5 million records. The CDB/RM includes information from billing forms including demographics, diagnoses, and procedures as captured by International Classification of Diseases, Ninth Revision (ICD‐9) codes, discharge disposition, and line item charge detail for the type of bed (eg, floor, ICU). Most hospitals also provide detailed charge information including pharmacy, imaging, blood products, lab tests, and supplies. Some hospitals do not provide any charge data. The Beth Israel Deaconess Medical Center and University of Washington institutional review boards reviewed and approved the conduct of this study.

We included all inpatients discharged by hospitalists or general internal medicine physicians from UHC hospitals between April 1, 2011 and March 31, 2012. We excluded minors, pregnant patients, and prisoners. One hundred fifty‐eight adult academic medical centers and affiliated hospitals submitted data throughout this time period. Our primary independent variable, IHT status, was defined by patients whose admission source was another acute care institution. ED admissions were defined as patients admitted from the AHS ED whose source of origination was not another hospital or ambulatory surgery site.

RESULTS

DISCUSSION

Our study of IHT patients ultimately discharged by hospitalists and general internists at US academic referral centers found significantly increased average LOS, costs, and in‐hospital mortality compared with patients admitted from the ED. The increased risk of mortality persisted after adjustment for patient characteristics and variables representing endogenous risk of mortality, and in more homogeneous subgroups after stratification by presence and timing of ICU utilization. These data confirm findings from single‐center studies and suggest that observations about the difference between IHT and ED populations may be generalizable across US academic hospitals.

Our work builds on 2 single‐center studies that examined mixed medical and surgical academic IHT populations from the late 1980s and early 1990s,[9, 10] and 1 studying surgical ICU patients in 2013.[17] These studies demonstrated longer average LOS, higher costs, and higher mortality rates (in both adjusted and unadjusted analyses). Our work confirmed these findings utilizing a more current, multicenter large dataset of IHT patients ultimately discharged by hospitalists and general internists. Our work is unique from a larger, more recent study[7] in that it focuses on patients transferred to academic health systems, and therefore has particular relevance to those settings. In addition, we divided patients into subpopulations based on the timing of ICU utilization, and found that in each of these populations, IHT remained independently associated with in‐hospital mortality.

Our analysis does not explain why the outcomes of IHTs are worse, but plausible contributing factors include that (1) patients chosen for IHT are at higher risk of death in ways uncaptured by established mortality risk scores, (2) referring, transferring, or accepting providers and institutions have provided inadequate care, (3) the transfer process itself involves harm, (4) socioeconomic bias in selection for IHT,[18] or (5) some combination of the above. Regardless of the causes of the worse outcomes observed in these outside‐hospital transfers, as these patients are colloquially known at accepting hospitals, they present challenges to everyone involved. Referring providers may feel a sense of urgency as these patients' needs exceed their management capabilities. The process is often time consuming and burdensome for referring and accepting providers because of poorly developed systems.[19] The transfer often takes patients further from their home and may make it more difficult for family to participate in their care. The transfer may delay care if the accepting institution cannot immediately accept the patient or if the time in transport is prolonged, which could result in decompensation at a critical juncture. For providers inheriting such patients, the stress of caring for these patients is compounded by the difficulty obtaining records about the prior hospitalization.[20] This frustrating experience is often translated into unfounded judgment of the institution that referred the patient and the care provided there.[21] It is important for hospitalists making decisions throughout the transfer process and for hospital leaders who determine staffing levels, measure the quality of care, manage hospital networks, or write hospital policy to appreciate that the transfer process itself may contribute to the challenges and poor outcomes we observe. Furthermore, regardless of the cause for the increased mortality that we observed, our findings imply that IHT patients require careful evaluation, management, and treatment.

Many accepting institutions have transfer centers that facilitate these transitions, utilizing protocols and templates to standardize the process.[22, 23] Future research should focus on the characteristics of these centers to learn which practices are most efficacious. Interventions to mitigate the known challenges of transfer (including patient selection and triage, handoff communication, and information sharing) could be tested by randomized studies at referring and accepting institutions. There may be a role for health information exchange or the development of enhanced pretransfer evaluation processes using telemedicine models; there is evidence that information sharing may reduce redundant imaging.[24] Perhaps targeted review of IHTs admitted to a non‐ICU portion of the hospital and subsequently transferred to the ICU could identify opportunities to improve triaging protocols and thus avert some of the bad outcomes observed in this subpopulation. A related future direction could be to create protected forumsusing the patient safety organization framework[25]to facilitate the discussion of interhospital transfer outcomes among the referring, transporting, and receiving parties. Lastly, future work should investigate the reasons for the different proportions of black patients in the ED versus IHT cohorts. Our finding that black race was associated with lower risk of mortality has been previously reported but may also benefit from more investigation.[26]

There are several limitations of our work. First, despite extensive adjustment for patient characteristics, due to the observational nature of our study it is still possible that IHTs differ from ED admissions in ways that were unaccounted for in our analysis, and which could be associated with increased mortality independent of the transfer process itself. We are unable to characterize features of the transfer process, such as the reason for transfer, differences in transfer processes among hospitals, or the distance and mode of travel, which may influence outcomes.[27] Because we used administrative data, variations in coding could incorrectly estimate the complexity or severity of illness on admission, which is a previously described risk.[28] In addition, although our dataset was very large, it was limited by incomplete charge data, which limited our ability to measure ICU utilization in our full cohort. The hospitals missing ICU charge data are of variable sizes and are distributed around the country, limiting the chance of systematic bias. Finally, in some settings, hospitalists may serve as the discharging physician for patients admitted to other services such as the ICU, introducing heterogeneity and bias to the sample. We attempted to mitigate such bias through our subgroup analysis, which allowed for comparisons within more homogeneous patient groupings.

In conclusion, our large multicenter study of academic health systems confirms the findings of prior single‐center academic studies and a large general population study that interhospital transfer patients have an increased average LOS, costs, and adjusted in‐hospital mortality than patients admitted from the ED. This difference in mortality persisted even after controlling for several other predictors of mortality. Our findings emphasize the need for future studies designed to clarify the reason for the increased risk and identify targets for interventions to improve outcomes for the interhospital transfer population.

Interhospital transfers (IHTs) to academic medical centers (AMCs) or their affiliated hospitals may benefit patients who require unique specialty and procedural services. However, IHTs also introduce a potentially risky transition of care for patients suffering from complex or unstable medical problems.[1] Components of this risk include the dangers associated with transportation and the disrupted continuity of care that may lead to delays or errors in care.[2, 3] Furthermore, referring and accepting providers may face barriers to optimal handoffs including a lack of shared communication standards and difficulty accessing external medical records.[3, 4, 5] Although some authors have recommended the creation of formal guidelines for interhospital transfer processes for all patients to mitigate the risks of transfer, the available guidelines governing the IHT triage and communication process are limited to critically ill patients.[6]

A recent study of a diverse patient and hospital dataset demonstrated that interhospital transfer patients have a higher risk of mortality, increased length of stay (LOS), and increased risk of adverse events as compared with non‐transfer patients.[7] However, it is unknown if these findings persist in the population of patients transferred specifically to AMCs or their affiliated hospitals (the combination is hereafter referred to as academic health systems [AHSs]). AMCs provide a disproportionate share of IHT care for complex patients and have a vested interest in improving the outcomes of these transitions.[8] Prior single‐center studies of acute care adult medical patients accepted to AMCs have shown that IHT is associated with a longer LOS, increased in‐hospital mortality, and higher resource use.[9, 10] However, it is difficult to generalize from single‐center studies due to the variation in referral practices, geography, and network characteristics. Additionally, AMC referral systems, patient mix, and utilization of hospitalists have likely changed substantially in the nearly 2 decades since those reports were published.

