Machine Learning Prediction of Pharmacogenetic Test Uptake Among Opioid-Prescribed Patients Using Electronic Health Records: A Retrospective Cohort Study

BackgroundOpioids are a widely prescribed class of medication for pain management. However, they have variable efficacy and adverse effects among patients, due to complex interplay between biological and clinical factors. Pharmacogenetic (PGx) testing can be utilized to match patients genetic profiles to individualize opioid therapy, improving pain relief and reducing the risk of adverse effects. Despite its potential, PGx uptake--utilization of PGx testing--remains low due to a range of barriers at the patient, health care provider, infrastructure, and financial levels. Since testing typically involves a shared decision between the provider and patient, predicting likelihood of patient undergoing PGx testing and understanding the factors influencing that decision can help optimize resource use and improve outcomes in pain management. ObjectiveTo develop machine learning (ML) models, identifying patients likelihood of PGx uptake based on their demographics, clinical variables, medication use, and social determinants of health (SDoH). MethodsWe utilized electronic health records (EHR) data from a single center healthcare system to identify patients prescribed opioids. We extracted patients demographics, clinical variables, medication use, and SDoH, and developed and validated ML models, including neural networks (NN), logistic regression (LR), random forests (RF), gradient boosting (XGB), naive bayes (NB), and support vector machines (SVM) for PGx uptake prediction based on procedure codes. We performed 5-fold cross validation (CV) and created an ensemble probability-based classifier using the best-performing ML models for PGx uptake prediction. Various performance metrics, uptake stratification analysis, and feature importance analysis were employed to evaluate the performance of the models. ResultsThe ensemble model using XGB and SVM-RBF classifiers had the highest C-statistics at 79.61%, followed by XGB (78.94%), and NN (78.05%). While XGB was the best-performing model, the ensemble model achieved a high accuracy (67.38%), recall (76.50%), specificity (67.25%), and negative predictive value (99.49%). The uptake stratification analysis using the ensemble model indicated that it can effectively distinguish across uptake probability deciles, where those in the higher strata are more likely to undergo PGx in real-world (6.59% in the highest decile compared to 0.12% in the lowest). Furthermore, SHAP value analysis using the XGB model indicated age, hypertension, and household income as the most influential factors for PGx uptake prediction. ConclusionsThe proposed ensemble model demonstrated a high performance in PGx uptake prediction among patients using opioids for pain. This model can be utilized as a decision support tool, assisting clinicians in identifying patients likelihood of PGx uptake and guiding appropriate decision-making.