Mathematical Sciences Department Statistics Seminar - Jianing Wang, Mass General Hospital, Biostatistics Center and Harvard Medical School (FL320)

Thursday, November 9, 2023
11:00 am to 12:00 pm
Floor/Room #


Mathematical Sciences Department

Statistics Seminar

Speaker: Jianing Wang, Massachusetts General Hospital Biostatistics Center and Harvard Medical School

Thursday, November 9, 2023

11:00 am - 12:00 pm

Fuller Labs 320

Title: A spatial capture-recapture approach for estimating opioid use disorder prevalence in small areas using linked health administrative data

Abstract: The evolving nature of the opioid epidemic highlights the need to estimate the overall and subpopulation prevalence of opioid use disorder (OUD). Local jurisdictions often seek localized estimates to provide actionable insights for resource allocation. However, it is challenging to use traditional estimators because this population is hard to reach under typical sampling frameworks and is incompletely observed in administrative data, which are the primary sources of data for opioid surveillance. The classic capture-recapture (CRC) framework has been widely used to overcome this issue. However, producing reliable local estimates under this framework by stratification suffers from small sample issues. Additionally, there might be geographical variation in data recording and integration which should be accounted for. Ignoring such spatial patterns in prevalence, especially the effect of spatial proximity, could limit the applicability of a stratified approach. Therefore, we propose a two-stage Bayesian hierarchical model with spatial smoothing using a CRC data structure to estimate OUD prevalence among small areas. We explicitly model the hidden prevalence of interest and the associated detection model that describes individual detection histories across data sources. An intrinsic conditional autoregressive model (ICAR) is specified to capture spatial dependence. Comparing our model to a traditional log-linear model stratified by geographic area, simulation studies demonstrate that our framework resolves convergence issues in the traditional method and accurately estimates area-specific prevalence with lower variance. To illustrate our method, we applied it to estimate city-level OUD prevalence in Massachusetts. Our modeling framework is the first to explicitly model localized prevalence variation under this specific data structure. This framework can also incorporate additional community characteristics vital for understanding OUD epidemic patterns, empowering local jurisdictions to identify spatial clusters and high-risk OUD areas for precise intervention deployment.

Bio: I am an Assistant Investigator in Biostatistics at Massachusetts General Hospital (MGH) Biostatistics Center and a member of the faculty in medicine at Harvard Medical School (HMS). I’m working on projects regarding community-based high-performance surveillance networks for drug use and the platform trial design for Amyotrophic Lateral Sclerosis (ALS). I earned a Ph.D. in Biostatistics and an MSc in Applied Statistics from Boston University. Before joining MGH, my research contributed to projects under the NIH HEAL (Helping to End Addiction Long-Term) Initiative, with a primary focus on leveraging large-scale healthcare surveillance data to predict the scope of hidden populations impacted by substance misuse and use disorders across various geographies, time periods, and vulnerable demographic groups. 



Mathematical Sciences