In 1.4 Million People, Long Sleep Looked Worse Than Short Sleep

A 2010 meta-analysis in SLEEP pooled 27 prospective cohorts covering 1,382,999 people and 112,566 deaths, then plotted habitual sleep duration against all-cause mortality.

The result is a U-shaped curve. Short sleepers showed roughly 12% higher mortality risk. Long sleepers showed roughly 30% higher.

The immediate reading — that nine hours of sleep is more dangerous than five — is almost certainly wrong, and the reason it is wrong is more instructive than the finding itself.

Consider what makes people sleep more. Undiagnosed cancer lengthens sleep, often well before diagnosis. So does heart failure, depression, chronic infection, frailty, and the medications used to treat all of them. In a cohort study, someone in the early stages of an undetected illness gets recorded as a long sleeper at baseline, then dies during follow-up, and the analysis dutifully attributes that death to their sleep duration.

This is reverse causation, and long sleep is close to a textbook case. The exposure is downstream of the outcome. Most of these studies adjust for self-reported health status at baseline, but self-report cannot capture disease that has not yet been diagnosed — which is exactly the disease doing the damage.

The short-sleep end of the curve is more likely to reflect genuine causation, because a plausible forward mechanism exists and is supported by experimental work: restricted sleep impairs glucose tolerance, raises blood pressure, disrupts appetite regulation and degrades immune response, and these effects appear in controlled laboratory studies where sleep is deliberately manipulated rather than merely observed.

So the honest interpretation is that the two ends of the same curve mean different things. One end is probably a cause. The other is probably a warning light — a signal that something else is wrong, which is useful information but not something you fix by sleeping less.

Two further limitations apply throughout. Sleep duration in these cohorts is almost entirely self-reported, and people are unreliable narrators of their own sleep, typically overestimating time asleep and misjudging quality. And duration is a crude summary of something multidimensional: timing, continuity, stage architecture and regularity all matter, and none are captured by a single number of hours.

The practical upshot is narrower than the headline suggests. Consistently short sleep is worth addressing. A consistent, unexplained increase in sleep need — particularly a change from your own baseline, in the absence of an obvious cause like training load or illness recovery — is worth mentioning to a doctor, not because the sleep is harming you but because of what it may be indicating.

Why this matters for tracking: the value of sleep tracking here is in the trend, not the absolute number. A stable eight hours means very little on its own; a drift from seven to nine over a few months, unexplained, is the pattern this literature suggests is worth noticing. Log the change, not the value.