Feasibility, reliability, and validity of using accelerometers to measure physical activities of patients with stroke during inpatient rehabilitation

Abstract

Promoting physical activities is important for medical and functional recovery after stroke. Therefore, an accurate and convenient measurement of physical activities is necessary to provide feedback on functional status and effects of rehabilitative interventions. We assessed the feasibility, reliability, and validity of wearing accelerometers to monitor physical activities of stroke patients by estimating energy expenditure. This was a prospective observational quantitative study conducted in an inpatient rehabilitation unit. Twenty-four patients with subacute stroke were enrolled. They wore accelerometers on wrists and ankles for three consecutive weekdays. The feasibility was evaluated by daily wear-time. The test-retest reliability was determined by intra-class correlation coefficient. The validity was evaluated by comparing accelerometeric data to behavior mappings using Mann-Whitney U test, Spearman's rho correlation coefficient (r) and Bland-Altman plots. Average wearing time for four accelerometers was 20.99 ± 3.28 hours per day. The 3-day accelerometer recording showed excellent test-retest reliability. For sedentary activities, wrist accelerometers showed higher correlation with direct observation than ankle accelerometers. For light to moderate activities, ankle accelerometers showed higher correlation with direct observation than wrist accelerometers. Overall, combined models of accelerometers showed higher correlation with direct observation than separate ones. Wearing accelerometers for 24 h may be useful for measuring physical activities in subjects with subacute stroke in an inpatient rehabilitation unit.

Fig 1. Bland-Altman plots showing energy expenditure agreements between accelerometers and behavior mappings.

Filled circles (•) show behavior mappings. Open circle (○), triangle (Δ), and square (□) represent sedentary (< 1.6 METs), light (1.6–2.9 METs), and moderate (3.0–5.9 METs) activities from acccelerometers. Vigorous (≥ 6 METs) activity was not observed. Behavior mapping showed the strongest agreement with energy expenditure (PAEE) of bUL+bLL without proportional bias (R² = 0.003, p = 0.052, Fig 1I). A relatively better agreement was observed in lower intensities than that in higher intensities. Dotted lines represent mean ± 1.96 SD of difference in PAEEs. Lightface lines represent regression line for difference in PAEEs.