Proximal rectus femoris release does not influence the effects of distal rectus femoris transfer on the knee in cerebral palsy

Abstract

Introduction: Single event multilevel surgery (SEMLS) purpose is to improve the cerebral palsy (CP) children’s gait by associating multiple surgeries on the same therapeutic time. It is therefore complex to isolate the effect of these actions in this multifocal context. To address this problem we chose to specifically identify the effect of hamstrings lengthening (HL) in CP children with crouch gait. The aims of this study were to describe the specific parameters influenced by HL and to classify the positive or not-positive effect of HL in SEMLS. Patients/materials andmethods: 42 CP children (12±3 years) were divided into two groups: 31 (G1=60 lower limbs (LL)) and 11 (G2=20 LL), respectively having followed and not-followed HL among all the associated surgeries. All patients had clinical gait analysis before and 1.9±0.8 years after surgery. The GDI is calculated [1]. All kinematic data (angles, velocities) were doublenormalized and conditioned in two vectors. A homogeneity test (G1 vs G2) selected the kinematic parameters influenced by HL (t-test, p<0.005). Principal component analysis identified theminimumdescriptors characterizing the effect of HL. Several classifiers (Regularized Discriminant Analysis (RDA) and linear or nonlinear Support Vector Machines (SVM)) were supervised by 6 experts’ opinions. Experts’ opinions were based on video and kinematic curves comparison between pre and post-surgery conditions. The classifiers performances in learning, validation (leave one out) and generalization were compared. Results: GDI results showed that 83% of the subjects of G1 were globally improved by SEMLS. Among all the kinematic data, 16 sub-vectors, significantly influenced by HL were selected. Their dimensionality was reduced by principal component analysis. The 6 experts have classified the effect of HL for 37 LL: 24 were positive and 13 not-positive. The classification method with the best performance was the linear SVMwith error rate 0% in learning, 5.4% in validation and 6.5% in generalization. In view of the classification system 1/3 of G1 LL were not improved by HL. Discussion and conclusions: This supervised classification and data conditioning techniques are able to categorize the specific effect of HL among all the associated performed procedures in two classes “positive effect” and “not-positive effect”. While 83% of patients were improved by SEMLS, HL had positively contributed to this improvement in only 70% of these cases. This methodology can be generalized to study the effect of other surgical procedures.