Reasons for exclusions are indicated. ED indicates emergency department; ICU, intensive care unit; and LOS, length of stay.
Three levels of occupancy overlaid with fractional polynomial regression fit and 95% CIs are shown.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Elliott DJ, Young RS, Brice J, Aguiar R, Kolm P. Effect of Hospitalist Workload on the Quality and Efficiency of Care. JAMA Intern Med. 2014;174(5):786–793. doi:10.1001/jamainternmed.2014.300
Copyright 2014 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.
Hospitalist physicians face increasing pressure to maximize productivity, which may undermine the efficiency and quality of care.
To determine the association between hospitalist workload and the efficiency and quality of inpatient care.
Design, Setting, and Participants
We conducted a retrospective cohort study of 20 241 admissions of inpatients cared for by a private hospitalist group at a large academic community hospital system between February 1, 2008, and January 31, 2011.
Daily hospitalist workload as measured by relative value units and patient encounters from the hospitalist billing records.
Main Outcomes and Measures
The main outcomes were length of stay (LOS), cost, rapid response team activation, in-hospital mortality, patient satisfaction, and 30-day readmission rates. Key covariates included hospital occupancy and patient-level characteristics.
The LOS increased as workload increased, particularly at lower hospital occupancy. For hospital occupancies less than 75%, LOS increased from 5.5 to 7.5 days as workload increased. For occupancies of 75% to 85%, LOS increased exponentially above a daily relative value unit of approximately 25 and a census value of approximately 15. At high occupancy (>85%), LOS was J-shaped, with significant increases at higher ranges of workload. After controlling for LOS, cost increased by $111 for each 1-unit increase in relative value unit and $205 for each 1-unit increase in census across the range of values. Changes in workload were not associated with the remaining outcomes.
Conclusions and Relevance
Increasing hospitalist workload is associated with clinically meaningful increases in LOS and cost. Although our findings should be validated in different clinical settings, our results suggest the need for methods to mitigate the potential negative effects of increased hospitalist workload on the efficiency and cost of care.
Hospital medicine is the fastest growing medical specialty in the United States.1,2 A major driver of this growth has been empirical evidence suggesting that hospitalists provide inpatient care that is more efficient, less costly, and of equal or higher quality than traditional models of care.3,4 Currently, hospitalist programs face growing pressure to increase productivity to compensate for declining revenue or to meet operational demands resulting from policy and practice changes, such as limitations on resident work hours, specialty comanagement, and decreased presence of primary care physicians in the hospital.5-7 Increased workloads for nurses and resident physicians have been associated6,8 with adverse events, leading to mandated workload reductions, but there is little empirical evidence illuminating the association of hospitalist workload and clinical outcomes.
Historically, benchmark recommendations for US hospitalist workload range from 10 to 15 patient encounters per day, but these figures lack empirical support.9 In a recent national survey of hospitalists,10 40% of respondents reported exceeding what they perceived as a safe workload at least monthly and that increased workload led to delays in care, poor communication between physicians and patients, delivery of unnecessary care, medication errors, and complications of care, including death. Although the correlations between these physician perceptions and patient outcomes are not known, increasing productivity requirements for hospitalists could undermine the gains in efficiency and quality that have contributed to the growth of hospital medicine. Spurred in large part by programs such as the Centers for Medicare and Medicaid’s Value-Based Purchasing Program, which seek to transform payment from fee-for-service models to payments based on quality and value, hospitals and hospitalists are under increasing pressure to manage both efficiency and quality.11 Therefore, understanding the association between physician workload and the efficiency and quality of care is essential to designing an optimal inpatient system for the current and future environment.
Our objective for the present study was to determine the association between hospitalist workload and the efficiency and quality of inpatient care. We hypothesized that increased physician workload would be associated with decreased efficiency measured by increased length of stay (LOS) and cost per case as well as decreased quality measured by in-hospital mortality, activation of the rapid response team (RRT), 30-day readmissions, and patient satisfaction. We also hypothesized that both hospital occupancy and physician continuity would affect the association between workload and clinical outcomes.
