Accuracy Of Behavioral Assessment With A Wearable Camera in Semi-structured And Free Living Conditions In Older Adults: 2307 Board #320 June 1 2
METHODS: 824 observations were analyzed (n=8; 5 female; 69±5.1 y; 169.1±5.2 cm; 69.5±12.5 kg) over a 20-minute Semi bout of activity (sit, stand, walk) while video recorded. Videos were analyzed by DO. An independent sample of 1499 observations were analyzed (n=5; 3 female, 71±5.1 y; 162.1±5.1 cm; 64.5±10.6 kg) during a three-hour bout of community-dwelling Free activity while a researcher performed DO. Both groups wore a WC placed at the sternum. WC images and DO were annotated independently for posture (sedentary, standing, movement), activity category (sedentary, walking, household, exercise/sport), and type (laundry, dishes, cooking, general cleaning). Cross tabs and Kappa statistics were run to assess accuracy between the WC images and the DO results across both conditions for observations. Time spent in each attribute was tested with the Wilcoxon signed-rank test.
RESULTS: Posture had a 96.4% (Kappa=0.93; SE=0.10; p<0.0001) and 93.1% (Kappa=0.89; SE=0.07; p<0.0001) agreement between the WC and DO during the Semi and Free conditions, respectively. For activity category, there was 76.7% (Kappa=0.45; SE=0.10; p<0.0001) and 94.6% (Kappa=0.92; SE=0.08; p<0.0001) agreement between the WC and DO during the Semi and Free conditions, respectively. For activity type, obtained from Free only, there was 100% (Kappa=1.0; SE=0.20; p<0.0001) agreement across measures. WC total time spent in seconds was not significantly different to that obtained from the DO for posture during Semi (186±47 vs. 199±50, z=0.9342) or Free conditions (1341±414 vs. 1400±418, z=0.3170). For activity category, time spent across each measurement type did not significantly differ for Semi (WC=162±38 vs. DO=171±44, z=0.7348) or Free conditions (WC=1082±322 vs. DO=1256±342, z=0.4806). For activity type, time estimates for Free only were not statistically different 827±333 for WC and 145±70 for DO, z=0.7459.
CONCLUSIONS: Results from this study suggest that there is high congruency between wearable cameras and direct observation for behavioral observations and time spent in posture, activity category and type.