Validity of a Novel Algorithm to Detect Bedtime, Wake Time, and Sleep Time in Adults

Abstract

Purpose: To determine accuracy of activPAL Technologies’ CREA algorithm to assess bedtime, wake time, and sleep time. Methods: As part of a larger study, 104 participants recorded nightly sleep logs (LOGs) and wore the activPAL accelerometer at the thigh and ActiGraph accelerometer at the hip for 24 hr/day, for seven consecutive days. For sleep LOGs, participants recorded nightly bed and daily wake times. Previously validated ActiGraph, proprietary activPAL, and the Winkler sleep algorithm were used to compute sleep variables. Eighty-seven participants provided 2+ days of valid data. Pearson correlations, paired samples t tests, and equivalency tests were used to examine relationships and differences between methods (activPAL vs. ActiGraph, activPAL vs. LOG, and activPAL vs. Winkler algorithm). Results: For screened data, moderately high to high correlations but significant mean differences were found between activPAL versus ActiGraph for bedtime (t86 = −6.80, p ≤ .01, r = .84), wake time (t86 = 4.80, p ≤ .01, r = .93), and sleep time (t86 = 7.99, p ≤ .01, r = .88). activPAL versus LOG comparisons also yielded significant mean differences and moderately high to high correlations for bedtime (t86 = −4.68, p ≤ .01, r = .82), wake time (t86 = 8.14, p ≤ .01, r = .93), and sleep time (t86 = 8.60, p ≤ .01, r = .72). Equivalency testing revealed that equivalency could not be claimed between activPAL versus LOG or activPAL versus ActiGraph comparisons, though the activPAL and Winkler algorithm were equivalent. Conclusion: The activPAL algorithm overestimated sleep time by detecting earlier bedtimes and later wake times. Because of the significant differences between algorithms, bedtime, wake time, and sleep time are not interchangeable between methods.