We conducted a retrospective cohort study of all patients admitted to a private hospitalist service between February 1, 2008, and January 31, 2011. All patients were admitted to a general medicine or step-down unit at Christiana Care Health System, an academic community health system in northern Delaware comprising Christiana Hospital, a 780-bed tertiary care hospital, and Wilmington Hospital, a 291-bed urban community hospital, which together have more than 55 000 annual admissions. With the exception of active cardiology and oncology subspecialty services, hospitalists admit most nonsurgical patients in our system. The study was approved by the Christiana Care Institutional Review Board.
Three hospitalist groups provided care in our system during the study period. We describe the experience of one private hospitalist group that had operated continuously since 1996. The group provided continuous 24-hour care to patients at both hospitals. The practice included 20 to 35 active providers each year, with 0 to 2 nurses providing support in the hospital. The group does not care for patients in the intensive care unit at Christiana Hospital but does at Wilmington Hospital. Most care was in a nonteaching setting, but the group covered 2 teaching teams during most of the study period.
Patients were eligible for inclusion if a member of the hospitalist group was the attending of record and there was a billing event on the first or second day of admission or if a physician from the group submitted an admission bill and a discharge bill, regardless of the designated attending of record. We excluded patients who were younger than 18 years, were admitted exclusively as observation status, were admitted directly to the intensive care unit, were transferred from another acute care hospital, were not discharged by the end of the study period, or had an LOS of less than 0.5 days or more than 30 days. See Figure 1 and Table 1.
We derived measures of physician workload from the billing records of the hospitalist group. Patients are cared for by different physicians with varying workloads during a hospitalization. To determine the association of workload with patient-level outcomes, we first calculated the total workload for each physician on each day of the study period from all clinical settings. We defined physician workload during each day of the study period in 2 ways: (1) the total number of generated relative value units (RVUs) and (2) the number of patients for whom the physician submitted a billable encounter (census). We standardized RVU values to 2011 Centers for Medicare and Medicaid values.12 Because the Centers for Medicare and Medicaid stopped reimbursing for consultation codes in January 2010, we mapped consultation codes across the study period to comparable visits based on recommendations from our billing advisors. We then assigned the workload value for each physician on a given day to every patient for whom the provider submitted a bill on that day. We repeated this for each day of the study period.
The primary efficiency outcomes were LOS and cost. The LOS was derived from the Christiana Care Health System data warehouse. Cost was derived using the Truven Health Analytics CareDiscovery, which converts hospital charge information to nationally standardized costs.13 The quality outcomes were in-hospital mortality, activation of the RRT, 30-day readmission rate, and patient satisfaction, all of which were obtained from the Christiana Care Health System data warehouse. Although RRT activation occurs for many reasons, it commonly is used as a proxy for patient decompensation.14 Readmissions within 30 days of discharge are driven by many factors but are associated with the quality of care at discharge.15 We report all-cause readmissions to Christiana Care, but our hospital system accounts for more than 75% of hospitalizations in northern Delaware. We also measured 7-day readmissions because a shorter time to readmission may be more sensitive to hospitalist workload. We measured patient satisfaction using responses to the Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) survey.16 The primary satisfaction outcome was the overall hospital rating; secondary measures included responses to physician-level questions about the physician’s level of concern, effectiveness of communication, and courtesy. We considered satisfaction as the patient reporting the “top box” score for each question consistent with how satisfaction is measured in the Value-Based Purchasing Program.11
We adjusted models for patient-, visit-, and hospital-level characteristics. Patient characteristics included age, sex, insurance status, race/ethnicity, and presence of a previous admission within 30 days. We adjusted for patient severity using expected measures for LOS, cost, and mortality from Truven Health Analytics CareDiscovery,13 which provides predicted values based on claims data for each patient using models developed in national data sets. We adjusted readmission models for comorbidities using the Elixhauser classification.17 Visit-level characteristics included the admitting hospital, teaching status, and admission day of the week.
We adjusted our models for provider continuity, hospital occupancy, and the turnaround time for a transthoracic echocardiogram by hospital for each day of the study period. Assessing continuity is challenging. Previous studies18,19 have measured continuity as the proportion of care given by a single provider, but these measures are potentially confounded by endogeneity, meaning that the measure of continuity changes with variation in the LOS. Additionally, they are defined at the visit level and not on a daily basis. In our models, we defined provider continuity on a daily basis as present when the same physician had submitted a bill for the patient on the previous day. In multiple sensitivity analyses using summary and daily definitions of continuity, the measured effect of continuity varied widely but did not influence the main effects of workload on the primary outcomes of interest. Therefore, we do not report the effect of continuity.
