A latent spatial factor approach for synthesizing opioid associated deaths and treatment admissions in Ohio counties

TL;DR

This paper develops a Bayesian spatial factor model to synthesize opioid death and treatment data at the county level in Ohio, revealing spatial patterns and social covariates associated with the epidemic's burden.

Contribution

It introduces a novel generalized spatial factor model that jointly analyzes death and treatment counts, accounting for spatial dependence and providing a unified measure of opioid epidemic burden.

Findings

Higher latent burden in southern Ohio counties

Treatment rates contribute to the burden in the south

Death rates contribute to the burden in the southwest

Abstract

Background: Opioid misuse is a major public health issue in the United States and in particular Ohio. However, the burden of the epidemic is challenging to quantify as public health surveillance measures capture different aspects of the problem. Here we synthesize county-level death and treatment counts to compare the relative burden across counties and assess associations with social environmental covariates. Methods: We construct a generalized spatial factor model to jointly model death and treatment rates for each county. For each outcome, we specify a spatial rates parameterization for a Poisson regression model with spatially varying factor loadings. We use a conditional autoregressive model to account for spatial dependence within a Bayesian framework. Results: The estimated spatial factor was highest in the southern and southwestern counties of the state, representing a higher burden of the opioid epidemic. We found that relatively high rates of treatment contributed to the factor in the southern part of the state; whereas, relatively higher rates of death contributed in the southwest. The estimated factor was also positively associated with the proportion of residents aged 18-64 on disability and negatively associated with the proportion of residents reporting white race. Conclusions: We synthesized the information in the opioid associated death and treatment counts through a spatial factor model to estimate a latent factor representing the consensus between the two surveillance measures. We believe this framework provides a coherent approach to describe the epidemic while leveraging information from multiple surveillance measures.