Evaluating Gait Variability Indices

Once all optimizations were performed, data from four previous publications using the original version of the GVI was extracted to compute the EGVI scores, and the results were compared.
When GVI<100, the gait variability is increased or decreased compared to the reference and GVI≥100 indicates a normal amount of gait variability. The EGVI differentiates low and high variability, and EGVI<100 indicates low variability, EGVI = 100 similar amount of variability as the reference group, EGVI>100 high variability.
To compare unbiased values of the GVI scores to the EGVI, data that corresponded to GVI>100 (magnitude problem), or GVI<100 and EGVI<100 (directional problem) were removed before calculating the coefficients of determination. Previously published data for individuals with Friedreich ataxia (FRDA dataset) [1 (link)], typically developing children (TD dataset) [5 (link)], older adults (OA dataset) [4 (link)], and for individuals with mild to moderate Parkinson’s Disease (PD dataset) [11 (link)] were re-analyzed using the same statistics as in the original publications (more details about statistics can be found in Figure A and Table A in S1 Appendix). For the FRDA dataset, Pearson’s correlations were used to investigate relationships between EGVI and FAPS, 8 m walk test time, Lower limb testing, ICARS, and PGD subscale [see Table A1 in S1 Appendix for details]. In the TD dataset, non-parametric rank tests (Spearman’s r) were carried out on the data of the 140 children and teenagers to evaluate the relationships between the EGVI and other spatiotemporal gait parameters. In the OA dataset, Pearson correlation coefficients investigated the relationship between EGVI and clinical measures of functional mobility and balance, including number of falls in the past year, Berg Balance Scale, Short Physical Performance Battery, Activities-Specific Balance Confidence, Timed Up and Go Test, Community Balance and Mobility Scale, Dynamic Gait Index and Functional Reach Test. For the PD dataset, the Pearson’s correlation coefficient was used to investigate the association between EGVI and the Mini-BESTest and the TUG. Due to a heteroscedastic distribution in the data, an inverse transformation was performed on the TUG scores (1/TUG). Like for the GVI in the article by Rennie et al. [11 (link)], the responsiveness of the EGVI was examined by ROC curve statistics to explore to what extent the index was able to discriminate between individuals classified Hoehn & Yahr 2 and 3. The main clinical feature that separates H&Y 2 and 3 is the manifestation of postural instability in those classified as H&Y 3 as opposed to 2. It was therefore hypothesized that those graded as H&Y 3 would have higher EGVI scores as decreasing balance capabilities are associated with higher gait variability [12 (link), 13 (link)].