Advanced Methods for Improved Prediction of Opioid-Related Overdose and Suicide Events from Administrative Health Care Data

• Activity Number: 414 - Risk Modeling and Regression Techniques
• Type: Contributed
• Date/Time: Thursday, August 12, 2021 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Ralph Ward* and Erin Weeda and David J Taber and Robert Neal Axon and Mulugeta Gebregziabher
• Companies: Ralph H Johnson VAMC Charleston SC and Ralph H Johnson VAMC Charleston SC and Ralph H Johnson VAMC Charleston SC and Ralph H Johnson VAMC Charleston SC and MUSC
• Keywords: Risk prediction; Decision support; Opioid epidemic; mGLMM model

Abstract

Veterans suffer disproportionate health impacts from the opioid epidemic, including overdose, suicide, and death. Prediction models based on electronic medical record data can be powerful tools for identifying patients at greatest risk. The Veterans Health Administration implemented one such model in 2018: the Stratification Tool for Opioid Risk Mitigation (STORM). In this study we propose improvements to the original STORM model design to improve risk prediction performance. We use a multivariate generalized linear mixed model (mGLMM) to produce separate predictions for overdose and suicide-related events (SRE) rather than a single prediction for combined outcomes. Further improvements include incorporation of additional data sources and new predictor variables. Our improved model had significantly better prediction performance in terms of AUC (84% vs. 77%), sensitivity (71% vs. 66%) and specificity (81% vs. 76%) compared to replicated STORM results. The mGLMM performed particularly well in identifying patients at risk for SREs, where 72% of actual events were accurately predicted among patients with the 100,000 highest risk scores compared with 49.7% for the replicated STORM m