The ongoing overdose crisis demands innovative approaches to risk assessment and prevention.

Dr. Rao's presentation, "Contextualizing Overdose Death Risk," showcased a collaborative research project aimed at understanding and modeling drug overdose risk factors. This effort was initiated through a workshop on data science and substance abuse encouraged by Daniel Feaster, Ph.D., professor in the Division of Biostatistics.

The project involved a multidisciplinary team, including Mengyu Liu, former Biostatistics Ph.D. student, Dr. Daniel Feaster, Viviana Horigian, Ph.D., professor and Director of Public Health Education, Imelda Moises, Ph.D., M.P.H., associate professor of Geography and Director of Global Health Studies, and collaborators from Washington State University and the Seattle Police Department (SPD).

The research team focused on developing a geospatial risk factor analysis and introducing the concept of "contextual vulnerability" to evaluate how these factors influence overdose death risk. 

Leveraging unique geolocated overdose event data provided by the SPD and geospatial hotspot detection with Risk Terrain Modeling (RTM), the analysis concentrated on King County, WA. The research intended to identify critical hotspots as a foundation for estimating individual contextual vulnerability using a pioneering multilevel model prediction approach.

“RTM, originally developed by criminologists at Rutgers University, is a spatial diagnostic tool used to identify environmental features that co-locate and interact to create behavior settings conducive to specific outcomes, in this case, overdose deaths,” explained Dr. Rao, who is also an emeritus professor at the Department of Public Health Sciences. This approach allowed the researchers to identify hotspots or areas with a high concentration of overdose deaths. 

Contextual factors such as neighborhood attributes, accessibility to health facilities, and housing quality, are of extreme importance in contributing to overdose death risk. Dr. Rao and the research team aimed to provide more accurate risk stratification and prediction models by analyzing these factors with individual-level data. 

“The innovation of contextual vulnerability lies in integrating individual data with geospatial hotspots,” Dr. Rao explained. Using generalized linear mixed models, the team could classify individuals as contextually vulnerable or not, based on their interaction with these high-risk environments. This classification was further refined through model averaging techniques from machine learning, improving the prediction of overdose death risk. 

Utilizing SPD data from multiple years, researchers demonstrated the non-random distribution of overdose events and the presence of identifiable hotspots. 

Findings revealed that incorporating contextual vulnerability into predictive models significantly improved prediction accuracy, especially for individuals classified within this group. This improvement was evident across multiple years of data, reinforcing the robustness of the approach. 

Dr. Rao emphasized the potential implications of their findings for targeted intervention strategies and personalized prevention measures. He suggested linking compositional and contextual factors could enhance overdose detection and response systems. This could incorporate digital systems and user geolocation to provide real-time risk alerts. 

“As the drug overdose epidemic continues to worsen in the U.S., alternative ways of thinking about overdose risk are needed,” said Dr. Rao. “This comprehensive approach offers a deeper understanding of the multifaceted nature of overdose risks and paves the way for more targeted and effective intervention strategies,” he continued.

Written by Deycha Torres Hernández, published on May 20, 2024.