Abstract P659: Cluster Analysis of County-Level Cardiovascular Disease Mortality in the United States

Abstract

Introduction: In the United States, cardiovascular disease (CVD) mortality has been declining since 1980s. However, the decline in CVD mortality is not uniform across regions and may show different trajectories that are driven by place-based social determinants of health (SDOH) such as physical and built environment, social-cultural environment, and health care system. We conducted a cluster analysis of county-level CVD mortality and SDOH indicators from 2009 to 2018. Hypothesis: We test the hypothesis that CVD mortality in US counties show different trajectories that can be grouped by clusters, and within each cluster, there are different SDOH variables that predict CVD mortality. Methods: We linked the Centers for Disease Control and Prevention's CVD mortality data with the Agency for Healthcare Research and Quality's public database on SDOH using county identifiers. CVD mortality rate was calculated as the number of deaths per 100,000 people in each county. There were 179 SDOH variables included in the model, classified into five different domains: the biological, behavioral, built and physical environment, health care system, and sociocultural environment domain. We employed a linear mixed-effects model fitted with a doubly penalized likelihood to induce sparsity for parameter estimation and effect selection. Results: We identified three clusters among a total of 3,136 counties. Overall, 1,182 counties were identified as cluster 1 that showed a consistent decline of CVD mortality over the past decade. The most important driving factor of CVD mortality in cluster 1 was population distribution - the percentage of population aged 15-17 (coefficient=7.032). Secondly, 1,116 counties were identified as cluster 2 that showed a decline between 2009 and 2012, and then a steady increase in CVD mortality after 2012. Within cluster 2, the number of mild drought months was the most predictive factor of CVD mortality (coefficient= -1.065), and the percentage of adults reporting no leisure-time physical activity was the second important predictor (coefficient= 0.646). Finally, 838 counties were identified as cluster 3 that showed a flattening trend in CVD mortality. Within cluster 3, the total number of obstetrics and gynecology, generalist, and primary care hospital residents plays the most important role (coefficient = -25.537). Each cluster has unique pattern and distinctive associations between certain SDOH variables and CVD mortality. Conclusions: In conclusion, we identified three intrinsic clusters that showed distinguishing trajectories of CVD mortality during the past decade. The SDOH indicators most predictive of CVD mortality were different within each cluster. This study suggests that place based SDOH drive substantial differences in CVD outcomes across regions.