We calculated daily hospital occupancy separately for each hospital as the average of the hourly occupancy of all non–intensive care unit inpatient medicine beds. We categorized low, medium, or high occupancy as less than 75%, 75% to 85%, and greater than 85%, respectively, based on the distribution in our system. To more fully reflect system inefficiencies, we calculated the mean time from ordering to completion (ie, turnaround time) of an echocardiogram. Similar tests of turnaround time have been associated with efficiency of care in other settings.20-22 Our hospitalist group identified echocardiogram from among the 5 most commonly ordered radiographic tests on general medicine units as the single test that most captures system inefficiencies not limited to occupancy.
We developed hierarchical models with clustering by provider and by patient to assess the association between RVU and census separately for each outcome. We used parametric log logistic accelerated failure-time methods for LOS, in-hospital mortality, RRT, and 30-day readmissions because the data violated proportional hazards assumptions.23,24 Patients were censored at death, at discharge, or at 30 days in the readmission analysis if not readmitted by then. The RVU, census, continuity, occupancy, and turnaround time varied by day during hospitalization. We tested RVU2 and census2 terms to test for nonlinearity as well as interactions of workload with hospital occupancy in all models. We dropped regression coefficients for the quadratic and/or interaction terms that were not statistically significant. We calculated a predicted value for each outcome based on the final models and used fractional polynomial regression to generate a smoothed curve of the predicted values.
We anticipated that workload near the time of discharge would be more strongly associated with readmission, so we entered workload in readmission models in 2 ways: (1) on the day of discharge and (2) for the mean of the last 3 days of the hospitalization. We similarly summarized occupancy and turnaround time in readmission models. We included discharge day of the week, continuity on the last day, and discharge disposition in readmission models.
We used semiparametric methods to analyze cost due to skewness even after eliminating patients with LOS less than 0.5 or greater than 30 days. We estimated model regression coefficients with linear regression and calculated SEs with bootstrap methods using 250 replicates stratified by hospitalization. Cost was fixed rather than time variant, but because the bootstrap unit was hospitalization (rather than patient), we used continuity, occupancy, and turnaround time values for the entire hospitalization. We modeled patient satisfaction in the same way, except we used logistic regression for the dichotomized top box rating. We also conducted a secondary analysis modeling patient satisfaction using ordinal regression across all values. We examined the percentage of missing data for the variables included in the LOS, cost, RRT, mortality, and readmission models. None were missing more than 5% of any variable, so we excluded observations with missing data from individual analyses. We conducted all analyses with Stata, version 12 (Stata Corp).
Overall, the hospitalist group provided care during 33 137 stays, of which 20 241 hospitalizations for 13 916 patients met study criteria. Reasons for exclusion appear in Figure 1. Table 1 contains patient and visit characteristics, and Table 2 and Table 3 depict workload distribution and unadjusted outcomes. The hospitalists had a mean (SD) of 15.5 (2.7) patient encounters generating 28.6 (4.7) RVUs per day.
The association between workload and LOS was nonlinear and nonuniform across hospital occupancy levels for RVU and census (P < .001 for interaction and quadratic terms for both RVU and census).
Figure 2 plots predicted LOS across the range of workload values stratified by 3 hospital occupancy levels (<75%, 75%-85%, and >85%). For occupancies less than 75%, LOS increased linearly from approximately 5.5 to 7.5 days across low to high workload values. For occupancies of 75% to 85%, LOS remained generally stable across lower workload values and then increased exponentially to approximately 8.0 days at high workload values. The transition threshold corresponded to a daily RVU of approximately 25 and a census value of approximately 15. For high occupancy (>85%), the change in LOS was J-shaped for both RVU and census, decreasing slightly to the midrange of values and then increasing significantly.
The RRT was activated in 548 hospitalizations (2.7%). The mortality rate was 1.4% (286 of 20 241) overall (Table 3). There were no statistically significant associations with workload and RRT or mortality (P > .20 for both RVU and census).