Hospitalists and general internists often manage the transfer acceptance processes for internal medicine services at receiving hospitals, helping to triage and coordinate care for IHT patients. As a result, it is important for hospitalists to understand the characteristics and outcomes of the IHT population. In addition to informing the decision making around transfer for a given patient, such an understanding is the foundation for helping providers and institutions begin to systematically identify and mitigate peritransfer risks.

We conducted this large multicenter study to describe the characteristics and outcomes of a current, nationally representative IHT patient population discharged by hospitalists and general internists at AHSs. To identify unique features of the IHT population, we compared patients transferred from another hospital to an AHS to those admitted to the AHS directly from the AHS's emergency department (ED). Based on our anecdotal experiences and the prior single‐center study findings in adult medical populations,[9, 10] we hypothesized that the IHT population would be sicker, stay in the hospital and intensive care unit (ICU) longer, and have higher costs and in‐hospital mortality than ED patients. Although there may be fundamental differences between the 2 groups related to disease and patient condition, we hypothesized that outcome differences would persist even after adjusting for patient factors such as demographics, disease‐specific risk of mortality, and ICU utilization.

PATIENTS AND METHODS

We conducted a retrospective cohort study using data from the University HealthSystem Consortium (UHC) Clinical Database and Resource Manager (CDB/RM). UHC is an alliance of 120 academic medical centers and 300 of their affiliated hospitals for the purposes of collaboration on performance improvement. Each year, a subset of participating hospitals submits data on all of their inpatient discharges to the CDB/RM, which totals approximately 5 million records. The CDB/RM includes information from billing forms including demographics, diagnoses, and procedures as captured by International Classification of Diseases, Ninth Revision (ICD‐9) codes, discharge disposition, and line item charge detail for the type of bed (eg, floor, ICU). Most hospitals also provide detailed charge information including pharmacy, imaging, blood products, lab tests, and supplies. Some hospitals do not provide any charge data. The Beth Israel Deaconess Medical Center and University of Washington institutional review boards reviewed and approved the conduct of this study.

We included all inpatients discharged by hospitalists or general internal medicine physicians from UHC hospitals between April 1, 2011 and March 31, 2012. We excluded minors, pregnant patients, and prisoners. One hundred fifty‐eight adult academic medical centers and affiliated hospitals submitted data throughout this time period. Our primary independent variable, IHT status, was defined by patients whose admission source was another acute care institution. ED admissions were defined as patients admitted from the AHS ED whose source of origination was not another hospital or ambulatory surgery site.

RESULTS

DISCUSSION

Our study of IHT patients ultimately discharged by hospitalists and general internists at US academic referral centers found significantly increased average LOS, costs, and in‐hospital mortality compared with patients admitted from the ED. The increased risk of mortality persisted after adjustment for patient characteristics and variables representing endogenous risk of mortality, and in more homogeneous subgroups after stratification by presence and timing of ICU utilization. These data confirm findings from single‐center studies and suggest that observations about the difference between IHT and ED populations may be generalizable across US academic hospitals.

Our work builds on 2 single‐center studies that examined mixed medical and surgical academic IHT populations from the late 1980s and early 1990s,[9, 10] and 1 studying surgical ICU patients in 2013.[17] These studies demonstrated longer average LOS, higher costs, and higher mortality rates (in both adjusted and unadjusted analyses). Our work confirmed these findings utilizing a more current, multicenter large dataset of IHT patients ultimately discharged by hospitalists and general internists. Our work is unique from a larger, more recent study[7] in that it focuses on patients transferred to academic health systems, and therefore has particular relevance to those settings. In addition, we divided patients into subpopulations based on the timing of ICU utilization, and found that in each of these populations, IHT remained independently associated with in‐hospital mortality.

Our analysis does not explain why the outcomes of IHTs are worse, but plausible contributing factors include that (1) patients chosen for IHT are at higher risk of death in ways uncaptured by established mortality risk scores, (2) referring, transferring, or accepting providers and institutions have provided inadequate care, (3) the transfer process itself involves harm, (4) socioeconomic bias in selection for IHT,[18] or (5) some combination of the above. Regardless of the causes of the worse outcomes observed in these outside‐hospital transfers, as these patients are colloquially known at accepting hospitals, they present challenges to everyone involved. Referring providers may feel a sense of urgency as these patients' needs exceed their management capabilities. The process is often time consuming and burdensome for referring and accepting providers because of poorly developed systems.[19] The transfer often takes patients further from their home and may make it more difficult for family to participate in their care. The transfer may delay care if the accepting institution cannot immediately accept the patient or if the time in transport is prolonged, which could result in decompensation at a critical juncture. For providers inheriting such patients, the stress of caring for these patients is compounded by the difficulty obtaining records about the prior hospitalization.[20] This frustrating experience is often translated into unfounded judgment of the institution that referred the patient and the care provided there.[21] It is important for hospitalists making decisions throughout the transfer process and for hospital leaders who determine staffing levels, measure the quality of care, manage hospital networks, or write hospital policy to appreciate that the transfer process itself may contribute to the challenges and poor outcomes we observe. Furthermore, regardless of the cause for the increased mortality that we observed, our findings imply that IHT patients require careful evaluation, management, and treatment.

Many accepting institutions have transfer centers that facilitate these transitions, utilizing protocols and templates to standardize the process.[22, 23] Future research should focus on the characteristics of these centers to learn which practices are most efficacious. Interventions to mitigate the known challenges of transfer (including patient selection and triage, handoff communication, and information sharing) could be tested by randomized studies at referring and accepting institutions. There may be a role for health information exchange or the development of enhanced pretransfer evaluation processes using telemedicine models; there is evidence that information sharing may reduce redundant imaging.[24] Perhaps targeted review of IHTs admitted to a non‐ICU portion of the hospital and subsequently transferred to the ICU could identify opportunities to improve triaging protocols and thus avert some of the bad outcomes observed in this subpopulation. A related future direction could be to create protected forumsusing the patient safety organization framework[25]to facilitate the discussion of interhospital transfer outcomes among the referring, transporting, and receiving parties. Lastly, future work should investigate the reasons for the different proportions of black patients in the ED versus IHT cohorts. Our finding that black race was associated with lower risk of mortality has been previously reported but may also benefit from more investigation.[26]

There are several limitations of our work. First, despite extensive adjustment for patient characteristics, due to the observational nature of our study it is still possible that IHTs differ from ED admissions in ways that were unaccounted for in our analysis, and which could be associated with increased mortality independent of the transfer process itself. We are unable to characterize features of the transfer process, such as the reason for transfer, differences in transfer processes among hospitals, or the distance and mode of travel, which may influence outcomes.[27] Because we used administrative data, variations in coding could incorrectly estimate the complexity or severity of illness on admission, which is a previously described risk.[28] In addition, although our dataset was very large, it was limited by incomplete charge data, which limited our ability to measure ICU utilization in our full cohort. The hospitals missing ICU charge data are of variable sizes and are distributed around the country, limiting the chance of systematic bias. Finally, in some settings, hospitalists may serve as the discharging physician for patients admitted to other services such as the ICU, introducing heterogeneity and bias to the sample. We attempted to mitigate such bias through our subgroup analysis, which allowed for comparisons within more homogeneous patient groupings.

