Knee osteoarthritis (OA) is a complex disease process that involves multiple, correlated mechanical factors. Modem gait analysis has the potential for providing important insight into the mechanical nature of the disease process of knee OA. Its clinical use, however, has been hindered by a lack of appropriate techniques for analysing large volumes of correlated gait data. A need has been addressed for a gait analysis technique to discriminate between groups based on multiple time varying and constant gait measures.

The hypothesis was that clinically relevant differences in the gait patterns of patients with knee OA existed. It was further hypothesized that these differences involved the interaction of several gait measures and could be detected with the use of two multivariate statistical techniques, principal component analysis and discriminant analysis. The major objective of this thesis was to develop a multidimensional gait analysis technique to discriminate between the gait patterns of groups of subjects. This technique would simultaneously consider multiple time varying and constant gait measures.

The multidimensional technique was applied to detect and interpret gait pattern differences between 63 normal subjects and 50 subjects with severe knee OA. Twelve biomechanical features described the major gait pattern changes with knee OA. A discriminant function based on these twelve features successfully separated the groups with a misclassification error rate of <6%. A single-valued discriminant score defined an OA severity index that quantified the abnormality of a subject's gait pattern. The discriminatory features were biomechanically interpreted in terms of the relative contributions of each gait measure, and in terms of the portion of the gait cycle responsible for the difference that was described.

The multidimensional technique was validated by using it to detect more subtle pre-operative gait pattern differences between patients who received unicompartmental knee replacements (UKR) and total knee replacements (TKR). The UKR patient data were introduced to the discrimination model that was developed with the normal and severe knee OA subjects. Their discriminant scores fell between the normal and OA scores. This indicated only mild gait pattern abnormalities in the UKR patients, as expected. A second discriminatory model was created with the UKR and TKR subjects and the features of the first model. A new discriminant function completely separated the groups. Five features described important pre-operative gait pattern differences between them.

The discriminatory features involved the interaction of a number of gait measures throughout the gait cycle and therefore represented multidimensional gait phenomena, indistinguishable with visual gait observation, and undetectable with univariate data analysis techniques. The results of this thesis indicated that multidimensional, correlated gait data could be reduced to an interpretable difference measure, sensitive to gait pattern changes associated with knee OA.