Predicting oxcarbazepine-induced hyponatremia in adult epilepsy patients: A multicenter machine learning analysis using real-world CDM data

PURPOSE: Oxcarbazepine (OXC) is a widely used antiseizure medication (ASM) associated with hyponatremia. This study aimed to assess the prevalence and risk factors for OXC-induced severe hyponatremia using machine learning (ML) models applied to multicenter real-world data standardized within the Observational Medical Outcomes Partnership-Common Data Model (OMOP-CDM). METHODS: We conducted a retrospective cohort study using OMOP-CDM data from two tertiary hospitals in South Korea. Adult epilepsy patients prescribed OXC were included, and severe hyponatremia was defined as a serum sodium concentration ≤128 mmol/L. Two prediction experiments were conducted: (1) single-institution training and external validation of an XGBoost model; and (2) multicenter training and evaluation of five machine learning algorithms, including XGBoost, random forest, support vector machine, logistic regression, and naïve Bayes. SHAP (SHapley Additive exPlanations) values were used for model interpretation. RESULTS: Among 2253 patients, the prevalence of severe hyponatremia was 8.4%. In Experiment 1, XGBoost showed strong internal performance (AUROC 0.82) but decreased external performance (AUROC 0.72). In Experiment 2, XGBoost trained on multicenter data achieved the highest AUROC (0.83) and F1-score (0.41), outperforming other models. SHAP analysis identified key predictors including valproate use, diuretics, high OXC dosage, age, and stroke history. Additional medications such as beta-blockers, calcium channel blockers, hypnotics, and other ASMs (e.g., levetiracetam, pregabalin, lacosamide) also contributed to risk. CONCLUSION: XGBoost demonstrated robust predictive performance for OXC-induced severe hyponatremia using multicenter CDM data. SHAP-based interpretation revealed clinically relevant risk factors, supporting the implementation of personalized monitoring strategies in epilepsy care.