Dissociable effects of self-reported daily sleep duration on high-level cognitive abilities

Abstract

Most people will at some point experience not getting enough sleep over a period of days, weeks, or months. However, the effects of this kind of everyday sleep restriction on high-level cognitive abilities-such as the ability to store and recall information in memory, solve problems, and communicate-remain poorly understood. In a global sample of over 10000 people, we demonstrated that cognitive performance, measured using a set of 12 well-established tests, is impaired in people who reported typically sleeping less, or more, than 7-8 hours per night-which was roughly half the sample. Crucially, performance was not impaired evenly across all cognitive domains. Typical sleep duration had no bearing on short-term memory performance, unlike reasoning and verbal skills, which were impaired by too little, or too much, sleep. In terms of overall cognition, a self-reported typical sleep duration of 4 hours per night was equivalent to aging 8 years. Also, sleeping more than usual the night before testing (closer to the optimal amount) was associated with better performance, suggesting that a single night's sleep can benefit cognition. The relationship between sleep and cognition was invariant with respect to age, suggesting that the optimal amount of sleep is similar for all adult age groups, and that sleep-related impairments in cognition affect all ages equally. These findings have significant real-world implications, because many people, including those in positions of responsibility, operate on very little sleep and may suffer from impaired reasoning, problem-solving, and communications skills on a daily basis.

Figures

Figure 1.

Scatterplots and histograms of (A) self-reported typical sleep duration per night in the past month versus age at test; (B) the number of hours slept the night prior to testing versus age at test; and (C) typical sleep duration versus previous night's sleep in the past month. Scatter plots were convolved with a Gaussian kernel to illustrate a 2D estimate of the probability distribution function of the data. Regression lines and associated statistics are shown in each plot. Histograms of each variable and their probability density function (PDF) estimates are shown above and to the right of each plot.

Figure 2.

Overall test performance versus self-reported typical sleep duration per night in the past month, with the predicted overall performance for our population sample in light blue (with 95% confidence intervals). To better illustrate the density of data points in the sample, the color of each point is scaled by a kernel density estimate.

Figure 3.

Predicted score performance as a function of sleep duration, for (A) STM, (B) Reasoning, (C) Verbal, and (D) Overall scores, in units of standard deviations. Although STM did not show a significant relationship with typical sleep duration, it is included for comparison. Shaded regions on top and bottom of the curve indicate 95% confidence intervals of the prediction. Vertical dashed lines indicate the location of the curves’ maxima, with shaded 95% confidence intervals prediction (except for STM, where they could not be calculated due to a nonsignificant quadratic term).

Figure 4.

The difference from predicted peak performance for (A) STM, (B) Reasoning, (C) Verbal, and (D) Overall scores, in units of standard deviations. Shaded regions above and below the curve indicate 95% confidence intervals of the difference. Vertical dashed lines (sparse dash, on the left) indicate where the lower confidence bound crosses the horizontal black line (i.e. x = 0, no difference from peak performance). Peak location is also marked with vertical dashed lines.

Figure 5.

Predicted (A) Reasoning and (B) Overall performance as a function of “sleep delta”—the amount of sleep the night prior to testing minus the typical amount of sleep. Shaded regions above and below the curve depict 95% confidence intervals. The x-coordinate of the peak of the curve is indicated with a vertical line, and shaded regions to either side indicate 95% confidence intervals on the location of this optimum.