Figure 3 plots cost across workload values. Cost increased with increasing workload and occupancy (P < .001 for both). For every unit increase in RVU, cost increased by $133, and for every unit increase in census, cost increased by $262. Cost increased by $1634 for successive categories of occupancy (<75%, 75%-85%, and >85%). Cost models included LOS, which was the strongest predictor of cost. However, workload remained significant after adjustment for LOS, with costs only mildly attenuated, increasing $111 for each 1-unit increase in RVU and $205 for each 1-unit increase in census.
Patient responses to the HCAHPS survey are presented in Table 3. There was no association between workload and patient satisfaction. Increasing occupancy was associated with a 12% decrease in top box hospital ratings (P = .03). The results were unchanged when results were analyzed as continuous or ordinal (data not shown).
The readmission rate was 18.2% (3640 of 19 955). There were no statistically significant associations of 30-day readmission with workload or occupancy (P > .10 for both) for the models that defined workload and occupancy as the value on the last day or as the mean of the last 3 hospital days. Similarly, there was no association between workload and 7-day readmissions (P > .40 for both RVU and census).
Table 4 summarizes the results for each outcome by workload, occupancy, and continuity. The interactions of workload and occupancy for LOS were simplified to reflect an overall difference in change in LOS.
This retrospective cohort study of more than 20 000 hospitalizations provides evidence that increased hospitalist workload is associated with reduced efficiency and higher costs. More important, the magnitude of the difference in LOS and cost revealed across the range of hospitalist workload—as much as 2 days and between $5000 and $7500—is clinically meaningful to hospitals, administrators, and payers. The LOS was particularly sensitive to workload at lower levels of hospital occupancy, and the association between workload and cost was generally consistent regardless of hospital occupancy. We did not find associations between workload and in-hospital mortality, RRT activation, 30-day readmissions, or patient satisfaction.
Our findings are consistent with those of previous research,10 suggesting that increasing workload may reduce hospitalist efficiency. Hospitalists spend most of their time in indirect patient care activities, including documentation, clinical decision making, and coordination of care with the patient, family, and other providers, all of which are critical to delivering efficient, high-quality care.25,26 In the setting of higher workloads, time allocated to these activities decreases while face-to-face time with patients remains unchanged.26 Thus, hospitalists have reported10 that high workloads delay admitting and discharging patients because these activities are time intensive. Similarly, hospitalists report10 that high workloads reduce their ability to adequately assess the patient and fully discuss the plan of care with patients and families. Likely as a result, hospitalists report27 ordering unnecessary procedures, testing, and consultations when busy, all of which drive increased costs.
We found that the effect of hospitalist workload on LOS varied according to hospital occupancy. The impact was linear throughout the range of workload at lower occupancy. At middle occupancies, the effect was small across lower ranges of workload but increased exponentially above the mean values for workload. At the highest occupancy, LOS slightly decreased as workload increased across lower ranges of workload but increased rapidly above an RVU of 25 and census of 15. This suggests that, as hospitals reach capacity, LOS is affected more by hospital factors, such as demands for nursing support or other ancillary services, than by physician factors at lower ranges of hospitalist workload.8,28 However, hospitalist workload becomes a driving factor above these thresholds. Notably, our findings include adjustment for the turnaround time for an echocardiogram, which our hospitalists identified as representative of system-level inefficiencies and delays that affect the efficiency of care that they provide. In contrast, the association of hospitalist workload with costs remained more consistent across levels of hospital occupancy, even after adjusting for LOS. This suggests that the patterns of care driving cost in the setting of higher hospitalist workload persist regardless of hospital occupancy.
Hospitalists have related10 workload excesses with prescribing errors, delays in response to abnormal test results, and even morbidity and mortality. We did not find an association between increased hospitalist workload and adverse events, including in-hospital mortality or RRT activations, although these events occurred rarely in our population, and our findings may reflect inadequate power. Additionally, although errors may occur commonly in medical practice, most errors do not lead to serious adverse events because of other protections or resilience within the patient and/or system.29,30 Future studies should consider analyses that address more granular measures potentially attributable to hospitalists.31
We also did not see differences in readmissions based on workload on either the final 1 or 3 days of hospitalization. Prior research32 suggests the existence of a tradeoff between LOS and readmission, and longer LOS associated with higher workloads in our study may have decreased rates of 30-day readmission. Previous research33,34 found little variation by hospitalist in 30-day readmission among Medicare patients, which is consistent with increasing recognition that readmissions likely reflect patient or system factors more than hospitalist factors. A recent study35 of intern workload demonstrated a slight increase in the odds of readmission as the number of discharges rose, but the effect was small and difficult to generalize to mature hospital medicine practices. Finally, assistance by discharge nurses and other office staff, as was available to our group, may have mitigated the impact of additional discharges in our sample.
