Developing and Evaluating a Machine-Learning Buprenorphine Care Discontinuity Prediction E-Tool (BUP-CARE)

About This Grant

Project Summary Over 81,000 Americans died from opioid overdose in 2023. Opioid overdose and opioid use disorder (OUD) impose significant societal and economic burdens, making sustained OUD treatment a critical public health priority. Buprenorphine, an evidence-based therapy for OUD, reduces overdose and mortality risks with >6 months of therapy. Yet, >50% of patients discontinue it within 6 months of initiation, increasing relapse and mortality risks. Primary care providers (PCP) prescribe >60% of buprenorphine prescriptions and are expected to prescribe more following elimination of the federal waiver requirement, highlighting their role in treatment retention. Although PCPs may recognize some patients at risk of discontinuation, they face time constraints, competing demands, and a lack of systematic clinical decision support (CDS) tools to integrate multi-faceted risk factors to efficiently identify patients at high risk of buprenorphine discontinuation. Accurate prediction of buprenorphine discontinuation can better inform tailored interventions, such as long-acting injectable buprenorphine and peer recovery support, to improve retention and outcomes. Prior studies using traditional regression-based methods have identified individual risk factors of buprenorphine discontinuation; however, little is known about whether these single factors accurately predict such risk. In addition, using these methods to predict buprenorphine discontinuation proves to be challenging because of their limited ability to handle complex relationships between predictors and outcomes. Machine learning offers an alternative, uncovering hidden patterns in complex data and generating precise prediction algorithms and risk stratification subgroups to guide clinical care and interventions. Our prior work has successfully applied machine learning to claims data to identify commercially insured patients at high risk of buprenorphine discontinuation. Leveraging our prior work, we propose to develop and evaluate a machine-learning buprenorphine care discontinuity prediction e-tool (BUP-CARE) for PCPs within healthcare systems to identify patients at high risk of buprenorphine care discontinuity. We have 2 specific aims. Aim 1 will use OneFlorida+ electronic health records (EHR) data from 2014 to 2025 in the PCORnet Common Data Model to develop and validate machine-learning prediction algorithms to identify patients at high risk of buprenorphine discontinuation. We will expand our previous work by incorporating advanced deep learning (e.g., recurrent neural network) for risk prediction and by including social determinants of health as predictors. Aim 2 will identify barriers and facilitators for developing a BUP-CARE CDS tool to alert front-line PCPs for patients at high risk of buprenorphine discontinuation. We will use an iterative user-centered design approach to enhance BUP-CARE’s functionality and usability. Our proposed research is highly innovative and clinically relevant in its use of a machine learning- based CDS tool to guide clinical practice and tailor evidence-based interventions, thereby optimizing resource allocation and reducing patients’ buprenorphine care discontinuity.