In conclusion, our large multicenter study of academic health systems confirms the findings of prior single‐center academic studies and a large general population study that interhospital transfer patients have an increased average LOS, costs, and adjusted in‐hospital mortality than patients admitted from the ED. This difference in mortality persisted even after controlling for several other predictors of mortality. Our findings emphasize the need for future studies designed to clarify the reason for the increased risk and identify targets for interventions to improve outcomes for the interhospital transfer population.

Interhospital transfers (IHTs) to academic medical centers (AMCs) or their affiliated hospitals may benefit patients who require unique specialty and procedural services. However, IHTs also introduce a potentially risky transition of care for patients suffering from complex or unstable medical problems.[1] Components of this risk include the dangers associated with transportation and the disrupted continuity of care that may lead to delays or errors in care.[2, 3] Furthermore, referring and accepting providers may face barriers to optimal handoffs including a lack of shared communication standards and difficulty accessing external medical records.[3, 4, 5] Although some authors have recommended the creation of formal guidelines for interhospital transfer processes for all patients to mitigate the risks of transfer, the available guidelines governing the IHT triage and communication process are limited to critically ill patients.[6]

A recent study of a diverse patient and hospital dataset demonstrated that interhospital transfer patients have a higher risk of mortality, increased length of stay (LOS), and increased risk of adverse events as compared with non‐transfer patients.[7] However, it is unknown if these findings persist in the population of patients transferred specifically to AMCs or their affiliated hospitals (the combination is hereafter referred to as academic health systems [AHSs]). AMCs provide a disproportionate share of IHT care for complex patients and have a vested interest in improving the outcomes of these transitions.[8] Prior single‐center studies of acute care adult medical patients accepted to AMCs have shown that IHT is associated with a longer LOS, increased in‐hospital mortality, and higher resource use.[9, 10] However, it is difficult to generalize from single‐center studies due to the variation in referral practices, geography, and network characteristics. Additionally, AMC referral systems, patient mix, and utilization of hospitalists have likely changed substantially in the nearly 2 decades since those reports were published.

Hospitalists and general internists often manage the transfer acceptance processes for internal medicine services at receiving hospitals, helping to triage and coordinate care for IHT patients. As a result, it is important for hospitalists to understand the characteristics and outcomes of the IHT population. In addition to informing the decision making around transfer for a given patient, such an understanding is the foundation for helping providers and institutions begin to systematically identify and mitigate peritransfer risks.

We conducted this large multicenter study to describe the characteristics and outcomes of a current, nationally representative IHT patient population discharged by hospitalists and general internists at AHSs. To identify unique features of the IHT population, we compared patients transferred from another hospital to an AHS to those admitted to the AHS directly from the AHS's emergency department (ED). Based on our anecdotal experiences and the prior single‐center study findings in adult medical populations,[9, 10] we hypothesized that the IHT population would be sicker, stay in the hospital and intensive care unit (ICU) longer, and have higher costs and in‐hospital mortality than ED patients. Although there may be fundamental differences between the 2 groups related to disease and patient condition, we hypothesized that outcome differences would persist even after adjusting for patient factors such as demographics, disease‐specific risk of mortality, and ICU utilization.

PATIENTS AND METHODS

We conducted a retrospective cohort study using data from the University HealthSystem Consortium (UHC) Clinical Database and Resource Manager (CDB/RM). UHC is an alliance of 120 academic medical centers and 300 of their affiliated hospitals for the purposes of collaboration on performance improvement. Each year, a subset of participating hospitals submits data on all of their inpatient discharges to the CDB/RM, which totals approximately 5 million records. The CDB/RM includes information from billing forms including demographics, diagnoses, and procedures as captured by International Classification of Diseases, Ninth Revision (ICD‐9) codes, discharge disposition, and line item charge detail for the type of bed (eg, floor, ICU). Most hospitals also provide detailed charge information including pharmacy, imaging, blood products, lab tests, and supplies. Some hospitals do not provide any charge data. The Beth Israel Deaconess Medical Center and University of Washington institutional review boards reviewed and approved the conduct of this study.

We included all inpatients discharged by hospitalists or general internal medicine physicians from UHC hospitals between April 1, 2011 and March 31, 2012. We excluded minors, pregnant patients, and prisoners. One hundred fifty‐eight adult academic medical centers and affiliated hospitals submitted data throughout this time period. Our primary independent variable, IHT status, was defined by patients whose admission source was another acute care institution. ED admissions were defined as patients admitted from the AHS ED whose source of origination was not another hospital or ambulatory surgery site.

RESULTS

DISCUSSION

Our study of IHT patients ultimately discharged by hospitalists and general internists at US academic referral centers found significantly increased average LOS, costs, and in‐hospital mortality compared with patients admitted from the ED. The increased risk of mortality persisted after adjustment for patient characteristics and variables representing endogenous risk of mortality, and in more homogeneous subgroups after stratification by presence and timing of ICU utilization. These data confirm findings from single‐center studies and suggest that observations about the difference between IHT and ED populations may be generalizable across US academic hospitals.

Our work builds on 2 single‐center studies that examined mixed medical and surgical academic IHT populations from the late 1980s and early 1990s,[9, 10] and 1 studying surgical ICU patients in 2013.[17] These studies demonstrated longer average LOS, higher costs, and higher mortality rates (in both adjusted and unadjusted analyses). Our work confirmed these findings utilizing a more current, multicenter large dataset of IHT patients ultimately discharged by hospitalists and general internists. Our work is unique from a larger, more recent study[7] in that it focuses on patients transferred to academic health systems, and therefore has particular relevance to those settings. In addition, we divided patients into subpopulations based on the timing of ICU utilization, and found that in each of these populations, IHT remained independently associated with in‐hospital mortality.

Our analysis does not explain why the outcomes of IHTs are worse, but plausible contributing factors include that (1) patients chosen for IHT are at higher risk of death in ways uncaptured by established mortality risk scores, (2) referring, transferring, or accepting providers and institutions have provided inadequate care, (3) the transfer process itself involves harm, (4) socioeconomic bias in selection for IHT,[18] or (5) some combination of the above. Regardless of the causes of the worse outcomes observed in these outside‐hospital transfers, as these patients are colloquially known at accepting hospitals, they present challenges to everyone involved. Referring providers may feel a sense of urgency as these patients' needs exceed their management capabilities. The process is often time consuming and burdensome for referring and accepting providers because of poorly developed systems.[19] The transfer often takes patients further from their home and may make it more difficult for family to participate in their care. The transfer may delay care if the accepting institution cannot immediately accept the patient or if the time in transport is prolonged, which could result in decompensation at a critical juncture. For providers inheriting such patients, the stress of caring for these patients is compounded by the difficulty obtaining records about the prior hospitalization.[20] This frustrating experience is often translated into unfounded judgment of the institution that referred the patient and the care provided there.[21] It is important for hospitalists making decisions throughout the transfer process and for hospital leaders who determine staffing levels, measure the quality of care, manage hospital networks, or write hospital policy to appreciate that the transfer process itself may contribute to the challenges and poor outcomes we observe. Furthermore, regardless of the cause for the increased mortality that we observed, our findings imply that IHT patients require careful evaluation, management, and treatment.

