Neuromuscular Disease Explainable Machine-Learning Prediction of Motor, Respiratory, And Functional Decline in Duchenne Muscular Dystrophy

Abstract

Duchenne muscular dystrophy (DMD) is a progressive neuromuscular disorder characterized by heterogeneous deterioration in motor performance, respiratory capacity, and upper-limb function. This study quantitatively examined longitudinal disease progression and evaluated explainable machine-learning models for predicting motor, respiratory, and functional decline. A retrospective longitudinal design was applied to 286 individuals with genetically confirmed DMD who contributed 1,247 clinical observations. Mean baseline age was 10.84 ± 4.21 years, 65.7% of participants were ambulatory, and 83.6% had documented corticosteroid exposure. Longitudinal outcomes included North Star Ambulatory Assessment (NSAA), six-minute walk distance (6MWD), timed-function measures, forced vital capacity (FVC), percentage-predicted FVC, and Performance of Upper Limb (PUL) scores. Significant deterioration occurred across all major domains, with NSAA decreasing by 2.41 points annually, 6MWD by 31.8 meters annually, FVC percentage predicted by 4.27 percentage points annually, and PUL by 1.86 points annually (all p < .001). During follow-up, 49 of 188 initially ambulatory participants (26.1%) lost independent ambulation, with a median estimated loss-of-ambulation age of 13.7 years. Previous domain-specific decline, baseline functional performance, chronological age, ambulatory status, and corticosteroid exposure emerged as important predictors. Gradient boosting achieved the strongest predictive performance, explaining 78% of variation in NSAA decline, 72% in FVC decline, and 75% in PUL decline. Classification AUC values reached 0.89 for motor, 0.86 for respiratory, and 0.88 for functional deterioration. Explainability analyses identified previous decline and baseline functional reserve as the dominant predictive features, demonstrating that DMD progression was quantitatively heterogeneous across patients and functional domains.