A data driven model for optimal orthosis selection in children with cerebral palsy

Abstract

A statistical orthosis selection model was developed using the Random Forest Algorithm (RFA). The model's performance and potential clinical benefit was evaluated. The model predicts which of five orthosis designs – solid (SAFO), posterior leaf spring (PLS), hinged (HAFO), supra-malleolar (SMO), or foot orthosis (FO) – will provide the best gait outcome for individuals with diplegic cerebral palsy (CP). Gait outcome was defined as the change in Gait Deviation Index (GDI) between walking while wearing an orthosis compared to barefoot (ΔGDI=GDIOrthosis−GDIBarefoot). Model development was carried out using retrospective data from 476 individuals who wore one of the five orthosis designs bilaterally. Clinical benefit was estimated by predicting the optimal orthosis and ΔGDI for 1016 individuals (age: 12.6 (6.7) years), 540 of whom did not have an existing orthosis prescription. Among limbs with an orthosis, the model agreed with the prescription only 14% of the time. For 56% of limbs without an orthosis, the model agreed that no orthosis was expected to provide benefit. Using the current standard of care orthosis (i.e. existing orthosis prescriptions), ΔGDI is only +0.4 points on average. Using the orthosis prediction model, average ΔGDI for orthosis users was estimated to improve to +5.6 points. The results of this study suggest that an orthosis selection model derived from the RFA can significantly improve outcomes from orthosis use for the diplegic CP population. Further validation of the model is warranted using data from other centers and a prospective study.