How many nights are needed? The short-term stability of intraindividual variability in sleep parameters derived from...

Individual sleep/wake patterns can be characterized by the intraindividual mean (iM) and variability (iSD), which captures night-to-night fluctuations. Most accelerometry-based sleep studies assess 7-14 nights, which may be insufficient for reliable variability estimates. This pre-registered, large-cohort study quantifies how many nights are needed for reliable sleep iM and iSD estimates using accelerometry.

Participants (n = 10 412, Mage = 37.75 ± 10.65, 50.00% women) wore a validated wrist-worn biometric device (WHOOP) for one year (3 700 492 person-nights). Total sleep time (TST), sleep onset, midsleep, sleep offset, wake after sleep onset (WASO), and sleep percentage were estimated. Reliability correlation coefficients compared reference values (calculated from all 363 nights) with 2-362 consecutive nights randomly selected per participant, with r > .80 considered reliable. 95% limits of agreement were calculated using the Bland-Altman method.

Reliable iM estimates (r > .80) were achieved with 2-7 nights of data, depending on the sleep metric: TST (7 nights), sleep-onset and midsleep (3), sleep-offset and sleep percentage (4), WASO (5). In contrast, reliable iSD estimates required substantially more nights: TST (43 nights), sleep-onset (44), midsleep and sleep-offset (41), sleep percentage (62), and WASO (65). iSDs estimated from 7-14 nights showed poor reliability (r = .50-.67) and wide 95% limits of agreement (e.g. TST: ±50.5 min for 7 nights).

Findings highlight the importance of tailoring study periods to the specific sleep metric of interest. Longer monitoring periods may improve diagnostic consistency and inform evidence-based recommendations for sleep assessments in clinical and research settings.