Development and validation of machine-learning algorithms predicting retention, overdoses, and all-cause mortality among US military Veterans treated with buprenorphine for opioid use disorder
Abstract: Aim: The aim of this study was to develop and validate machine-learning algorithms predicting retention, overdoses, and all-cause mortality among US military veterans treated with buprenorphine for opioid use disorder (B-MOUD). Methods: This study used a prognostic cohort study design and data from the U.S. Veterans Healthcare Administration on veterans initiating B-MOUD from fiscal years 2006-2020. Candidate predictors (n=114) were measured over the 12 months before B-MOUD initiation. Veterans’ B-MOUD episodes were randomly divided into training (80%; n=45,238) and testing samples (20%; n=11,309). Candidate algorithms (multiple logistic regression, least absolute shrinkage and selection operator regression, random forest (RF), gradient boosting machine (GBM), and deep neural network (DNN)) were used to build and validate a classification model to predict six different binary outcomes important to persons in B-MOUD treatment: 1) B-MOUD retention, 2) any overdose, 3) opioid-related overdose, 4) overdose death, 5) opioid overdose death, and 6) all-cause mortality. To balance the outcome categories, various over- and under-sampling techniques were employed. Model performance was assessed using standard classification statistics [e.g., area under the receiver operating characteristic curve (AUC-ROC)]. Results: Training and testing samples had similar characteristics (93% male, 78% White, mean age was 46.5 years). In the testing sample, the GBM model slightly outperformed other models in predicting B-MOUD retention (AUC-ROC=0.72). Generally, RF models outperformed other models in predicting any overdose (AUC-ROC=0.77) and opioid overdose (AUC-ROC=0.77). RF and GBM outperformed other models for overdose death (AUC-ROC=0.74 for both), and RF and DNN outperformed other models for opioid overdose death (RF AUC-ROC=0.79; DNN AUC-ROC=0.78). RF and GBM also outperformed other models for all-cause mortality (AUC-ROC=0.76 for both). Conclusions: Machine-learning algorithms, particularly RF, GBM, and DNN, can accurately predict B-MOUD retention, overdoses, and all-cause mortality among Veterans initiating B-MOUD with moderate predictive performance.