Part 1 of 3 · Next: AI is the new microscope · The future of brain medicine
My interest in sleep
I started studying neuroscience to learn what the brain could teach us about building better AI systems. At first I was mostly interested in cognition. Working memory, attention, reasoning and executive function seemed like obvious places to borrow from. Later I got curious about affect and mood. Did they have a role to play in AI, or were they legacy routines from evolution that we could do without? The chapters on sleep I usually skipped. After all, the action was in the waking brain, or so I thought.
That changed when I went through bouts of burnout and depression. I couldn't fall asleep, and when I did, I woke several times a night and lay awake. Clinicians call this onset and maintenance insomnia. The worse I slept, the worse my brain worked. Sleep, I realised, was critical for the brain. Duh.
I went back to the chapters I'd skipped, read books like Why We Sleep1 and listened to every Huberman episode and sleep podcast I could find. I got obsessed with my own sleep and tried various supplements, behavioral protocols and sleep gadgets to steer it in the right direction. Fixing my sleep fixed my brain. Since then I've worked with leading sleep scientists and interviewed researchers and clinicians in neurology and psychiatry. Again and again, they pointed me to the link between sleep and various brain disorders.
Along the way I saw the link back to AI. Memory consolidation and dreaming happen during sleep, and maybe AI systems will need to sleep too. But the bigger idea ran the other way. Instead of using the brain to build AI, I could point AI at sleep and use it to understand and improve brain health.
This three-part series asks three questions. What can sleep tell us about brain health? What applications is AI unlocking, and what is holding it back? What does this mean for the future of brain medicine?
I'm not a scientist or a clinician. I'm a founder building in this space, so these are my views, but I try to stay as close to the science as I can.
Sleep and brain health
A clinical sleep study, polysomnography (PSG), records various biosignals through a single night: brain activity, eye movements, muscle tone, leg movements, heart rhythm, breathing and blood oxygen.
It is the standard test for sleep apnea, where the airways keep closing during sleep. Apnea is common. A modelled estimate puts 936 million adults aged 30 to 69 worldwide at the lab threshold, 425 million of them in moderate or severe form2. Simpler home tests now handle many straightforward cases3. Together with a daytime nap test, a sleep study is also how doctors diagnose narcolepsy4. It is also required to confirm REM sleep behaviour disorder (RBD), where people act out their dreams because REM sleep loses its normal muscle paralysis5. It is the main test for other parasomnias and for periodic limb movements too, though not for insomnia, the most common sleep complaint.
Some of these disorders reach far beyond sleep, and RBD is the clearest example. In 1,280 people with isolated RBD followed at 24 specialist centres, about 6% a year went on to parkinsonism or dementia, roughly three in four within 12 years by projection, as Parkinson's disease, Lewy body dementia or multiple system atrophy6. A night in the lab confirms RBD, and that confirmation can come years before the movement or memory problems that lead to the brain disease being diagnosed.
The physicians and researchers I spoke to kept telling me that sleep medicine reaches much further than sleep disorders, and I wanted to see how far. Sleep health has several measurable dimensions7. Building on that, I broke a night into 17 components, from how long you sleep to the fine rhythms in the brain's signal. Then I had AI agents search PubMed, which returned 110,682 articles linking sleep to brain, mental and body conditions, and pull the strongest evidence for each component and condition. Figure 1 shows the result.
The full map, linking every component to every condition, needs a larger screen.
Sleep fingerprint
Dementia
Years before diagnosis, the sleep EEG of people who go on to develop dementia tends to look older than their age, and their 24-hour activity rhythm is flatter. Apnea and unusually short or long sleep also carry risk, but these are shared with many conditions, and no single sleep-stage percentage has held up across cohorts.
Early warningWhy
In long-term community studies, each 10 years by which the sleep EEG looked older than its owner carried 39% higher dementia risk, and a flatter daily activity rhythm about 1.5 times the risk, while standard sleep-stage measures pooled across five cohorts showed no link. A genetic analysis using UK Biobank data found no sign that sleep traits cause dementia, and the observed links weakened once heart and blood-vessel risk factors were accounted for, so much of the signal is probably early disease or shared risk. No trial has shown that improving sleep prevents dementia.
Sleep fingerprint
Alzheimer's
In Alzheimer's disease there is less deep sleep and less REM, more time awake and a later first REM period, and the loss of deep sleep becomes clear only once dementia has set in. In healthy adults, disrupting slow waves for one night raises amyloid in spinal fluid, and in early Alzheimer's, weaker spindle activity over the temporal lobes goes with faster cognitive decline. Broken sleep and apnea are shared with many conditions.
Both waysWhy
Long-term studies link diagnosed sleep apnea to about 45% higher risk of Alzheimer's and insomnia to about 49% higher. In short experiments in healthy adults, the more slow waves were disrupted in one night, the more amyloid rose in spinal fluid, and one dose of a sleep drug lowered it. Genetic studies do not back a causal role, since an insomnia signal largely disappeared once education and cholesterol were accounted for. In the other direction, people with Alzheimer's had fewer than half as many neurons in a sleep-promoting part of the hypothalamus at autopsy, and people with fewer of these neurons had more broken sleep before death. The sleep-to-disease direction is shown only for biomarkers over hours, and no trial has shown that treating sleep slows Alzheimer's.
Sleep fingerprint
Mild cognitive impairment
In mild cognitive impairment the night is already shorter and more broken, with about half an hour less sleep, 20 minutes more time awake and slightly less REM. Deep sleep is still largely preserved at this stage, so most of the signal is the shared core of broken sleep, and apnea carries about 1.65 times the odds of impairment.
Early warningWhy
Long-term studies show that apnea and a flatter 24-hour activity rhythm come before mild cognitive impairment. Older women with an apnea-hypopnea index of 15 or more had about 1.85 times the odds of developing impairment or dementia within five years. Evidence that sleep causes the impairment is thin, and most of the apnea link comes from studies measured at a single point in time. Trials are small; across 10 trials, melatonin added about two points on a 30-point cognitive screen, short of a definitive effect.
Sleep fingerprint
Parkinson's
The defining sign is REM sleep behaviour disorder: the muscle paralysis that normally comes with dreaming is lost, and people act out their dreams, often years before tremor or slowness. When it occurs on its own, not explained by medication or narcolepsy, and is confirmed on video-polysomnography, it is close to specific for the alpha-synuclein diseases (Parkinson's, Lewy body dementia and multiple system atrophy). The rest of the Parkinson's night, with less deep sleep and REM, more waking and more apnea and limb movements, is largely shared.
Early warningWhy
Long-term studies of people with isolated REM sleep behaviour disorder confirmed on polysomnography show that about 6% a year develop parkinsonism or dementia, roughly three in four within 12 years. Most of them already carry misfolded alpha-synuclein in spinal fluid or skin, so the sleep sign is the disease showing itself early rather than a cause. In a placebo-controlled trial of treating apnea in Parkinson's, positive airway pressure improved a 30-point cognitive screen by about one point, which missed significance in its main analysis.
Sleep fingerprint
Lewy body dementia
Lewy body dementia shares Parkinson's hallmark, dream enactment with loss of REM muscle paralysis, which often starts years before memory problems. Its other distinctive sign is a slowed EEG, seen mostly over the back of the head while awake, with delta waves in REM in some patients; apnea and limb movements are common but about as common in Parkinson's without dementia.
Early warningWhy
In people with dementia followed to autopsy, a history of dream enactment raised the odds of confirmed Lewy body disease about six-fold, and the 2017 diagnostic criteria made it a core feature, with REM sleep without paralysis on polysomnography counting as a biomarker. Long-term studies of REM sleep behaviour disorder show that about a quarter of those who convert develop Lewy body dementia, so the sleep sign comes first. No trial shows that treating sleep changes the course of the disease, and the 2017 criteria suggest clonazepam or melatonin only to reduce injuries from the behaviours.
Sleep fingerprint
Stroke
After a stroke, sleep apnea is the dominant finding, with about 71% having an apnea-hypopnea index above 5 and 30% above 30, at similar rates in the first month and after three months. Before a stroke the warning signs are general rather than specific, such as unusually short or long sleep.
Both waysWhy
Long-term studies link apnea to later stroke, most clearly in men, where the highest apnea quartile carried almost three times the risk, and genetic analysis supports insomnia as a cause, with about 31% higher odds. The largest CPAP trial, in people who already had heart or brain-vessel disease, did not reduce strokes, though nightly use averaged only 3.3 hours. Part of the breathing problem appears to follow the stroke itself. In one cohort followed from the first days after a stroke, central apneas and Cheyne-Stokes breathing fell by three months while obstructive apneas did not, which suggests the obstructive part was mostly there before. Pooled small trials after stroke suggest that treating apnea lowers the odds of a repeat event.
Sleep fingerprint
Epilepsy
In focal epilepsy, deep non-REM sleep with its large slow waves is when epileptic discharges are most easily triggered, while REM sleep is the most protective state across epilepsies. The night is also less efficient and less stable, with focal epilepsy losing some REM, and the odds of obstructive apnea are about twice those of healthy people.
Both waysWhy
In 209 newly treated patients, self-reported short sleep went with about 2.3 times the risk of a seizure recurrence within a year. In the other direction, sleep efficiency is already low before any anti-seizure drug, which points to the epilepsy itself disturbing sleep. Trials are few and small, and the seizure benefit of CPAP comes only from observational comparisons.
Sleep fingerprint
Multiple sclerosis
In MS, pooled studies find the night about 38 minutes shorter and 6 points less efficient, though sleep-lab studies alone find no difference in length, with more waking and more periodic limb movements, and restless legs syndrome is several times as common as in people without MS. Apnea is frequent, but its excess over controls is small, and worse sleep goes with greater disability.
ConsequenceWhy
The changes are measured in people who already have MS, and worse sleep and more apnea go with greater disability, which fits the disease disturbing sleep, although age and inactivity could explain part of that link. The one signal in the other direction comes from adults with MS recalling short sleep as teenagers, about 1.4 times the odds, a result that recall bias could produce. No trial has tested whether treating sleep changes relapses or disability.