Many accepting institutions have transfer centers that facilitate these transitions, utilizing protocols and templates to standardize the process.[22, 23] Future research should focus on the characteristics of these centers to learn which practices are most efficacious. Interventions to mitigate the known challenges of transfer (including patient selection and triage, handoff communication, and information sharing) could be tested by randomized studies at referring and accepting institutions. There may be a role for health information exchange or the development of enhanced pretransfer evaluation processes using telemedicine models; there is evidence that information sharing may reduce redundant imaging.[24] Perhaps targeted review of IHTs admitted to a non‐ICU portion of the hospital and subsequently transferred to the ICU could identify opportunities to improve triaging protocols and thus avert some of the bad outcomes observed in this subpopulation. A related future direction could be to create protected forumsusing the patient safety organization framework[25]to facilitate the discussion of interhospital transfer outcomes among the referring, transporting, and receiving parties. Lastly, future work should investigate the reasons for the different proportions of black patients in the ED versus IHT cohorts. Our finding that black race was associated with lower risk of mortality has been previously reported but may also benefit from more investigation.[26]

There are several limitations of our work. First, despite extensive adjustment for patient characteristics, due to the observational nature of our study it is still possible that IHTs differ from ED admissions in ways that were unaccounted for in our analysis, and which could be associated with increased mortality independent of the transfer process itself. We are unable to characterize features of the transfer process, such as the reason for transfer, differences in transfer processes among hospitals, or the distance and mode of travel, which may influence outcomes.[27] Because we used administrative data, variations in coding could incorrectly estimate the complexity or severity of illness on admission, which is a previously described risk.[28] In addition, although our dataset was very large, it was limited by incomplete charge data, which limited our ability to measure ICU utilization in our full cohort. The hospitals missing ICU charge data are of variable sizes and are distributed around the country, limiting the chance of systematic bias. Finally, in some settings, hospitalists may serve as the discharging physician for patients admitted to other services such as the ICU, introducing heterogeneity and bias to the sample. We attempted to mitigate such bias through our subgroup analysis, which allowed for comparisons within more homogeneous patient groupings.

In conclusion, our large multicenter study of academic health systems confirms the findings of prior single‐center academic studies and a large general population study that interhospital transfer patients have an increased average LOS, costs, and adjusted in‐hospital mortality than patients admitted from the ED. This difference in mortality persisted even after controlling for several other predictors of mortality. Our findings emphasize the need for future studies designed to clarify the reason for the increased risk and identify targets for interventions to improve outcomes for the interhospital transfer population.

Cirrhosis affects up to 3% of the population and is 1 of the 10 most common causes of death in the United States.[1, 2, 3, 4] Paracentesis procedures are frequently performed in patients with liver disease and ascites for diagnostic and/or therapeutic purposes. These procedures can be performed safely by trained clinicians at the bedside or referred to interventional radiology (IR).[2, 3, 4]

National practice patterns show that paracentesis procedures are increasingly referred to IR rather than performed at the bedside by internal medicine or gastroenterology clinicians.[5, 6, 7] In fact, a recent study of Medicare beneficiaries showed that inpatient and outpatient paracentesis procedures performed by radiologists increased by 964% from 1993 to 2008.[7] Reasons for the decline in bedside procedures include the increased availability of IR, lack of sufficient reimbursement, and the time required to perform paracentesis procedures.[5, 6, 7, 8] Surveys of internal medicine and family medicine residents and gastroenterology fellows show trainees often lack the confidence and experience needed to perform the procedure safely.[9, 10, 11] Additionally, many clinicians do not have expertise with ultrasound use and may not have access to necessary equipment.

Inconsistent certification requirements may also impact the competence and experience of physicians to perform paracentesis procedures. Internal medicine residents are no longer required by the American Board of Internal Medicine (ABIM) to demonstrate competency in procedures such as paracentesis for certification.[12] However, the Accreditation Council for Graduate Medical Education (ACGME) requirements state that internal medicine programs must offer residents the opportunity to demonstrate competence in the performance of procedures such as paracentesis, thoracentesis, and central venous catheter insertion.[13] The American Board of Family Medicine (ABFM) does not outline specific procedural competence for initial certification.[14] The ACGME states that family medicine residents must receive training to perform those clinical procedures required for their future practices but allows each program to determine which procedures to require.[15] Due to this uncertainty, practicing hospitalists are likely to have variable training and competence in bedside procedures such as paracentesis.

We previously showed that internal medicine residents rotating on the hepatology service of an academic medical center performed 59% of paracentesis procedures at the bedside.[16] These findings are in contrast to national data showing that 74% of paracentesis procedures performed on Medicare beneficiaries were performed by radiologists.[7] Practice patterns at university hospitals may not be reflected in this data because the study was limited to Medicare beneficiaries and included ambulatory patients.[7] In addition to uncertainty about who is performing this procedure in inpatient settings, little is known about the effect of specialty on postparacentesis clinical outcomes.[16, 17]

The current study had 3 aims: (1) evaluate which clinical specialties perform paracentesis procedures at university hospitals; (2) model patient characteristics associated with procedures performed at the bedside versus those referred to IR; and (3) among patients with a similar likelihood of IR referral, evaluate length of stay (LOS) and hospital costs of patients undergoing procedures performed by different specialties.

METHODS

We performed an observational administrative database review of patients who underwent paracentesis procedures in hospitals participating in the University HealthSystem Consortium (UHC) Clinical Database from January 2010 through December 2012. UHC is an alliance of 120 nonprofit academic medical centers and their 290 affiliated hospitals. UHC maintains databases containing clinical, operational, financial, and patient safety data from affiliated hospitals. Using the UHC database, we described the characteristics of all patients who underwent paracentesis procedures by clinical specialty performing the procedure. We then modeled the effects of patient characteristics on decision‐making about IR referral. Finally, among patients with a homogeneous predicted probability of IR referral, we compared LOS and direct costs by specialty performing the procedure. The Northwestern University institutional review board approved this study.

RESULTS

Predicted Probabilities of IR Referral

Figure 1 presents the distribution of predicted probabilities for IR referral. Predicted probabilities were low overall, with very few patients having an equal chance of referralthe standard often used in comparative effectiveness analyses from observational data. Figure 1 indicates that IR referral probabilities were clustered in an unusual bimodal distribution. The cluster on the left, which centers around a 15% predicted probability of IR referral, consists of discharges with patient characteristics that were associated with a very low chance of an IR paracentesis. We therefore used this distribution to conduct comparative analyses of admission outcomes between clinical specialty groups, choosing to examine patients with a 20% or greater chance of IR referral.

