Transforming data for harm reduction progress: Lessons on predictive analytics from the US-based PROVIDENT project

Author: Brandon Marshall, Jesse Yedinak, Claire Pratty, Yu Li, Abigail Cartus, William Goedel, Benjamin Hallowell, Cathy Schultz, Melissa Basta, Jennifer Ahern, Daniel Neill, Bennett Allen, Magdalena Cerdá

Theme: Epidemiology & Public Health Research Year: 2023


The worsening overdose crisis in the United States means that harm reduction
organizations are being expected to provide services to growing, more diverse, and geographically
dispersed populations. Now more than ever, harm reduction programmes need accessible and
timely information to inform allocation of services that are in growing demand. Predictive analytic
techniques such as machine learning could help harm reduction programmes identify communities
in highest need.

Yet, there are no examples of the responsible and ethical use of forecasting models
for harm reduction service provision. In this context, we implemented PROVIDENT— a partnership
between researchers, a state health department, and local harm reduction organizations in the US
state of Rhode Island—to develop, implement, and evaluate a machine learning-based forecasting
tool to predict future overdose deaths at the neighborhood level. We then worked collaboratively to
guide harm reduction resources to communities the model identified as being at highest risk of
future overdose deaths.

Key successes included the validation of a forecasting model that
successfully predicted the top 20% of neighborhoods statewide in which over 40% of overdose
deaths occurred in the six months following predictions, and the implementation of an interactive
web tool to visualize the model predictions and guide resource allocation.


Key challenges included addressing ethical concerns about the reallocation of urgently needed overdose prevention
resources based on modeled predictions, translation of machine learning predictions to frontline
harm reduction practice, and limited organizational capacity in light of unprecedented and
escalating COVID-19, economic, and housing crises. In this presentation, we will discuss how we
addressed these challenges using a community-engaged research approach and the usefulness of
machine learning models for harm reduction funding, policy, and praxis.

Disclosure of Interest Statement:

The authors disclose no conflicts of interest. Note: the Rhode
Island Department of Health (RIDOH) and the Rhode Island Executive Office of Health and Human
Services (EOHHS) are not responsible for any analyses, opinions or conclusions contained in this
document. The views expressed in this report are those of the authors and do not represent the
official positions or policy of the RIDOH or EOHHS

Download abstract Download poster