Abstract Introduction Measuring sleep timing information through passive phone-usage metrics could increase sample sizes and reduce participant burden for clinical and research use. A required step is evaluating how phone-based measures compare with established methods. We investigated the relationship between phone usage and recorded sleep-onset and offset timing across 30 days in college students. Methods 202 college students (35% Female) completed 30 days of at-home monitoring with wrist actigraphy, daily sleep diaries, and phone-usage tracking. All screen-on events longer than 10 secs (to reduce notification-only events) were identified and averaged in 10-minute bins within a 5-h window surrounding both sleep onset and offset times, not including naps, as defined using actigraphy and sleep diaries. For each participant, phone use within this time window was normalized, mean phone use during the 30 min before and after each transition was calculated, and paired t-tests were run. To evaluate whether the difference scores could identify sleep onset or offset, we determined the percentage of participants whose scores exceeded one standard deviation from the mean difference across each transition. Results At sleep onset, mean phone use dropped from 7.5% ±4.0 (SD) in the 30 minutes before to 1.1% ±1.4 after (p.001). At sleep offset, phone use increased from 2.6% ±1.6 before to 5.3% ±2.0 after (p.001). Mean difference scores were –6.4 ±4.2% for sleep onset and 2.7 ±2.8% for sleep offset. Only ~65% of participants for sleep onset and ~55% for sleep offset had differences in percentage use within one standard deviation of the mean: ~35% or ~45% (respectively) of participants had differences outside this range. Conclusion As expected, phone use showed a clear group-level shift across sleep-wake transitions. However, large individual differences limit the ability to use these difference scores for classifying sleep onset or offset on an individual level. Phone-use difference scores around sleep transitions based on these methods are not yet reliable individual-level markers of sleep timing in this population. Higher-resolution data and personalized normalization may improve predictions. Support (if any) NIH R01GM105018, NEC, Samsung Electronics
Hollister et al. (Fri,) studied this question.