From Prediction to Practice: Ethical Considerations for the Integration of Machine Learning in Community-Based Overdose Prevention

Adelya Urmanche, PhD – Assistant Professor, Department of Psychology, Appalachian State University; Brenda Curtis, PhD – Chief, Technology and Translational Research Unit, National Institute on Drug Abuse; Celia Fisher, PhD – Marie Ward Doty University Chair in Ethics, Center for Ethics Education, Fordham University

Abstract: The integration of machine learning (ML) into community-level public health practice offers new possibilities for enhancing data-driven decision-making. Predictive analytics can inform the distribution of scarce public health resources, yet whether ML-driven forecasts should guide intervention placement—rather than historical trends—remains an open question. As spatiotemporal overdose forecasting improves, some service providers have begun using ML-driven predictions to allocate overdose prevention and outreach services at the community level, introducing a novel model of data-informed public health practice. However, ethical considerations surrounding ML-based resource allocation remain underexplored. This qualitative empirical study investigates the ethical dimensions of using ML to guide overdose prevention service delivery. Between June 2024 and February 2025, we conducted semi-structured interviews with 25 harm reduction providers across the New York City metropolitan region. Using a visual vignette to illustrate ML-informed resource allocation, we explored provider perspectives on fairness, transparency, and implementation challenges related to this new model of public health practice. Findings reveal providers’ ethical concerns related to distributive and procedural justice, data ownership, privacy, and proportionality. Participants questioned whether ML models might displace community expertise, exacerbate inequities in service access and overdose outcomes, or misrepresent populations. Others saw ML as an advocacy lever to justify increased service to marginalized groups, such as gender and sexual minorities. Skepticism emerged in organizations with strong community ties, where providers doubted ML’s added value. This study provides early insights into the ethical implications of ML-driven public health practice, offering guidance for policymakers and practitioners considering predictive analytics for overdose prevention.

Keywords: public health ethics, machine learning and artificial intelligence, overdose prevention