Figure 1

Post hoc analysis revealed that the biggest factor driving low predicted probability of IR referral was whether patients experienced an ICU stay at any time during hospitalization. Among the discharges with a predicted probability 0.2 (n=26,615 discharges), there were only 87 discharges with ICU stays (0.3%). For the discharges with predicted probability <0.2 (n=7676), 91.9% (n=7055) had an ICU admission. We therefore used a threshold of 0.2 or greater to present the most comparable LOS and direct cost differences.

DISCUSSION

This study showed that internal medicine‐ and family medicine‐trained clinicians perform approximately half of the inpatient paracentesis procedures at university hospitals and their affiliates. This confirms findings from our earlier single‐institution study[16] but contrasts with previously published reports involving Medicare data. The earlier report, using Medicare claims and including ambulatory procedures, revealed that primary care physicians and gastroenterologists only performed approximately 10% of US paracentesis procedures in 2008.[7] Our findings suggest that practices are different at university hospitals, where patients with severe liver disease often seek care. Because we used the UHC database, it was not possible to determine if the clinicians who performed paracentesis procedures in this study were internal medicine or family medicine residents, fellows, or attending physicians. However, findings from our own institution show that the vast majority of bedside paracentesis procedures are performed by internal medicine residents.[16]

Our findings have implications for certification of internal medicine and family medicine trainees. In 2008, the ABIM removed the requirement that internal medicine residents demonstrate competency in paracentesis.[12] This decision was informed by a lack of standardized methods to determine procedural competency and published surveys showing that internal medicine and hospitalist physicians rarely performed bedside procedures.[5, 6] Despite this policy change, our findings show that current clinical practice at university hospitals does not reflect national practice patterns or certification requirements, because many internal medicine‐ and family medicine‐trained clinicians still perform paracentesis procedures. This is concerning because internal medicine and family medicine trainees report variable confidence, experience, expertise, and supervision regarding performance of invasive procedures.[9, 10, 21, 22, 23, 24] Furthermore, earlier research also demonstrates that graduating residents and fellows are not able to competently perform common bedside procedures such as thoracentesis, temporary hemodialysis catheter insertion, and lumbar puncture.[25, 26, 27]

The American Association for the Study of Liver Diseases (AASLD) recommends that trained clinicians perform paracentesis procedures.[3, 4] However, the AASLD provides no definition for how training should occur. Because competency in this procedure is not specifically required by the ABIM, ABFM, or ACGME, a paradoxical situation occurs in which internal medicine and family medicine residents, and internal medicine‐trained fellows and faculty continue to perform paracentesis procedures on highly complex patients, but are no longer required to be competent to do so.

In earlier research we showed that simulation‐based mastery learning (SBML) was an effective method to boost internal medicine residents' paracentesis skills.[28] In SBML, all trainees must meet or exceed a minimum passing score on a simulated procedure before performing one on an actual patient.[29] This approach improves clinical care and outcomes in procedures such as central venous catheter insertion[30, 31] and advanced cardiac life support.[32] SBML‐trained residents also performed safe paracentesis procedures with shorter hospital LOS, fewer ICU transfers, and fewer blood product transfusions than IR procedures.[16] Based on the results of this study, AASLD guidelines regarding training, and our experience with SBML, we recommend that all clinicians complete paracentesis SBML training before performing procedures on patients.

Using our propensity model we identified patient characteristics that were associated with IR referral. Patients with a liver transplant were more likely to be cared for in IR. This may be due to a belief that postoperative procedures are anatomically more complex or because surgical trainees do not commonly perform this procedure. The current study confirms findings from earlier work that obese and female patients are more likely to be referred to IR.[16] IR referral of obese patients is likely to occur because paracentesis procedures are technically more difficult. We have no explanation why female patients were more likely to be referred to IR, because most decisions appear to be discretionary. Prospective studies are needed to determine evidence‐based recommendations regarding paracentesis procedure location. Patients with more comorbidities (eg, ICU stay, awaiting liver transplant, coagulation disorders) were more likely to undergo bedside procedures. The complexity of patients undergoing bedside paracentesis procedures reinforces the need for rigorous skill assessment for clinicians who perform them because complications such as intraperitoneal bleeding can be fatal.

Finally, we showed that LOS was similar but hospital direct costs were $800 to $1300 lower for patients whose paracentesis procedure was performed by medicine or gastroenterology/hepatology compared to IR. Medical subspecialties and surgery procedures were more expensive than IR, consistent with the higher LOS seen in these groups. IR procedures add costs due to facility charges for space, personnel, and equipment.[33] At our institution, the hospital cost of an IR paracentesis in 2012 was $361. If we use this figure, and assume costs are similar across university hospitals, the resultant cost savings would be $10,257,454 (for the procedure alone) if all procedures referred to IR in this 2‐year study were instead performed at the bedside. This estimate is approximate because it does not consider factors such as cost of clinician staffing models, which may differ across UHC hospitals. As hospitals look to reduce costs, potential savings due to appropriate use of bedside and IR procedures should be considered. This is especially important because there is no evidence that the extra expense of IR procedures is justified due to improved patient outcomes.

This study has several limitations. First, this was an observational study. Although the database was large, we were limited by coding accuracy and could not control for all potential confounding factors such as Model for End‐Stage Liver Disease score,[34, 35] other specific laboratory values, amount of ascites fluid removed, or bedside procedure failures later referred to IR. However, we do know that only a small number of second, third, or fourth procedures were subsequently referred to IR after earlier ones were performed at the bedside. Additionally the UHC database does not include patient‐specific data, and therefore we could not adjust for multiple visits or procedures by the same patient. Second, we were unable to determine the level of teaching involvement at each UHC affiliated hospital. Community hospitals where attendings managed most of the patients without trainees could not be differentiated from university hospitals where trainees were involved in most patients' care. Third, we did not have specialty information for 9434 (9.7%) procedures and had to exclude these cases. We also excluded a large number of paracentesis procedures in our final outcomes analysis. However, this was necessary because we needed to perform a patient‐level analysis to ensure the propensity and outcomes models were accurate. Finally, we did not evaluate inpatient mortality or 30‐day hospital readmission rates. Mortality and readmission from complications of a paracentesis procedure are rare events.[3, 4, 36] However, mortality and hospital readmission among patients with liver disease are relatively common.[37, 38] It was impossible to link these outcomes to a paracentesis procedure without the ability to perform medical records review.

In conclusion, paracentesis procedures are performed frequently by internal medicine‐ and family medicine‐trained clinicians in university hospitals. Because of these findings regarding current practice patterns, we believe the ACGME, ABIM, and ABFM should clarify their policies to require that residents are competent to perform paracentesis procedures before performing them on patients. This may improve supervision and training for paracentesis procedures that are already occurring and possibly encourage performance of additional, less costly bedside procedures.