The burden of increasing workloads may be temporized by a system with high levels of provider continuity.36 We evaluated the effect of continuity in several ways, including our novel measure of continuity defined as whether the provider had seen the patient on the previous day. This value avoids endogeneity, which has hindered previous assessments of continuity.18 We consistently found statistically significant associations between continuity, LOS, and cost, but these results varied significantly depending on the chosen definition of continuity. Regardless, the measures did not affect the primary association of workload and outcomes, suggesting that continuity does not mitigate the negative effect of workload on efficiency outcomes.
Our study has several limitations. First, we examined a single established hospitalist group at a large, well-resourced academic community hospital. The association of workload and outcomes is likely context dependent, and our results may not apply to hospitalist programs in other settings.31,37 Second, we did not account for physician differences such as experience, because our intent was to understand the association of workload with outcomes at the practice level accounting for patient and hospital characteristics. However, we accounted for physician variance by clustering all models by physician as a random effect. The variance by physician was small, increasing the overall model error variance by only 3%, with little effect on estimated regression coefficients. Third, we identified patients who received care by the hospitalist at the beginning and end of the hospitalization and may have excluded patients who deteriorated and were transferred to another service, causing us to underestimate the rate of adverse events. Finally, although we have accounted for a comprehensive list of covariates in all models, inherent to retrospective observational study designs, we cannot exclude confounding by unmeasured variables.
Although our findings should be confirmed in other settings, the increase in LOS and cost observed in our system as hospitalist workload increases has several important implications. Most hospitalist programs are subsidized by hospital systems, and incentives based on productivity are common. Programs that employ or support hospitalist practices should be aware that policies and incentives that increase workload to minimize short-term costs may undermine larger system efforts targeting efficiency and costs of care. At a minimum, incentive programs should balance productivity, efficiency, and quality measures. Additionally, our findings highlight the need for hospitals and hospitalist groups to have mechanisms to account for fluctuations in census. Certainly practices need to have appropriate coverage schemes and support structures, but our results suggest that hospitals should be prepared to increase services to handle high clinical loads as well, perhaps through additional care coordination or discharge support for providers.
Overall, we found that increasing hospitalist workload was associated with clinically meaningful increases in LOS and cost but did not appear to affect mortality, RRT activation, 30-day readmission, or patient satisfaction. Although our findings require validation in different clinical settings given the likely variability of these associations across systems, our results suggest that incentives aimed at increasing workload may lead to inefficient and costly care. In systems that incentivize physicians based on productivity, consideration should be given to including measures of efficiency and quality.
Corresponding Author: Daniel J. Elliott, MD, MSCE, Department of Medicine, Christiana Care Health System, Room 2E75B, 4755 Ogletown-Stanton Rd, Newark, DE 19718 (email@example.com).
Published Online: March 31, 2014. doi:10.1001/jamainternmed.2014.300.
Author Contributions: Drs Elliott and Kolm had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Elliott, Young, Brice.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Elliott, Young, Aguiar, Kolm.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Elliott, Kolm.
Obtained funding: Elliott.
Administrative, technical, or material support: Elliott, Brice, Aguiar.
Study supervision: Elliott, Young.
Conflict of Interest Disclosures: During the study period and the evaluation period, Dr Brice served as division chief, hospital medicine, Christiana Care Health System and president and managing physician, Christiana Medical Group, PA. She has served as a consultant for IPC, The Hospitalist Company, Inc. No other disclosures were reported.
Funding/Support: This work was supported by internal funding from the Chairs Leadership Council at Christiana Care Health System.
Role of the Sponsor: The Chairs Leadership Council at Christiana Care Health System had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Additional Contributions: Seema Sonnad, PhD, Christiana Care Value Institute, and Sandy Schwartz, MD, Department of Medicine, University of Pennsylvania, provided critical reviews of an earlier version of this article. The contributors received no financial compensation.
Create a personal account or sign in to: