Statistical analysis

Lower limb kinematics and temporospatial data were calculated using the Plug-In-Gait model (Vicon, Oxford Metrics). Step-length (SL) and SSWS were normalised against average leg-length for each participant. Leg-length was defined as the distance between the anterior superior iliac spine and the medial malleolus on the same side, and the arithmetic mean of the left and right leg-length formed the average leg-length value used for normalisation. The gait deviation index (GDI) was calculated using the template provided by its authors[28]. For the amputee group, the GDI was calculated for six trials per limb per participants and averaged to obtain the value used in subsequent analyses. A representative trial from the left and right limb was used from each able-bodied participant, and contributed to the normative database required for the calculation of the GDI. In doing so, the GDI distribution for the able-bodied participants has a mean value of 100, with every 10 points below equal to one standard deviation away from the mean. The average of the three TUGTs was used in further statistical analyses. The time taken to stand (t_stand) was derived from the TUGT. Both the summary scales and individual questions from the PEQ were utilised.

Normalcy of data was assessed using the Anderson-Darling test. Summary statistics were calculated using measures appropriate to their distribution - mean and standard deviation for normal distributions, and median and interquartile range for non-normal distributions. Analysis of variance was used to compare results between the able-bodied group, transtibial amputee group and transfemoral amputee group for normally distributed data (Table ​(Table1).1). Kruskal-Wallis one-way analysis of variance was used for data that did not conform to a normal distribution. A P-value of less than 0.05 was considered significant.

Summary of results

Spearman’s rank correlation coefficient, ρ, was used to determine the relationships between the GDI and participant characteristics, performance-based measures and self-report measures. Strict significant criteria for the correlation coefficient were required to minimise the chance of coincidental findings, possible due to the large number of relationships investigated in this study[29]. Significance was set at P ≤ 0.001, or |ρ| ≥ 0.70.

Stepwise regression analyses were used to determine the major predictors of the GDI (dependent variable), with participant characteristics (as listed in Table ​Table11 and including aetiology), performance-based measures and responses from the PEQ used as independent variables in the regression models. The alpha-to-enter and alpha-to-exclude were set to 0.2 to accommodate the small sample size[30]. Predicted R² values were calculated using a leave one out cross-validation protocol. Three types of regression analyses were performed for various reasons. The GDI was the dependent variable in all models.

All independent variables: A purely explorative model, including all independent variables to determine the best possible predictors of the GDI.

Omission of SL relationships: Clinical utility requires that reliance on instrumentation be minimised. Of the outcome measures adopted in this study, with the exception of the GDI, instrumentation was required only for the calculation of SL. Other measures needed little more than a stopwatch to obtain. SL relationships were omitted from the second regression analysis to minimise the need for instrumentation and consider applicability.

Forced inclusion of walking speed relationships, omission of SL relationships: Walking speed is often considered a robust measure of functional ability[1] in population groups with movement disorders. This was investigated in the final regression analysis by forcing the inclusion of walking speed relationships as independent variables.

Since frustration is known to affect self-efficacy[31], responses to self-report measures will differ between participants reporting frustration and participants reporting an absence of frustration. The PEQ contains within it questions relating to frustration. To account for differences in self-efficacy, participants were separated based upon the presence (n = 16) and absence of frustration (n = 4) as measured by the frustration questions in the PEQ [Larger studies (n = 135) by our group have shown that approximately 75% of LLAs experience some form of frustration as measured by the PEQ]. Regression analyses were performed using only participants reporting frustration. The small sample size prohibited separate analysis of the participants who were not frustrated.