Sleep fingerprint
Migraine
People with migraine have slightly less REM sleep, and restless legs are more common; children with migraine also sleep less and wake more. About half of patients have attacks on a daily pattern, with few attacks between 11 pm and 7 am. The restless legs link and attack timing are the more distinctive parts.
Both waysWhy
Long-term studies show that sleep problems come before migraine. In a Norwegian population, people with insomnia had about twice the risk of migraine 11 years later, and US health records link a sleep apnea diagnosis to about 1.85 times the risk of a new migraine diagnosis. The same Norwegian cohort shows the reverse as well, with migraine at baseline carrying about 1.9 times the odds of insomnia 11 years later, though other headaches did nearly as much. Genetic analysis gives only weak, preliminary support for apnea as a cause, and only for migraine with aura. Trials of sleep treatment are few and small, and whether treating sleep prevents attacks is not settled.
Sleep fingerprint
Brain injury
After a brain injury, sleep becomes broken and irregular, with more waking, lower efficiency and larger night-to-night swings, and many people need more sleep than before without realising it. Chronic moderate-to-severe injury shows more deep sleep rather than less. In mild injury, objective sleep is often close to normal even when people report poor sleep.
ConsequenceWhy
The changes follow the injury, and people with mild injury keep reporting worse sleep for more than a year even when recordings look normal. Because an injury is an event rather than an inherited trait, genetic studies cannot easily test the question, and long-term studies mostly ask how sleep after injury affects recovery. Trials treat the consequence; online CBT for insomnia lowered insomnia scores by 3.5 points more than sleep education in veterans with mild injury, and no trial yet shows better brain recovery.
Sleep fingerprint
Depression
A depressed night takes longer to start and breaks up more often. That broken sleep is shared with most psychiatric conditions. REM that arrives early and carries more eye movements leans toward depression, but it is not unique to it and shrinks in the best-controlled studies.
Both waysWhy
Long-term studies find that people with insomnia are more than twice as likely to become depressed later, and genetic studies point both ways, with insomnia-linked genes raising depression risk and depression-linked genes raising insomnia risk, though one genetic study found neither. The strongest evidence comes from trials. Treating insomnia with CBT-I cut new depression roughly in half in adults over 60 (12% against 26% over three years) and by about 40% in young people (10% against 18% over a year), but each result rests on a single trial, and only about half of the older adults completed three-year follow-up.
Sleep fingerprint
Bipolar disorder
In mania sleep shortens sharply and REM comes early, while in bipolar depression the deep slow-wave sleep thins. Between episodes people tend to sleep longer and less regularly than others, and that, with the swing between phases, is the more bipolar-specific part; short sleep and early REM are shared with depression.
Early warningWhy
Sleep loss is the early sign of mania that patients recall most often (a median of 77% across 17 studies), and in 250 young people seeking mental health care, longer sleep on a wrist sensor roughly doubled the risk of progressing to bipolar or psychotic disorder. One genetic study found that insomnia-linked genes raise bipolar risk, with no sign of the reverse. The only trial is a 58-person pilot in which bipolar-adapted CBT-I cut manic relapses (5% against 32% over six months), so whether protecting sleep prevents episodes is untested at scale.
Sleep fingerprint
Anxiety
An anxious night is somewhat shorter and less efficient than normal, and irregular sleep timing goes with more anxiety. Nothing on the overnight recording is specific to anxiety; the loss of deep sleep once credited to anxiety comes mainly from PTSD.
Both waysWhy
Experiments give the clearest arrow. Cutting or breaking up sleep in healthy people raised anxiety by a medium amount across 154 studies, and trials that improved sleep lowered anxiety by a similar amount, with bigger sleep gains bringing bigger mental health gains overall. Long-term studies agree, with insomnia roughly tripling the odds of a later anxiety disorder, though that estimate is imprecise. The reverse arrow, anxiety disturbing sleep, is widely assumed in clinics but less directly tested, and no clean genetic study of anxiety alone was found.
Sleep fingerprint
PTSD
A PTSD night is shorter and more broken, with less deep slow-wave sleep and, in some studies, more eye movements during REM. Broken, shallow sleep is shared with depression and anxiety. The REM changes lean towards PTSD, though studies disagree on them.
Both waysWhy
Sleep before trauma predicts who does worse. In 1,059 soldiers, each extra hour of self-reported sleep before deployment went with about a fifth lower odds of PTSD symptoms worsening afterwards. Genetic studies find that insomnia-linked variants raise PTSD risk by about a third, with a much smaller effect in the reverse direction. In people who already have PTSD, small trials show CBT-I easing PTSD symptoms right after treatment, but no trial in this set tests whether treating sleep soon after trauma prevents PTSD.
Sleep fingerprint
Schizophrenia
The most distinctive feature is a shortage of sleep spindles, the brief 11-16 Hz bursts of light N2 sleep, on top of longer time to fall asleep and more time awake. The spindle deficit persists after accounting for medication, but its size varies widely between patients. On wrist sensors, patients spend about two extra hours in bed, mostly in studies of people taking antipsychotics.
UnclearWhy
In UK Biobank, an unhealthy self-reported sleep pattern (long sleep, evening preference, napping, trouble getting up) went with about 75% higher risk of a later schizophrenia diagnosis, but only 231 people developed it, and early illness may have changed sleep first. Treating insomnia slightly reduced paranoia and hallucinations in 3,755 students without a psychotic disorder. In people with schizophrenia, CBT-I improves insomnia without a clear effect on overall symptoms.
Sleep fingerprint
ADHD
In adults, lab recordings look close to normal, and the differences show up on wrist sensors and questionnaires. In children, pooled studies find a few minutes' longer time to fall asleep and more frequent awakenings. A later body clock is the most ADHD-leaning feature; the loss of continuity is shared with many other conditions.
Both waysWhy
Genetic studies disagree. Two find that insomnia-linked variants raise ADHD risk, while others find no effect or point the other way, from ADHD to broken sleep. In a trial in 244 children, a brief behavioural sleep programme modestly reduced ADHD symptoms, with a third to a half of the gain running through better sleep. That is one trial with unblinded parent ratings, so this verdict is provisional.
Sleep fingerprint
Autism
Autistic children take longer to fall asleep and sleep less, with more time in light N1 sleep. One large clinical study also found slightly less REM that starts later, though a pooled analysis found no REM difference. The continuity loss is shared with other conditions. Later, smaller REM would run opposite to the early REM of mood disorders, making it the more autism-leaning feature if it holds up.
Early warningWhy
Sleep problems show up before diagnosis. In an Australian birth cohort, about an hour more night sleep at 6 months went with 22% lower odds of an autism diagnosis by age 11.5, though sleep was parent-reported and shared causes are likely. Prolonged-release melatonin adds about half an hour of sleep a night over placebo, but no trial here shows that better sleep changes autism itself.
Sleep fingerprint
Alcohol use
Two to four weeks into abstinence from alcohol, the night is somewhat shorter, with less slow-wave sleep and REM that arrives earlier and takes a slightly larger share of the night. Other substances look different and are far less studied. Early, heavy REM is shared with depression, and in a small 1994 study of veterans it also predicted who would relapse.
Both waysWhy
Genetic studies find that insomnia-linked variants raise the risk of alcohol dependence and of starting cannabis. In the other direction, laboratory studies in healthy adults show that alcohol before bed delays and shortens REM sleep from about two drinks, with larger effects at higher doses, though total sleep and time awake did not change on average. Genetic studies find no sign that drinking or cannabis causes insomnia, so the alcohol-to-sleep arrow is shown for the night's architecture rather than for insomnia as a disorder, while smoking and insomnia seem to drive each other. CBT-I reduces insomnia in people who drink heavily, by about 5.5 points on a 28-point scale, but has not yet been shown to change drinking.
Sleep fingerprint
High blood pressure
Breathing pauses and a low share of deep slow-wave sleep. In people who complain of insomnia, the risk rises only when a sleep study confirms under six hours of sleep.
CauseWhy
Long-term studies tie breathing pauses and low deep sleep to new high blood pressure, and insomnia adds risk only when measured sleep is short. Trials give the firmest answer. Treating apnea with CPAP lowers night-time systolic pressure by about 3.5 mmHg, more in people whose pressure was not controlled, so the effect is real but small. Genetic studies back insomnia as a cause.
Sleep fingerprint
Cardiovascular disease
Deep oxygen dips with each breathing pause (the hypoxic burden) and large heart-rate surges after each pause. Insomnia with short measured sleep also adds risk.
UnclearWhy
Long-term studies are consistent. Sleep apnea goes with about 1.8 times the risk of cardiovascular events, and about 2.5 times when severe. Yet trials of CPAP in people who already had heart disease found no overall benefit, and the lower risk in people who used it four or more hours a night is not a randomized comparison. A 2026 re-analysis of those trials suggests CPAP helps people with large oxygen dips or heart-rate surges and may harm low-risk people, which still needs a trial designed to test it.
Sleep fingerprint
Atrial fibrillation
Oxygen drops during breathing pauses and more time awake after falling asleep. Short, broken sleep is shared with high blood pressure and heart disease.
Both waysWhy
Large sleep-clinic cohorts tie oxygen-heavy apnea and more time awake in the night to new atrial fibrillation, and a recorded insomnia diagnosis goes with 14% higher risk across 1.8 million people. Genetic analysis finds no effect of short sleep, and a CPAP trial did not cut time spent in the rhythm. Daily pacemaker data show each one preceding the other within a day, so the link likely runs both ways, though that evidence covers only the short term.
Sleep fingerprint
Type 2 diabetes
Less deep slow-wave sleep, weaker coupling between slow waves and sleep spindles, and irregular bed and wake times. The deep-sleep link to blood sugar is the most diabetes-specific part; apnea and irregular timing are shared with heart disease.
CauseWhy
In a lab experiment, suppressing deep sleep without shortening it lowered insulin sensitivity in healthy young adults, and genetic studies name insomnia as a risk factor for type 2 diabetes. Long-term studies add irregular sleep and severe apnea. Treating apnea is the weak link. Across 14 trials CPAP did not lower long-term blood sugar, so the causal case rests on insomnia and lost deep sleep rather than on apnea.
