Data and Method: in this study, we merge the state-level unemployment rates with the Nationwide Inpatient Sample (NIS) from 1993 to 2011 by hospital location and the time of patient admission. Post-2011 NIS no longer has hospital location information and is thus not included in this study. Patient outcomes are measured by three indicators including the probability of hospital acquired infection, length of stay and inpatient mortality among high risk patients. Probit models and nonlinear regression models are used. Marginal effects are derived and reported. Our data is well suited for this kind of study. First, it captures accurately the local economic environment based on both the state and the month. Second, it has detailed information on patient demographics, diagnosis, severity, and comorbidities which reduces the omitted variable bias. Third, the study period of 19 years helps estimate the long term impact of local economic conditions on patient outcomes.
Results: We find that local unemployment is significantly associated with patient outcomes with varying magnitude across different indicators and different insurance types. Although the exact mechanism is unclear, we propose several hypotheses including capacity constraints, supply-side inducement and demand-side inducement.