A machine learning system was able to predict relapse, defined as non-prescribed opioid use and treatment discontinuation, among people with opioid use disorder (OUD) receiving outpatient treatment with medications for opioid use disorder (MOUD). The system was able to predict next-day use of non-prescribed opioids, medication non-adherence, and treatment retention. The system used data from electronic health records, ecological momentary assessment (EMA), and deep learning in a study with 62 people. EMA gathers repeated, real-time data on an individual’s experiences, behaviors, and emotions.

Study participants were asked to respond to EMA prompts related to their mental health and substance use three times daily. Data was collected on sleep, pain, stress, cravings, withdrawal symptoms, substances last used, MOUD adherence, mood, location, and social context.

Recent substance use (alcohol or nicotine) was a top-performing predictor that the person would use a non-prescribed opioid. The researchers concluded that self-reported EMA data and deep learning are effective for supporting the forecasting of actionable outcomes in those receiving MOUD. “These insights will enable the development of personalized dynamic risk profiles and just-in-time adaptive interventions (JITAIs) to mitigate high-risk OUD outcomes,” the researchers stated.

These findings were reported in A Longitudinal Observational Study With Ecological Momentary Assessment And Deep Learning To Predict Non-Prescribed Opioid Use, Treatment Retention, And Medication Nonadherence Among Persons Receiving Medication Treatment For Opioid Use Disorder by Michael V. Heinz, George D. Price, Avijit Singh, and colleagues. The study included 62 adults who completed 14,322 EMA observations. A participant was considered to be “retained in treatment” if for 84 days they had no long gaps in their MOUD coverage. Adherence was defined as having the expected daily amount of medication.

The full text of A Longitudinal Observational Study With Ecological Momentary Assessment And Deep Learning To Predict Non-Prescribed Opioid Use, Treatment Retention, And Medication Nonadherence Among Persons Receiving Medication Treatment For Opioid Use Disorder was published on May 10, 2025, by the Journal of Substance Use and Addiction Treatment. A free copy is available (accessed November 10, 2025).

For more information, contact: Michael V. Heinz, M.D., Postdoctoral Fellow, Center for Technology and Behavioral Health, Geisel School of Medicine, Dartmouth College, 1 Medical Center Drive, Lebanon, New Hampshire 03756; Email: michael.v.heinz@dartmouth.edu; Website: https://geiselmed.dartmouth.edu/epidemiology/profile/michael-heinz-md/