How to choose and interpret similarity indices to quantify the variability in gait joint kinematics

Repeatability and reproducibility indices are often used in gait analysis to validate models and assess patients in their follow-up. When comparing joint kinematics, their interpretation can be ambiguous due to a lack of understanding of the exact sources of their variations. This paper studied four indices (Root Mean Square Deviation, Mean Absolute Variability, Coefficient of Multiple Correlation, and Linear Fit Method) in relation to five confusing-factors: joints’ range of motion, sample-by-sample amplitude variability, offset, time shift and curve shape. A first simulation was conducted to test the mathematics behind each index. A second simulation tested the influence of the curve shape on the indices using a Fourier’s decomposition. The Coefficient of Multiple Correlation and the Linear Fit method Coefficients were independent from the range of motion. Different Coefficients of Multiple Correlation were found among different joints, leading to misinterpretation of the results. The Linear Fit Method coefficients should not be adopted when time shift increases. Root Mean Square Deviation and Mean Absolute Variability were sensitive to all the confusing-factors. The Linear Fit Method coefficients seemed to be the most suitable to assess gait data variability, complemented with Root Mean Square Deviation or Mean Absolute Variability as measurements of data dispersion.