Abstract
Although hospitalists (generalist inpatient physicians) have overlapping functions and expertise with specialist physicians, practice differences and resulting patient outcomes between the physician types are not well understood. We investigate how hospitalists and inpatient specialists differ in their treatment strategies and how such differences are associated with patient outcomes. Our identification strategy focuses on a subset of common conditions treated by both hospitalists and specialists within a single hospital such that the setting and organizational policies are the same for both physician types.
We use a two‐stage design to disentangle differences in both practice patterns and patient outcomes. In the first stage, we explore whether hospitalists and specialists differ in the number of laboratories and drugs ordered using multilevel negative binomial generalized linear models, accounting for both the distribution of the data and the nesting of patients within providers. We control for patient characteristics, provider workload, and details of the admission including length of stay and any use of intensive care.
In the second stage, we model the likelihood that a patient experiences an adverse event or a readmission. These multilevel logistic models expand upon the first‐stage control variables to include the number and types of laboratories and drugs ordered.
We leverage data on physicians and patients in a large urban pediatric hospital between January 1, 2009, and August 31, 2015. By focusing on 57 diagnoses commonly treated by both hospitalists and specialists, we are able to distinguish differences in treatment patterns and outcomes for similar patient cases.
Practice patterns differ significantly between hospitalists and specialists, and these differences are subsequently associated with differences in patient outcomes. Over the median 3‐day hospital stay, hospitalists ordered 5.0 fewer laboratories (31.3%, P < .001) compared to specialists for conditions identified as well‐known (ie, conditions with clear diagnosis and treatment protocols) and 6.3 fewer laboratoriess (45.1%, P < .001) for less well‐known conditions. Drug orders follow a similar pattern, with hospitalists placing 2.0 fewer orders (11.4%, P = .002) for well‐known conditions and 8.9 fewer orders (46.9%, P = .002) in less well‐known conditions. Specialists are associated with a 3.2 percentage point higher likelihood of patients’ experiencing an adverse drug event (P < .001) and a 5.3 percentage point higher probability of a patient experiencing any adverse event (P < .001). We find no differences in readmission rates.
Patients are likely to receive substantially different treatment based on whether they are treated by hospitalists or specialists. Compared to specialists, hospitalists order fewer laboratories and drugs. These patterns are associated with increased costs and reduced adverse events, with no corresponding change in readmission rates.
Hospitalists are able to perform similarly or better than specialists on key quality outcomes, despite fewer orders for laboratories and drugs. Efforts to mitigate medical overuse must recognize these differences in practice across specialties, so as to better target the source of inefficient treatment strategies.