These projects harness advanced analytic and machine-learning techniques to forecast health outcomes and inform targeted interventions. Our projects leverage diverse data sources—such as Medicaid claims, electronic health records, criminal justice, and EMS datasets—to build predictive models for opioid overdose and other high-stakes events. Working alongside public health departments, healthcare systems, and community stakeholders, we develop and validate tools that accurately identify at-risk populations, enabling proactive, data-driven strategies to improve patient safety and healthcare quality across multiple domains.
Risk Prediction Research Projects
Active Risk Prediction Projects
Machine-Learning Prediction and Reducing Overdoses with EHR Nudges (mPROVEN)
Funder: NIH/NIDA; PI: Gellad
The mPROVEN study uses a machine learning risk prediction tool embedded in the electronic health record to identify and support patients at high risk of opioid overdose.
Developing and Evaluating a Machine-Learning Opioid Prediction & Risk-Stratification E-Platform (DEMONSTRATE)
Funder: NIH/NIDA; PI: Lo-Ciganic
The DEMONSTRATE study evaluates how clinical decision support tools can improve the safety and effectiveness of opioid prescribing in real-world healthcare settings.
Machine Learning and Opioid Overdoses in Allegheny County
Funder: RK Mellon Foundation; PI: Gellad
This developed and validated a machine learning algorithm—using linked Medicaid, human services, and criminal justice data—to predict and flag individuals at high risk of fatal opioid overdose within 90 days of jail release.
Completed Risk Prediction Projects
PDMP Overdose Data to Action (OD2A) Opioid Overdose Surveillance
Funder: CDC; PI: Gellad
This study linked prescription drug monitoring program data with related records in Pennsylvania to develop a real-time machine learning model that identifies individuals at high risk of opioid overdose.
Using Machine Learning to Predict Problematic Prescription Opioid Use and Opioid Overdose
Funder: NIH/NIDA; PI: Gellad
This study developed and validated machine learning algorithms—using a combination of Medicaid claims and clinical data—to identify patients at high risk of problematic opioid use and overdose, and compared the accuracy of claims‑only models to those that include clinical data.