Sleep fingerprint
Obesity
Short sleep, with deep and REM sleep falling over the years as weight rises. In obesity, extra weight drives much of the apnea, and apnea may add a little weight back.
Both waysWhy
Genetic studies show that excess weight causes apnea, and a year-long trial of a weight-loss drug cut breathing pauses by about 20 an hour more than placebo. Genetic studies also hint that apnea adds a little weight back. For sleep length, a trial in short sleepers with overweight found that sleeping about an hour longer cut daily intake by 270 kcal.
Sleep fingerprint
Kidney disease
Sleep apnea and short, broken sleep. Restless legs are common once kidney disease is present, and the two-way link with apnea is the most kidney-specific part.
Both waysWhy
Long-term studies show apnea coming before new kidney disease and kidney disease coming before new apnea. Genetic studies support short sleep and insomnia as causes of kidney disease, find no clear effect of apnea on kidney disease or filtration rate, and find some effect of poor kidney function on apnea. The four trials of CPAP did not slow the decline in kidney function.
Sleep fingerprint
Death, any cause
Less REM sleep, and sleep that is too short or too long. Many features of the night predict death, so this reads as a general marker of frailty rather than one specific pattern.
Early warningWhy
Long-term studies are large and consistent. Across 79 cohorts, short sleep went with 14% and long sleep with 34% higher mortality, and less REM sleep predicts death in two separate cohorts. Genetic studies in UK Biobank found no causal effect of sleep length or insomnia on death. Trials of CPAP show no effect on deaths overall, and the 37% lower mortality in a large 2025 review rests mostly on non-randomized studies and was funded by a CPAP maker.
Sleep fingerprint
Inflammation
Sleep apnea, and several nights in a row of short sleep, each with higher inflammation markers in the blood. The links with short or disturbed sleep are small, and sleep stages themselves are barely tested against inflammation.
Both waysWhy
Experiments show cause in one direction. Three or more nights of sleep cut to about four hours raised blood inflammation markers, while a single night did not. Treating apnea is less clear, since trials against a sham device have not shown lower markers even though trials against other controls show modest falls. Genetic studies find small effects of inflammatory proteins on sleep timing and napping, and some effects the other way, so the arrow may also run from inflammation to sleep. Body weight likely accounts for part of the apnea link.
Brain activity
Deep sleep
Linked to 18 of 26 conditions, 3 with strong evidence (A or B). Choose one to see its fingerprint.
- B Zhang 2022 · doi
- B Zhang 2020 · doi
- B Zhang 2019 · doi
- C Baglioni 2016 · doi
- C Mantua 2018 · doi
- C Henckaerts 2025 · doi
- C Barbato 2026 · doi
- C Chan 2016 · doi
- C Oliva 2026 · doi
- C Berger 2022 · doi
- C Tasali 2008 · doi
- C Reither 2021 · doi
- C Kristjansson 2025 · doi
- D Himali 2023 · doi
- D Yeh 2022 · doi
- D Juginovic 2025 · doi
- D Kwon 2015 · doi
- D Chen 2022 · doi
Brain activity
Spindles
Linked to 8 of 26 conditions, 1 with strong evidence (A or B). Choose one to see its fingerprint.
Brain activity
EEG slowing
Linked to 6 of 26 conditions, 1 with strong evidence (A or B). Choose one to see its fingerprint.
Brain activity
REM amount
Linked to 16 of 26 conditions, 4 with strong evidence (A or B). Choose one to see its fingerprint.
Brain activity
REM latency
Linked to 11 of 26 conditions, 4 with strong evidence (A or B). Choose one to see its fingerprint.
Brain activity
REM density
Linked to 6 of 26 conditions, 0 with strong evidence (A or B). Choose one to see its fingerprint.
REM paralysis
REM without atonia
Linked to 4 of 26 conditions, 2 with strong evidence (A or B). Choose one to see its fingerprint.
Breathing
Apnea events
Linked to 24 of 26 conditions, 11 with strong evidence (A or B). Choose one to see its fingerprint.
- A Benning 2025 · doi
- B Li 2024 · doi
- B Ungvari 2025 · doi
- B Ren 2025 · doi
- B Maggi 2024 · doi
- B Seiler 2019 · doi
- B Craciun 2025 · doi
- B Peppard 2000 · doi
- B Koh 2024 · doi
- B Punjabi 2009 · doi
- B Nadeem 2013 · doi
- C Lin 2016 · doi
- C Toubasi 2026 · doi
- C Chen 2025 · doi
- C Kao 2025 · doi
- C Stubbs 2016 · doi
- C Kao 2025 · doi
- C Zhang 2017 · doi
- C Sedky 2013 · doi
- C Simou 2018 · doi
- C Zhang 2022 · doi
- C Wang 2025 · doi
- D Terzaghi 2013 · doi
- D Stubbs 2016 · doi
Breathing
Oxygen drops
Linked to 7 of 26 conditions, 0 with strong evidence (A or B). Choose one to see its fingerprint.
Continuity
Time to fall asleep
Linked to 9 of 26 conditions, 6 with strong evidence (A or B). Choose one to see its fingerprint.
Continuity
Waking in the night
Linked to 16 of 26 conditions, 10 with strong evidence (A or B). Choose one to see its fingerprint.
- B Zhang 2022 · doi
- B Wei 2024 · doi
- B Zhang 2020 · doi
- B Yeh 2022 · doi
- B Toubasi 2026 · doi
- B Grima 2016 · doi
- B Ho 2024 · doi
- B Baglioni 2016 · doi
- B Baglioni 2016 · doi
- B Chan 2016 · doi
- C Baglioni 2016 · doi
- C Stanyer 2021 · doi
- C Xian 2025 · doi
- C Baglioni 2016 · doi
- C Heinzinger 2025 · doi
- D Sanders 2024 · doi
Continuity
Arousals
Linked to 9 of 26 conditions, 3 with strong evidence (A or B). Choose one to see its fingerprint.
Duration
Short sleep
Linked to 25 of 26 conditions, 15 with strong evidence (A or B). Choose one to see its fingerprint.
- B Oye-Somefun 2026 · doi
- B Zhang 2022 · doi
- B Wei 2024 · doi
- B Zhang 2020 · doi
- B Wang 2021 · doi
- B Grima 2016 · doi
- B Zhai 2015 · doi
- B Chan 2016 · doi
- B Chen 2020 · doi
- B Oliva 2026 · doi
- B Dai 2024 · doi
- B Tasali 2022 · doi
- B Ricardo 2017 · doi
- B Ungvari 2025 · doi
- B Smagula 2016 · doi
- C Winsor 2021 · doi
- C Toubasi 2026 · doi
- C Stanyer 2021 · doi
- C Barbato 2026 · doi
- C Baglioni 2016 · doi
- C Zhang 2019 · doi
- C Xian 2025 · doi
- C Pejovic 2026 · doi
- C Heinzinger 2025 · doi
- C Johnson 2021 · doi
Duration
Long sleep
Linked to 8 of 26 conditions, 4 with strong evidence (A or B). Choose one to see its fingerprint.
Timing
Regularity
Linked to 13 of 26 conditions, 3 with strong evidence (A or B). Choose one to see its fingerprint.
Timing
Circadian rhythm
Linked to 18 of 26 conditions, 1 with strong evidence (A or B). Choose one to see its fingerprint.
