Using machine learning to estimate health spillover effects.

We develop a nonparametric model to study health spillover effects of policy interventions. We use double/debiased machine learning to estimate the model using data from 74 hospitals in Rio de Janeiro, Brazil, and examine cross-patient spillover effects during the COVID-19 pandemic. The pandemic forced hospitals to develop new protocols to offer intensive care to both COVID and non-COVID patients. Our results show that the need to care for COVID patients affects health outcomes of non-COVID patients. Controlling for a number of confounders, we find that mortality rates and length of stay of non-COVID ICU patients increase when hospitals simultaneously offer intensive care to both types of patients. Policy simulations suggest that an increase in the number of ICU beds can counter morbidity spillover, but it is unlikely to be a feasible approach to counter mortality spillover.