Development and validation of a community-level social determinants of health index for drug overdose deaths in the HEALing Communities Study

Abstract

Introduction: Social determinants of health (SDoH), such as socioeconomic status, education level, and food insecurity, are believed to influence the opioid crisis. While global SDoH indices such as the CDC's Social Vulnerability Index (SVI) and Area Deprivation Index (ADI) combine the explanatory power of multiple social factors for understanding health outcomes, they may be less applicable to the specific challenges of opioid misuse and associated outcomes. This study develops a novel index tailored to opioid misuse outcomes, tests the efficacy of this index in predicting drug overdose deaths across contexts, and compares the explanatory power of this index to other SDoH indices.

Methods: Focusing on four HEALing Communities Study (HCS) states (Kentucky, Massachusetts, New York and Ohio; encompassing 4269 ZIP codes), we identified multilevel SDoH potentially associated with opioid misuse and aggregated publicly available data for each measure. We then leveraged a random forest model to develop a composite measure that predicts age-adjusted drug overdose mortality rates based on SDoH. We used this composite measure to understand HCS and non-HCS communities in terms of overdose risk across areas of varying racial composition. Finally, we compared variance in drug overdose deaths explained by this index to variance explained by the SVI and ADI.

Results: Our composite measure included 28 SDoH measures and explained approximately 89 % percent of variance in age-adjusted drug overdose mortality across HCS states. Health care measures, including emergency department visits and primary care provider availability, were top predictors within the index. Index accuracy was robust within and outside of HCS communities and states. This measure identified high levels of overdose mortality risk in segregated communities.

Conclusions: Existing SDoH indices fail to explain much variation in area-level overdose mortality rates. Having tailored composite indices can help us to identify places in which residents are at highest risk based on their composite contexts. A comprehensive index can also help to develop effective community interventions for programs such as HCS by considering the context in which people live.

Keywords: Drug overdose mortality; Machine learning; Opioid misuse; Public health; Social determinants.