Background
Turning manoeuvres are an essential component of mobility and are vital for effective real-world navigation. Turning is more challenging than straight-line walking, involving complex cognitive functions to execute multi-segment co-ordination. Therefore, people with cognitive impairment (PwCI) may be more susceptible to impaired turning performance. Inertial measurement units (IMUs) can be used to quantify turning performance; however, IMU-based algorithms have not yet been validated for PwCI, or across dementia disease subtypes.
Research question
Is a custom-built algorithm for accurately detecting turn start and end valid for use in PwCI and in different dementia disease subtypes?
Methods
Sixty-six PwCI due to Alzheimer’s disease, Lewy body disease and vascular dementia, along with 23 cognitively healthy older adults (controls) were included. Participants wore an IMU on their lower back while completing six 10-m intermittent walks, segmented by 180° turns. A 2D colour video camera was used as the reference system. Videos were reviewed by two independent blinded raters annotating turn start and end. Agreement (intra-class correlation (ICC (2,1)), Spearman’s rho and Limits of agreement) and error (Root mean square error; RMSE and bias) between the raters (rater 1 vs. 2) and the algorithm (rater vs. algorithm) were evaluated.
Results
There was excellent agreement (rater-rater and rater-algorithm) for detecting turn start and end for PwCI and across dementia disease subtypes (rho = 1.00, ICC = 1.00). The error between raters was lower (RMSE < 0.72 s, bias < 0.41 s) than the error between raters and algorithm (RMSE < 1.29 s, bias < 1.4 s). Error was lowest for controls (RMSE < 0.94 s), followed by AD (RMSE < 1.21 s) and LBD (RMSE < 1.29 s).
Significance
Key findings suggest that this algorithm can detect turn start and end using an IMU in PwCI in agreement with a reference system (video ratings). Future research should consider the clinical application of turning assessment in PwCI, such as its ability to differentiate dementia disease subtypes to support accurate diagnosis.