Primary Submission Category: Applications in Health and Biology
Provider-Level Heterogeneity in Inpatient Efficiency Under Resource Strain: A Multilevel Quasi-Experimental Analysis
Authors: Eliot Weinstein, Matthew Cerasale,
Presenting Author: Eliot Weinstein*
Length of stay (LOS) is a central measure of inpatient efficiency, yet isolating provider‑level contributions is challenging because encounters are nested within complex hospital systems and confounded by fluctuations in patient mix and resource availability. Using encounter‑level data from adult inpatient admissions at an urban tertiary medical center, we leverage a hospital‑defined “Jeopardy” status (periods of heightened resource strain requiring additional provider coverage) as a quasi‑experimental shock to examine how LOS responds to system stress and, novelly, how this response varies across physicians within a hospital. We estimate multilevel generalized linear mixed models to characterize population‑level and physician‑specific effects of Jeopardy on LOS. These models let us quantify heterogeneity in provider efficiency under strain, identifying clinicians whose LOS patterns diverge from peers. Motivated by consistent associations between ICU use and LOS, we also outline a multilevel two‑stage residual inclusion (2SRI) strategy that treats Jeopardy as an instrument for encounter‑level ICU admission to estimate the causal effect of ICU use on LOS among encounters whose ICU status is shifted by resource strain. Together, these approaches allow us to characterize system‑level impacts of resource strain and physician‑level variation in responsiveness to that strain, offering a rigorous foundation for studying provider efficiency under high-demand conditions.