Geographically specific associations between county-level socioeconomic and household distress and mortality from drug poisoning, suicide, alcohol, and homicide among working-age adults in the United States

Abstract

Background: Mortality rates from drug poisoning, suicide, alcohol, and homicide vary significantly across the United States. This study explores localized relationships (i.e., geographically specific associations) between county-level economic and household distress and mortality rates from these causes among working-age adults (25-64).

Methods: Mortality data were from the National Vital Statistics System for 2014-2019. County-level socioeconomic distress (poverty, employment, income, education, disability, insurance) and household distress (single-parent, no vehicle, crowded housing, renter occupied) were from the 2009-2013 American Community Survey. We conducted Ordinary Least Squares (OLS) regression to estimate average associations and Geographically Weighted Regression (GWR) to estimate localized spatial associations between county-level distress and working-age mortality.

Results: In terms of national average associations, OLS results indicate that a one standard deviation increase in socioeconomic distress was associated with an average of 6.1 additional drug poisoning deaths, 3.0 suicides, 2.1 alcohol-induced deaths, and 2.0 homicides per 100,000 population. A one standard deviation increase in household distress was associated with an average of 1.4 additional drug poisonings, 4.7 alcohol-induced deaths, and 1.1 homicides per 100,000 population. However, the GWR results showed that these associations vary substantially across the U.S., with socioeconomic and household distress associated with significantly higher mortality rates in some parts of the U.S than others, significantly lower rates in other parts of the U.S., and no significant associations in others. There were also some areas where distress overlapped to influence multiple causes of death, in a type of compounded disadvantage.

Conclusions: Socioeconomic and household distress are significant and substantial predictors of higher rates of drug poisoning mortality, suicide, alcohol-induced deaths, and homicide in specific regions of the U.S. However, these associations are not universal. Understanding the place-level factors that contribute to them can inform geographically tailored strategies to reduce rates from these preventable causes of death in different places.

Keywords: Alcohol; Drug poisoning; Geographically weighted regression; Homicide; Suicide; Working age mortality.

Fig. 1

County-level mortality rates from drug poisoning, suicide, alcohol-induced death, and homicide among U.S. working age adults (ages 25–64). Note: Quantiles method with 7 categories in ArcGIS are used to categorize mortality rate from each cause of deaths.

Fig. 2

Local Indicators of Spatial Association (LISA) for county-level mortality rates from drug poisoning, suicide, alcohol-induced deaths, and homicide among working age adults. Note: ‘High-High’ denotes counties with above average mortality rates that are surrounded by neighboring counties with above average mortality rates. ‘Low-Low’ denotes counties with below average mortality rates that are surrounded by neighboring counties with below average mortality rates. ‘High-Low’ denotes counties with above average mortality rates surrounded by counties with below average mortality rates. ‘Low-High’ denotes counties with below average mortality rates surrounded by counties with above average mortality rates.

Fig. 3

Results from geographically weighted regression models identifying factors significantly related to mortality from drug poisonings, sucides, alcohol-induced deaths, and homicides, 2014-19. Note: Mortality rates are age adjusted. Significance is based on an alpha <.05, and a VIF score lower than 5. Quantile classification is used to categorize relationships between social vulnerability and each cause of death in most maps. In cases where there were a small number of counties with negative associations and where the range of negative values were small, we either dichotomized the negative coefficients (Fig. 3A1 or used only one category to show the counties with negative coefficients).