Objective: To determine if a novel, location-agnostic algorithm provides consistent turn-speed measures in people with Parkinson’s (PwP) across sternum, lower back, and thigh locations.

Background: Objective assessment of turn speed is essential in PwP, as turning performance is a sensitive indicator of disease progression and a predictor of fall risk. In PwP, turning is often characterised by en bloc rotation, shuffling, and freezing of gait. These features, combined with leg-swing noise and soft-tissue artefacts at different carry locations, can degrade turn-speed accuracy. A location-agnostic algorithm is required to ensure that data reflect true turning performance rather than carry-location noise, facilitating reliable remote monitoring in home-based settings.

Method: This study analysed data from the WearGait-PD dataset (N=57 PwP; Modified Hoehn & Yahr: 2.1±0.5; Age: 65.6±8.6y; Disease duration: 7.6±5.8y; 61% male). Participants performed self-paced and hurried-paced 180-degree turn walk tests, wearing four synchronised inertial measurement units (IMUs) on the sternum, lower back, and lateral thighs. A novel, location-agnostic turn-detection algorithm extracted turn speed from each IMU. Consistency was evaluated using Bland-Altman analysis and Mean Absolute Percentage Error (MAPE), relative to the lower back reference. The Two One-Sided Test (TOST) verified statistical equivalence across locations (equivalence bounds: 5 deg/s).

Results: The algorithm demonstrated high internal consistency across wear locations. Bland-Altman analysis revealed a near-zero mean bias and narrow 95% limits of agreement, confirming a high degree of agreement, with no systematic over- or underestimation of turn speed. Measurement variance was uniform, with no significant proportional error. Absolute error remained low across both sternum and thigh locations compared to the lower back reference (MAPE<6.1%; Sternum=3.8%). TOST confirmed statistical equivalence for all pairings (p<0.01), falling within the zone of clinical indifference.

Conclusion: Near-zero bias and high internal consistency confirm the algorithm is robust across sternum-to-thigh locations. This wear-location invariance is essential for data integrity in longitudinal clinical trials, supporting patient flexibility and adherence while ensuring that changes in turn speed reflect true disease progression rather than carry-location noise.

« Back to 2026 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/robust-location-agnostic-turn-speed-estimation-to-ensure-consistent-remote-monitoring-of-gait-impairment-in-parkinsons-disease/