Machine Learning Identifies Distinct Biomechanical Gait Phenotypes in Pediatric and Adolescent Ataxia: Establishing a Clinical Decision Model

Background: Standard clinical scales for pediatric and adolescent ataxia lack biomechanical granularity, limiting precision rehabilitation. This study aimed to identify compensatory gait phenotypes using unsupervised machine learning and establish a robust, highly accessible clinical decision model. Methods: Spatiotemporal gait data from 51 youths (31 ataxia and 20 healthy controls) were analyzed. To ensure pathological specificity, Principal Component Analysis (PCA) and hierarchical clustering were applied exclusively to 13 biomechanical variables from the ataxia cohort (n = 31) to extract underlying domains and identify patient subgroups. Healthy controls were subsequently used as a normative reference. A Classification and Regression Tree (CRT) algorithm was developed for clinical translation. Results: Two distinct phenotypes reflecting the evolution of compensatory strategies were identified: a “Rapid Rhythm” strategy (n = 24) and a severe “Prolonged Stance” strategy (n = 7). Unlike previous assumptions, the phenotypes strongly correlated with clinical severity (Scale for the Assessment and Rating of Ataxia (SARA) scores: 9.79 vs. 16.78, p = 0.012) and exhibited significantly different gait speeds (p < 0.001). The CRT model identified the stance phase duration as the primary discriminator. A recalibrated critical cut-off of >69.68% effectively classified the severe Prolonged Stance phenotype. This threshold sits distinctly above the healthy pediatric norm, achieving an overall cross-validated accuracy of 96.8%, with 100% specificity. Conclusions: Gait phenotypes in pediatric and adolescent ataxia represent progressive stages of neuromechanical compensation driven by disease severity. The established 69.68% stance-phase threshold provides clinicians with a powerful, single-variable biomechanical red flag to identify severe pathological gait and guide phase-specific precision rehabilitation.