Predicting gait kinematics in youth with cerebral palsy using clinically informed machine learning algorithms

Gait deviations are common in youth with Cerebral Palsy (CP), with the change in gait pattern during growth/development being influenced by a variety of individual and treatment factors. The goal of this study is to use Machine Learning (ML) to evaluate the relationship between youth and treatment-related factors to predict changes in gait kinematics.

Kinematic gait data acquired from instrumented gait analysis (IGA) were collected from ambulatory youth with spastic CP (GMFCS I-III) during sequential visits. Twenty Gradient Boosting Regressor (GBR) models were trained to predict the change and the follow-up score of ten gait outcomes (Gait Profile Score (GPS), and Gait Variable Scores (GVS)) based on treatment (presence or absence of orthopedic surgery) and patient factors (age, initial gait, motor function).

The study included 702 evaluation pairs (GMFCS: I [18%], II [56%], III [25%]). A baseline instrumented gait analysis was performed at an average age of 10.9 ± 3.8 years, with a subsequent analysis 2.1 ± 1.0 years later. More improvement in GPS was observed in younger (<11 years) compared to older youth (p<0.001), and in those that had intervening orthopedic surgery (p<0.001). Baseline gait pattern, motor function, and proximal positioning (trunk and pelvis) also influenced changes in gait.

Improvement in gait kinematics in youth with CP is significantly influenced by age, orthopedic surgery, and baseline factors. Our findings provide a ML framework for improving clinical decision support systems to aid providers in predicting gait changes for ambulatory youth with CP.