- B Daghlas 2021 · doi
- C Yang 2026 · doi
- C Xiao 2022 · doi
- C Leng 2020 · doi
- C Lyall 2018 · doi
- C Cheung 2023 · doi
- C Coogan 2017 · doi
- C Rossignol 2011 · doi
- C Feng 2023 · doi
- C Liu 2025 · doi
- C Zhang 2022 · doi
- C Feng 2023 · doi
- C Xiao 2022 · doi
- D Yang 2026 · doi
- D Gao 2023 · doi
- D Zalai 2020 · doi
- D Zheng 2024 · doi
- D Chen 2022 · doi
Key findings for Dementia 8 papers
- BEEG slowing Across five community cohorts, each 10 years by which the sleep EEG looked older than the person's age went with 39% higher risk of later dementia, and the link held after accounting for apnea and the APOE e4 gene. HR 1.39 (95% CI 1.21-1.59) per 10-year brain age index; 1.31 (1.14-1.50) after AHI and comorbidity, 1.22 (1.02-1.45) after APOE e4; IPD meta-analysis of MESA, ARIC, FHS, MrOS, SOF; 7,105 people, 1,088 cases; home PSG, central EEG channels; Fine-Gray with death as competing risk.Sun 2026 · JAMA Netw Open · doi
- BWhole night In five US home-sleep cohorts, four of them shared with the brain-age study, 44 standard sleep measures were condensed into nine composites, and none of them was linked to later dementia among 5,471 people, although longer, more consolidated sleep and stronger slow oscillation-spindle coupling went with better thinking. No composite associated with incident dementia (Cox; n=5,471, 1,134 cases, median follow-up 5-19 y); global cognition: quantity/efficiency composite beta 0.03 (0.004-0.06), SO-spindle coupling beta 0.04 (0.003-0.07); n=5,958; Sleep and Dementia Consortium (ARIC, CHS, FHS, MrOS, SOF).Yiallourou 2026 · Neurology · doi
- BShort sleep Across 17 cohorts of 1.34 million people, habitually sleeping under 7 hours carried about 18% higher dementia risk and sleeping over 8 hours about 28% higher, compared with 7 to 8 hours. RR 1.18 (1.09-1.28) for <7 h; RR 1.28 (1.15-1.43) for >8 h; meta-analysis of 17 prospective cohorts, n=1,344,170; duration mostly self-reported; moderate to substantial heterogeneity.Oye-Somefun 2026 · PLoS One · doi
- BWhole night In 502,383 UK Biobank adults, most self-reported sleep traits went with later dementia, but genetic analysis found no causal link, and the associations weakened after accounting for cardiovascular risk factors. Observational: almost all sleep measures associated with dementia (p<0.001) except insomnia, attenuated after vascular-risk adjustment; two-sample MR null for dementia and Alzheimer disease; 7,668 dementia cases; self-reported sleep.Guo 2024 · Neurology · doi
Show 4 moreShow fewer
- BApnea events Across 15 studies of about 5.2 million people, those with sleep apnea had about 34% higher risk of later dementia, with little variation between studies, while vascular dementia showed no clear link. All-cause dementia HR 1.34 (1.17-1.53), I2 30%; Alzheimer's HR 1.28 (1.16-1.41), I2 22%; vascular dementia HR 1.36 (0.88-2.10), NS; Parkinson's HR 1.64 (1.47-1.82); n=5,207,312; observational, apnea ascertainment not given in abstract; no genetic or trial evidence in this analysis.Li 2024 · Folia Neuropathol · doi
- BLong sleep Pooling cohort studies, people who reported sleeping more than 8 hours had about 43% higher risk of dementia and 66% higher risk of Alzheimer's than those sleeping 7 to 8 hours, while short sleep showed no clear link to dementia in this analysis. Long (>8 h) vs 7-8 h: all-cause dementia RR 1.43 (1.21-1.69), I2 62%; AD RR 1.66 (1.44-1.91), I2 0%; short (<7 h) not associated with dementia or AD here, unlike Oye-Somefun 2026 (short RR 1.18); 76 cohort studies overall; self-reported duration; reverse causation from prodromal disease not excluded.Zhang 2025 · J Neurol · doi
- CCircadian rhythm In ARIC adults around 79 who wore a heart-monitor patch for about 12 days, a flatter 24-hour activity rhythm went with about 1.5 times the risk of dementia over a median of only 3 years, so part of the signal may be disease already under way. HR 1.54 (1.32-1.78) per 1-SD lower relative amplitude; intradaily variability HR 1.19 (1.02-1.38) per SD; late acrophase HR 1.45 (1.01-2.07); n=2,183, 176 cases; accelerometer in Zio XT ECG patch; no subtype data, no objective sleep-disorder measures.Wang 2025 · Neurology · doi
- CWhole night A sleep foundation model reading a full clinical polysomnogram, together with the person's age and sex, correctly ordered about 85% of patient pairs by who was diagnosed with dementia first, in a sleep-clinic population rather than the general public. C-index 0.85 for dementia (SleepFM fine-tuned with age and sex appended; Stanford Sleep Clinic PSG linked to EHR diagnoses); pretraining on ~585,000 h PSG from ~65,000 participants; authors note a demographics-only baseline 'performs well'; no prospective community validation for dementia.Thapa 2026 · Nat Med · doi
Key findings for Alzheimer's 13 papers
- BApnea events Across prospective cohorts, people with sleep apnea had about 45% higher risk of later Alzheimer's disease, and people with insomnia about 49% higher. OSA HR 1.45 (1.24-1.69); insomnia HR 1.49 (1.27-1.74); all-cause dementia OSA HR 1.33 (1.09-1.61); meta-analysis of 39 prospective cohorts; exposure definitions vary across studies; MR for OSA null or inverse (Zhao 2025, not re-verified here).Ungvari 2025 · Geroscience · doi
- BDeep sleep A meta-analysis of 24 polysomnography studies found that people with Alzheimer's spent about 5% of the night in deep sleep against about 10% in healthy controls; less deep sleep tracked worse cognitive scores only in studies scored with the older rules. SWS% SMD -0.86 (-1.14 to -0.58), 25 comparisons, 644 AD vs 659 controls, I2 81%; study means 4.7% vs 10.0%; larger with AASM scoring (-1.70) than R&K (-0.65); MMSE meta-regression p=0.075 overall, p=0.004 in R&K-scored studies only, so the abstract's 'significantly associated' overstates the pooled result; no publication bias on Egger test; case-control.Zhang 2022 · Transl Psychiatry · doi
- BREM amount Pooled across 25 polysomnography comparisons, people with Alzheimer's spent about 14% of their sleep in REM, against about 18% in healthy controls. REM% SMD -0.77 (-1.14 to -0.39), 25 comparisons, 668 AD vs 711 controls, I2 90%; study means 14.0% vs 17.7%; larger with AASM scoring (SMD -2.17, 5 comparisons) than R&K (-0.39); MMSE meta-regression p=0.071 overall, p<0.001 in R&K-scored studies only; case-control, mostly mild to moderate AD, OSA exclusion mostly unreported.Zhang 2022 · Transl Psychiatry · doi
- BREM latency Across 23 polysomnography comparisons, the first REM period came later in people with Alzheimer's, after about 111 minutes against about 93 in healthy controls. REM latency SMD 0.35 (0.13-0.58), 23 comparisons, 609 AD vs 646 controls, I2 70%; study means 111.4 vs 92.8 min; significant under both AASM (0.59) and R&K (0.25) scoring; case-control.Zhang 2022 · Transl Psychiatry · doi
Show 9 moreShow fewer
- BTime to fall asleep People with Alzheimer's took longer to fall asleep in the sleep lab, about 23 minutes against 15 in healthy controls, and this difference varied less between studies than most others. Sleep latency SMD 0.45 (0.29-0.61), 20 comparisons, 588 AD vs 610 controls, I2 37%; study means 22.6 vs 14.9 min; between-study meta-regression found larger differences in older samples (p=0.008) and samples with fewer men (p=0.010), a study-level pattern, not a patient-level one; case-control.Zhang 2022 · Transl Psychiatry · doi
- BWaking in the night Across 16 polysomnography comparisons, people with Alzheimer's spent about 104 minutes awake after first falling asleep against about 74 in healthy controls, but the gap appeared only in studies that gave everyone a night to get used to the lab. WASO SMD 0.74 (0.38-1.10), 16 comparisons, 540 AD vs 541 controls, I2 87%; study means 104.4 vs 73.8 min; with adaptation night SMD 0.97, without 0.02 (4 comparisons, NS), subgroup difference p<0.001; sleep efficiency SMD -0.96 (-1.36 to -0.57), 20 comparisons, means 71.3% vs 80.6%; case-control.Zhang 2022 · Transl Psychiatry · doi
- BShort sleep On a night of polysomnography, people with Alzheimer's slept about 5.5 hours against about 6.2 hours in healthy controls, pooled across 23 comparisons. TST SMD -0.60 (-0.86 to -0.34), 23 comparisons, 644 AD vs 660 controls, I2 79%; study means 331 vs 370 min; present with and without an adaptation night; lab-measured sleep in diagnosed patients, not habitual sleep duration before diagnosis.Zhang 2022 · Transl Psychiatry · doi
- BArousals People with Alzheimer's woke up more often during a night in the sleep lab, about 16 times against 14 in healthy controls, across 15 comparisons. Number of awakenings SMD 0.55 (0.25-0.86), 15 comparisons, 383 AD vs 415 controls, I2 74%; counts awakenings, not the EEG arousal index; prospectively, higher actigraphic fragmentation went with incident AD, HR 1.22 (1.03-1.44) per SD (Lim 2013 Sleep, doi 10.5665/sleep.2802; 737 older adults, 97 cases).Zhang 2022 · Transl Psychiatry · doi
- CWhole night Genetic variants linked to insomnia went with higher Alzheimer's risk, but the link largely vanished once education and cholesterol were taken into account. IVW MR OR 3.02 (1.28-7.10) insomnia to AD (UK Biobank instruments, FinnGen outcomes); largely attenuated in multivariable MR adjusting for education and LDL.Guo 2026 · Alzheimers Dement · doi
- CWhole night In a randomized trial in 38 cognitively healthy adults aged 45 to 65, one evening dose of the insomnia drug suvorexant lowered amyloid-beta in spinal fluid by about 10 to 20% over the following day, and at the higher dose lowered phosphorylation at one tau site by 10 to 15%, while two other tau sites did not change. Placebo (n=13) vs suvorexant 10 mg (n=13) or 20 mg (n=12); CSF every 2 h for 36 h via lumbar catheter; Abeta -10-20% from 5 h post-dose; pT181/T181 -10-15% at 20 mg only; pS202 and pT217 unchanged; acute surrogate outcome, not a disease trial.Lucey 2023 · Ann Neurol · doi
