Over 2,000 Tennesseans died in 2019 from overdose – can we predict overdose risk anywhere in the state to guide prevention? Our study on this question is out now in the Journal of the American Medical Informatics Association.
Since 2012, Tennessee has participated in a comprehensive Prescription Drug Monitoring Program called the Controlled Substance Monitoring Database (CSMD). To date, there have been no tested, rigorously evaluated prediction algorithms using those data in Tennessee to guide overdose prevention. Thanks to a rare and extraordinary collaboration with the Tennessee Department of Health (TDH) we were able to develop these models in a secure, privacy-preserving, and scalable way using CSMD data, the Hospital Discharge Data System data, publicly available community-level data, and Vital Statistics data.
Our TDH collaboration is not important because of data – it only matters because of the people (like Melissa McPheeters, Ben Tyndall, Charlotte Cherry, Sanura Latham, and Allison Roberts), the expertise, the experience, and the potential uses of scalable predictive models of overdose risk downstream. The VUMC team includes lead author Michael Ripperger – application developer for the lab – as well as Sarah Lotspeich, Drew Wilimitis, Carrie Fry, Matthew Lenert, Katey Robinson, and Cindy Chen.
The open access manuscript, available here, demonstrates how ensemble learning on comprehensive state data might inform risk for overdose events that are extremely rare at statewide scale. To read more about our wonderful team and collaboration with TDH, check out this story written up in the VUMC Reporter.