Predicting Early Non-Response to Buprenorphine Treatment for Opioid Use Disorder: A Machine Learning Approach

Le résumé fourni par la source

Objective: Early non-response to buprenorphine treatment for opioid use disorder (OUD), including elevated craving and illicit opioid use in the first weeks of treatment, robustly predicts later recovery outcomes. Yet, little is known about who is at risk of early non-response, and identifying predictors could help classify at-risk individuals even earlier in care. We trained and tested risk prediction models for early non-response to buprenorphine treatment. Method: Data were from two harmonized clinical trials of buprenorphine treatment for OUD (N=562). We used elastic net and random forest models to predict two previously-validated definitions of early non-response: 1) 2+ days of opioid use in the first three weeks, and 2) a score of 2+ on a 0-10 craving visual analog scale in the first week. Predictors included pre-treatment demographics and clinical variables (e.g., polysubstance use, co-occurring disorders). Results: Models demonstrated weak predictive performance for both outcomes, with areas under the receiver operating characteristic curve (AUCs; 0.590–0.656) below the benchmark for acceptable performance (AUC>0.7). The strongest predictors differed by the two outcome definitions; opioid use severity indicators (e.g., heroin use frequency) were strong predictors of non-response based on opioid use, while craving, withdrawal, and psychosocial functioning were strong predictors of non-response based on craving. Conclusions: Pre-treatment characteristics showed limited predictive value for early non-response to buprenorphine, aligning with prior work using machine learning models to predict later outcomes in OUD treatment. There is likely more value in attending to medication response in the first weeks after initiation as a prognostic marker.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.