In-lab Assessment of a Sternum-Worn Accelerometer to Measure Gait Speed in Healthy Individuals
Nunzio Camerlingo1, Miles Welbourn1, Dimitrios John Psaltos1, Andrew Messere1, Hao Zhang1, Charmaine Demanuele1, Mar Santamaria1, Lukas Adamowicz1, Fikret Isik Karahanoglu1, Xuemei Cai1
1Digital Science and Translational Imaging, Pfizer
Objective:
To assess the validity of gait speed derived using a sternum-worn accelerometer, thus supporting the deployment of a single device to measure multiple biosignals, e.g., gait and ECG.
Background:
Gait assessment is essential to understand, diagnose, and monitor numerous neurological disorders. In contrast to conventional clinical evaluations, wearable sensors used in free-living environments grant more objective measurements. While wearing the widespread lumbar-worn accelerometers may be burdensome over the long-term, sternum-worn devices, already validated for at-home vitals monitoring, may be more comfortable. 
Design/Methods:

During two in-lab visits, twenty healthy individuals (13 females, mean±sd age: 33.9±9.1 years, BMI: 24.4±4.67kg/m2) were instrumented with a 6-sensor set (APDM), including lumbar-worn and sternum-worn accelerometers. Participants underwent two assessments: A1) walking three 20-ft laps on an instrumented mat (GAITRite), at normal, slow, and fast speeds; A2) freely performing in-lab 20-min activities with frequent walking bouts. Gait speed was extracted by sternum-worn and lumbar-worn accelerometers (test devices) using the in-house built GaitPy algorithm, and compared against the values collected by GAITRite and APDM (reference devices for A1 and A2, respectively). Agreement was evaluated via Bland-Altman analysis, Pearson’s correlation (R), and intraclass correlation coefficient (ICC).

Results:

In A1, sternum-worn device underestimated gait speed compared to GAITRite, but the values were highly correlated, and the ICC showed moderate-to-good agreement (R: 0.91, 0.81, 0.66; bias (m/s): -0.137, -0.162, -0.209; ICC: 0.59, 0.50, 0.43, for slow, normal, and fast speeds, respectively). In A2, bias=0.055, R=0.82, and ICC= 0.78 demonstrated excellent agreement between gait speed derived from APDM and sternum-worn device. Sternum-worn and lumbar-worn accelerometers showed comparable performances.

Conclusions:

Sternum-worn devices can reliably measure gait speed both in scripted walking tasks and naturalistic walks. They can be deployed to simultaneously capture gait parameters and vital signs, thus opening new opportunities for more in-depth, free-living studies of patients with neurological disorders.

10.1212/WNL.0000000000202383