99 Modeling recurrent overdose patterns to guide real-world harm-reduction placement decisions

Objectives/Goals: To quantify the association between proximity to harm-reduction vending machines and recurrent nonfatal overdose using recurrent-event survival models, estimate effects on time to reoverdose and hazard, and identify modifiable predictors to guide machine sitting. Methods/Study Population: Retrospective cohort of Milwaukee County nonfatal overdoses, Jan 2018 to Dec 2023 (28,108 events; 20,300 people). Records were geocoded, deduplicated, linked within person, and tagged as single or recurrent. Exposure: residence or event within a 30 min walk of a harm-reduction vending machine. Recurrence summarized with Kaplan-Meier curves and mean cumulative function. An interrupted time series (ITS) assessed pre- vs post-HRVM expansion. For inference, we compared Andersen-Gill, Prentice-Williams-Peterson, and Cox ML models, selecting AG gap time. Covariates: age, sex, race, jurisdiction, location type, disposition, proximity. Assumptions checked; hazards reported with 95% CIs. Results/Anticipated Results: 2018 to 2023: 28,108 nonfatal overdoses among 20,300 people; 3,803 had repeats (11,611 events). Inter-event times shortened with each recurrence. Kaplan–Meier for first to second overdose showed small, consistent separation favoring those within 30 min of HRVMs; mean cumulative functions rose more slowly near HRVMs. The selected Andersen–Gill gap-time model found no meaningful short-term effect of proximity (aHR ≈ 1.03, 95% CI 1.00–1.06). Younger age and public or commercial locations predicted faster recurrence. County ITS suggested stabilization after HRVM expansion without a clear early reduction. Discussion/Significance of Impact: Early HRVM rollout yields small reductions in cumulative recurrence without a meaningful short-term hazard change. Results support MOUD linkage and mobile outreach targeting younger people and public settings. The recurrent-event and ITS framework is generalizable for sitting and monitoring.