Cirrhosis affects up to 3% of the population and is 1 of the 10 most common causes of death in the United States.[1, 2, 3, 4] Paracentesis procedures are frequently performed in patients with liver disease and ascites for diagnostic and/or therapeutic purposes. These procedures can be performed safely by trained clinicians at the bedside or referred to interventional radiology (IR).[2, 3, 4]

National practice patterns show that paracentesis procedures are increasingly referred to IR rather than performed at the bedside by internal medicine or gastroenterology clinicians.[5, 6, 7] In fact, a recent study of Medicare beneficiaries showed that inpatient and outpatient paracentesis procedures performed by radiologists increased by 964% from 1993 to 2008.[7] Reasons for the decline in bedside procedures include the increased availability of IR, lack of sufficient reimbursement, and the time required to perform paracentesis procedures.[5, 6, 7, 8] Surveys of internal medicine and family medicine residents and gastroenterology fellows show trainees often lack the confidence and experience needed to perform the procedure safely.[9, 10, 11] Additionally, many clinicians do not have expertise with ultrasound use and may not have access to necessary equipment.

Inconsistent certification requirements may also impact the competence and experience of physicians to perform paracentesis procedures. Internal medicine residents are no longer required by the American Board of Internal Medicine (ABIM) to demonstrate competency in procedures such as paracentesis for certification.[12] However, the Accreditation Council for Graduate Medical Education (ACGME) requirements state that internal medicine programs must offer residents the opportunity to demonstrate competence in the performance of procedures such as paracentesis, thoracentesis, and central venous catheter insertion.[13] The American Board of Family Medicine (ABFM) does not outline specific procedural competence for initial certification.[14] The ACGME states that family medicine residents must receive training to perform those clinical procedures required for their future practices but allows each program to determine which procedures to require.[15] Due to this uncertainty, practicing hospitalists are likely to have variable training and competence in bedside procedures such as paracentesis.

We previously showed that internal medicine residents rotating on the hepatology service of an academic medical center performed 59% of paracentesis procedures at the bedside.[16] These findings are in contrast to national data showing that 74% of paracentesis procedures performed on Medicare beneficiaries were performed by radiologists.[7] Practice patterns at university hospitals may not be reflected in this data because the study was limited to Medicare beneficiaries and included ambulatory patients.[7] In addition to uncertainty about who is performing this procedure in inpatient settings, little is known about the effect of specialty on postparacentesis clinical outcomes.[16, 17]

The current study had 3 aims: (1) evaluate which clinical specialties perform paracentesis procedures at university hospitals; (2) model patient characteristics associated with procedures performed at the bedside versus those referred to IR; and (3) among patients with a similar likelihood of IR referral, evaluate length of stay (LOS) and hospital costs of patients undergoing procedures performed by different specialties.

METHODS

We performed an observational administrative database review of patients who underwent paracentesis procedures in hospitals participating in the University HealthSystem Consortium (UHC) Clinical Database from January 2010 through December 2012. UHC is an alliance of 120 nonprofit academic medical centers and their 290 affiliated hospitals. UHC maintains databases containing clinical, operational, financial, and patient safety data from affiliated hospitals. Using the UHC database, we described the characteristics of all patients who underwent paracentesis procedures by clinical specialty performing the procedure. We then modeled the effects of patient characteristics on decision‐making about IR referral. Finally, among patients with a homogeneous predicted probability of IR referral, we compared LOS and direct costs by specialty performing the procedure. The Northwestern University institutional review board approved this study.

RESULTS

Predicted Probabilities of IR Referral

Figure 1 presents the distribution of predicted probabilities for IR referral. Predicted probabilities were low overall, with very few patients having an equal chance of referralthe standard often used in comparative effectiveness analyses from observational data. Figure 1 indicates that IR referral probabilities were clustered in an unusual bimodal distribution. The cluster on the left, which centers around a 15% predicted probability of IR referral, consists of discharges with patient characteristics that were associated with a very low chance of an IR paracentesis. We therefore used this distribution to conduct comparative analyses of admission outcomes between clinical specialty groups, choosing to examine patients with a 20% or greater chance of IR referral.

Figure 1

Post hoc analysis revealed that the biggest factor driving low predicted probability of IR referral was whether patients experienced an ICU stay at any time during hospitalization. Among the discharges with a predicted probability 0.2 (n=26,615 discharges), there were only 87 discharges with ICU stays (0.3%). For the discharges with predicted probability <0.2 (n=7676), 91.9% (n=7055) had an ICU admission. We therefore used a threshold of 0.2 or greater to present the most comparable LOS and direct cost differences.

DISCUSSION

This study showed that internal medicine‐ and family medicine‐trained clinicians perform approximately half of the inpatient paracentesis procedures at university hospitals and their affiliates. This confirms findings from our earlier single‐institution study[16] but contrasts with previously published reports involving Medicare data. The earlier report, using Medicare claims and including ambulatory procedures, revealed that primary care physicians and gastroenterologists only performed approximately 10% of US paracentesis procedures in 2008.[7] Our findings suggest that practices are different at university hospitals, where patients with severe liver disease often seek care. Because we used the UHC database, it was not possible to determine if the clinicians who performed paracentesis procedures in this study were internal medicine or family medicine residents, fellows, or attending physicians. However, findings from our own institution show that the vast majority of bedside paracentesis procedures are performed by internal medicine residents.[16]

Our findings have implications for certification of internal medicine and family medicine trainees. In 2008, the ABIM removed the requirement that internal medicine residents demonstrate competency in paracentesis.[12] This decision was informed by a lack of standardized methods to determine procedural competency and published surveys showing that internal medicine and hospitalist physicians rarely performed bedside procedures.[5, 6] Despite this policy change, our findings show that current clinical practice at university hospitals does not reflect national practice patterns or certification requirements, because many internal medicine‐ and family medicine‐trained clinicians still perform paracentesis procedures. This is concerning because internal medicine and family medicine trainees report variable confidence, experience, expertise, and supervision regarding performance of invasive procedures.[9, 10, 21, 22, 23, 24] Furthermore, earlier research also demonstrates that graduating residents and fellows are not able to competently perform common bedside procedures such as thoracentesis, temporary hemodialysis catheter insertion, and lumbar puncture.[25, 26, 27]

The American Association for the Study of Liver Diseases (AASLD) recommends that trained clinicians perform paracentesis procedures.[3, 4] However, the AASLD provides no definition for how training should occur. Because competency in this procedure is not specifically required by the ABIM, ABFM, or ACGME, a paradoxical situation occurs in which internal medicine and family medicine residents, and internal medicine‐trained fellows and faculty continue to perform paracentesis procedures on highly complex patients, but are no longer required to be competent to do so.

In earlier research we showed that simulation‐based mastery learning (SBML) was an effective method to boost internal medicine residents' paracentesis skills.[28] In SBML, all trainees must meet or exceed a minimum passing score on a simulated procedure before performing one on an actual patient.[29] This approach improves clinical care and outcomes in procedures such as central venous catheter insertion[30, 31] and advanced cardiac life support.[32] SBML‐trained residents also performed safe paracentesis procedures with shorter hospital LOS, fewer ICU transfers, and fewer blood product transfusions than IR procedures.[16] Based on the results of this study, AASLD guidelines regarding training, and our experience with SBML, we recommend that all clinicians complete paracentesis SBML training before performing procedures on patients.

