A Bayesian Spatio-Temporal Top-Down Framework for Estimating Opioid Use Disorder Risk Under Data Sparsity

TL;DR

This paper introduces a Bayesian hierarchical spatio-temporal framework to estimate county-level opioid use disorder rates in the US, effectively handling data sparsity and quantifying uncertainty for better targeted interventions.

Contribution

It presents a novel top-down Bayesian approach that leverages state and county covariates to estimate small-area OUD rates when direct data is unavailable.

Findings

Accurately estimated county-level OUD rates for over 3,000 US counties.

Demonstrated improved uncertainty quantification in data-sparse regions.

Validated model performance through simulation studies.

Abstract

County-level estimates of opioid use disorder (OUD) are essential for understanding the influence of local economic and social conditions. They provide policymakers with the granular information needed to identify, target, and implement effective interventions and allocate resources appropriately. Traditional disease mapping methods typically rely on Poisson regression, modeling observed counts while adjusting for local covariates that are treated as fixed and known. However, these methods may fail to capture the complexities and uncertainties in areas with sparse or absent data. To address this challenge, we developed a Bayesian hierarchical spatio-temporal top-down approach designed to estimate county-level OUD rates when direct small-area (county) data is unavailable. This method allows us to infer small-area OUD rates and quantify associated uncertainties, even in data-sparse environments using observed state-level OUD rates and a combination of state and county level informative covariates. We applied our approach to estimate OUD rates for 3,143 counties in the United States between 2010 and 2025. Model performance was assessed through simulation studies.