AI Model Predicts ALS Progression and Healthcare Needs

Amyotrophic lateral sclerosis (ALS) is a complex and heterogeneous neurodegenerative disease, making it challenging to predict key clinical milestones like the need for assistive devices. This paper introduces a novel, digital-twin-inspired framework that uses temporal machine learning to provide individualized predictions of functional decline and healthcare utilization for ALS patients. The framework begins by harmonizing a comprehensive longitudinal dataset, including diagnosis records, ALSFRS-R assessments, activities of daily living, and demographic information. It then employs correlation-based clustering to identify coherent functional domains (bulbar, upper limb, axial, lower limb, respiratory) and uses generalized additive mixed models to characterize their nonlinear decline. A temporal machine learning model further predicts longitudinal functional decline, capturing stage-dependent disease progression. Crucially, Cox proportional hazards modeling identified lower limb function as a strong predictor for earlier wheelchair access. Building on these insights, the digital twin-inspired time-to-event (TTE) model generates individualized survival curves, dynamically predicting wheelchair-free survival. This scalable, interpretable, and clinically actionable approach has significant implications for proactive care planning, clinical trial stratification, and precision medicine in ALS.