Evaluating the predictive performance of different data sources to forecast overdose deaths at the neighborhood level with machine learning in Rhode Island.

The study aimed to assess the predictive performance of various data sources in forecasting fatal overdoses in Rhode Island neighborhoods. The ultimate goal was to provide a template for other jurisdictions to utilize predictive analytics in directing overdose prevention resources. To achieve this, the researchers evaluated seven different combinations of data from six administrative sources, including the American Community Survey (ACS), built environment, emergency medical services, prescription drug monitoring program, carceral release, and historical fatal overdose data. They used two machine learning approaches, linear regressions and random forests, to predict fatal overdoses in Rhode Island census block groups (CBGs) from 2019 to 2021. The results showed that linear models trained on ACS data combined with one other data source performed well, and their performance was comparable to models trained on all available data. Specifically, models that included emergency medical service, prescription drug monitoring program, or carceral release data with ACS data were able to capture a significant percentage of statewide overdoses. These models were able to identify neighborhoods that were at high risk of fatal overdoses, making it possible to prioritize them for overdose prevention efforts. The study found that using a simple model trained on publicly available ACS data combined with only one other administrative data source was sufficient to achieve the desired goals. This suggests that other jurisdictions can replicate this approach, using readily available data to forecast overdose risk and direct prevention resources effectively. The findings of this study provide a valuable starting point for other jurisdictions interested in utilizing predictive analytics to combat the opioid epidemic. Link: https://pubmed.ncbi.nlm.nih.gov/40164400/