Abstract
INTRODUCTION: Prevalence estimates of opioid use disorder (OUD) at local levels are critical for public health planning and surveillance, yet largely unavailable across the US especially at the local county level. METHODS: We used a Bayesian evidence synthesis approach to estimate the prevalence of OUD for 57 counties across New York State for 2017-2019 and compare rates of OUD across counties as well as assess the extent of undiagnosed OUD. We developed a generative model to assess conditional probabilistic relations between different subgroups of the OUD population defined by diagnosis, treatment, and overdose fatality. RESULTS: Average OUD prevalence from 2017 to 2019 ranged from 2.4 % (Westchester County) to 8.3 % (Sullivan County). Overall OUD prevalence estimates were relatively stable year to year, from 4.5 % in 2018 and 4.6 % in both 2017 and 2019. The Bayesian evidence synthesis estimate is consistently higher than the percentage diagnosed in Medicaid, by age and sex. By 2019, the estimated proportion of OUD that was undiagnosed ranged from 16.7 % in Clinton County to 62.7 % in Onondaga County. Counties with relatively high overdose death rates and low buprenorphine prescription percentages had the highest estimated level of undiagnosed OUD. CONCLUSION: OUD prevalence varied considerably across the state. We identified counties with high OUD and overdose levels and a high proportion of undiagnosed OUD, providing insight into areas of the state in need of rapid expansion of services for people with OUD. Bayesian evidence synthesis approaches for OUD prevalence estimation represent a reliable and rigorous approach to providing local areas with information on OUD magnitude.