Assessing User Engagement With an Interactive Mapping Dashboard for Overdose Prevention Informed by Predictive...

Area of Science:

• Public Health
• Data Science
• Epidemiology

Background:

• Predictive modeling aids in identifying high-risk neighborhoods for overdose deaths.
• Machine learning models, like PROVIDENT in Rhode Island, predict overdose risk at the census block group (CBG) level.
• This predictive capability supports harm reduction resource allocation by community organizations.

Purpose of the Study:

• To evaluate if CBGs identified by the PROVIDENT model received increased user engagement on an online dashboard.
• To assess the utility of a predictive overdose forecasting and resource planning dashboard for community organizations.

Main Methods:

• Utilized modified Poisson regression to estimate prevalence ratios, adjusting for confounding CBG-level characteristics.
• Analyzed CBG-level data from Rhode Island (N=809) between November 2021 and July 2024.
• The exposure was whether a CBG was prioritized by the PROVIDENT model and displayed on the interactive dashboard.

Main Results:

• Dashboard users were 1.0 to 2.4 times more likely to engage with CBGs prioritized and shown on the dashboard.
• Adjustments were made for prior predictions, engagement, overdose counts, and demographic factors.
• This indicates a significant association between model prioritization and user interaction with the dashboard.

Conclusions:

• Interactive mapping tools combined with predictive modeling can effectively support harm reduction organizations.
• These tools aid in resource allocation to neighborhoods identified as high-risk for future overdose deaths.
• The PROVIDENT model's dashboard demonstrates potential for enhancing community-based public health interventions.