| Summary: | Incidence of osteoarthritis (OA) is steadily increasing amongst the developed world, with the knee being the most commonly affected joint. Knee OA is a complex, progressive and multifactorial disease which can result in severe disability, pain, and reduced quality of life. Numerous biomechanical changes have been associated with OA disease progression within both the affected and unaffected joints. Total knee replacement (TKR) is a common surgical intervention which aims to replace the degenerated articular surfaces. As longevity of the prostheses have improved, TKR surgery is being recommended to an increasingly younger population. There is, however, a growing body of evidence to suggest a proportion of patients exhibit several functional limitations following surgery. Measuring functional changes is challenging, and numerous studies suggest patient-reported changes in physical function aren’t reflective of objectively measured changes. This study builds upon techniques to objectively assess biomechanical function during level gait using three-dimensional stereophotogrammetry, with an aim to quantify biomechanical changes that occur as a result of late-stage OA, and measure and summarise functional changes following TKR surgery. Firstly, the appropriateness of principal component analysis (PCA) and the Cardiff Dempster-Shafer Theory (DST) classifier to reduce and summarise level gait biomechanics is investigated within a cohort of 85 OA and 38 non-pathological (NP) subjects. The validity of previously adopted rules for retaining principal components (PCs) is assessed; namely the application of Kaiser’s rule, and a factor loading threshold of ±0.71. Through the reconstruction of biomechanical waveforms using individual PCs, it is demonstrated that this rule discards biomechanical features which can accurately distinguish between OA and NP gait biomechanics. The currently accepted definitions of two control parameters of the DST classifier, which define the shape of the sigmoid activation function, are shown to introduce a bias under certain conditions. New definitions are proposed and tested, which result in an increase in classification accuracy. The robustness of the leave-one-out (LOO) cross-validation algorithm to assess the performance of the classification is investigated, and findings suggest little benefit of retaining larger cohorts within the cross-validation set. Training bodies of different sizes are investigated, and their ability to classify the remaining data is evaluated. Results indicated that a training body of ten subjects in each group resulted in high classification accuracy (92% ± 2.5%), and improvements in accuracy then began to steadily plateau. The techniques developed thus far are then adopted to classify the hip, knee and ankle biomechanics of 41 OA and 31 NP subjects, to describe the biomechanical characteristics of late-stage OA. There were numerous methodological changes within this section of the study, and it was proved necessary to recalculate new PCs using this cohort. These new PCs were contextualised and used to classify OA biomechanics, resulting in a LOO classification accuracy of 98.6%. Anecdotally, the single misclassified subject had late-stage OA, but reported only mild functional impairments. The biomechanical features which consistently distinguished OA gait are ranked and discussed. The trained DST classifier was used to quantify the biomechanical function of 22 subjects pre and 12-months post-TKR surgery. In contrast to previous findings using the DST technique, biomechanical improvements varied, with no clear group of improvers. Five subjects were classified as NP post-operatively, seven were classified as “non-dominant OA”, and ten as “dominant OA”. Objectively measured function was significantly correlated with two out of nine patient-reported outcome measures both before surgery, and in all nine post-operatively. This might explain discrepancies in the literature between patient-reported and objectively measured changes. A retrospective analysis explored pre-operative predictors highlighted knee and ankle coronal plane angulation at heel strike, ankle range of motion, and timing of peak knee flexion as potential predictors of post-operative function.
|
|---|
