Predict to Prevent: Dynamic Spatiotemporal Analyses of Opioid Overdose to Guide Pre-Emptive Public Health Responses PROJECT SUMMARY/ABSTRACT Opioid overdose (OD) fatalities have reached crisis levels in all socioeconomic and geographic communities in the US. By utilizing a first-of-its-kind statewide Public Health Data Warehouse (PHD) with multiple linked administrative datasets and state-of-the-art Bayesian spatiotemporal models, we are in a unique position to fill in the fundamental gaps in the field’s ability to rapidly identify current OD patterns, predict future OD epidemics, and evaluate the effectiveness of public health and clinical interventions. In Massachusetts (MA), the State Legislature enacted policy in 2015 that provided authorization to the MA Department of Public Health (MDPH) to develop a massively linked administrative dataset to allow public health officials and policymakers to better understand the extent of and contributors to the opioid OD epidemic. The PHD Warehouse, representing 98% of the MA population, currently links data from 25+ distinct sources (e.g., death records, all-payer claims, post-mortem toxicology, hospital discharges, and the prescription monitoring program). Supported by strong preliminary studies demonstrating the power of the PHD and our strong partnership with MDPH, we aim to develop a new population health analytic framework to support opioid OD control in MA that can be generalizable to other parts of the country. Our Specific Aims are to: 1) Develop a Bayesian multilevel spatiotemporal model to identify individual, interpersonal, community, and societal factors that contribute to opioid OD; 2) develop an efficient Bayesian spatiotemporal model to identify time- space OD clusters, and extend the model to construct a dynamic predictive model; and, 3) evaluate and predict policy and intervention effects through model-based simulation studies to provide practical guidance and decision-making support to public health officials. Aims 1, 2 and 3 can be easily adopted and reproduced by users in other public health jurisdictions and sectors to foster cross-sector, cross-agency opioid OD control. Our approach is innovative due to the use of PHD and sophisticated Bayesian spatiotemporal modeling approaches. The proposed study is highly significant, because it is conceptualized to improve current and future public health practice, facilitating data-driven and evidence-based implementation science interventions in the locations at greatest risk and at the time when they are most needed. Our results can immediately and significantly influence opioid OD prevention policies and practices, guiding pre-emptive public health and clinical responses. We will develop our visualization tools, analytical approaches, and related code, in collaboration with MDPH and our Community Advisory Board (CAB), to enhance PHD capabilities and improve dissemination of findings. Our tools, approaches, and code will also be made available for national dissemination, providing paradigm shifting approaches to address the opioid crisis. Our research directly addresses NIDA’s goal to “Develop new and improved strategies to prevent drug use and its consequences.”

Predict to Prevent: Dynamic Spatiotemporal Analyses of Opioid Overdose to Guide Pre-Emptive Public Health Responses NARRATIVE Novel spatiotemporal forecasting models that rely on massively linked administrative data are needed to forecast future opioid overdose spikes and assess public health intervention success. Guided by the social ecological model, we plan to develop a comprehensive approach to identify individual, interpersonal, community and societal factors that contribute to opioid overdose, efficiently detect overdose hotspots, and develop dynamic forecasting models for timely prediction and prevention of future opioid overdose epidemics. We will disseminate blueprints, visualization tools, and code to other jurisdictions to facilitate replication of our approach.