Including environmental factors into Multiple Sclerosis relapse prediction models: A comparative study

Multiple Sclerosis (MS) is a chronic neuroinflammatory disease marked by the presence of relapses - episodes of worsening symptoms typically lasting from days to weeks. Although several studies have reported associations between environmental factors (such as temperature, air pollution, and gaseous pollutants) and the risk of MS relapse, the nature of these relationships remains unclear. In this study, the predictive value of environmental data in forecasting relapses was explored through the development of a framework that used four weeks of longitudinal clinical and environmental data (i.e., PM 10 , PM 2.5 , O 3 , NO 2 , SO 2 , and CO) to predict relapse occurrence in the fifth week. Various combinations of relapse and relapse-free groups, feature engineering strategies, and feature subsets were evaluated using data from 814 individuals. In addition, a range of candidate Machine Learning (ML) and Deep Learning (DL) models, along with different feature selection strategies, were tested. The ML models included logistic regression, random forest, and extreme gradient boosting, while the DL models comprised a one-dimensional convolutional neural network and a recurrent neural network. Feature selection strategies involved random forest importance-based selection and Backward Feature Selection (BFS). Among the tested models, the best-performing configuration was a random forest classifier combined with BFS achieving the highest Area Under the ROC curve of 0.869 (95% CI: 0.833–0.898) across different settings, with PM 10 and NO 2 emerging as predictively informative variables in certain settings. In addition, comparative analyses of different feature subsets demonstrated that incorporating longitudinal and environmental information improved the model’s predictive performance. These findings highlight the advantages of combining temporal clinical and environmental data with non-linear modeling approaches for relapse forecasting in MS.