Patient sample. GMS indicates General Medicine Service.
Change because of the number of team admissions on the patient's admission day. The asterisk indicates a significant difference (P<.05) from the reference group of 0 to 3 admissions for inpatient mortality and length of stay only. The dagger indicates a significant difference (P<.05) from the reference group of 0 to 3 admissions for both length of stay and total costs.
Change because of the average team census during the patient's stay. The asterisk indicates a significant difference (P<.05) from the reference group of 0 to 3 patients as the average census for length of stay only. The dagger indicates a significant difference (P<.05) from the reference group of 0 to 3 patients as the average census for both length of stay and total costs.
Ong M, Bostrom A, Vidyarthi A, McCulloch C, Auerbach A. House Staff Team Workload and Organization Effects on Patient Outcomes in an Academic General Internal Medicine Inpatient Service. Arch Intern Med. 2007;167(1):47-52. doi:10.1001/archinte.167.1.47
House staff work-hour regulations have required residency programs to reengineer inpatient services. However, few data describe how house staff workload on a patient's day of admission or on subsequent hospital days influences patient outcomes.
Retrospective cohort analysis of 5742 adults admitted to an academic general medical service between July 1, 1998, and June 30, 2001.
After multivariate risk adjustment for patient severity and other structural factors, we found that 2 different measures of house staff workload significantly affected patient outcomes. House staff workload increases on the day of admission, defined as each additional team admission on a patient's admission day, increased length of stay (difference, 3.09%; 95% confidence interval [CI], 2.22%-3.96%), total costs (difference, 2.31%; 95% CI, 1.29%-3.33%), and risk of inpatient mortality (odds ratio, 1.09; 95% CI, 1.02-1.15). Patients had an even higher mortality risk when more than 9 patients were admitted to their team on their admission day. In contrast, house staff workload increases during the patient's entire stay, defined as every additional patient added to the team average census, reduced length of stay (difference, −5.30%; 95% CI, −4.54% to −6.07%) and total costs (difference, −5.11%; 95% CI, −4.20% to −6.00%). Reductions in length of stay and costs were most striking when the team average census exceeded 15 patients.
Our findings suggest that higher house staff workload on admitting days—when fewer backup resources are available—increases resource use and may increase inpatient mortality. Conversely, a higher average team census was associated with reduced resource use, perhaps reflecting service-level adaptations to workload. Future studies should confirm these findings in larger trials.
Residency training programs are constantly reinventing themselves to meet local needs. New duty-hour reduction requirements for house staff have led to substantial reengineering of residency team structures.1,2 Residency programs ideally want to ensure that these changes improve or have no effect on patient outcomes, such as death, readmission, length of stay, and total costs. However, there are few studies that address these topics.3
To our knowledge, the only published study4 of internal medicine house staff workloads found that increases in the number of admissions by an intern had no effect on inpatient mortality, reduced total hospital charges, and had a nonlinear effect on length of stay (the first few admissions increased length of stay and subsequent admissions reduced length of stay). These findings are surprising given the literature describing an increased risk of 30-day mortality5- 7 and longer lengths of stay8 with high nursing workloads. There are also few studies3 of internal medicine house staff team organization and effects on patient outcomes. One study9 of changes to a house staff short call/“night-float” system found that length of stay and test ordering were reduced after implementation, while other studies found that cross-coverage systems led to longer lengths of stay10 and more preventable adverse events,11 and that the combined implementation of cross-coverage and night-float systems increased the rate of complications and test delays.12 Other studies have focused on the date of admission in relation to internal medicine house staff schedules. These studies examined admission early in the academic year (ie, the “July phenomenon”13- 16), service switch-day discontinuities,11 and admissions on weekends.17,18
To better understand the influence of house staff team workload and organization on patient outcomes, we performed a retrospective analysis of outcomes for more than 5000 patients admitted to a large academic medical center over 3 years. We examined variations in team workload and organization using administrative patient data combined with house staff scheduling data, and then correlated these variations with individual patients to determine their independent effects on inpatient mortality, 30-day readmission, length of stay, and total costs.
This study examined data from the General Medicine Service (GMS) at University of California, San Francisco's Moffitt-Long Hospital, a 525-bed tertiary care center and university hospital. The GMS cares for a wide range of general medical diagnoses, and provides care to patients admitted to the intensive care unit (ICU) in conjunction with intensivists who operate an ICU comanagement service. Patients with primary diagnoses of cardiovascular, neurologic, or oncologic origin are admitted to separate services.