- DSpindles In early clinical Alzheimer's, spindle activity over the temporal lobes was lower than in controls, and people with lower temporal spindle density declined faster on a cognitive screen over up to 7 years. Temporal spindle density vs MoCA slope beta 0.61 (0.35-0.87), FDR p<0.001; spindle power beta 0.56 (0.22-0.88); 47 AD vs 43 controls cross-sectional, 26 AD and 25 controls longitudinal; ambulatory scalp EEG with temporal leads, which standard PSG does not include.Bender 2025 · Neurology · doi
- DDeep sleep In 17 healthy adults, the more slow-wave activity was disrupted during one night, the more amyloid-beta rose in spinal fluid the next morning; sleep duration and efficiency showed no such link. Correlation of SWA disruption with Abeta40 increase r=0.61 (p=0.009), crossover vs sham night; specific to Abeta (not tau, total protein, YKL-40, hypocretin); tau rose with poorer actigraphic sleep efficiency over 6 nights (r=0.54); age 35-65.Ju 2017 · Brain · doi
- DWaking in the night In 45 older adults followed to autopsy, those with Alzheimer's disease had fewer than half as many neurons in a sleep-promoting nucleus of the hypothalamus, and in the whole group fewer of these neurons went with more fragmented sleep on wrist actigraphy. Galanin-immunoreactive intermediate nucleus (proposed human VLPO homologue) neurons 2,348 in AD (n=12) vs 5,220 without; difference -2,872 (SE 829), P=0.001; -3,128 (SE 928) adjusted for age and sex; actigraphic fragmentation kRA -0.0016 per 1,000 neurons (SE 0.0005), P=0.005 (Table 2; text prints 0.0005), adjusted for age, sex and lag; -0.0013 (P=0.023) with actigraphy within 1 y of death (n=19), -0.00009 (P=0.81) beyond 1 y; slope did not differ by AD status; AD vs non-AD fragmentation itself only a trend (P=0.15); Rush Memory and Aging Project, mean age at death 89.2; cross-sectional clinicopathological.Lim 2014 · Brain · doi
Key findings for Mild cognitive impairment 9 papers
- BREM amount Across the Alzheimer's continuum, lower sleep efficiency and less REM were already present at the mild cognitive impairment stage, while loss of deep sleep appeared only once dementia had set in. Network meta-analysis of PSG, actigraphy and self-report across MCI and AD dementia; SE and REM% reduced in MCI and more so in AD dementia; SWS reduced only in AD dementia; pooled effect sizes not given in abstract; REM reduction in MCI also found by Wei 2024 (MD -2.69 points, 12 studies, doi 10.1016/j.sleep.2023.12.010) and D'Rozario 2020 (10 studies, doi 10.1016/j.smrv.2020.101308).Zhang 2026 · Sleep Med Rev · doi
- BApnea events Across 22 studies and 73,733 people, sleep apnea and mild cognitive impairment went together: apnea carried about 1.65 times the odds of impairment, and about 39% of people with impairment had apnea. OR 1.65 (1.39-1.96); OSA prevalence in MCI 39% (23-55), I2 99%; MCI prevalence in OSA 39% (24-53); mostly cross-sectional, so direction cannot be read from it.Ren 2025 · Sleep Med · doi
- BShort sleep Compared with healthy older adults, people with mild cognitive impairment slept about 34 minutes less on objective recording, spent about 20 minutes more awake during the night and had about 8 percentage points lower sleep efficiency; REM was 2.7 points lower. TST MD -34.44 min; SE -7.96 points; WASO +19.61 min; SL +6.97 min; N1 +2.72 points; N3 -0.78 points; REM -2.69 points; 15 studies, n=771; objective measurement (PSG or actigraphy).Wei 2024 · Sleep Med · doi
- BWaking in the night Compared with healthy older adults, people with mild cognitive impairment spent about 20 more minutes awake during the night on objective recording, and their sleep efficiency was about 8 points lower. WASO MD +19.61 min; SE MD -7.96 points; 15 studies, n=771, PSG or actigraphy; CIs not in abstract; an earlier meta-analysis of 10 studies (n=430) also found more WASO and lower SE in MCI (D'Rozario 2020 Sleep Med Rev, doi 10.1016/j.smrv.2020.101308); case-control.Wei 2024 · Sleep Med · doi
Show 5 moreShow fewer
- BTime to fall asleep People with mild cognitive impairment took about 7 minutes longer to fall asleep than healthy older adults on objective recording. Sleep latency MD +6.97 min; 15 studies, n=771, PSG or actigraphy; CI and number of studies for this outcome not in abstract; D'Rozario 2020 (10 studies, n=430, doi 10.1016/j.smrv.2020.101308) also found longer sleep onset latency; small absolute difference; case-control.Wei 2024 · Sleep Med · doi
- BLong sleep Across cohort studies, older adults who reported sleeping more than 8 hours had about 23% higher risk of later cognitive decline than those sleeping 7 to 8 hours, with little variation between studies. RR 1.23 (1.12-1.36), I2 12%, long (>8 h) vs 7-8 h; short (<7 h) RR 1.27 (1.12-1.42); outcome is decline on cognitive tests such as the MMSE, not a clinical MCI diagnosis; self-reported duration; meta-analysis of 76 cohort studies overall; long sleep may partly mark disease already under way.Zhang 2025 · J Neurol · doi
- CApnea events In 298 older women followed for about five years, those with an apnea-hypopnea index of 15 or more had about 1.85 times the odds of developing mild cognitive impairment or dementia; oxygen drops mattered, while sleep fragmentation and sleep duration did not. AOR 1.85 (1.11-3.08) for AHI>=15; ODI>=15 AOR 1.71 (1.04-2.83); >7% of sleep in apnea or hypopnea AOR 2.04 (1.10-3.78); arousal index, WASO and TST not associated; SOF in-home PSG, mean age 82; outcome is MCI or dementia combined.Yaffe 2011 · JAMA · doi
- CCircadian rhythm In 91,517 UK Biobank adults who wore a wrist accelerometer, a flatter 24-hour activity rhythm went with about 23% higher risk of dementia or mild cognitive impairment over up to 7.5 years, while day-to-day regularity of the rhythm showed no link. HR 1.23 (1.16-1.29) per 1-SD lower relative amplitude; amplitude HR 1.32 (1.17-1.49); intradaily variability HR 1.14 (1.05-1.24); interdaily stability not associated; 555 incident MCI or dementia; adjusted for AD genetic risk.Haghayegh 2024 · JMIR Public Health Surveill · doi
- CWhole night Across 10 randomized trials in adults 65 and older with mild cognitive impairment or dementia, melatonin added about 12 minutes of sleep and about 1.8 points on a 30-point cognitive screen, a modest effect the authors judged short of definitive. MMSE MD +1.8 (0.9-2.7), I2 65%; TST MD +12.4 min (8.1-16.7), I2 72%; 10 RCTs, n=516; sleep efficiency and circadian markers not significantly changed.Mdluli 2025 · Age Ageing · doi
Key findings for Parkinson's 12 papers
- AREM without atonia In 1,280 people with isolated REM sleep behaviour disorder confirmed by polysomnography at 24 centres, about 6.3% a year developed parkinsonism or dementia, an estimated 73.5% by 12 years; more muscle activity during REM sleep at baseline went with faster conversion. Phenoconversion 6.3%/y to an overt neurodegenerative syndrome (parkinsonism or dementia); Kaplan-Meier 73.5% at 12 y; mean follow-up 4.6 y; 82.5% male, mean age 66; REM atonia loss HR 1.54; quantitative motor testing HR 3.16; olfactory deficit HR 2.62; International RBD Study Group.Postuma 2019 · Brain · doi
- BREM without atonia Pooled across 15 studies, about 80% of people with isolated RBD tested positive for misfolded alpha-synuclein in spinal fluid and about 75 to 79% in skin, far more often than people whose RBD had another cause; a positive skin biopsy did not by itself predict faster conversion. CSF aSyn-SAA positivity 80% (68-88), I2 71%; skin SAA 74.8% (53.2-88.5); skin p-syn 78.5% (70.4-84.9); OR 112 (20-629) iRBD vs other RBD; phenoconversion rate ratio biopsy-positive vs negative 1.28 (0.68-2.41).Iftikhar 2024 · J Sleep Res · doi
- BDeep sleep A meta-analysis of 63 polysomnography studies found shorter, less efficient sleep in Parkinson's, with less deep sleep and less REM, a later first REM period and higher apnea and periodic limb movement indices than in controls; patients with and without RBD did not differ in sleep architecture. 63 PSG case-control studies; reduced TST, SE, N2%, SWS, REM%; increased WASO, N1%, REM latency, AHI, PLMI; heterogeneity from sex, age, disease duration, RBD, medication, cognition, adaptation night.Zhang 2020 · Sleep Med Rev · doi
- BShort sleep In a meta-analysis of 63 polysomnography studies, people with Parkinson's slept less during a night in the sleep lab than controls. TST reduced vs controls; random-effects SMD; pooled size, CI and number of contributing studies not in abstract (full text not accessed); heterogeneity from sex, age, disease duration, RBD, medication, cognition and adaptation night; lab sleep in diagnosed, mostly treated patients.Zhang 2020 · Sleep Med Rev · doi
Show 8 moreShow fewer
- BWaking in the night In a meta-analysis of 63 polysomnography studies, people with Parkinson's spent more time awake after falling asleep and had lower sleep efficiency than controls. WASO increased and SE reduced vs controls; random-effects SMD; pooled size, CI and number of contributing studies not in abstract (full text not accessed); heterogeneity from sex, age, disease duration, RBD, medication, cognition and adaptation night; PD with and without RBD did not differ in sleep continuity.Zhang 2020 · Sleep Med Rev · doi
- BREM amount In a meta-analysis of 63 polysomnography studies, REM took up a smaller share of the night in Parkinson's than in controls, and patients with and without REM sleep behaviour disorder did not differ. REM% reduced vs controls; random-effects SMD; pooled size, CI and number of contributing studies not in abstract (full text not accessed); heterogeneity from sex, age, disease duration, RBD, medication, cognition and adaptation night; no remarkable difference in sleep architecture between PD with and without RBD.Zhang 2020 · Sleep Med Rev · doi
- BREM latency In a meta-analysis of 63 polysomnography studies, the first REM period came later in people with Parkinson's than in controls. REM latency increased vs controls; random-effects SMD; pooled size, CI and number of contributing studies not in abstract (full text not accessed); heterogeneity from sex, age, disease duration, RBD, medication, cognition and adaptation night.Zhang 2020 · Sleep Med Rev · doi
