Predicting Opioid Cost Burden through Integrated PBM and SDOH Modeling: An Explainable AI Framework

AbstractThe opioid epidemic continues to impose a profound human and financial burden on the U.S. healthcare system. While previous studies have examined prescribing behavior or social determinants of health (SDOH) separately, few have quantitatively integrated these domains with Pharmacy Benefit Management (PBM) cost levers to explain geographic variation in opioid-related spending. This study develops an interpretable, county-level machine-learning framework linking Medicare Part D opioid-prescribing data (2013–2023) with socioeconomic, behavioral, and healthcare-access indicators from the County Health Rankings dataset.Using a Random Forest regression pipeline with automated feature selection and five-fold cross-validation, the final model achieved R² = 0.9698 and RMSE = $445.36, explaining nearly all observed variation in opioid cost per capita. Comparative analysis showed that SDOH-only models explained 34% of variance, utilization-only models 95%, and the integrated PBM + SDOH model 97%, confirming that economic, behavioral, and access factors jointly shape fiscal burden. Global SHAP and permutation-importance analyses identified cost per claim, claims per 1,000 population, and opioid-prescribing rate as dominant predictors, amplified by structural vulnerabilities such as unemployment, income inequality, and low mental-health-provider density.Counterfactual simulations quantified the potential financial impact of PBM and SDOH interventions. PBM levers—including formulary tightening, prescriber-utilization guidance, and cost-per-claim reduction—produced short-term savings of approximately 4–6%. SDOH improvements, such as expanded primary-care and mental-health capacity and reductions in behavioral-risk prevalence, generated larger compounding effects, with integrated PBM + SDOH reforms yielding national savings of roughly 17% (≈$23 billion annually). Geographic patterns aligned with CDC dispensing-rate distributions, with the highest predicted cost burden concentrated in Utah, Kentucky, and Alabama.This integrated analytical framework demonstrates how combining PBM program design with SDOH modeling can inform equitable policy, optimize healthcare spending, and guide targeted interventions in the ongoing U.S. opioid crisis.Keywords: opioid prescribing; social determinants of health; pharmacy benefit management; Medicare Part D; machine learning; public health policy.