A Multivariate Deep-Learning Approach for Stratifying Amyotrophic Lateral Sclerosis Patients Based on Temporal Dynamics

Amyotrophic lateral sclerosis (ALS) exhibits marked heterogeneity in disease progression, complicating prognosis and personalized care. Data-driven clustering approaches may enhance prognostic precision, guide clinical trial design, and inform personalized treatment planning. In this work, we introduce a novel multivariate deep-learning approach for ALS patient stratification, leveraging the temporal dynamics embedded in longitudinal clinical records. We apply this approach to the PRO-ACT clinical trial ALS dataset to automatically stratify patients into homogeneous subgroups based on clinical characteristics observed during the first 6 months of the trial. Starting with over 5300 subjects, we identify four clusters (C1-C4, Silhouette score=0.23) reflecting distinct phenotypes: (C1) rapid progressors with steep functional decline in all domains; (C2) intermediate, limb-dominant patients exhibiting moderate motor deterioration; (C3) bulbar-onset patients with early speech and swallowing deficits; and (C4) slow progressors maintaining function over extended periods. These groups also exhibit divergent trajectories in respiratory measurements and serum creatinine, variables not included in the clustering approach but previously recognized as prognostic markers, and in the 6 months following the initial observation period, suggesting robust differentiation of patients also in terms of future progression. Although applied to a clinical trial population, which may introduce bias due to selection criteria, our approach effectively identifies clinically relevant phenotypes from the first 6 months of observation, revealing characteristic, nonlinear progression patterns. Extending this approach to real-world data, also including additional features (such as imaging or molecular biomarkers), holds the potential to further refine subgroup definitions, deepen the understanding of ALS heterogeneity, and advance precision medicine.