The Impact of U.S. County-Level Factors on COVID-19 Morbidity and Mortality

Itzhak, N.; Shahar, T.; Moskovitch, R.; Shahar, Y.

Background: The effect of socioeconomic factors, ethnicity, and other variables, on the frequency of COVID-19 cases [morbidity] and induced deaths [mortality] at sub-population, rather than at individual levels, is only partially understood. Objective: To determine which county-level features best predict COVID-19 morbidity and mortality for a given county in the U.S. Design: A Machine-Learning model that predicts COVID-19 mortality and morbidity using county-level features, followed by a SHAP-values-based importance analysis of the predictive features. Setting: Publicly available data from various American government and news websites. Participants: 3,071 U.S. counties, from which 53 county-level features, as well as morbidity and mortality numbers, were collected. Measurements: For each county: Ethnicity, socioeconomic factors, educational attainment, mask usage, population density, age distribution, COVID-19 morbidity and mortality, air quality indicators, presidential election results, ICU beds. Results: A Random Forest classifier produced an AUROC of 0.863 for morbidity prediction and an AUROC of 0.812 for mortality prediction. A SHAP-values-based analysis indicated that poverty rate, obesity rate, mean commute time to work, and proportion of people that wear masks significantly affected morbidity rates, while ethnicity, median income, poverty rate, and education levels, heavily influenced mortality rates. The correlation between several of these factors and COVID-19 morbidity and mortality, from 4/2020 to 11/2020 shifted, probably due to COVID-19 being initially associated with more urbanized areas, then with less urbanized ones. Limitations: Data are still coming in. Conclusions: Ethnicity, education, and economic disparity measures are major factors in predicting the COVID-19 mortality rate in a county. Between-counties low-variance factors (e.g., age), are not meaningful predictors. Differing correlations can be explained by the COVID-19 spread from metropolitan to less metropolitan areas.