Using our propensity model we identified patient characteristics that were associated with IR referral. Patients with a liver transplant were more likely to be cared for in IR. This may be due to a belief that postoperative procedures are anatomically more complex or because surgical trainees do not commonly perform this procedure. The current study confirms findings from earlier work that obese and female patients are more likely to be referred to IR.[16] IR referral of obese patients is likely to occur because paracentesis procedures are technically more difficult. We have no explanation why female patients were more likely to be referred to IR, because most decisions appear to be discretionary. Prospective studies are needed to determine evidence‐based recommendations regarding paracentesis procedure location. Patients with more comorbidities (eg, ICU stay, awaiting liver transplant, coagulation disorders) were more likely to undergo bedside procedures. The complexity of patients undergoing bedside paracentesis procedures reinforces the need for rigorous skill assessment for clinicians who perform them because complications such as intraperitoneal bleeding can be fatal.

Finally, we showed that LOS was similar but hospital direct costs were $800 to $1300 lower for patients whose paracentesis procedure was performed by medicine or gastroenterology/hepatology compared to IR. Medical subspecialties and surgery procedures were more expensive than IR, consistent with the higher LOS seen in these groups. IR procedures add costs due to facility charges for space, personnel, and equipment.[33] At our institution, the hospital cost of an IR paracentesis in 2012 was $361. If we use this figure, and assume costs are similar across university hospitals, the resultant cost savings would be $10,257,454 (for the procedure alone) if all procedures referred to IR in this 2‐year study were instead performed at the bedside. This estimate is approximate because it does not consider factors such as cost of clinician staffing models, which may differ across UHC hospitals. As hospitals look to reduce costs, potential savings due to appropriate use of bedside and IR procedures should be considered. This is especially important because there is no evidence that the extra expense of IR procedures is justified due to improved patient outcomes.

This study has several limitations. First, this was an observational study. Although the database was large, we were limited by coding accuracy and could not control for all potential confounding factors such as Model for End‐Stage Liver Disease score,[34, 35] other specific laboratory values, amount of ascites fluid removed, or bedside procedure failures later referred to IR. However, we do know that only a small number of second, third, or fourth procedures were subsequently referred to IR after earlier ones were performed at the bedside. Additionally the UHC database does not include patient‐specific data, and therefore we could not adjust for multiple visits or procedures by the same patient. Second, we were unable to determine the level of teaching involvement at each UHC affiliated hospital. Community hospitals where attendings managed most of the patients without trainees could not be differentiated from university hospitals where trainees were involved in most patients' care. Third, we did not have specialty information for 9434 (9.7%) procedures and had to exclude these cases. We also excluded a large number of paracentesis procedures in our final outcomes analysis. However, this was necessary because we needed to perform a patient‐level analysis to ensure the propensity and outcomes models were accurate. Finally, we did not evaluate inpatient mortality or 30‐day hospital readmission rates. Mortality and readmission from complications of a paracentesis procedure are rare events.[3, 4, 36] However, mortality and hospital readmission among patients with liver disease are relatively common.[37, 38] It was impossible to link these outcomes to a paracentesis procedure without the ability to perform medical records review.

In conclusion, paracentesis procedures are performed frequently by internal medicine‐ and family medicine‐trained clinicians in university hospitals. Because of these findings regarding current practice patterns, we believe the ACGME, ABIM, and ABFM should clarify their policies to require that residents are competent to perform paracentesis procedures before performing them on patients. This may improve supervision and training for paracentesis procedures that are already occurring and possibly encourage performance of additional, less costly bedside procedures.

Cirrhosis affects up to 3% of the population and is 1 of the 10 most common causes of death in the United States.[1, 2, 3, 4] Paracentesis procedures are frequently performed in patients with liver disease and ascites for diagnostic and/or therapeutic purposes. These procedures can be performed safely by trained clinicians at the bedside or referred to interventional radiology (IR).[2, 3, 4]

National practice patterns show that paracentesis procedures are increasingly referred to IR rather than performed at the bedside by internal medicine or gastroenterology clinicians.[5, 6, 7] In fact, a recent study of Medicare beneficiaries showed that inpatient and outpatient paracentesis procedures performed by radiologists increased by 964% from 1993 to 2008.[7] Reasons for the decline in bedside procedures include the increased availability of IR, lack of sufficient reimbursement, and the time required to perform paracentesis procedures.[5, 6, 7, 8] Surveys of internal medicine and family medicine residents and gastroenterology fellows show trainees often lack the confidence and experience needed to perform the procedure safely.[9, 10, 11] Additionally, many clinicians do not have expertise with ultrasound use and may not have access to necessary equipment.

Inconsistent certification requirements may also impact the competence and experience of physicians to perform paracentesis procedures. Internal medicine residents are no longer required by the American Board of Internal Medicine (ABIM) to demonstrate competency in procedures such as paracentesis for certification.[12] However, the Accreditation Council for Graduate Medical Education (ACGME) requirements state that internal medicine programs must offer residents the opportunity to demonstrate competence in the performance of procedures such as paracentesis, thoracentesis, and central venous catheter insertion.[13] The American Board of Family Medicine (ABFM) does not outline specific procedural competence for initial certification.[14] The ACGME states that family medicine residents must receive training to perform those clinical procedures required for their future practices but allows each program to determine which procedures to require.[15] Due to this uncertainty, practicing hospitalists are likely to have variable training and competence in bedside procedures such as paracentesis.

We previously showed that internal medicine residents rotating on the hepatology service of an academic medical center performed 59% of paracentesis procedures at the bedside.[16] These findings are in contrast to national data showing that 74% of paracentesis procedures performed on Medicare beneficiaries were performed by radiologists.[7] Practice patterns at university hospitals may not be reflected in this data because the study was limited to Medicare beneficiaries and included ambulatory patients.[7] In addition to uncertainty about who is performing this procedure in inpatient settings, little is known about the effect of specialty on postparacentesis clinical outcomes.[16, 17]

The current study had 3 aims: (1) evaluate which clinical specialties perform paracentesis procedures at university hospitals; (2) model patient characteristics associated with procedures performed at the bedside versus those referred to IR; and (3) among patients with a similar likelihood of IR referral, evaluate length of stay (LOS) and hospital costs of patients undergoing procedures performed by different specialties.

METHODS

We performed an observational administrative database review of patients who underwent paracentesis procedures in hospitals participating in the University HealthSystem Consortium (UHC) Clinical Database from January 2010 through December 2012. UHC is an alliance of 120 nonprofit academic medical centers and their 290 affiliated hospitals. UHC maintains databases containing clinical, operational, financial, and patient safety data from affiliated hospitals. Using the UHC database, we described the characteristics of all patients who underwent paracentesis procedures by clinical specialty performing the procedure. We then modeled the effects of patient characteristics on decision‐making about IR referral. Finally, among patients with a homogeneous predicted probability of IR referral, we compared LOS and direct costs by specialty performing the procedure. The Northwestern University institutional review board approved this study.

