Classifying healthcare facilities as predictors of COVID-19 mortality rates in US counties (2020–2021)
Edwin M McCulley, Jana A Hirsch, Alina Schnake-Mahl, Brisa Sanchez, Gina S Lovasi, Usama BilalAbstract
Background
The COVID-19 pandemic disproportionately impacted vulnerable populations, with contextual factors like healthcare accessibility influencing mortality. However, limited evidence exists on which types of healthcare facilities affect COVID-19 death rates.
Methods
We examined which facility types were statistically associated with, and improved prediction of, county-level COVID-19 mortality (2020–2021) using over dispersed Poisson models and healthcare facility data from the 2020 National Establishment Time Series database. Five feature selection strategies guided model construction: a theory-driven approach, three data-driven methods [Least Absolute Shrinkage and Selection Operator (LASSO), stepwise, and random forest], and a synthesized strategy integrating shared predictors.
Results
Based on Quasi-Akaike’s Information Criterion (QAIC), LASSO and stepwise models offered the best fit. Across methods, consistent predictors of county-level COVID-19 mortality rates included pharmacies/drug stores, hospitals and major medical centers, emergency medical transport, offices and clinics of health practitioners, and urgent care facilities. Data-driven strategies also selected chiropractors, highlighting potential confounding bias.
Conclusions
Our classification approach highlights facility types associated with COVID-19 mortality, offering insight into how healthcare infrastructure may influence pandemic-related health outcomes. These findings can support descriptive characterizations of local medical environments, generate hypotheses, and guide future research aimed at improving population health during public health emergencies.