Background: Complex clinical gait analysis results can be expressed as single number gait deviations by applying multivariate processing methods. The original Movement Deviation Profile (MDP) quantifies the deviation of abnormal gait using the most trusted nine dynamic joint angles of lower limbs. Research question: Which subset of joint angles maximises the ability of the MDP to separate abnormal gait from normality? What is the effect of using the best subset in a large group of patients, and in individuals? Methods: A self-organising neural network was trained using normal gait data from 166 controls, and then the MDP of 1923 patients with cerebral palsy (3846 legs) was calculated. The same procedure was repeated with 511 combinations of the nine joint angles. The standardised distances of abnormal gait from normality were then calculated as log-transformed Z-scores to select the best combination. A mixed design ANOVA was used to assess how removing the least discriminating angle improved the separation of patients from controls. The effect of using the optimal subset of angles was also quantified for each individual leg by comparing the change in MDP to the independent FAQ levels of patients. Results: Removal of hip rotation significantly (p<0.0005) increased the separation of the patient group from normality (ΔZ-score 0.24) and also at FAQ levels 7-10 (ΔZ-score 0.38, 0.27, 0.22, 0.14). The MDP of individual patients changed in a wider range of -4.65 to 1.12 Z-scores and their change matched their independent FAQ scores, with less functional patients moving further from, and more functional patients moving closer to normality. Significance: In existing gait databases we recommend excluding hip rotation from data used to calculate the MDP. Alternatively, the calculation of hip rotation can be improved by post-hoc correction, but the ultimate solution is to use more accurate and reliable models of hip rotation.
Bibliographical notePublisher Copyright:
- Hip rotation
- Movement deviation profile
- Neural network
- Sensitivity analysis