In 2023, Kentucky had the fifth-largest drug overdose fatality rate in the United States, and 79% of those deaths involved opioids. This project develops a statewide surveillance system to monitor and respond to the opioid crisis, collecting data from a variety of sources and agencies around Kentucky.
The project uses predictive analytics and ML techniques to forecast future trends of opioid overdoses in different areas of Kentucky, providing forecasts across geographical levels to identify high-risk regions. Results are shared via dashboard for stakeholder action.
Past Covariates:
Future Covariates: Monthly temperature and precipitation
Static Covariates: Regional characteristic indicators
| Model | Type | Notes |
|---|---|---|
| Linear Regression | Baseline | - |
| N-Linear | Single-layer neural network | Best overall (RMSE) |
| Temporal Fusion Transformer | Deep learning with self-attention | Best with all covariate types |
Best covariates: Medicaid OUD treatment data, DOC intake/release data, nearby regional trend averaging. N-Linear performed best at the Area Development District level.
The prediction dashboard is available at: rador-ky.uky.edu
Paper accepted to 2025 AMIA Symposium: arXiv:2410.16500
Funded by NIH (R01DA057605-01, R01DA057605-01S1, R01DA057605-01S2). Uses the Forecaster platform for model training and evaluation.