Application of a PCA-Biplot methodology in understanding gait temporal and spatial parameters during aging

INTRODUCTION Gait variables are highly correlated and have a temporal dependence [1]. Principal component analysis (PCA) has been applied successfully to extract information from highly correlated data [2]. Interpretation of the PCs’ is challenging. An alternative is to use a graphical display called Biplot [3] that gives insights into relationships and trends of complex data sets. Our goal was to demonstrate the role played by gait variables in women during life span with PCA-biplot. METHODS There were 453 subjects in the study: 151 elderly females (71.6±5.0 yrs), 152 adults (44.7±5.4 yrs) and 150 young (21.7±4.1 yrs) subjects. Subjects were able to walk independently. Gait parameters were collected using a computerized carpet (GAITRite, CIR systems, USA) at subjects’ preferred walking speed. PCs were first obtained followed by the PCA-biplot. The axes are represented by PC1 and PC2; the PC scores of the groups were averaged and represented by symbols and the variables by vectors. The origin of the vectors is the mean for all variables. Variables’ with higher vector length indicate higher variance. The smaller the relative angle between two variables’ vectors the higher their correlation. The projection of the groups onto the variables’ vectors indicates their spatial distance in relation both to each other and to a set of variables; projections above the mean indicates higher value for that specific variable. In addition, the highest the distance along the variable more important is that variable in group separation [3]. RESULTS Figure 1 show that PC1 is heavily weighted by stance time, cadence, velocity and double support time and PC2 by step length and swing time. The highest relative variance is attributed to swing time. Velocity and double support show higher negative correlation. Elderly females walk with lower velocity, smaller step length, shorter swing time, higher cadence and longer double support time compared to the other two groups. The greatest distances between the groups are along the variable step length; therefore, step length and PC2 are important in group separation. With the objective to interpret the behavior of the groups in the 2-PC model, Table 1 was built with the result of the projection of the groups on the variables represented by symbols; when the projection falls well above the mean or below the mean, two plus and two negative signs were marked, respectively; when the projection falls somewhere between only one plus or one minus sign was marked. From the table we can conclude that as we get older we seek for a more stable gait pattern, since we decrease our speed and increase our cadence and time spent in stance. The behavior of the young group is opposite to the elderly’s and the adult group is somewhere between. Figure 1: PCA-biplot of the gait variables and the average score of the female elderly, adult and young groups (N=453).