Peer effects among hospitalized patients: Evidence from roommate assignments
Background: The tendency for an individual to adopt the behaviors of others within their social contexts has been reported in many health outcomes (smoking, obesity, mental health, mortality, etc.). Social externalities in health are important for policy because an exogenous change in individual behavior will produce social multipliers thus propagating a desired policy change beyond the initially targeted group. Empirical studies of peer effects face identification challenges including unobserved peer selection, endogeneity/ reflection, and common exposure to shared environmental influences. Social externalities in acute care settings have not been examined.
Data: The paper uses original data, collected for this study, on hospital roommate assignments and clinical outcomes of adult patients. The sample includes 1,401 females and 1,251 males who were discharged from a large urban teaching hospital during 6/1/11-12/31/11 and who had at least one roommate throughout the duration of the entire duration of the hospital stay. All data were obtained from the hospital's electronic medical records, including a new measure of patient clinical condition that is automatically generated for each patient using a metric of 26 clinical variables (test results, nursing assessment inputs) and continuously updated, in real time, using a novel clinically validated proprietary algorithm adopted by the hospital.
Model: Using the continuous measure of the patient’s clinical condition, the study estimates a standard lagged linear-in-means peer effect model where the patient's clinical condition score at discharge (t) is regressed on the average admission clinical condition scores (t-1) of the roommates weighted by the proportion of stay spent with each roommate, conditional on the patient's own clinical condition at admission (t-1) and room fixed effects.
Roommate assignments: Patients are assigned to rooms based on gender, diagnosis, and care needs. A set of balancing tests show that, conditional on a set of observable patient characteristics (gender, diagnosis) and room fixed effects, roommate assignments were plausibly exogenous.
Identification: the identifying source of variation is variation in the clinical condition of the roommates. Conditional randomization of roommate assignments deals with unobserved selection, and using a pre-exposure measure of the roommate's clinical condition deals with confounding due to reflection and exposure to the shared environment.
Results: The paper finds evidence of peer effects for females, but not for males. Females sharing a room with sicker patients on average had a poorer clinical condition at discharge (0.5 point lower discharge patient condition score for every 1 point reduction in the average admission score of the roommates). Female patients with severely ill roommates had an over two-fold increase in the risk of being readmitted back to the hospital after being discharged, and significantly greater odds of being at risk for hospital mortality. These effects appear to operate through psychological pathways are not driven by competition for nursing care. I do not observe similar effects for males.