Patients studied were 18 years or older admitted to the GMS between July 1, 1998, and June 30, 2001. To retain a focus on patients admitted for acute medical illness, we excluded 171 patients admitted for elective chemotherapy or for short-stay procedures, providing a base sample of 8325 admissions. We further excluded 2400 subsequent admissions of patients already admitted once during the study period to minimize uncontrolled clustering effects. Other exclusions were 23 admissions for which hospitalization and death data could not be reconciled (date of death would precede date of admission), 149 admissions that could not be matched to a team, and 11 admissions that had missing data. Thus, our final sample included 5742 admissions of unique individuals (Figure 1).
The GMS teams were composed of 1 attending physician, 1 resident physician, and 1 or 2 interns. Resident admission volumes on teams with 2 interns were capped at 10 admissions, while those with 1 intern were capped at 5 admissions. When teams exceeded their caps, patients were admitted by a cross-covering resident physician and redistributed to a team admitting the next morning. During the study, the number of GMS teams admitting per day (between 1 and 2) and on service (between 4 and 8) varied because of perceived service needs and staffing availability. On days with 2 admitting teams, admissions alternated between the 2 teams. During this study, there were no set caps on team patient censuses on nonadmitting days and no policy for reassigning patients on busier teams to less-busy teams.
Patient-level data were drawn from a Transition Systems Inc (Boston, Mass) cost accounting system that collects data abstracted from patient charts at hospital discharge and contains information regarding sociodemographics, principal International Classification of Diseases, Ninth Revision (ICD-9), diagnosis, length of stay, costs, attending physician at discharge, and ICU stays during hospitalization. Admission and discharge dates, but not time information, were available. Patient death in relation to hospitalization was determined from the California State Death Index (for patients admitted in 1998) and from Social Security death indexes (for patients admitted from 1999-2001 or for those who resided outside California).
Team structure and personnel information were obtained from team rosters maintained by the GMS and hospital residency training programs. These data were then merged with patient-level data; patients were associated with a team scheduled to admit patients on the patient's admission day. In 41.7% of cases, there was only 1 team scheduled to admit patients. In another 36.5% of cases, 2 admitting teams existed on the patient admission day, and were linked by finding a match between the listed Transition Systems Inc database discharging physician and 1 of the attending physicians for the 2 admitting teams. Remaining cases were matched to teams using procedures that are available from the authors. Daily team and GMS patient censuses were retrospectively constructed based on the admissions and discharges of all patients (including multiple admissions) associated with the specified team and the overall GMS.
Study outcomes included inpatient mortality, 30-day readmission, length of stay, and total costs. Our data captured deaths and readmission to Moffitt-Long Hospital only. To account for skewness of data and to stabilize the variance of residuals, we logarithmically transformed length of stay and total costs.
Our primary predictors were as follows: (1) team workload on a patient's admitting day, measured as the number of admissions to the team on each patient's day of admission; and (2) team workload during a patient's hospitalization, measured by the average number of patients on the patient's team during the patient's hospitalization. All workload variables were calculated within the data set.
By using identical methods, we also calculated a set of secondary team workload metrics, including the number of discharges on a patient's admission day, number of discharges on a patient's discharge day, number of admissions on a patient's discharge day, and total patients on each team over a calendar month.
Team organization measures included the total number of GMS teams working on the patient's admission day, the number of admitting teams on the patient's admission day, and the number of interns on the patient's admitting team. Team organization measures were calculated using the house staff schedule described earlier.
Key adjusters included patient sociodemographic factors (age, sex, ethnicity, and insurance type), severity adjustment by diagnosis-related group weights, ICU stays, and 3 principal diagnoses (human immunodeficiency virus, cancer, and pneumonia) found in preliminary analyses to have significant effects on the study outcomes. We controlled for changes over the academic year (ie, the July phenomenon13- 16) through quarterly categorical variables using January to March as the reference group, and for discontinuities in care due to team personnel switches11 by including the number of days between the patient's admission and the prior intern and the prior resident physician/attending physician switch day. We also controlled for hospitalist attending physicians,19- 22 the year of admission, admission during a winter holiday schedule (which uses a skeletal staff), and admission occurring on a weekend.17,18 All models included the average GMS daily patient census during the patient's hospital stay, the GMS patient census on the patient's day of admission and discharge, admissions to other GMS teams on the patient's day of admission, and discharges from other GMS teams on the patient's day of discharge.
We characterized our key variables using univariate techniques to assess for skewness of data and missing values, and to examine alternative analytic forms (eg, logarithmic transformation). Bivariate comparisons between our outcomes and predictor variables were performed using χ2 analyses for categorical data and t tests for continuous data. We then constructed multivariate models using forward stepwise automated methods and manual entry of variables statistically associated with our outcomes in bivariate analyses. Variance inflation factors were checked to assess multicollinearity. Variables were then retained based on their statistical association with the outcome of interest, the observed effects of confounding, minimal collinearity, or to retain face validity.
