What is actually being claimed
The strong version, widely repeated, is that sleep is when the brain and body carry out essential maintenance, that this capacity degrades with age, and that the degradation drives neurodegeneration and cardiometabolic disease. It follows, on this account, that protecting sleep protects against those outcomes.
Parts of that are well supported. Sleep architecture does change with age, and the changes are measurable rather than inferred. Other parts are extrapolation from animal models or from associations whose direction is unresolved. This review separates them.
It also distinguishes normal age-related change from a sleep disorder. Sleeping less deeply at seventy than at twenty is not a disease. Obstructive sleep apnoea and chronic insomnia are conditions with diagnostic criteria and treatments, and they are the part of this field where clinical action is clearly indicated.[1]
What changes with age, and how it is measured
Sleep is not one state. Polysomnography, the reference method, distinguishes light non-REM sleep, deep slow wave sleep and REM sleep, using brain electrical activity, eye movement and muscle tone together. This is a laboratory measurement requiring electrodes and scoring, and it is the only method that measures stages rather than estimating them.
With age, several things change consistently. The proportion of slow wave sleep falls, in some people substantially. Sleep becomes more fragmented, with more brief arousals across the night. Circadian phase advances, so sleep and wake times drift earlier. Total sleep time falls modestly, though less than commonly assumed, and daytime napping increases.
The mechanistic hypothesis that has attracted most attention concerns clearance of metabolic waste from brain tissue during sleep, with the suggestion that this process is more active during deep sleep and that its failure contributes to protein accumulation in neurodegenerative disease. The foundational work here is in rodents. Human evidence is indirect and the magnitude and even the existence of the process at the scale claimed remains actively contested. It is a hypothesis worth following, not an established fact.
Measurement outside the laboratory is the field's practical weakness. Wearables and phone applications estimate stages from movement and heart rate variability. They track total sleep time and timing reasonably, and they estimate stage composition poorly against polysomnography. A large part of what the public believes about their own sleep architecture comes from devices that cannot reliably measure it.
What the human evidence shows
The association between self-reported sleep duration and mortality in large cohorts has a consistent and awkward shape. Both short and long sleep are associated with higher mortality, producing a U-shaped curve that has been reproduced across many populations.
The long sleep arm of that curve is almost certainly not what it appears. Long sleep duration is strongly associated with existing illness, depression, frailty and undiagnosed disease, all of which increase time in bed. It is best read as a marker of poor health rather than a cause of it, and this is one of the clearest examples of reverse causation in the whole of preventative medicine.
The short sleep arm is harder to dismiss but is not clean either. Short sleep clusters with shift work, deprivation, pain, stimulant use, caring responsibilities and untreated mental illness, and adjustment cannot remove those. Experimental sleep restriction studies in laboratories do show real physiological effects on glucose handling, appetite regulation and inflammatory markers over days, which supports a causal contribution, but those are short exposures measured with surrogate outcomes.
Sleep fragmentation and reduced slow wave sleep have been associated in cohort studies with cognitive decline and with markers of neurodegeneration. The direction here is genuinely ambiguous, because the pathology in question is known to disrupt sleep regulation early in its course, potentially years before diagnosis.
The treatment evidence is the most sobering part. Obstructive sleep apnoea is common, underdiagnosed and unquestionably associated with cardiovascular disease. Randomised trials of treating it have improved symptoms, daytime sleepiness and quality of life, which are genuine benefits, but have not consistently reduced cardiovascular events, with adherence to therapy a persistent complicating factor. For chronic insomnia, cognitive behavioural therapy is the recommended first line treatment in UK practice and has good randomised evidence for improving sleep itself.[1] Whether treating insomnia changes long-term disease outcomes has not been established.
The limitations that hold the grade at C
| Limitation | Why it matters for the grade |
|---|---|
| Reverse causation | Illness disrupts sleep, often years before it is diagnosed, which inflates every association. |
| Self-reported duration | People are poor estimators of their own sleep, and the error correlates with mood and health. |
| Consumer devices cannot stage sleep | The most widely used measurements are the least accurate for the variable of interest. |
| Neutral outcome trials | Treating sleep apnoea has not consistently improved cardiovascular endpoints in randomised trials. |
| Rodent-derived clearance hypothesis | The mechanism most often cited in public discussion is not established in humans at the claimed scale. |
| Confounding cluster | Short sleep travels with shift work, deprivation, pain and untreated mental illness. |
A practical implication follows. Anxiety about imperfect sleep, driven by device readouts that cannot measure what they claim to measure, is itself a well recognised route into insomnia. Treating a wearable score as a health outcome is a category error with a plausible mechanism for harm.
What would change the grade
Grade B would follow from randomised trials in which a sleep intervention with demonstrated adherence improved a pre-registered clinical endpoint, whether cardiovascular, cognitive or metabolic, over meaningful follow-up. Trials in sleep apnoea designed around the adherence problem are the most likely route.
Grade A would require replication of that result in an independent population.
Progress would also come from better ambulatory measurement. If sleep stages could be measured accurately outside a laboratory at cohort scale, the associations could be tested against the variable that actually matters rather than against self-reported hours in bed. See also our review of time-restricted eating, where circadian timing is the shared mechanism.