Machine Learning for Predicting Multiple Sclerosis Progression Using Clinical and Environmental Data

Multiple Sclerosis (MS) has become an increasing focus of research in recent years, driven by its growing prevalence and substantial impact on public health. It is a lifelong, chronic neuroinflammatory disorder that progressively impairs neurological functions, often resulting in significant disability. Although therapeutic advances have helped slow disease activity, MS remains a chronic condition, and understanding the drivers of progression is critical for long-term management and for improving patients’ quality of life. Consequently, the study of progression factors has become central to both clinical practice and research. This work forms part of the BRAINTEASER project, which aims to bring AI-driven care to patients with MS and Amyotrophic Lateral Sclerosis (ALS) across three European countries (Spain, Italy, and Portugal) with clinical and environmental data. Within this framework, this work investigates the role of environmental exposures in MS progression and prediction by integrating them with clinical and demographic data. The work is organized around three main aims. Aim 1, addressed in Chapter 3, focuses on the development of a comprehensive longitudinal dataset by integrating clinical and demographic data with multiple types of environmental exposures, including both air pollutants and weather conditions. This constitutes a crucial step toward analyzing the influence of environmental factors on MS progression. To capture both short- and long-term effects, environmental features were aggregated across multiple temporal windows, ranging from one week to several months prior to each patient visit. Since missing data represent a key challenge in medical research, advanced imputation techniques tailored for longitudinal datasets were employed to ensure robust and reliable analyses. Aim 2, addressed in Chapter 4, investigates the relationship between environmental exposures and disease progression using the Expanded Disability Status Scale (EDSS), the most widely adopted clinical measure of disability in MS. For this purpose, Continuous-Time Markov Models (CTMM) were applied to model transitions between disability states, taking advantage of their ability to handle irregular follow-up intervals and to capture both progression and potential improvement. This approach provides a robust framework for examining how exposures may influence the dynamic course of disability over time. Aim 3, addressed in Chapter 5, focuses on forecasting Multiple Sclerosis Severity Score (MSSS) classes through sequential data modeling using one of the integrated datasets developed in this research. This study pursues two main objectives. The first, and most central, is to forecast MSSS class, a more recent and refined measure of MS progression compared to EDSS. The second is to assess the contribution of environmental exposures to this predictive task, specifically, to determine their relative importance compared with clinical and demographic variables. This analysis is particularly relevant for forecasting the severity class at a patient’s next clinical visit, an outcome of direct practical value for clinicians. The most notable finding is that several environmental exposures exert a stronger-than-expected influence on transitions between disability states when progression is considered (Aim 2). In addition, Aim 3 highlights the key potential of environmental exposures, in some cases surpassing well-established demographic and clinical features, traditionally regarded as top predictors in MS research, for forecasting outcomes at the next visit.