All models used 2-level mixed-effect models to determine the effect of team workload and organization,23- 25 and included all team workload and organization variables plus adjustment variables. Team workload and organization measures were treated as fixed effects, while the effect of each team was treated as random to account for clustering within teams. Average patient census and number of admissions were modeled as continuous and categorical variables. Logistic regression models were used for dichotomous outcome measures of inpatient mortality and 30-day readmission, and log-linear models were used for continuous outcome measures of length of stay and total costs. We calculated odds ratios with 95% confidence intervals for categorical outcomes, and percentage change with 95% confidence intervals for continuous outcomes. All statistical analyses were performed using 2 programs (PROC MIXED and PROC GLIMMIX) in SAS statistical software, version 9.1, for Windows (SAS Institute Inc, Cary, NC); and 2-tailed P<.05 was considered statistically significant in all analyses.
To explore the robustness of our findings, we performed subgroup analyses (not presented) on 921 patients with ICU stays and 828 patients presumed to be redistributed via night-float services (ie, patients whose discharging attending physician name was matched to the attending physician of a team on service but not admitting on the admission day [more information is available from the authors]). Patients with ICU stays receive additional services from the ICU consultation team that could lighten the workload of the admission team. Redistributed patients, unlike nonredistributed patients, may have been misallocated by our matching algorithms.
Most of our study patients were older than 65 years, female, and white, and had either Medicare or commercial insurance (Table 1). More than 5% of the patients died during their hospitalization, and nearly 8% were readmitted to Moffitt-Long Hospital within 30 days of discharge. The median length of stay was 4 days, and the median total cost of the hospitalization was $4319 (Table 1).
Patients in our sample were admitted to teams that averaged 5.8 admissions per call cycle, 44.2 admissions during a full month, and a daily census of 10.1 patients (Table 2). At the service level, the GMS average daily census was 44.8 patients (SD, 11.3 patients) during the study period, with a mean of 7.8 admissions (SD, 3.3 admissions) and 7.8 discharges (SD, 3.5 discharges) per day.
Each additional team admission on a patient's admission day increased length of stay, total costs, and risk of inpatient mortality (Table 3 and Table 4). This mortality risk increases substantially when admitting more than 9 patients (Figure 2). In contrast, each additional patient on the team's average census during a patient's hospitalization reduced length of stay and total costs (Table 3). Reductions were most marked when the average census was more than 15 patients (Figure 3).
Other workload and organizational metrics had inconsistent effects on our key outcomes after adjusting for our primary workload predictors. Other workload findings were that each additional admitting team's discharge and each additional team working on the day of admission reduced length of stay and total costs, while each additional patient admitted by the team during the patient's admission month increased length of stay and total costs (Table 3). Each additional admission to the patient's team on the discharge day also increased the patient's total costs. In addition, admission to a team with 2 interns instead of 1 increased the risk of 30-day readmission (Table 4).
Subgroup analyses of redistributed and nonredistributed patients consistently demonstrated the same associations with our primary workload predictors, suggesting that the threat of misallocation bias was limited in our cohort. Subgroup analyses of patients with ICU stays and those without ICU stays also demonstrated similar associations with our primary workload predictors, except patients with ICU stays had no significant admission day effects and neither group had significant mortality effects.
In this study of patients admitted to a GMS at an academic medical center, patients had increased resource use and worsened clinical outcomes if admitted to house staff teams during a day with a high admission load. However, patients have reduced resource use and no differences in readmission or mortality if cared for by teams with a high average census during the patient's stay. These findings have potentially important implications for residency training programs and hospital administrators.
Other studies of inpatient care organization have demonstrated improvements in clinical outcomes when more nurses are available5- 7 or when care is provided in a setting that systematizes greater physician and nurse availability, such as “high-intensity” ICUs26 or hospitalist systems.19,21 These studies suggest that increased house staff workload should worsen patient outcomes, and are consistent with our observation that busier admission days are associated with higher resource use and potentially higher risk for inpatient mortality. Admission workup activity is extensive; more admissions reduce the time spent by teams on any 1 admitted patient, potentially leading to inaccurate initial clinical assessment or pushing workup activity onto subsequent days, leading to longer lengths of stay and higher total costs. At the highest admission loads, this observation may be in part because of the service's method of redistributing patients after the busiest admitting days. Despite this, our admission day metrics remained statistically significant and similar in magnitude even when we examined redistributed and nonredistributed patients separately, suggesting that redistribution (and misallocation bias) played a small role, if any, in our results.
In contrast, higher workload on later hospital days actually decreased a patient's length of stay and costs, while producing no consistent effects on mortality or readmission. We believe this counterintuitive finding, which mirrors the findings of the prior study4 on internal medicine house staff workload, represents the secondary response of teams or the inpatient service as a whole. Unlike admission volume, which cannot be modified by teams, team average census is a work measure that teams can control by adapting their daily tasks, such as skipping teaching conferences to perform patient care tasks,27 to meet workload demands. High workload may also increase pressure to discharge patients quickly or focus attention of social workers and discharge planners on teams with higher patient volumes. This effect is robust, because it is found in all primary and secondary analyses.