- BApnea events Pooled across cohort studies, people with sleep apnea had about 1.6 times the risk of a later Parkinson's diagnosis, though apnea within Parkinson's looks largely secondary to age, sex and weight. Incident PD HR 1.64 (1.47-1.82), I2 25%; meta-analysis of 15 studies, n=5,207,312 overall (number contributing to PD not in abstract); apnea ascertainment not given in abstract; prodromal PD could raise apnea risk, so direction is unsettled; within PD, OSA (pooled prevalence 45%, 17 studies, 1,448 patients) went with older age, male sex, higher BMI and less RBD (Maggi 2023 Eur J Neurol, doi 10.1111/ene.16109).Li 2024 · Folia Neuropathol · doi
- CApnea events In 94 people with Parkinson's, sleep apnea and below-normal cognition, six months of positive airway pressure versus placebo nasal strips improved a 30-point cognitive screen by 1.0 point in the main analysis, which missed significance, and by 1.4 points after adjustment. COPE-PAP RCT; MoCA ITT between-group difference 1.00 (-0.06 to 2.05) primary; adjusted ITT 1.44 (0.09-2.79); per-protocol adjusted 1.43 (0.05-2.81); PAP n=48, control n=46.Lajoie 2025 · Sleep · doi
- CWhole night An AI model reading only breathing signals during sleep ranked a random person with Parkinson's above a random person without it about 90% of the time on held-out data and 85% on external data, and its score tracked disease severity; most people tested already had a diagnosis. AUC 0.90 held-out, 0.85 external; n=7,671 from US hospitals and public datasets; MDS-UPDRS correlation R=0.94; also from radio-wave breathing at home; detection of established PD, so medication and disease effects cannot be separated from early signal.Yang 2022 · Nat Med · doi
- CCircadian rhythm In 2,930 older men followed for 11 years, those with the weakest 24-hour activity rhythm were about three times as likely to develop Parkinson's as those with the strongest, and still about 2.4 times after excluding cases diagnosed in the first two years. OR 3.11 (1.54-6.29) lowest vs highest amplitude quartile; 2.40 (1.15-5.00) excluding first 2 y; OR per 1-SD lower amplitude 1.77 (1.30-2.41); MrOS wrist actigraphy; 78 incident PD (2.7%), physician diagnosis; men only.Leng 2020 · JAMA Neurol · doi
- CLimb movements A meta-analysis of polysomnography studies found a higher periodic limb movement index in Parkinson's than in controls, and about 20% of people with Parkinson's have restless legs syndrome, a figure entangled with dopaminergic treatment. PLMI increased vs controls (pooled size and number of studies not in abstract; medication status a source of heterogeneity); RLS prevalence 20% of 6,990 PD patients from 46 studies with no control arm, associated with dopaminergic-therapy variables, not with ferritin or iron (Maggi 2024 Sleep Med, doi 10.1016/j.sleep.2024.07.015); RLS is a clinical diagnosis, not the PSG PLM index.Zhang 2020 · Sleep Med Rev · doi
Key findings for Lewy body dementia 6 papers
- BREM without atonia The 2017 international diagnostic criteria made dream enactment a core feature of Lewy body dementia, because it is found in 76% of autopsy-confirmed cases versus 4% of other dementias. They also count REM sleep without paralysis on polysomnography as a biomarker; together with a history of dream enactment, it makes a synuclein disease at least 90% likely. Fourth DLB Consortium report; RBD core clinical feature (76% vs 4% at autopsy); PSG-confirmed REM sleep without atonia = indicative biomarker, sufficient with RBD history for probable DLB; >=90% likelihood of synucleinopathy; posterior slow-wave EEG = supportive biomarker; mimics (confusional awakenings, severe OSA, PLMs) must be excluded.McKeith 2017 · Neurology · doi
- BREM without atonia Pooling long-term RBD cohorts, the estimated risk of a neurodegenerative disease was about one in three at 5 years and over 90% by 14 years; among those who converted, 43% developed Parkinson's and 25% Lewy body dementia. Kaplan-Meier fitted risk 33.5% at 5 y, 82.4% at 10.5 y, 96.6% at 14 y (few patients followed that long); mean conversion 31.95% over 4.75 y; converters 43% PD, 25% DLB (share of converters, not prevalence).Galbiati 2018 · Sleep Med Rev · doi
- CREM without atonia In 234 people with dementia followed to autopsy, a clinical history of dream enactment raised the odds of confirmed Lewy body disease about six-fold, more than any other core feature, and adding it raised diagnostic sensitivity from 85% to 90% without losing specificity. OR ~6 (no CI in abstract); 98 intermediate/high-likelihood DLB vs 136 non-DLB at autopsy; RBD by clinical history, not PSG; sensitivity 85% to 88-90%, specificity 73% unchanged.Ferman 2011 · Neurology · doi
- DREM without atonia Video-polysomnography in 35 people with mild Lewy body dementia confirmed RBD in half of those with enough REM sleep; most were unaware they acted out dreams, and a standard questionnaire caught only half the cases while flagging others whose night events were really apnea or limb movements. RBD on vPSG in 50% with sufficient REM; 72.7% unaware; RBD preceded cognitive onset in 63.7%; Mayo Sleep Questionnaire sensitivity 50%, specificity 66.7%; sleep-wake abnormalities in 75%: occipital slowing on waking EEG 34.4%, delta in REM 19.2%, absent spindles/K-complexes 12.9%, slow spindles 12.9%, RSWA 44%.Fernandez-Arcos 2019 · Sleep · doi
Show 2 moreShow fewer
- DEEG slowing In 66 people with isolated RBD, a machine-learning model using the REM-sleep EEG from one baseline polysomnogram ordered who would convert to an alpha-synuclein disease sooner about 72% of the time, with EEG slowing carrying the most weight; it needs confirmation in larger cohorts. Harrell's C 0.723 +/- 0.113; Uno's C 0.741; 18 converters within 2.7 +/- 1.0 y; random survival forest, 4-fold cross-validation; single centre (Innsbruck); outcome any alpha-synucleinopathy, not DLB specifically.Cesari 2024 · Brain Sci · doi
- DLimb movements In 29 people with Lewy body dementia on video-polysomnography, apnea (31%) and periodic limb movements (61%) were frequent but about as common as in matched Parkinson's patients without dementia, while confusional night events that mimic RBD were far more common. Sleep apnea 30.7% vs 34.8% in 29 matched PD; PLMs 60.9% vs 50.0%; disruptive motor behaviours 69.6% vs 26.9% (p=0.008); confusional events 30.3% vs 3.8% (p=0.020).Terzaghi 2013 · Mov Disord · doi
Key findings for Stroke 8 papers
- BApnea events Across 89 studies of 7,096 people after stroke or mini-stroke, 71% had an apnea-hypopnea index above 5 and 30% above 30, with similar rates in the first month and after three months. AHI>5 71% (66.6-74.8); AHI>30 30% (24.4-35.5); mean AHI 26; full PSG or AASM type I-IV devices; high heterogeneity; no control group, and AHI>5 is also common in healthy older adults.Seiler 2019 · Neurology · doi
- BWhole night In 502,383 UK Biobank adults, genetic analysis supported insomnia as a cause of stroke, with about 31% higher odds, while the observed link with self-reported insomnia was small once cardiovascular risk factors were accounted for. MR OR 1.31 (1.13-1.51) insomnia to any stroke; observational insomnia HR 1.05 (1.01-1.11) after vascular-risk adjustment; 10,334 strokes; self-reported insomnia.Guo 2024 · Neurology · doi
- BApnea events In 2,717 people with moderate-to-severe apnea and existing heart or brain-vessel disease, CPAP for an average of 3.7 years did not reduce strokes or other cardiovascular events; average use was only 3.3 hours a night. SAVE RCT; primary composite HR 1.10 (0.91-1.32); no effect on any individual endpoint including stroke; AHI 29.0 to 3.7 on CPAP; mostly men with minimal sleepiness; secondary prevention.McEvoy 2016 · N Engl J Med · doi
- BShort sleep Across prospective cohorts, short sleep carried about 33% higher risk of a first stroke and long sleep about 71% higher, though long sleep may partly reflect illness already present. RR 1.33 (1.19-1.49) short; RR 1.71 (1.50-1.95) long; U-shaped dose-response; 15 incidence reports in 20 articles; duration mostly self-reported.Wang 2021 · Sleep Med · doi
Show 4 moreShow fewer
- BLong sleep Across 12 prospective cohorts of 559,252 people, those reporting long sleep had about 45% higher risk of stroke, and in the authors' own British cohort people who reported persistently long sleep had about twice the risk of those with persistently average sleep. Pooled RR long 1.45 (1.30-1.62), I2 54%; 1.53 (1.42-1.65), I2 0% excluding one study; short RR 1.15 (1.07-1.24); 11,695 strokes, follow-up 7.5-35 y; long defined as >=8-9 h, all self-reported, 6 studies fatal stroke only; EPIC-Norfolk long sleep HR 1.46 (1.08-1.98), n=9,692, 346 strokes; short-to-long change over 4 y HR 3.75 (1.17-12.05).Leng 2015 · Neurology · doi
- CApnea events Pooling 21 randomized trials in 1,457 stroke survivors, positive airway pressure went with about half the odds of a repeat vascular event and better neurological recovery, but not with more independence in daily life; the trials were small. Recurrent vascular events OR 0.45 (0.28-0.78); neurological deficit -0.30 (-0.47 to -0.14), reported as OR but behaves as SMD; functional independence 0.25 (-0.11 to 0.60), NS; 21 RCTs, n=1,457 (PAP 753, control 704).Sampsonas 2025 · Sleep Med Rev · doi
- CApnea events In 5,422 community adults followed for about 9 years, men in the highest apnea quartile had almost three times the risk of ischemic stroke, and risk rose about 6% for each extra event per hour in the mild-to-moderate range; in women a link appeared only at higher apnea levels. Men OAHI>19 adjusted HR 2.86 (1.1-7.4); +6% (2-10%) per unit OAHI for OAHI 5-25; women: increased risk only at OAHI>25; Sleep Heart Health Study, home PSG, untreated, 193 ischemic strokes.Redline 2010 · Am J Respir Crit Care Med · doi
