CONTEXT: Predictive modeling can identify neighborhoods at elevated risk of future overdose death and may assist community organizations’ decisions about harm reduction resource allocation. In Rhode Island, PROVIDENT is a research initiative and randomized community intervention trial that developed and validated a machine learning model that predicts future overdose at a census block group (CBG) level. The PROVIDENT model prioritizes the top 20th percentile of CBGs at highest risk of future overdose death over the subsequent 6-month period. In CBGs assigned to the trial intervention arm, these predictions are then displayed for partnering community organizations via an interactive mapping dashboard.
OBJECTIVE: To evaluate whether CBGs prioritized by the PROVIDENT model were associated with increased user engagement via an online dashboard for fatal overdose forecasting and resource planning.
DESIGN: We estimated prevalence ratios using modified Poisson regression models, adjusted for CBG-level characteristics that may confound the relationship between model predictions and dashboard engagement. SETTING: We used CBG-level data in Rhode Island (N = 809) from November 2021 to July 2024.
INTERVENTION; Our exposure of interest was whether each CBG was prioritized by the PROVIDENT model and shown as prioritized on the interactive mapping dashboard.
MAIN OUTCOME MEASURE: Our primary outcome was whether a dashboard user from any partnering community organization engaged (eg, clicked, interacted with dashboard elements, or completed assessment or planning surveys) with each CBG on the interactive mapping dashboard. RESULTS: After adjusting for previous model predictions and dashboard engagement, nonfatal overdose counts, and distribution of race and ethnicity, poverty, unemployment, and rent burden, dashboard users were 1.0 to 2.4 times as likely to engage with CBGs prioritized by the PROVIDENT model that were shown as prioritized on the dashboard as compared to CBGs that were prioritized by the PROVIDENT model that were blinded on the dashboard.
CONCLUSIONS: Interactive mapping tools with predictive modeling may be useful to support community-based harm reduction organizations in the allocation of resources to neighborhoods predicted to be at high risk of future overdose death.
Assessing user engagement with an interactive mapping dashboard for overdose prevention informed by predictive modeling in Rhode Island
Journal of Public Health Management and Practice [Epub 2025 Jul 18]. doi: 10.1097/PHH.0000000000002200.