Automated identification of clinically meaningful biomechanical phenotypes in cerebral palsy through multicenter gait data.

Background: Cerebral palsy is the most prevalent motor disability in childhood, encompassing various movement disorders that affect walking. Researchers have described gait patterns in cerebral palsy, but these are often subjective and based on clinician experience. This study introduces an automated approach to objectively identify clinically meaningful biomechanical phenotypes in cerebral palsy and test it on multicenter gait data. Utilizing instrumented gait analysis, this research aims to improve treatment strategies for gait dysfunction. This study addresses whether classification algorithms can objectively identify clinically meaningful gait patterns and if severe gait deviations are more frequent in advanced forms of cerebral palsy.

Methods: Two novel classification algorithms (sagittal and transverse planes) were developed and automated in Python. These were based on previous work and refined using clinical expertise and data from four motion analysis centers in the Shriners Children's system, including 700 patients with cerebral palsy. The patient's gait data was applied to the treatment algorithms, and the percentage of each phenotype is presented.

Findings: Novel sagittal and transverse plane gait phenotype algorithms were created. When applied to the cerebral palsy cohort, we found that more severe gait deviations, or combinations of deviations, were more apparent in the more severe forms of cerebral palsy.

Interpretations: Classifying a patient's biomechanical phenotype provides valuable insights into therapeutic interventions. The results allow for the automation of data-driven classification algorithms, leading to efficient, accurate, and reliable classifications of biomechanical phenotypes that support evidence-based, personalized treatment decisions and clinical management.