Background: The published literature shows no automated method valid for isolating adults out-of-bed wear in continuously worn (24h/day) accelerometer data. We developed an automated algorithm (AA) for continuous hip-worn protocol Actigraph (GT3X+) data that separates out-of-bed wear from in-bed wear and non-wear periods.
Methods: In 95 participants of the Raine study (mean±SD, 21.9±0.57 years, 54% female), we examined agreement with a referent method (i.e., ‘getting into’ and ‘getting out’ of bed times visually identified by two independent raters, R1 and R2).
Results: The mean difference (AA – R1) in out-of-bed wear was 7 min/day (95% Limits of Agreement: 220 to 234 min) on days valid by both methods. Agreement for the classification of each minute as out-of bed wear (yes/no) was excellent (Îº > 0.75) for 89% of participants (AA vs R1) and poor (Îº<0.4) for only 1%, with a median Îº=0.86, comparable to the agreement between raters R1 and R2 (median Îº= 0.94). Here, the AA performed better than previous algorithms validated in children (median Îº =0.77) and adolescents (Îº =0.66) in our dataset. Agreement in classification as in-bed wear (yes/no) for AA vs R1 was excellent for only 51% of participants (median Îº =0.80). Out-of-bed minutes were identified as such 94% of the time while 9% and 11% of in-bed worn minutes were classified as nonwear and out-of-bed wear, respectively.
Discussion: The algorithm is useful and better than existing automated alternatives, for isolating out-of-bed wear but requires improvement to specifically isolate in-bed wear periods.