A Joint Survival Modeling and Therapy Knowledge Graph Framework to Characterize Opioid Use Disorder Trajectories

Area of Science:

• Computational biology
• Clinical informatics
• Public health

Background:

• Opioid use disorder (OUD) involves complex transitions between onset, remission, and relapse.
• Prescription opioid exposure is a common pathway to developing OUD.
• Integrated electronic health record (EHR) and survey data offer opportunities for stage-specific risk modeling.

Purpose of the Study:

• To develop a multi-stage framework for modeling time-to-onset, time-to-remission, and time-to-relapse in OUD.
• To identify high-confidence predictors for each transition using longitudinal EHR and survey data.
• To map identified risk factors to potential therapeutic interventions.

Main Methods:

• Utilized the All of Us Research Program's linked EHR and survey data.
• Derived longitudinal predictors including clinical conditions and survey concepts (e.g., event counts, cumulative exposures).
• Applied regularized Cox models to identify significant predictors for each OUD transition stage.

Main Results:

• Chronic pain, mental health conditions (anxiety, depression), and polysubstance use (tobacco, cannabis) were prominent predictors of OUD onset and relapse.
• Tobacco dependence during remission and other remission-coded conditions were associated with successful remission.
• A therapy knowledge graph was constructed to link risk factors to candidate OUD treatments.

Conclusions:

• A predictive framework for OUD transitions was successfully developed using integrated data.
• Specific clinical and behavioral factors significantly influence OUD onset, remission, and relapse.
• The findings support therapeutic prioritization and personalized intervention strategies for OUD management.