0816 Identifying Time in Bed with Chest-Worn Accelerometry

Abstract Introduction Chest-worn accelerometers can capture both acceleration and posture, but their use in detecting and measuring time in bed (TIB) has been limited. We investigated whether TIB can be estimated from chest-worn accelerometers using both acceleration and posture data without using sleep diary information. Methods We developed a method to classify TIB using data from the Zio XT Patch (iRhythm, San Francisco, CA, USA) onboard accelerometer in the Atherosclerosis Risk in Communities (ARIC) Study. The method was optimized by maximizing its F1 score using minute level data relative to the in-bed status which was estimated from simultaneously collected hip-worn actigraphy data using a published method among ARIC participants (n=405, age 78.5±4.7 years). The method was then used to estimate whether a person was in or out of bed among all ARIC participants (n=2,470, age 79.2±4.6 years) and derived the average TIB, TIB midpoint, and the standard deviation of TIB midpoint across days. To evaluate contextual validity of the new TIB summaries, we examined their cross-sectional associations with demographic and clinical characteristics, and longitudinal associations with incident dementia. Results Participants contributed an average of 5.7 days (IQR: 5, 6.6) to the optimization process. The best algorithm had a F1 score of 0.843. Compared to the TIB estimated from the hip-worn device, TIB identified from the chest-worn device were on average 34.5 minutes shorter (95% CI: 28.4, 40.6), began 5.4 minutes earlier (95% CI: 0.3,10.5), and ended 36 minutes earlier (95% CI: 32.9,40.2). TIB was cross-sectionally associated with sex, race, comorbidities, and cognitive function. During a median 5.2 years (IQR: 4.5-5.7) of follow-up, 316 participants developed dementia. Compared to participants who spent a moderate amount of TIB (8.5 to < 9.75 hours/day), the cumulative incidence ratio for dementia was 1.61 (95% CI: 1.15,2.27) for the lowest tertile and 1.37 (95% CI: 1.01,1.85) for the highest tertile of TIB. An earlier in-bed midpoint and lower day-to-day variability in TIB midpoint were marginally associated with a reduced dementia incidence. Conclusion This study suggests that chest-worn accelerometers provide reasonable and useful estimates of TIB, opening a new opportunity for sleep assessment in older adults. Support (if any)