Why sleep has a U-shaped mortality curve, not a "more is better" one

Sleep is the factor in this model most likely to produce a result people want to argue with. Not because short sleep is scored as harmful — nearly everyone accepts that — but because the model also penalises long sleep, and that strikes many people as obviously wrong. The reasoning behind it is worth working through, because the answer involves one of the most instructive problems in observational epidemiology.

The evidence for the U-shape

The foundational reference is a 2010 meta-analysis by Cappuccio and colleagues in the journal Sleep, which pooled data from sixteen prospective studies covering more than a million participants. It found elevated all-cause mortality at both ends of the duration distribution: short sleep (typically defined as under six hours) and long sleep (typically over nine hours), relative to a reference of around seven hours.

That pattern has replicated widely since, in different countries, age groups, and study designs. The location of the trough is fairly stable across analyses, generally landing in the seven-to-eight-hour range for adults. The consistency of the finding is not really in dispute. What the two arms of the curve mean is a different question, and they don't mean the same thing.

Why short sleep is harmful

This half has a clear and well-supported mechanistic story, backed by experimental work rather than just observation. Controlled sleep-restriction studies — where healthy volunteers are deliberately kept short of sleep for days or weeks — produce measurable, reproducible changes:

• Impaired glucose metabolism. Even a few nights of restriction reduces insulin sensitivity measurably in healthy young adults.
• Elevated blood pressure and reduced overnight dipping, a pattern independently associated with cardiovascular risk.
• Raised inflammatory markers including CRP and IL-6.
• Impaired immune response, including reduced antibody response to vaccination.
• Cognitive and mood effects — degraded attention, emotional regulation, and reaction time, with accident risk consequences that are immediate rather than long-term.

Because these effects show up in randomised experimental designs, the causal claim is much stronger here than for most factors in this calculator. Short sleep doesn't merely correlate with poor outcomes; restricting sleep produces the intermediate physiological changes that lead to them.

Why long sleep shows up as a risk marker

The long-sleep arm is where the interesting problem lives, and the honest answer is that much of it is probably not causal.

The leading explanation is reverse causation. Long sleep duration is a symptom of a wide range of conditions: depression, chronic inflammatory disease, cancer, heart failure, sleep apnoea, chronic fatigue, and the effects of numerous medications. In an observational cohort, someone in the early undiagnosed stages of a serious illness may sleep ten hours a night because they are ill. When they die three years later, the data records a long sleeper who died — and the arithmetic produces an association that reverses the actual direction of causation.

Supporting this reading: studies that exclude early follow-up deaths, or that adjust more aggressively for baseline health status and depression, generally find the long-sleep association attenuates. It also tends to weaken more than the short-sleep association does under such adjustments.

But — and this is why the model doesn't simply discard it — the association often persists at reduced magnitude even after these adjustments. Several possible explanations remain live: residual confounding from undetected illness, long sleep as a marker of poor sleep quality requiring more time in bed for the same restorative effect, or genuine effects of extended sedentary time and circadian disruption. There is no consensus.

Regularity may matter as much as duration

A more recent line of research, and one the calculator does not currently capture, concerns sleep regularity — the consistency of your sleep and wake times, independent of how long you sleep.

Several analyses using wearable data have found that irregular sleep timing predicts mortality and cardiometabolic outcomes, in some cases more strongly than duration does. Someone averaging seven hours by sleeping five hours on weeknights and ten at weekends appears to be in a worse position than someone sleeping a consistent seven — despite identical averages.

The mechanistic account involves circadian misalignment. The body's clock coordinates hormone release, core temperature, and metabolic processes on an approximately 24-hour cycle, and it takes time to shift. Large swings in timing produce a state functionally similar to recurring mild jet lag, with the metabolic consequences that implies. The occupational literature on shift workers, who show elevated rates of cardiometabolic disease and certain cancers, points in the same direction.

We haven't added regularity as an input because self-reported schedule variability is unreliable and because the evidence base, while growing, is younger and thinner than the duration literature. But if you're deciding what to change, moving toward a consistent wake time is plausibly worth more than adding thirty minutes to your average.

The condition hiding in this factor

Worth stating explicitly, because it's common and widely undiagnosed: obstructive sleep apnoea involves repeated breathing interruptions during sleep, producing oxygen desaturation and fragmented sleep architecture. It's strongly associated with hypertension, atrial fibrillation, stroke, and type 2 diabetes, and estimates suggest a large share of cases are never diagnosed.

It's relevant here because it distorts this input in a specific way. Someone with untreated apnoea may report a perfectly normal eight hours and score neutrally on this factor while getting sleep of substantially degraded quality. The calculator cannot detect this. The classic signals are loud snoring, witnessed pauses in breathing, waking with a dry mouth or headache, and persistent daytime sleepiness despite adequate time in bed. If several of those apply, a sleep study is worth more than anything this calculator can tell you, because treatment is effective and the untreated condition is consequential.

Individual variation and what to do about it

Sleep need genuinely varies between people, with a heritable component. A small minority appear to function well on substantially less than seven hours; the more common situation is people who believe they belong to that group and don't, since chronic partial sleep restriction impairs the very self-assessment used to judge it.

The more useful test than counting hours is functional: do you wake without an alarm at roughly the same time, and are you alert through the day without relying on caffeine to stay functional? Reliably falling asleep within a few minutes of lying down, and needing several hours on weekends to catch up, are both signs of accumulated debt rather than efficiency.

How the model scores it

Two design decisions worth surfacing. The positive range here (+2) is smaller than exercise's (+3), because sleep's upside is largely about avoiding harm rather than an open-ended benefit — there's no evidence that engineering your way to more sleep within the healthy range yields further gains. And the long-sleep penalty is deliberately the smallest non-zero modifier anywhere in the model, reflecting our reading that the causal evidence there is the weakest of any relationship the calculator scores.