What question did this study set out to answer?

April 11, 2026PLOS mental health.0 citationsOpen Access

Subgroup identification of disparities in buprenorphine discontinuation in opioid-use disorder: A Virtual Twins machine learning approach using nationwide United States claims data, 2006–2022

KZKehe Zhang PJPaula A Jaimes-BuitronThe University of Texas at San Antonio WGWei GuoGuangdong Provincial Academy of Environmental Science

Key Points

The study aims to identify and quantify disparities in buprenorphine treatment retention for opioid use disorder.
Utilized a two-stage machine learning approach with longitudinal claims data from 2006-2022.
Trained classification models in Stage one to estimate treatment discontinuation disparities.
Employed regression models, decision trees, and neural networks in Stage two for subgroup identification.
Assessed model performance using C-statistics and precision-recall curves.
71% of treatment episodes resulted in discontinuation within 180 days.
Disparities were identified based on insurance type, geographic region, mental health status, age, and SVI.
Higher discontinuation risk found in privately insured older adults and younger publicly insured individuals.
Utilization of psychiatric services reduced discontinuation risks across all subgroups.

Abstract

Buprenorphine retention is crucial for effective treatment of opioid use disorder (OUD), yet disparities in treatment discontinuation persist. This study aims to identify and quantify disparities in buprenorphine treatment retention using a machine learning framework adapted from Virtual Twins approach focusing on disparities related to sex, age, insurance type, geographic region, mental health status and community-level Social Vulnerability Index (SVI). Using a nationwide longitudinal cohort, we applied a two-stage machine learning approach. In Stage one, we trained classification models to estimate the counterfactual differences in treatment discontinuation across disparity types. Model performance was assessed using C-statistics and precision-recall curve. In Stage two, we employed regression models, decision trees and neural networks with Shapley Additive Explanations, to identify subgroups most vulnerable to the disparities and key contributing factors. Among 303,528 treatment episodes from 131,169 patients aged 18–85, 71% discontinued treatment within 180 days. Early medication adherence (3-month proportion of days covered) was the strongest predictor. Significant disparities emerged based on insurance, region, mental health status, age, and SVI. Higher discontinuation risk was observed among privately insured older adults, patients from high-SVI areas in the South without mental health diagnoses, and younger publicly insured individuals lacking psychiatric services. Psychiatric service utilization consistently mitigated discontinuation risks across subgroups. Limitations include the absence of race/ethnicity in the claims data, the inability to capture concurrent medications and initial buprenorphine dose, and lack of formal uncertainty estimates for the quantified disparities. The Virtual Twins analytical framework enabled identification of vulnerable subgroups and quantification of disparities attributable to specific risk factors. Interventions that prioritize early adherence, expand access to psychiatric services, and address structural barriers in high-SVI Southern communities and insurance-defined risk groups may improve equity in buprenorphine retention.

问 AI

Bookmark

View Full Paper