How much sleep you actually need, and how the range was set

Seven hours. Studies that tracked large groups of adults for years found the lowest risk of dying, and of heart attack and stroke, at about 7 hours of sleep a day, with risk climbing on both sides of it.¹ Each hour below that adds around 6% to the risk of death, each hour above about 13%.¹

It is a curve, not a rule

Plot risk against hours slept and you get a U. It falls as sleep rises toward seven hours, flattens across the bottom, then climbs again beyond it. The same shape turned up for every outcome the largest of these analyses looked at, and it barely differed between men and women.¹

The per-hour figures are worth reading in both directions. For death from any cause, risk rose by a factor of 1.06 for each hour below seven, and 1.13 for each hour above: 6% and 13%.¹ For stroke the lopsidedness is sharper, 1.05 per hour below against 1.18 above, so 5% each hour short and 18% each hour long.¹

A recommended range is a summary of that curve. A shallow dip near seven hours becomes, in translation, a bracket with hard edges.

One thing this evidence cannot tell you is how any official bracket was drawn. There is no guideline document here, so no panel, no criteria and no vote can be traced. What can be shown is the evidence such a panel would have been reading.

The long side of the curve rises faster

This is the part that surprises people, and it holds in every analysis here.

The earliest one followed adults for between 4 and 25 years and put the risk of death at 1.12 for short sleepers and 1.30 for long ones.² A later and much larger analysis of long sleep put death at 1.39, stroke at 1.46, cardiovascular disease at 1.25 and diabetes at 1.26.⁴ High blood pressure was the one outcome with no link at all, at 1.01.⁴

Its companion analysis of short sleep found death at 1.12, diabetes at 1.37, obesity at 1.38 and high blood pressure at 1.17.³ So the headline comparison is 1.39 against 1.12: a 39% higher death rate on the long side, 12% on the short one.^3,4

Higher death rate against people sleeping about 7 hours% higher risk

Long sleep39%

Short sleep12%

That companion analysis also found something the headlines usually drop. The relationship with death only became a straight line below six hours, and for every other outcome no dose-response showed up at all.³ Short sleep, in other words, does its damage at the extreme rather than steadily across the range.

A review that gathered the earlier reviews summarized both slopes. Among short sleepers, an hour less per day went with 3% to 11% more risk across death, coronary heart disease, osteoporosis, stroke and type 2 diabetes.⁵ Among long sleepers, an hour more went with 7% to 17%.⁵

Why the long-sleep number is the weaker evidence

Sleeping eleven hours is not really a behavior. It is often a symptom, of illness, of depression, of something not yet diagnosed. A study that measures sleep once at the start and then counts deaths for twenty years cannot tell a cause from a warning sign.

One review went looking for the mechanism and found a seam instead. In the observational half, the extreme of long sleep went with higher C-reactive protein, a blood marker that rises when the body is inflamed, and with higher interleukin-6, a signaling molecule that drives that response.⁷ Shorter sleep went with higher C-reactive protein but not interleukin-6, and the extreme of short sleep showed no link to C-reactive protein at all.⁷

Then the same review turned to the experiments, where sleep was taken away on purpose. Neither depriving people of sleep entirely nor restricting it moved C-reactive protein, interleukin-6 or TNF-alpha.⁷

That is one review contradicting itself across its two halves, on the same markers. It does not mean sleep length is irrelevant. It does mean the inflammation story usually told to explain the curve did not reproduce when anyone manipulated sleep deliberately.

What the restriction experiments do show

The laboratory work answers a narrower question, and answers it well: what happens to performance when sleep is cut.

The trial here put 70 healthy adults through two 5-day stretches restricted to 4 hours a day, with a single night between them.⁶ That middle night was the variable. Participants were assigned to sleep opportunities ranging from none at all up to 12 hours, and one group was never restricted.⁶

Two findings matter. Impairment on tests of attention and reaction time built up across both restriction weeks instead of leveling off.⁶ And the recovery night barely helped. Its effect was slight and brief, and whether it offered more sleep, less or none, performance through the second week was governed by the sleep the participant had banked beforehand.⁶

That is the sharpest available answer to anyone who says they run fine on five hours. The deficit accumulates, one long lie-in does not clear it, and the history you arrive with sets the level you perform at.

It is not, however, a measurement of how much sleep a given person needs. The trial assigned recovery doses. It never estimated how much need varies from one person to the next.

What this does not settle

How wide that variation is. Nothing in this evidence measures it. The restriction trial found that sleep history dominated the response, which is a different claim from everyone requiring the same number of hours.

Whether long sleep causes any of what it predicts. The authors of the long-sleep analysis closed by asking exactly that, and called for studies to find out whether the relationship is causal and whether anything can be done about it.⁴

Whether short sleep's association is causal either. The straight-line relationship with death appeared only below six hours, and no other outcome showed a dose-response at all.³

How much of this is length rather than quality. Sleep was recorded by questionnaire in these studies, and hours a person reports lying in bed are not hours of measured sleep.²

And where any official bracket came from. Answering that needs a guideline document, and there is none in this evidence.