Sleep and the Long Game

Sleep occupies a strange position in longevity. Everyone agrees it matters. The epidemiology is enormous. And the single most-quoted finding in that epidemiology is one that most people get backwards.

The U-shaped curve, and its awkward asymmetry

The foundational meta-analysis pooled 16 prospective studies containing 27 independent cohort samples, 1,382,999 participants, and 112,566 deaths, with follow-up from 4 to 25 years (Cappuccio et al., 2010).

The pooled relative risks for all-cause mortality:

• Short sleep: RR 1.12 (95% CI 1.06–1.18), P < 0.01
• Long sleep: RR 1.30 (95% CI 1.22–1.38), P < 0.0001

Long sleep carried a substantially higher mortality risk than short sleep. That result is reproduced constantly and misread constantly, because the intuitive interpretation — that sleeping ten hours is worse for you than sleeping five — is almost certainly wrong.

Long sleep is a classic marker of underlying illness. Undiagnosed cancer, heart failure, depression, sleep apnea and chronic inflammation all increase time in bed. In a cohort study, those people appear as long sleepers who die sooner. The sleep did not cause the death; the disease caused both.

The authors also found significant heterogeneity between studies for both associations, and every study measured sleep by questionnaire at a single baseline. Self-reported sleep duration correlates imperfectly with measured sleep.

A dose-response curve is not a causal curve. This one is the standard example.

Why short sleep is the more interesting half

Short sleep is harder to explain away, because researchers can measure it decades before the outcome.

The Whitehall II cohort followed 7,959 people for a mean of 24.6 years, recording 521 dementia cases (mean diagnosis age 77.1). Sleeping six hours or less, compared with seven (Sabia et al., 2021):

• At age 50: HR 1.22 (95% CI 1.01–1.48)
• At age 60: HR 1.37 (95% CI 1.10–1.72)
• At age 70: HR 1.24 (95% CI 0.98–1.57)

Persistent short sleep across all three ages versus persistent normal sleep: HR 1.30 (95% CI 1.00–1.69).

Long sleep showed no consistent association with dementia here — a useful contrast with the mortality data, and consistent with long sleep being an illness marker rather than a risk factor.

The authors address reverse causation directly. Dementia has a long preclinical phase, and disturbed sleep is one of its early features. Measuring sleep at specific ages decades before diagnosis mitigates this; it does not eliminate it. Notice which estimate lost significance: age 70, the one closest to diagnosis. That pattern is what you would expect if the associations are not purely artefacts of incipient disease — but it is suggestive, not decisive.

The same cohort examined multimorbidity, defined as accumulating two or more chronic diseases. Among participants free of multimorbidity at baseline, sleeping five hours or fewer carried (Sabia et al., 2022):

• At age 50: HR 1.30 (95% CI 1.12–1.50, P < 0.001), 7,864 participants, mean 22.6 years follow-up
• At age 60: HR 1.32 (1.13–1.55, P < 0.001)
• At age 70: HR 1.40 (1.16–1.68, P < 0.001)

And a finding that clarifies the mechanism: short sleep at age 50 was not associated with mortality among those who already had chronic disease. The authors read this as evidence that the sleep-mortality link operates through disease onset rather than disease progression.

That is a meaningfully different model of how sleep matters. It is not that poor sleep kills sick people faster. It is that poor sleep makes healthy people sick.

What might be happening during deep sleep

The most striking mechanistic finding of the past fifteen years came from mice.

Researchers measured the brain’s interstitial space — the fluid-filled gaps between cells — during sleep and wakefulness. Cortical interstitial volume fraction was 13–15% while awake and 22–24% during sleep or anesthesia, a roughly 60% increase. That expansion produced a striking increase in convective exchange between cerebrospinal fluid and interstitial fluid (Xie et al., 2013).

Functionally: beta-amyloid was cleared twofold faster in sleeping mice than in awake mice (n = 23–29, P < 0.05).

The authors proposed that the restorative function of sleep may be a consequence of enhanced removal of potentially neurotoxic waste that accumulates during waking.

Two caveats, stated plainly. This is a rodent study. And a plausible mechanism connecting short sleep to dementia is not evidence that short sleep causes dementia in humans — the mechanism makes the epidemiological association more credible without confirming it.

The number, and what it is not

Across these studies the reference category is seven hours. That is where risk is lowest in the mortality meta-analysis, and it is the comparator in both Whitehall II analyses.

But “seven hours” is a population mean, and it is time asleep, not time in bed. Sleep efficiency is rarely perfect, so seven hours of sleep typically requires somewhat more time in bed. And individual requirement varies.

What to do with this

• Protect the floor before optimizing anything else. Six hours or less at age 50 and 60 is where the dementia hazard ratios appear; five hours or less is where multimorbidity risk climbs. Getting off that floor is the intervention with evidence behind it.
• Do not try to shorten long sleep. The long-sleep mortality signal is very likely reverse causation. If you consistently need nine or more hours and still feel unrested, that is a reason to see a physician — for sleep apnea, thyroid function, depression, anaemia — not a reason to set an alarm.
• Treat sleep as a midlife investment. The Whitehall II design measured sleep at 50 and 60 and outcomes decades later. This is not a system where you can catch up at 70.
• Get apnea ruled out if you snore, wake unrefreshed, or have a large neck circumference. It is common, treatable, and it corrupts sleep quality independent of duration.
• Give the same weight to regularity as to duration. The persistent-short-sleep pattern across ages 50, 60 and 70 carried the standalone hazard ratio of 1.30. Chronic pattern, not occasional short nights.
• Hold the causality loosely. These are observational cohorts with self-reported sleep and acknowledged reverse-causation risk. The evidence is strong enough to act on because the intervention is free and low-risk, not because the case is closed.

The most defensible summary: sleep is not one of the levers where we have randomized evidence of life extension. It is one where a large, consistent body of prospective data, a credible mechanism, and a zero-cost intervention all point in the same direction. That combination is enough.