Naturalistic sleep tracking in a longitudinal cohort: how long is long enough?

BackgroundDespite broad interest in the health implications of sleep duration, traditional measurements via polysomnography or actigraphy are often limited to one or a few nights per person. Given the potential variability of sleep duration over time, inferential uncertainty remains an important issue for relatively short observation windows. MethodsWe describe potential limitations of shorter duration sleep tracking by sub-sampling from longer-term observation windows, using a combined approach of simulated data from known distributions, in addition to real-world data (30-365 nights) from over 35,000 participants who provided informed consent to participate in the Apple Heart and Movement Study and elected to contribute sleep data to the study. ResultsSimulations demonstrate that the magnitude of deviation from truth, defined using all available observations per individual, as well as the presence and direction of bias, depended on the sub-sample size, the type of simulated distribution (Gaussian versus skewed), and the summary statistics of interest, such as central tendency (mean, median) and dispersion (standard deviation (SD), interquartile range). For example, the SD computed from n=7 observations from a simulated normal distribution (7+1 hours) showed a median 6.7% under-estimation bias (IQR 24% under- to 14.7% over-estimation). Real-world sleep duration data, when under-sampled and compared to longer observations within-participant, showed similar SD bias at 7 nights, and similar convergence rates approaching the true value (based on 90 nights) as longitdunal sample number increases. Shapiro-Wilk tests for normality and log-normality show that 64% of simulated log-normal (skew) distributions fail to reject normality at n=7 samples, while real-world sleep duration data most commonly failed both normality and log-normality tests. Finally, simulated cohorts with sleep durations of 7+1 hours mixed with a subset of 6+1 hours sleepers showed that a random single-night observation of "short sleep" (6 hours) is more likely from random variation of a 7-hour sleeper, than from an actual 6-hour sleeper. Extending the observation to n=7 nights mitigates this mis-classification risk. ConclusionThe results of simulations and empiric data patterns suggests that longer duration tracking provides important and tangible benefits to reduce bias and uncertainty in sleep health research that historically relies on small observation windows.