RESULTS

Predicted Probabilities of IR Referral

Figure 1 presents the distribution of predicted probabilities for IR referral. Predicted probabilities were low overall, with very few patients having an equal chance of referralthe standard often used in comparative effectiveness analyses from observational data. Figure 1 indicates that IR referral probabilities were clustered in an unusual bimodal distribution. The cluster on the left, which centers around a 15% predicted probability of IR referral, consists of discharges with patient characteristics that were associated with a very low chance of an IR paracentesis. We therefore used this distribution to conduct comparative analyses of admission outcomes between clinical specialty groups, choosing to examine patients with a 20% or greater chance of IR referral.

Figure 1

Post hoc analysis revealed that the biggest factor driving low predicted probability of IR referral was whether patients experienced an ICU stay at any time during hospitalization. Among the discharges with a predicted probability 0.2 (n=26,615 discharges), there were only 87 discharges with ICU stays (0.3%). For the discharges with predicted probability <0.2 (n=7676), 91.9% (n=7055) had an ICU admission. We therefore used a threshold of 0.2 or greater to present the most comparable LOS and direct cost differences.

DISCUSSION

This study showed that internal medicine‐ and family medicine‐trained clinicians perform approximately half of the inpatient paracentesis procedures at university hospitals and their affiliates. This confirms findings from our earlier single‐institution study[16] but contrasts with previously published reports involving Medicare data. The earlier report, using Medicare claims and including ambulatory procedures, revealed that primary care physicians and gastroenterologists only performed approximately 10% of US paracentesis procedures in 2008.[7] Our findings suggest that practices are different at university hospitals, where patients with severe liver disease often seek care. Because we used the UHC database, it was not possible to determine if the clinicians who performed paracentesis procedures in this study were internal medicine or family medicine residents, fellows, or attending physicians. However, findings from our own institution show that the vast majority of bedside paracentesis procedures are performed by internal medicine residents.[16]

Our findings have implications for certification of internal medicine and family medicine trainees. In 2008, the ABIM removed the requirement that internal medicine residents demonstrate competency in paracentesis.[12] This decision was informed by a lack of standardized methods to determine procedural competency and published surveys showing that internal medicine and hospitalist physicians rarely performed bedside procedures.[5, 6] Despite this policy change, our findings show that current clinical practice at university hospitals does not reflect national practice patterns or certification requirements, because many internal medicine‐ and family medicine‐trained clinicians still perform paracentesis procedures. This is concerning because internal medicine and family medicine trainees report variable confidence, experience, expertise, and supervision regarding performance of invasive procedures.[9, 10, 21, 22, 23, 24] Furthermore, earlier research also demonstrates that graduating residents and fellows are not able to competently perform common bedside procedures such as thoracentesis, temporary hemodialysis catheter insertion, and lumbar puncture.[25, 26, 27]

The American Association for the Study of Liver Diseases (AASLD) recommends that trained clinicians perform paracentesis procedures.[3, 4] However, the AASLD provides no definition for how training should occur. Because competency in this procedure is not specifically required by the ABIM, ABFM, or ACGME, a paradoxical situation occurs in which internal medicine and family medicine residents, and internal medicine‐trained fellows and faculty continue to perform paracentesis procedures on highly complex patients, but are no longer required to be competent to do so.

In earlier research we showed that simulation‐based mastery learning (SBML) was an effective method to boost internal medicine residents' paracentesis skills.[28] In SBML, all trainees must meet or exceed a minimum passing score on a simulated procedure before performing one on an actual patient.[29] This approach improves clinical care and outcomes in procedures such as central venous catheter insertion[30, 31] and advanced cardiac life support.[32] SBML‐trained residents also performed safe paracentesis procedures with shorter hospital LOS, fewer ICU transfers, and fewer blood product transfusions than IR procedures.[16] Based on the results of this study, AASLD guidelines regarding training, and our experience with SBML, we recommend that all clinicians complete paracentesis SBML training before performing procedures on patients.

Using our propensity model we identified patient characteristics that were associated with IR referral. Patients with a liver transplant were more likely to be cared for in IR. This may be due to a belief that postoperative procedures are anatomically more complex or because surgical trainees do not commonly perform this procedure. The current study confirms findings from earlier work that obese and female patients are more likely to be referred to IR.[16] IR referral of obese patients is likely to occur because paracentesis procedures are technically more difficult. We have no explanation why female patients were more likely to be referred to IR, because most decisions appear to be discretionary. Prospective studies are needed to determine evidence‐based recommendations regarding paracentesis procedure location. Patients with more comorbidities (eg, ICU stay, awaiting liver transplant, coagulation disorders) were more likely to undergo bedside procedures. The complexity of patients undergoing bedside paracentesis procedures reinforces the need for rigorous skill assessment for clinicians who perform them because complications such as intraperitoneal bleeding can be fatal.

Finally, we showed that LOS was similar but hospital direct costs were $800 to $1300 lower for patients whose paracentesis procedure was performed by medicine or gastroenterology/hepatology compared to IR. Medical subspecialties and surgery procedures were more expensive than IR, consistent with the higher LOS seen in these groups. IR procedures add costs due to facility charges for space, personnel, and equipment.[33] At our institution, the hospital cost of an IR paracentesis in 2012 was $361. If we use this figure, and assume costs are similar across university hospitals, the resultant cost savings would be $10,257,454 (for the procedure alone) if all procedures referred to IR in this 2‐year study were instead performed at the bedside. This estimate is approximate because it does not consider factors such as cost of clinician staffing models, which may differ across UHC hospitals. As hospitals look to reduce costs, potential savings due to appropriate use of bedside and IR procedures should be considered. This is especially important because there is no evidence that the extra expense of IR procedures is justified due to improved patient outcomes.

This study has several limitations. First, this was an observational study. Although the database was large, we were limited by coding accuracy and could not control for all potential confounding factors such as Model for End‐Stage Liver Disease score,[34, 35] other specific laboratory values, amount of ascites fluid removed, or bedside procedure failures later referred to IR. However, we do know that only a small number of second, third, or fourth procedures were subsequently referred to IR after earlier ones were performed at the bedside. Additionally the UHC database does not include patient‐specific data, and therefore we could not adjust for multiple visits or procedures by the same patient. Second, we were unable to determine the level of teaching involvement at each UHC affiliated hospital. Community hospitals where attendings managed most of the patients without trainees could not be differentiated from university hospitals where trainees were involved in most patients' care. Third, we did not have specialty information for 9434 (9.7%) procedures and had to exclude these cases. We also excluded a large number of paracentesis procedures in our final outcomes analysis. However, this was necessary because we needed to perform a patient‐level analysis to ensure the propensity and outcomes models were accurate. Finally, we did not evaluate inpatient mortality or 30‐day hospital readmission rates. Mortality and readmission from complications of a paracentesis procedure are rare events.[3, 4, 36] However, mortality and hospital readmission among patients with liver disease are relatively common.[37, 38] It was impossible to link these outcomes to a paracentesis procedure without the ability to perform medical records review.

In conclusion, paracentesis procedures are performed frequently by internal medicine‐ and family medicine‐trained clinicians in university hospitals. Because of these findings regarding current practice patterns, we believe the ACGME, ABIM, and ABFM should clarify their policies to require that residents are competent to perform paracentesis procedures before performing them on patients. This may improve supervision and training for paracentesis procedures that are already occurring and possibly encourage performance of additional, less costly bedside procedures.