- DApnea events In 161 people studied in the first days after a first stroke or mini-stroke, central apneas and Cheyne-Stokes breathing fell by three months while obstructive apneas did not, which the authors read as central events following the stroke and obstructive events predating it. Portable respiratory recording 48-72 h after admission and at 3 months (86 retested); AHI 22.4±17.3 to 16.9±13.8 and central apnea index 6.2±10.2 to 3.3±7.6 (both p<0.05); obstructive apnea index unchanged; Cheyne-Stokes 42/161 (26.1%) acute vs 6/86 at 3 months; AHI>10 in 71.4% acute; no link with lesion location; not full PSG; 75 of 161 not retested; includes TIA.Parra 2000 · Am J Respir Crit Care Med · doi
Key findings for Epilepsy 7 papers
- BREM amount Across 24 polysomnography studies, focal epilepsy showed less REM and lower sleep efficiency, and generalized epilepsy more deep sleep and lower efficiency; efficiency was low even in patients not yet on anti-seizure drugs. 789 patients, 599 controls; focal: REM% and SE reduced; generalized: SWS increased, SE reduced; SE reduced in ASM-untreated and treated; ASM may restore REM in generalized but not focal epilepsy; effect sizes not in abstract.Yeh 2022 · Seizure · doi
- BWaking in the night Children with epilepsy slept less efficiently than healthy children, about 83% of time in bed against 89%, and meta-analyses in adults found lower efficiency in both focal and generalized epilepsy, including in patients not yet treated. SE SMD -0.71 (-1.23 to -0.19), 6 studies (5 PSG, 1 actigraphy), I2 39%; mean SE 83% vs 89%; parent-reported night waking SMD 0.42 (0.16-0.68); 19 studies, 901 children with epilepsy vs 1,470 controls overall; adults: Yeh 2022 (24 PSG studies, doi 10.1016/j.seizure.2022.01.014) SE lower in focal and generalized epilepsy, treated and untreated; Lehner 2022 (22 studies, doi 10.1016/j.smrv.2022.101689) SE lower in idiopathic generalized epilepsy; values from author manuscript.Winsor 2021 · Sleep Med Rev · doi
- CWhole night Across 42 studies of 1,458 patients, REM sleep was the most protective state, with about eight times fewer focal seizures and about three times fewer generalized seizures than wakefulness, after accounting for time spent in each state. Rate vs wake indexed to duration: focal seizures 7.83x fewer, generalized 3.25x fewer, focal interictal discharges 1.11x fewer in REM; ~1% or less of seizures in REM; review of 42 scalp and intracranial studies.Ng 2013 · Epilepsy Res Treat · doi
- CShort sleep In 209 newly treated patients at one Chinese centre, self-reported short sleep at baseline went with about 2.3 times the risk of a seizure recurrence within 12 months, with the lowest risk around 8 hours a night. Adjusted HR 2.282 (1.436-3.628); poor sleep quality (PSQI) HR 1.985 (1.321-2.984); poor quality plus short sleep HR 3.13 (1.779-5.507); U-shape with nadir 8 h/day; 103 recurrences; self-reported sleep.Zhong 2024 · Epilepsia · doi
Show 3 moreShow fewer
- CApnea events People with epilepsy had about twice the odds of obstructive sleep apnea compared with controls, about a third had at least mild apnea, and those treated with CPAP had better seizure control in observational comparisons. OR 2.36 (1.33-4.18) vs healthy controls; OSA prevalence 33.4% (20.8-46.1); CPAP-treated vs untreated better seizure control OR 5.26 (2.04-13.5), non-randomized; no difference by seizure type or number of drugs.Lin 2016 · Sleep Breath · doi
- CArousals Across 11 studies, people with epilepsy showed more non-REM instability, measured as a higher cyclic alternating pattern rate, and anti-seizure drugs reduced it. 209 patients, 197 controls; CAP rate increased, A1% decreased; A3% increased in sleep-related epilepsy; ASMs decreased CAP rate and lengthened phase B.Yeh 2022 · Sleep · doi
- DDeep sleep In eight patients with intracranial electrodes, comparing equal stretches of non-REM sleep with and without high-amplitude widespread slow waves, 79% of epileptic spikes fell in the slow-wave stretches, which suggests that deep-sleep slow waves, rather than non-REM sleep as a whole, are what activate the discharges. 79% of spikes and 65% of HFOs during high-amplitude widespread slow waves vs matched control segments; density highest at the up-to-down transition; n=8 drug-resistant focal epilepsy; 442 spikes, 8,487 HFOs.Frauscher 2015 · Brain · doi
Key findings for Multiple sclerosis 7 papers
- BLimb movements Across 75 studies of 15,411 people with MS, about 28% had restless legs syndrome, compared with 8% of controls. Pooled RLS prevalence 28% (24-33) vs 8% (6-10) in controls; women 30%, men 22%; RLS is a clinical diagnosis, not the PSG periodic limb movement index.Zali 2023 · Neurol Sci · doi
- BLimb movements In case-control studies, restless legs syndrome was about 4.5 times as common in MS as in healthy controls, and it was not clearly tied to disability, disease duration or MS type. OR 4.535 (3.043-6.759) from 9 case-control studies; mean prevalence 27.5% (range 13.2-65.1%) across 19 studies.Ozdogar 2021 · Mult Scler Relat Disord · doi
- BLimb movements A meta-analysis of 13 polysomnography studies found lower sleep efficiency, more waking and higher periodic limb movement and limb-movement arousal indices in MS than in controls. 13 PSG case-control studies; reduced N2% and SE; increased WASO, PLMI and PLM arousal index; heterogeneity partly explained by age and daytime sleepiness.Zhang 2023 · J Clin Sleep Med · doi
- BWaking in the night Across polysomnography studies, people with MS spent more time awake after falling asleep and had lower sleep efficiency than controls, and a larger 2026 meta-analysis put the efficiency gap at about 6 points. WASO SMD 0.46 (0.03-0.90), 4 studies, 163 MS vs 113 controls, I2 67%; SE SMD -0.42 (-0.72 to -0.12), 11 studies, 311 vs 229, I2 63%; TST and AHI did not differ in this analysis; Toubasi 2026 SE WMD -6.38 (-9.71 to -3.06), 13 studies, 741 MS, I2 86%; values from PMC full text, Table 2.Zhang 2023 · J Clin Sleep Med · doi
Show 3 moreShow fewer
- CShort sleep Across 40 studies of 3,264 people with MS and 679 controls, people with MS slept about 38 minutes less with 6 points lower efficiency and a slightly higher apnea index, though an earlier polysomnography meta-analysis found no difference in total sleep; 44% had obstructive sleep apnea, and a higher apnea index went with more disability. TST WMD -37.70 min (-62.27 to -13.12), 13 studies, 728 MS, I2 93%; high-quality subgroup -12.66 (CI as published -37.76 to -3.43); Zhang 2023 PSG meta-analysis TST SMD -0.14 (-0.46 to 0.17), 9 studies, NS (doi 10.5664/jcsm.10304); SE WMD -6.38 (-9.71 to -3.06), 13 studies, I2 86%; AHI WMD +1.76 (0.20-3.32); OSA prevalence 44% (33-56) with no matched control rate; AHI vs EDSS r=0.41 (0.26-0.54), SE vs EDSS r=-0.16; cross-sectional.Toubasi 2026 · J Cent Nerv Syst Dis · doi
- CWhole night People with MS report much worse sleep quality than healthy controls on a standard questionnaire, though studies varied widely. PSQI standardized difference 1.056 (0.758-1.372), I2 94.6%, 10 studies; poor sleepers OR 2.31 (0.82-6.35), NS, 4 studies.Moradi 2024 · Mult Scler Relat Disord · doi
- CShort sleep In a Swedish case-control study of 2,075 people with MS and 3,164 controls, those with MS more often recalled sleeping under 7 hours a night as teenagers, about 1.4 times the odds. OR 1.4 (1.1-1.7) short sleep (<7 h) in adolescence; low sleep quality OR 1.5 (1.3-1.9); results similar excluding shift workers; retrospective recall, population-based case-control.Akerstedt 2023 · J Neurol Neurosurg Psychiatry · doi
Key findings for Migraine 7 papers
- BCircadian rhythm Across 8 studies of 5,385 people, about half of those with migraine had attacks on a daily pattern, with a clear low between 11 pm and 7 am, and urinary melatonin was lower in migraine and lower still during attacks. Circadian attack pattern 50.1% (2,698/5,385); trough 23:00-07:00; circannual peak April-October; chronotype variable; 110 of 168 migraine susceptibility genes are clock-controlled genes.Benkli 2023 · Neurology · doi
- BLimb movements Restless legs syndrome affected about 17% of people with migraine versus 7% without; the odds were about 3.8 times higher in case-control studies but only 1.25 times in population surveys. Pooled OR 3.77 (2.73-5.21), I2 50%, case-control; OR 1.25 (1.11-1.41) cross-sectional; 15 studies; RLS clinical diagnosis, not PSG limb movements.Wang 2019 · Sleep Med · doi
- CApnea events In matched US health records of 393,728 adults, a diagnosis of obstructive sleep apnea went with about 1.85 times the risk of a new migraine diagnosis, and about 1.39 times in a stricter sensitivity analysis. HR 1.85 (1.79-1.90); test-negative design HR 1.39 (1.28-1.49); TriNetX, 1:1 propensity matched; diagnosis codes, no PSG; morning 'sleep apnoea headache' could be coded as migraine.Chen 2025 · Headache · doi
- CWhole night In 15,268 headache-free Norwegian adults, insomnia at baseline went with about 40% higher risk of headache 11 years later and about twice the risk of migraine. RR 1.4 (1.2-1.7) any headache; migraine RR 2.0 (1.3-3.1); tension-type RR 1.5 (1.1-2.1); HUNT-2 to HUNT-3; questionnaire-based DSM-IV insomnia proxy and ICHD-2 headache diagnoses.Odegard 2011 · Headache · doi
Show 3 moreShow fewer
- CREM amount Polysomnography shows slightly less REM sleep in adults with migraine and a larger reduction in children, who also slept less and woke more, but only 6 adult and 5 child studies exist. REM% g -0.22 (-0.41 to -0.04) adults; g -0.71 (-1.34 to -0.10) children; children TST g -1.37 (-2.66 to -0.10), wake g 0.52, sleep onset latency g -0.37; adult PSQI g 0.75 (0.54-0.96).Stanyer 2021 · Neurology · doi
