Validity of an automated algorithm to identify waking and in-bed wear time in hip-worn accelerometer data collected with a 24 h wear protocol in young adults

© 2016 Institute of Physics and Engineering in Medicine.Researchers are increasingly using 24 h accelerometer wear protocols. No automated method has been published that accurately distinguishes 'waking' wear time from other data ('in-bed', non-wear, invalid days) in young adults. This study examined the validity of an automated algorithm developed to achieve this for hip-worn Actigraph GT3X + 60 s epoch data. We compared the algorithm against a referent method ('into-bed' and 'out-of-bed' times visually identified by two independent raters) and benchmarked against two published algorithms. All methods used the same non-wear rules. The development sample (n = 11) and validation sample (n = 95) were Australian young adults from the Raine pregnancy cohort (54% female), all aged approximately 22 years. The agreement with Rater 1 in each minute's classification (yes/no) of waking wear time was examined as kappa (?), limited to valid days (10 h waking wear time per day) according to the algorithm and Rater 1. Bland-Altman methods assessed agreement in daily totals of waking wear and in-bed wear time. Excellent agreement (? > 0.75) was obtained between the raters for 80% of participants (median ? = 0.94). The algorithm showed excellent agreement with Rater 1 (? > 0.75) for 89% of participants and poor agreement (? <0.40) for 1%. In this sample, the algorithm (median ? = 0.86) performed better than algorithms validated in children (median ? = 0.77) and adolescents (median ? = 0.66). The mean difference (95% limits of agreement) between Rater 1 and the algorithm was 7 (-220, 234) min d-1 for waking wear time on valid days and -41 (-309, 228) min d-1 for in-bed wear time. In this population, the automated algorithm's validity for identifying waking wear time was mostly good, not worse than inter-rater agreement, and better than the evaluated published alternatives. However, the algorithm requires improvement to better identify in-bed wear time.