A Comparative Effectiveness Study on Opioid Use Disorder Prediction Using Artificial Intelligence and Existing Risk Models

Abstract

ABSTRACTObjectiveTo compare the effectiveness of multiple artificial intelligence (AI) models with unweighted Opioid Risk Tool (ORT) in opioid use disorder (OUD) prediction.Materials and MethodsThis is a retrospective cohort study of deidentified claims data from 2009 to 2020. The study cohort includes 474,208 patients. Cases are prescription opioid users with at least one diagnosis of OUD or at least one prescription for buprenorphine or methadone. Controls are prescription opioid users with no OUD diagnoses or buprenorphine or methadone prescriptions. Cases and controls are matched based on age, sex, opioid use duration and longitudinal data availability. OUD prediction performance of logistic regression (LR), random forest (RF), XGBoost, long short-term memory (LSTM), transformer, our proposed AI model for OUD prediction (MUPOD), and the unweighted ORT were assessed using accuracy, precision, recall, F1-score and AUC.ResultsData includes 474,208 patients; 269,748 were females with an average age of 56.78 years. On 100 randomly selected test sets including 47,396 patients, MUPOD can predict OUD more efficiently (AUC=0.742±0.021) compared to LR (AUC=0.651±0.025), RF (AUC=0.679±0.026), XGBoost (AUC=0.690±0.027), LSTM (AUC=0.706±0.026), transformer (AUC=0.725±0.024) as well as the unweighted ORT model (AUC=0.559±0.025).DiscussionOUD is a leading cause of death in the United States. AI can be harnessed with available claims data to produce automated OUD prediction tools. We compared the effectiveness of AI models for OUD prediction and showed that AI can predict OUD more effectively than the unweighted ORT tool.ConclusionEmbedding AI algorithms into clinical care may assist clinicians in risk stratification and management of patients receiving opioid therapy.