As a non-invasive technique, gait analysis holds potential for auxiliary screening and early detection of knee osteoarthritis (OA), and for assessing knee function before and after therapeutic or surgical intervention. However, the interpretation of gait data is controversial and has hindered its application to clinical decision making. A significant barrier to clinical use of gait information is the successful reduction and analysis of gait waveforms. This thesis addresses the need for a statistically based method to discriminate and classify subjects based on gait waveforms. The hypothesis was that principal component models of normal gait could detect differences in the gait pattern of OA patients that are clinically relevant. The objectives were to develop principal component models of normal gait data, to implement the methodology and investigate its ability to detect and interpret differences from normal gait, and to validate the methodology relative to two clinical measures:​ static knee alignment and the Knee Society Score (KSS).

Principal component models were developed for eight knee kinematic and kinetic gait waveforms of a group of 30 normal elderly subjects. The number of principal components required in the models varied from 2 to 4, while the proportion of the variation explained ranged from 60 to 95%. The models were designed to characterise normal gait waveform data by establishing statistical distance measures to indicate the similarity of gait waveforms to the average trajectory of the normal subjects. Each model consisted of a set of loading vectors, principal component scores and residuals. The loading vectors revealed the structure of the model and the scores and residuals were used as the distance measures about which confidence intervals were developed.

Pre-operative and post-operative gait data from 13 unicondylar arthroplasty (UCA) patients were used to demonstrate the application of the principal component models to pathological gait data. The principal component models assessed the overall statistical similarity of patient waveforms to normal using the two distance measures. These measures revealed whether the gait waveforms were within normal limits and the structure of the models revealed the portion of the gait cycle responsible for the difference.

As part of the validation of the methodology, the adduction moment principal component model was compared to the static coronal plane alignment. A correlation between the post-operative adduction moment principal component one score and the hip knee ankle (HKA) angle existed (r²=0.58, p<0.01), but there was no pre-operative correlation.

A gait score was developed to indicate the overall assessment of the kinematic and kinetic gait measures by the principal component models. This gait score was shown to agree with the clinical status as measured by the Knee Society Score (pre-op:​ r_s=0.86:​ post-op:​ r_s=0.73). Thus, the differences in gait pattern detected by the principal component models were clinical relevant.

The study indicated that data from the entire gait cycle can be reduced to simple measures which are sensitive to changes in gait pattern associated with OA and its treatment (UCA). The principal component models allow objective assessments of patient gait data, and interpretation of the assessments through interrogation of the underlying structure of the models.