Physician Consults in the Hospital
Abstract
Physician consultations are an important yet understudied facet of hospital care delivery. Consults occur when the primary attending physician formally requests the help of another physician in the diagnosis or treatment of a patient. Because most non-surgical, hospitalized patients are treated by hospitalists or other general medicine physicians, consults are the main mechanism by which specialty care is delivered in the inpatient setting. Previous work finds that more than half of all hospitalizations among traditional Medicare beneficiaries involved at least one consultation, and more than 20% involved consultations of two or more specialties. Consults could improve quality of care if they increase diagnostic accuracy, change patient management, or provide access to treatments and services not otherwise available. On the other hand, consults could represent “flat-of-the-curve” medicine if they do not improve care, result in unnecessary tests and procedures, and increase patients’ length of stay in the hospital.Despite their importance and frequency, there is little high-quality evidence on the effects of consults on health and utilization outcomes. Because consults are ostensibly requested when the primary team has a question or needs help, they are highly correlated with severity of illness and clinical information not commonly observed in administrative datasets. To address the endogeneity of consults, I conduct an instrumental variable analysis using detailed electronic medical record data from a large health system in Pennsylvania over 2017-2022. Specifically, I exploit variation in demand for infectious disease (ID) consults during the first 24 hours of admission to the general medicine service. I show that patients are less likely to receive a consult when consultants are busier due to consult requests made by other physicians. Estimation of the effect of an ID consult on 30-day readmission via OLS gives a 2 percentage point (pp) increase, which reflects omitted variable bias.