Development and validation of a risk prediction nomogram for suboptimal valproic acid concentrations in pediatric...

This study aims to identify independent risk factors affecting suboptimal valproic acid (VPA) blood concentration levels by performing a multifactorial logistic regression analysis on clinical data from 1569 pediatric epilepsy patients treated at our institution between January 2020 and December 2024. A nomogram risk prediction model was developed and validated to provide a scientifically grounded tool for individualizing drug regimens and aiding clinical decision-making in pediatric epilepsy.

A total of 1569 pediatric epilepsy patients aged 1-18 years, who met the inclusion criteria, were included in the study. Demographic data (age, gender) and laboratory parameters (blood ammonia, platelet count, blood urea, total bilirubin, total cholesterol, triglycerides, ALT, AST, etc.) were collected. Multifactorial logistic regression was employed to identify variables significantly associated with suboptimal VPA blood concentrations (defined as <50 µg/mL or >100 µg/mL). A nomogram was constructed based on the regression coefficients, and the model's performance was evaluated using receiver operating characteristic (ROC) curves (AUC), calibration curves, and decision curve analysis (DCA) to assess its discriminatory power, calibration, and clinical utility.

The multivariate analysis identified blood ammonia (OR = 1.128, 95% CI 1.051-1.210, P = 0.0009), platelet count (OR = 1.180, 95% CI 1.133-1.229, P < 0.001), blood urea (OR = 2.101, 95% CI 1.375-3.210, P = 0.0006), total bilirubin (OR = 1.413, 95% CI 1.234-1.617, P < 0.001), total cholesterol (OR = 1.637, 95% CI 1.134-2.362, P = 0.0084), triglycerides (OR = 139.790, 95% CI 24.913-784.390, P < 0.001), ALT (OR = 1.152, 95% CI 1.082-1.226, P < 0.001), and AST (OR = 0.918, 95% CI 0.861-0.980, P = 0.0097) as independent risk factors for suboptimal VPA blood concentrations. The nomogram model demonstrated excellent discrimination (AUC = 0.82) and calibration (Brier Score post-calibration = 0.1712), with DCA revealing a high net benefit, suggesting potential clinical applicability.

The nomogram model developed in this study, based on eight significant clinical indicators, demonstrated good discrimination (AUC = 0.82) and calibration. It serves as a visual tool to assist clinicians in identifying high-risk patients early. This model provides a methodological reference for individualized VPA dosing in children and lays the groundwork for future multicenter validation.