Early Prediction of Multiple Sclerosis Disability Progression via Multimodal Foundation Model Benchmarks

TL;DR

This study explores the use of multimodal foundation models to predict multiple sclerosis disability progression early by integrating clinical and digital data, showing improved accuracy over traditional methods.

Contribution

It introduces a multimodal transformer model that combines clinical and digital data for early MS progression prediction, demonstrating its superiority over unimodal models.

Findings

Multimodal transformer outperforms unimodal models in prediction accuracy.

Integrating digital data improves early prediction of MS disability.

AUROC of 0.63 indicates moderate predictive performance.

Abstract

Early multiple sclerosis (MS) disability progression prediction is challenging due to disease heterogeneity. This work predicts 48- and 72-week disability using sparse baseline clinical data and 12 weeks of daily digital Floodlight data from the CONSONANCE clinical trial. We employed state-of-the-art tabular and time-series foundation models (FMs), a custom multimodal attention-based transformer, and machine learning methods. Despite the difficulty of early prediction (AUROC 0.63), integrating digital data via advanced models improved performance over clinical data alone. A transformer model using unimodal embeddings from the Moment FM yielded the best result, but our multimodal transformer consistently outperformed its unimodal counterpart, confirming the advantages of combining clinical with digital data. Our findings demonstrate the promise of FMs and multimodal approaches to extract predictive signals from complex and diverse clinical and digital life sciences data (e.g., imaging, omics), enabling more accurate prognostics for MS and potentially other complex diseases.