Our additional findings provide intriguing and hypothesis-generating findings that deserve further study. Having more teams available reduces length of stay and total costs even after adjusting for admission workload and team average census, perhaps pointing out second-order benefits of workload reduction, such as more time for teaching and learning. While short-term workload (ie, team average census during patient stay) improved efficiency, teams that were busier over the entire monthlong rotation provided less efficient care. This finding suggests that internal efficiency can be increased in the short-term, but fatigue may accumulate within teams over time. Alternatively, higher monthly workloads may represent higher overall GMS workload and lower hospital efficiency. Patients on 1-intern teams have fewer readmissions because of more direct patient care provided by supervising resident physicians. Subgroup analyses suggest that patients with ICU stays are not affected by house staff workload on admission, because of additional assistance provided to these patients. However, increases in team average patient census reduce length of stay and total costs for all patients, suggesting that pressures to reduce workload affect patients regardless of their setting.
Our results may provide basis for local policy decisions, particularly if confirmed in other settings. For example, if each patient's team admitted 1 fewer patient, there would have been a savings of $117 and 0.15 hospital days per patient. At Moffitt-Long Hospital, these reductions would have saved $216 000 to $236 000 and 268 to 298 bed days per year. Such savings could help defray the costs of supporting additional physician or nonphysician coverage for reorganized house staff services.28
Our study has several limitations. First, we are circumspect about our mortality findings because our sample size is not powered to detect mortality differences, and because we depended on administrative data for risk adjustment. However, patients were admitted to teams essentially at random, making unmeasured severity a less likely influence than sample size limitations. Second, our methods are subject to misallocation bias resulting from redistribution of patients admitted to teams over their admission cap. However, redistribution would tend to bias our admission workload findings toward a null result. Furthermore, our sensitivity analyses were robust in redistributed and nonredistributed subsets of patients. Third, our 30-day readmission outcome does not detect hospitalizations at other hospitals, but is consistent with other studies.19,21 Fourth, it is possible that team workload and organization effects may be different during subsequent hospitalizations of patients. Finally, as a study of a GMS at 1 academic medical center, our findings may not be applicable to other settings, such as smaller institutions with limited resources or other (ie, surgical or pediatric) services with different workload or house staff training systems.
Our findings suggest that house staff workload on admitting and later days has important effects on patient outcomes. Programs seeking to minimize total costs and lengths of stay may want to find ways to reduce team admission loads, while maximizing availability of other resources on nonadmitting days. Whether these mechanisms will be available from within house staff training programs in an era of strict adherence to work-hour regulation seems unlikely, and may further catalyze inpatient care system change. In an era when hospital organizational structures are in considerable flux, further studies are required to prospectively determine the impact of house staff team workload and organization on patient outcomes. Balancing the clinical and economic outcomes with available resources and the educational impact of changes in the organization of house staff teams will be important tasks for training programs and teaching hospitals in coming years.
Correspondence: Michael Ong, MD, PhD, Division of General Internal Medicine and Health Services Research, Department of Medicine, University of California, Los Angeles, 911 Broxton Ave, First Floor, Los Angeles, CA 90024 (email@example.com).
Accepted for Publication: September 13, 2006.
Author Contributions:Study concept and design: Ong and Auerbach. Acquisition of data: Vidyarthi and Auerbach. Analysis and interpretation of data: Ong, Bostrom, McCulloch, and Auerbach. Drafting of the manuscript: Ong, Vidyarthi, and Auerbach. Critical revision of the manuscript for important intellectual content: Ong, Bostrom, Vidyarthi, McCulloch, and Auerbach. Statistical analysis: Ong, Bostrom, McCulloch, and Auerbach. Obtained funding: Auerbach. Study supervision: Vidyarthi and Auerbach.
Financial Disclosure: None reported.
Funding/Support: This study was partially supported by a VA Ambulatory Care Fellowship (Dr Ong); and by research and training grant K08 HS11416-02 from the Agency for Healthcare Research and Quality (Dr Auerbach).
Role of the Sponsor: The funding bodies had no role in data extraction and analyses, in the writing of the manuscript, or in the decision to submit the manuscript for publication.
Previous Presentations: Portions of this study were presented at the Society for General Internal Medicine Annual Meeting; May 14, 2004; Chicago, Ill, and at the AcademyHealth Annual Research Meeting; June 25, 2006; Seattle, Wash.
Acknowledgment: We thank Robert Wachter, MD, for comments on an earlier draft of the paper.