Clinical prediction of sodium valproate-induced movement disorders in hospitalized patients: a nomogram-based model using real-world data.

Abstract

Introduction: Drug-induced movement disorders (DIMDs) are often underrecognized and challenging to diagnose and manage in clinical practice. Sodium valproate (VPA), a widely prescribed antiepileptic drug, causes DIMDs. Predictive modeling based on electronic medical records and machine learning algorithms offers a promising approach to improve early identification of adverse drug reactions (ADRs) and enhance clinical safety.

Aim: This study aimed to conduct real-world active surveillance of VPA-associated DIMDs using hospital information system data, to identify independent risk factors, and to develop a clinically applicable prediction model for early warning and intervention.

Method: In this retrospective case-control study, data were collected from hospitalized patients prescribed sodium valproate between 2018 and 2022. DIMD cases were identified through automated alerts generated by the ADE Active Surveillance and Assessment System II, and subsequently confirmed using the Naranjo scale for manual causality assessment. A clinical prediction model was developed using LASSO logistic regression and was presented as a nomogram. Model performance was evaluated using area under the receiver operating characteristic (ROC) curve, calibration curve, decision curve analysis (DCA), and clinical impact curve (CIC).

Results: Of the 6692 patients screened, 98 were confirmed to have DIMDs, yielding an incidence rate of 1.46%. Four independent protective factors were identified: higher lymphocyte count, red blood cell count, serum sodium concentration, and co-administration of levetiracetam. The nomogram demonstrated good discrimination (AUC = 0.774), acceptable calibration (Brier score = 0.194), and strong clinical utility across a threshold probability range of 10-70%.The results of the external validation suggest that the efficiency of the model remained stable and showed no significant decline. It had better predictive ability in high- and medium-risk groups and had significant potential for clinical application. The incidence of DIMDs was significantly higher among patients receiving VPA for therapeutic purposes than among those receiving VPA prophylactically.

Conclusion: VPA-associated DIMDs are common in hospitalized patients. The prediction model based on routine clinical indicators enabled the identification of high- and middle-risk individuals, thereby facilitating timely monitoring and targeted interventions to reduce the burden of movement-related ADRs.