Best Combination of Gait Measures Discriminating Multiple Sclerosis from Healthy Controls Using Body-Worn Inertial Sensors

Background: Gait deficits are common in Multiple Sclerosis (MS) but poorly captured by stopwatch-timed tests or rating scales. Body-worn inertial sensors can detect gait abnormalities in people with MS who have normal walking speed. However, key challenges in using body-worn inertial sensors to monitor gait characteristics are an excessive number of measures and a lack of consensus on the most useful measures for MS.

Objectives: This study aimed to determine the best combination of gait measures to discriminate MS from healthy control (HC) subjects.

Methods: We used two datasets, Study I as a development and validation dataset, and Study II as an independent dataset to test the generalizability of the proposed model. Study I recruited 14 MS and 17 HC, and Study II recruited 9 MS and 7 HC. Participants were instructed to complete the 6-minute walk test at their fastest speed to cover as much distance as possible by walking back-and-forth along a 20-m straight walkway (Study I) or a 15-m straight walkway (Study II). Subjects wore inertial sensors (Opals) attached to both feet, sternum, and lumbar regions. LASSO (5-fold, cross-validated least absolute shrinkage and selection operator) was applied as a feature-selection method on 70% of the training dataset, followed by logistic regression on the remaining 30% of Study I. To test the generalizability of the proposed model, it was applied to the Study II independent data. The area under the curve (AUC) of receiver operator characteristic (ROC) curves was used to evaluate the discriminate ability of the proposed model.

Results: From 36 gait measures, LASSO selected 6 measures from the training dataset: stride time, double support time standard deviations (%), turn duration, total number of turns, elevation at mid-swing, and toe-out angle standard deviation. Logistic regression with the above 6 gait measures resulted in AUC=1 (sensitivity=1 and specificity=1) when applied on the validation dataset (30% of Study I). The proposed model applied to a totally independent dataset (Study II) resulted in AUC=0.92 (sensitivity=0.89, specificity=1).

Conclusions: The best combination of gait measures for accurate classification of MS gait from HC gait during the 6-minute walk test did not include gait speed. These findings pave the way for a better understanding of gait deficits in MS to support informed clinical decision-making about the status of the disease or the effect of an intervention.