2.3. Data Analysis

From the output of both workflows, eight joint angles (pelvic tilt, pelvic list/obliquity, pelvic rotation, hip flexion, hip abduction, hip rotation, knee flexion, and ankle dorsiflexion) and the FPA were extracted, and time was normalised to 100% of the gait cycle using custom scripts in MATLAB. Additionally, for each subject, the mean kinematic waveform of the three gait trials for each limb was calculated.

The kinematics estimated with the two workflows were statistically compared using the non-parametric paired sample t-test option of the 1-dimensional statistical parametric package [25] in MATLAB. A Bonferroni correction was applied for multiple comparisons, and α was set to 0.0056 (9 pairwise comparisons). The root mean square difference (RMSD) was calculated between individual trials’ kinematic waveform from PiG and mGen for each subject and averaged to quantify the effect of the chosen approach on the joint angle estimates. Similarly, the coefficient of determination (R²) was calculated for individual trials with the linear fit method between PiG and mGen and averaged per subject for each of the nine kinematic variables as a measure of curve similarity.

The GPS was also calculated for all subjects using the data from the age-matched controls and the mean kinematic waveform outputs from both mGen and PiG. The GPS is calculated separately for each limb and is based on the Gait Variable Score (GVS), which is the root mean square (RMS) difference between the gait vector of a patient and the mean gait vector of a non-pathological group. These gait vectors include nine key kinematic variables: pelvic tilt, pelvic list/obliquity, pelvic rotation, hip flexion, hip abduction, hip rotation, knee flexion, ankle dorsiflexion, and foot progression angle. Formally, the GVS and GPS were calculated as described in Equations (1) and (2), respectively.

where GVSi is the Gait Variable Score of the kinematic variable i, t is a specific instance in the gait cycle, T is the total number of points in the cycle, xi,t is the value of kinematic variable i at instance t, and xi,tref is the mean of that variable at the same time point for a reference population.

where N is the total number of kinematic variables used and i is the ith kinematic variable used.

Data were checked for normality, and appropriate statistical tests were used subsequently. Consistency in GPS estimated with mGen and PiG was assessed using Spearman’s correlation coefficient (ρ). A Wilcoxon rank test was performed to determine significant differences between the two groups of GPS estimates. Additionally, the Symmetry Index (SI) [26] was used to calculate left/right limb asymmetry in GPS values for both models and compared using the Pearson’s correlation coefficient. To test the agreement between models when estimating changes in GPS (Δ_tGPS) after FDO-SEMLS, a Bland–Altman analysis was used. Δ_tGPS were normally distributed, and a paired sample t-test was used to determine if model estimates of Δ_tGPS were significantly different. To understand the contributing factors to any differences in Δ_tGPS estimated by both models, the above-described analyses were applied also to the nine components of the GPS.

To determine a classification of improvement (responders) from 3D gait analysis data using both mGen and PiG estimates of GPS, change in GPS (Δ_tGPS) from pre- to post-FDO-SEMLS for each model was calculated, after which a criterion of Δ_tGPS ≤ −1.6 (minimal clinically important difference for GPS [27]) was applied. This 1.6° threshold reflects the mean difference between GPS values of children classified in adjacent levels of the Functional Assessment Questionnaire measure of functional mobility. Participants with Δ_tGPS values not satisfying this criterion were classified as non-responders.

The process for clinically adjudging participants as responders or non-responders to FDO-SEMLS was based on consensus between two consultant orthopaedic surgeons and the team of gait laboratory staff (including clinical scientists and physiotherapists). This process involved a review of clinical examination findings and clinical dictations as well as video recordings looking at components of the Edinburgh Gait Score (EGS) [28], a validated visual analysis scale, that pertained to the desired correction from the FDO. In this regard, the video analysis emphasized hip rotation (using the knee progression angle in stance with consideration of transverse plane alignment as a surrogate) and the foot progression angle in stance. One clinical scientist completed the classifications, and these decisions were reviewed by the consultant orthopaedic surgeons. The measured kinematics and report from 3D gait analysis were not used in this process. Only the limbs that underwent the surgery were considered for each participant.