Machine learning-based prediction of disability progression in multiple sclerosis using clinical, performance-based, and patient-reported outcomes

Abstract

Introduction: While clinical predictors of disability progres­ sion in people with multiple sclerosis (pwMS) are well-estab­ lished, the contribution of patient-reported outcome measures (PROMs) remains underexplored. Objectives/Aims: To develop a machine learning (ML) model for predicting disability progression-defined as a change in EDSS over three years-using clinical data, performance-based tests, and PROMs. Methods: Data were collected at baseline and at a 3-year fol­ low-up, including clinical&demographic outcomes (type of MS, baseline EDSS, sex), performance-based measures (Timed Results: Data from 182 pwMS (mean EDSS±:SD: 2.28±1.9, 69% female) were analyzed. Linear Regression demonstrated the best performance, achieving the lowest error rates and high­ est predictive accuracy (MAE: 0.321, MSE: 0.185, RMSE: 0.430, R 2 : 0.951). It was closely followed by Random Forest, which also showed strong predictive power (MAE: 0.347, MSE: 0.246, RMSE: 0.495, R 2 : 0.935. Decision Tree and SVR yielded less accurate predictions, with higher error rates and lower R 2 values (MAE: 0.432 and 0.547; R 2 : 0.879 and 0.844, respec­ tively). Visual inspection of prediction outputs confirmed that Linear Regression and Random Forest models closely tracked actual EDSS scores over time, particularly in the third year. Conclusion: Linear Regression's superior metrics suggest a strong linear relationship, likely driven by baseline EDSS,