Tides Need STEMMED: A Locally Operating Spatiotemporal Mutually Exciting Point Process with Dynamic Network for Improving Opioid Overdose Death Prediction.

This article presents STEMMED, a novel spatiotemporal mutually exciting point process with dynamic network model designed to improve public health surveillance and forecasting of opioid overdose deaths (OODs) across local communities and drug types. STEMMED captures dynamic, personalized triggering effects between OOD events by modeling interactions among community–drug pairs, incorporating nodal features (e.g., population, poverty rate, treatment coverage) and individual-level characteristics (e.g., age, sex, race). Applied to Massachusetts data from 2015 to 2022, STEMMED reveals evolving OOD dynamics, including a growing link between fentanyl and psychostimulants and spatial clustering centered around Boston, and outperforms established forecasting models in accuracy and timeliness. The study also evaluates data-sharing policies, finding that drug type–based collaboration among communities enhances forecasting performance more than geographic-based sharing. STEMMED’s forecasting system reduces detection delays by over 50%, offering actionable insights for targeted, timely public health interventions while acknowledging limitations related to causal inference, data privacy, and model scalability.