Relapse prediction and individualized treatment-effect modeling in relapsing multiple sclerosis: a systematic review

Abstract

Individualized prediction of relapse-related outcomes may support treatment selection and monitoring in relapsing multiple sclerosis. However, existing studies differ substantially in clinical purpose, target outcomes, modeling approaches, and validation strategies. We conducted a systematic review in accordance with PRISMA 2020. PubMed, Web of Science, IEEE Xplore, and Scopus were searched for English-language records published from 1 January 2010 to 14 January 2026. Studies developing or evaluating models for relapse or relapse-related outcomes in relapsing multiple sclerosis were included. Data on study design, predictors, modeling approach, target outcome, calibration, and validation were extracted. Owing to clinical and methodological heterogeneity, a structured narrative synthesis was performed. The protocol was registered in PROSPERO (CRD42024625392). Fourteen studies were included: five conventional relapse-prognosis studies, four individualized treatment-effect prediction studies, and five exploratory relapse-related studies. Clinically interpretable models based on structured clinical data generally demonstrated moderate discrimination but more transparent validation. Studies reporting very high predictive performance were commonly based on smaller samples, high-dimensional data, or limited independent validation. Calibration and external validation were inconsistently reported. Relapse-related prediction in multiple sclerosis is feasible, but current evidence remains heterogeneous and insufficiently validated for routine clinical implementation. Progress will require harmonized outcome definitions, consistent calibration reporting, transparent model evaluation, and external validation across clinically diverse populations.