Effect of sample size and reduced number of principle components on the Gillette Gait Index

The Gillette Gait Index uses principle components analysis of 16 variables to determine the deviation of an individual's gait compared to a normal control set. Previous literature has not reported on the effects of altering the size of the control set used to create the principle components, or described the effects of using less than the maximum number of principle components, 16, to calculate the Gillette Gait Index (GGI). Calculations of the GGI were determined for a group of 24 able-bodied normal subjects and 24 cerebral palsy subjects using 128 control subjects allotted into 15 subsets of varying sizes, from N = 16-128. A minimum of 40 controls were needed for GGI estimates to achieve less than approximately 20% error, and 96 controls were needed for less than 10% error, if all 16 principle components were used. With smaller control sets, an alternative method to increase the accuracy would be to use only those principle components that represent 95% of the variance. Caution must still be used when describing differences in GGI among groups, or changes in an individual's GGI over time. In addition, absolute changes in GGI should always be reported, as differences as great as 150 were seen in cerebral palsy patients across control groups, even when greater than 40 controls are used to create the principle components.