Machine Learning Prediction and Reducing Overdoses With Electronic Health Record Nudges (mPROVEN) in the Primary Care...

Area of Science:

• Public Health
• Health Informatics
• Clinical Trials

Background:

• Opioid overdose is a leading preventable cause of death in the US.
• Current risk assessment methods are imprecise, leading to misclassification of patient risk.
• Machine learning (ML) and electronic health record (EHR) interventions can improve overdose risk prediction and reduction.

Purpose of the Study:

• To evaluate the Machine Learning Prediction and Reducing Overdoses With EHR Nudges (mPROVEN) clinical trial.
• To test if integrating an ML overdose risk model with EHR nudges improves prescribing behaviors and reduces overdose risk.
• To assess the impact on evidence-based prescribing and patient overdose outcomes.

Main Methods:

• Pragmatic cluster randomized controlled trial in primary care practices.
• Adult patients identified with elevated overdose risk by ML algorithm.
• Three arms: usual care, EHR risk flag only, EHR risk flag + nudges (active choice, accountable justification alerts).
• Primary outcome: composite measure of safer opioid prescribing at 4 months.
• Intention-to-treat analysis using linear mixed-effects models.

Main Results:

• Enrollment for the primary analysis cohort (n=798) completed in May 2025.
• Enrollment for secondary analyses completed in December 2025 (n=1662).
• Primary cohort analyses began in January 2026, with results expected by mid-2027.

Conclusions:

• The mPROVEN study is a pioneering pragmatic randomized controlled trial.
• It integrates ML-based opioid overdose risk prediction with behavioral nudges in a large EHR system.
• The study aims to reduce opioid overdose risk through a scalable, low-touch intervention in primary care.