Abstract
INTRODUCTION: No prediction models exist for the success for buprenorphine initiation in opioid-naïve patients or in transition from other opioids in patients treated for chronic pain. OBJECTIVES: To create a prediction model for the successful use of buprenorphine to treat chronic pain. METHODS: Stepwise Akaike information criterion prediction modeling procedures were applied to a harmonized participant-level data set of 10 enriched enrollment randomized withdrawal clinical trials of buprenorphine submitted to the Food and Drug Administration. Available baseline factors and nine patient-reported outcomes were considered to predict success with the titration (10 studies) and maintenance of benefit after randomization (5 studies). Patient-reported outcomes were modeled separately given inconsistent use across studies. RESULTS: No prediction model reached an area under the receiver operator curve ≥0.70, the threshold for clinical usefulness. Successful initiation or transition of buprenorphine was accomplished in 3541 of 6052 (58.7%) participants, and 614 of 877 (70.0%) completed the 12-week maintenance phase with no difference between opioid-experienced and opioid-naïve participants. Only a medical history of obesity and baseline pain were retained in the overall titration model and only baseline pain in the maintenance model. Only brief pain inventory and subject opioid withdrawal scores were retained in the titration subsets containing those measures. CONCLUSION: No clinically useful prediction models of clinical benefit were identified, but a few covariates may be of interest in future studies of the initiation of buprenorphine in opioid-naïve patients or of transition from other opioids to buprenorphine. The lack of a predictor supports considering a trial of buprenorphine in clinically relevant scenarios for patients without known opioid use disorder, including careful monitoring and an a priori plan to deal with any problems that may occur.