Understanding Complex Patterns in Social, Geographic, and Economic Inequities in COVID-19 Mortality at the County Level in the US Using Generalized Additive Models

TL;DR

This paper demonstrates how generalized additive models (GAMs) can be used to analyze and visualize complex, changing patterns of COVID-19 mortality across US counties, accounting for social, economic, and geographic factors.

Contribution

It introduces three novel applications of GAMs for documenting COVID-19 mortality inequities, including dynamic relationship modeling, spatiotemporal smoothing, and covariate association estimation conditioned on risk surfaces.

Findings

GAMs reveal changing associations between mortality and county-level factors.

Spatiotemporal smoothing produces a dynamic risk surface highlighting geographic disparities.

Conditional modeling shows how socio-environmental factors influence mortality risk.

Abstract

I present three types of applications of generalized additive models (GAMs) to COVID-19 mortality rates in the US for the purpose of advancing methods to document inequities with respect to which communities suffered disproportionate COVID-19 mortality rates at specific times during the first three years of the pandemic. First, GAMs can be used to describe the changing relationship between COVID-19 mortality and county-level covariates (sociodemographic, economic, and political metrics) over time. Second, GAMs can be used to perform spatiotemporal smoothing that pools information over time and space to address statistical instability due to small population counts or stochasticity resulting in a smooth, dynamic latent risk surface summarizing the mortality risk associated with geographic locations over time. Third, estimation of COVID-19 mortality associations with county-level covariates conditional on a smooth spatiotemporal risk surface allows for more rigorous consideration of how socio-environmental contexts and policies may have impacted COVID-19 mortality. Each of these approaches provides a valuable perspective to documenting inequities in COVID-19 mortality by addressing the question of which populations have suffered the worst burden of COVID-19 mortality taking into account the nonlinear spatial, temporal, and social patterning of disease.