Do self-reported and device-measured sleep give different estimates of sleep duration's link to cardiovascular risk?

Why a questionnaire and a sleep monitor can tell different stories

A self-report asks what someone remembers about their sleep. A device — polysomnography in a sleep lab, an actigraph worn on the wrist, or a smartwatch — records movement or brain activity and estimates sleep from that. These are measuring different things, and they can disagree for ordinary reasons: people misjudge how long they were awake in bed, and devices can misclassify quiet wakefulness as sleep. In one community study, the correlation between objective and self-reported sleep duration was only weak [1]. A separate validation study found that smartwatch-recorded sleep differed from diary-reported sleep by roughly 7 to 21 minutes depending on the watch generation, with the oldest device showing a statistically significant gap and newer ones not [3]. That is a small average gap, but it shows the two sources are not interchangeable even when they land close together.

The practical consequence is that a study's answer depends partly on which clock it used. If self-reports systematically round sleep toward a familiar number, or if devices capture time in bed that the sleeper does not count as sleep, the same person can be classified into different sleep-duration bins by the two methods. That matters when the bins are narrow — the community study used categories as fine as 4–5 hours and 5–6 hours [1].

Both methods show risk rising at short sleep, but the curve's shape differs

In the community-based Sleep Heart Health Study, which followed about 5,000 adults for an average of 11 years, the pattern depended on the measure. With device-measured sleep, death rates from all causes and from cardiovascular disease fell steadily as sleep lengthened, and the strongest association was with sleeping 5 hours or less [1]. With self-reported sleep, the pattern was J-shaped: both short sleep (4 hours or less) and long sleep (more than 8 hours) on weekdays and weekends carried higher risk than 7–8 hours [1]. So the two methods agree that short sleep is risky, but only the self-report data flag long sleep as a problem in that cohort.

A much larger UK Biobank analysis of about 69,000 adults for cardiovascular disease and up to 77,000 for mortality found broadly similar dose-response curves across methods, with an inverse J-shape that was especially pronounced for device-based measures [2]. The lowest estimated risk fell at about 7.2 hours for self-reported sleep, 7.7 hours for device-measured sleep period time, and somewhere between 6.8 and 9.3 hours for device-measured total sleep time [2]. In other words, the optimal point shifts a little depending on the metric, but it stays in the 7–9 hour neighborhood.

Device measures can produce substantially stronger associations

The UK Biobank comparison gives a concrete sense of the size gap. Taking 5 hours as the reference, 7 hours of device-measured sleep period time was associated with a 30% lower risk of death, while 7 hours of device-measured total sleep time was associated with 17% lower risk, and 7 hours of self-reported sleep with 14% lower risk [2]. The confidence interval for the self-reported estimate reached 1.00, meaning it was compatible with no difference at all [2]. The authors concluded that associations were substantially stronger for sleep period time — the whole window from falling asleep to waking — than for the other metrics [2].

That does not mean self-report is useless. It means the device metric that includes time awake during the night may capture something the other measures smooth over. The same study suggests sleep period time could be the more relevant target for future device-based recommendations [2]. Which metric a study chooses is therefore part of the result, not a neutral technical detail.

What this means if you are trying to read your own risk

The consistent message across these studies is that 7–9 hours is the range where risk looks lowest, regardless of how sleep was measured [2]. Short sleep is the more consistently flagged problem: the community study found the strongest mortality associations at 5 hours or less of device-measured sleep [1], and a smaller study of 180 Pakistani adults found that short sleep was associated with increased coronary heart disease risk, with participants averaging 6.7 hours by self-report and 6.3 hours by actigraphy [4]. Self-reported short and long sleep have also been linked to specific cardiovascular conditions — heart failure, angina, heart attack, stroke — and to diabetes, with some differences between men and women in which conditions reached significance [5].

The honest summary is that the two methods mostly converge on the same advice but not on the same curve. Self-report is easier to collect and captures how people experience their sleep; devices give a more granular and often stronger signal, especially when they measure the full sleep period [2]. Neither is a direct readout of cardiovascular risk, and the weak correlation between them [1] means a person's questionnaire answer and their wearable's number can both be "true" while pointing at different estimates.

Sources used in this answer

Related research questions