- CWhole night In 19,271 Norwegian adults free of insomnia at baseline, those with migraine had about 1.9 times the odds of insomnia 11 years later, though other headaches carried a similar excess. Migraine OR 1.9 (1.7-2.1), 711 of 2,413 developed insomnia; migraine on 7 or more days a month OR 2.2 (1.7-2.7); migraine without chronic musculoskeletal pain OR 1.7 (1.5-2.0); non-migrainous headache OR 1.7 (1.5-1.8); adjusted for age, sex, HADS anxiety-depression and GI complaints; HUNT-2 to HUNT-3; questionnaire migraine (IHS criteria) and DSM-IV insomnia proxy; same cohort as the insomnia-to-migraine finding.Odegard 2013 · J Headache Pain · doi
- DApnea events Genetic analysis suggests sleep apnea may slightly raise the risk of migraine with aura, about 15% higher odds, a preliminary finding. IVW OR 1.147 (1.016-1.295); multivariable MR 1.184 (1.028-1.364); reverse-direction signals negated in multivariable MR; European ancestry; OSA GWAS instruments are weak.Wang 2025 · Nat Sci Sleep · doi
Key findings for Brain injury 6 papers
- BRegularity Across 23 actigraphy studies of 881 people with brain injury and 769 controls, injury meant longer time to fall asleep, more waking, lower efficiency and larger night-to-night swings in sleep time and waking. 23 actigraphy case-control studies; increased SOL and WASO, decreased SE; increased night-to-night variability in TST and WASO; mixed severity; effect sizes not in abstract.Zong 2025 · Sleep · doi
- BShort sleep In mild brain injury specifically, 69 studies found no objective difference in sleep duration or efficiency at any stage of recovery, and small differences in sleep stages disappeared once biased studies were removed, although people kept reporting worse sleep for over a year. Objective TST and SE MD not significant at any phase (<1 w to >=1 y); adults N1 +1.28%, N3 -1.16% (-2.08 to -0.23) at 1-12 m, REM -3.37% at >=1 y, none robust to removing selection-biased studies; PSQI +2.26 to +2.86 points at every phase; 45% of studies at low risk of bias.Roberts 2025 · Sleep Med Rev · doi
- BWaking in the night Across 16 polysomnography studies of community-dwelling people with brain injury of mixed severity, sleep was less efficient and shorter, with more waking than in controls, while sleep stages were similar. WASO SMD 0.60 (0.33-0.87); SE SMD -0.47 (-0.89 to -0.06); TST SMD -0.37 (-0.59 to -0.16); REM trend SMD -0.22 (-0.45 to 0.01); 637 TBI, 567 controls.Grima 2016 · J Clin Sleep Med · doi
- CLong sleep Six months after a first brain injury, 42 patients needed about 8.3 hours of sleep a day on actigraphy versus 7.1 in controls, 57% were objectively sleepy by day versus 19% of controls, and patients markedly underestimated both problems when asked. Sleep need per 24 h 8.3 +/- 1.1 vs 7.1 +/- 0.8 h (p<0.0001); objective EDS 57% vs 19%; mean sleep latency 8.7 vs 12.1 min; intracranial haemorrhage strongest risk factor; prospective controlled, 42 of 60 enrolled followed.Imbach 2015 · Brain · doi
Show 2 moreShow fewer
- CDeep sleep In chronic brain injury more than six months after the event, 14 polysomnography studies showed more deep sleep, driven by moderate-to-severe injury, while mild injury showed no change in sleep stages. 14 PSG case-control studies; SWS increased overall; meta-regression: moderate-severe TBI with elevated SWS, reduced N2 and reduced SE; mild TBI no significant architecture change.Mantua 2018 · Sleep Med Rev · doi
- CWhole night In 106 military service members and veterans with mild brain injury and insomnia, six weeks of internet-guided CBT for insomnia lowered insomnia scores by 3.5 points more than online sleep education, though fewer than half completed follow-up. ISI between-group difference -3.5 (-6.5 to -0.4), p=0.03, Cohen d -0.32; randomized 3:1; 125 randomized, 106 at baseline, 50 post-intervention, 41 at 3 months.Malarkey 2024 · JAMA Netw Open · doi
Key findings for Depression 12 papers
- AWhole night Pooling long-term studies, people with insomnia were more than twice as likely to develop depression later. Summary RR 2.31 (95% CI 1.90-2.81); umbrella review of 25 meta-analyses of prospective cohorts (63 summary estimates, 29 outcomes); observational, outcomes often symptom scales, residual confounding by subthreshold symptoms possible.Wu 2023 · Public Health · doi
- BWhole night In 708 young people aged 15 to 25 with insomnia and mild depressive symptoms, six weeks of app-based CBT-I lowered new major depression over the next year from 18% to 10%, compared with app-based health education. HR 0.58 (95% CI 0.38-0.87); assessor-blind RCT, n=708 Chinese youth (mean age 22.1), insomnia disorder plus subclinical depression; new MDD by telephone clinical interview over 12 months; NNT 10.9; insomnia mediated the effect.Chen 2025 · PLoS Med · doi
- BWhole night In 291 adults over 60 with insomnia and no major depression in the past year, CBT-I halved new or returning major depression over three years compared with sleep education (12% against 26%). HR 0.51 (95% CI 0.29-0.88); single-site assessor-blinded RCT vs active sleep education therapy, n=291 (mean age 70.1); DSM-5 interview; follow-up extended from 24 to 36 months by protocol change, 52-57% completed 36 months, more dropout in the CBT-I arm.Irwin 2022 · JAMA Psychiatry · doi
- BWhole night Genetic studies suggest the arrow runs both ways: gene variants linked to insomnia raised the odds of major depression by about 30%, and variants linked to depression raised the odds of insomnia by a similar amount. IVW OR 1.31 (95% CI 1.25-1.37) insomnia to MDD; OR 1.37 (1.14-1.64) MDD to insomnia; bidirectional two-sample MR, PGC and UK Biobank summary data. Gao 2019 (Eur Psychiatry, PMID 31234011) found no significant effect in either direction for MDD; pleiotropy and sample overlap cannot be excluded.Sun 2022 · Epidemiol Psychiatr Sci · doi
Show 8 moreShow fewer
- BWaking in the night Across 53 studies with wrist sensors, people with depression spent more time awake after first falling asleep and slept less efficiently than healthy controls, a small to moderate difference. WASO SMD 0.37 (95% CI 0.22-0.52); sleep efficiency SMD -0.41 (-0.56 to -0.25); actigraphy case-control meta-analysis, 4,000 depressed (MDD or depressive symptoms) vs 7,115 controls; heterogeneous devices and samples. On PSG, Baglioni 2016 (PMID 27416139) found major depression altered on 10 of 11 variables, continuity included.Ho 2024 · J Affect Disord · doi
- BTime to fall asleep An umbrella review of overnight-recording meta-analyses across 27 brain and mental disorders rated the evidence for longer time to fall asleep in major depression as highly suggestive, the second-highest of five levels, and wrist-sensor studies show the same small delay. Credibility class highly suggestive for increased sleep latency and reduced sleep efficiency in MDD (REM latency only suggestive); umbrella review of 27 meta-analyses, 465 PSG case-control studies; pooled values in full text. Actigraphy: Ho 2024 (PMID 38851435) sleep latency SMD 0.23 (0.12-0.33), 53 studies.Zhang 2022 · Psychol Med · doi
- BArousals On wrist sensors, people with depression woke up more often during the night than healthy controls, a moderate difference. Nocturnal awakenings SMD 0.58 (95% CI 0.29-0.88); actigraphy meta-analysis, 53 studies, 4,000 depressed vs 7,115 controls; actigraphic awakenings are movement-defined wake bouts, not EEG arousals.Ho 2024 · J Affect Disord · doi
- BCircadian rhythm In a genetic study drawing on nearly 700,000 people, gene variants linked to being a morning person, scaled to a one-hour earlier sleep midpoint, went with 23% lower odds of major depression. MR IVW OR 0.77 (95% CI 0.63-0.94) per 1-h earlier sleep midpoint; stringent PGC-defined MDD OR 0.73 (0.54-1.00); UK Biobank hospital codes OR 0.64 (0.39-1.06), ns; up to 340 chronotype loci (UK Biobank and 23andMe, n=697,828), scaled by accelerometer midpoint in 85,502; MDD 170,756 cases vs 329,443 controls, European ancestry; Egger intercept null. Observational: Ho 2024 (PMID 38851435) actigraphic rhythm amplitude SMD -0.33 (-0.57 to -0.09).Daghlas 2021 · JAMA Psychiatry · doi
- BRegularity On wrist sensors, people with depression have slightly less regular day-to-day rest and activity than controls, and in UK Biobank, regular sleepers had 38% lower risk of a new depression diagnosis over 7.5 years than irregular sleepers. Interdaily stability SMD -0.17 (95% CI -0.28 to -0.05), actigraphy meta-analysis, 53 studies. Li 2025 (Psychol Med, PMID 40814280): sleep regularity index 71 or more vs 51 or less, HR 0.62 (0.52-0.73), n=79,666, 1,646 incident cases, single 7-day baseline; residual confounding possible.Ho 2024 · J Affect Disord · doi
- BShort sleep Pooling seven long-term studies, adults who reported short sleep had about 31% higher risk of later depression than normal sleepers, and long sleepers about 42% higher. Short sleep RR 1.31 (95% CI 1.04-1.64), I² 0%; long sleep RR 1.42 (1.04-1.92), I² 0%; meta-analysis of 7 prospective studies, 25,271 participants for short sleep; self-reported duration; observational.Zhai 2015 · Depress Anxiety · doi
- CWhole night In 3,247 middle-aged and older adults with one night of home sleep recording, only a larger share of light N1 sleep predicted new depressive symptoms over about five years; REM and deep sleep did not. OR 1.06 per percentage point of N1 (95% CI 1.02-1.10); prospective cohort (Sleep Heart Health Study), single unattended in-home PSG, 225 incident cases of self-reported depressive symptoms over mean 5.3 years; exact onset timing missing.Jiang 2024 · J Affect Disord · doi
- CREM latency In drug-free depressed adults without other disorders, REM sleep starts earlier and carries more eye movements than in healthy people, but a 2024 meta-analysis of depressed adults found both differences lost significance once six poorly matched studies were dropped. SMD (pooled values in full text); 43 articles, 46 REM latency and 22 REM density comparisons, random effects, no publication bias. Contested: Ricciardiello 2024 (Sleep Med, PMID 39515262, 15 studies) found the REM latency and density effects nullified after removing 6 poorly selected case-control studies; Baglioni 2016 (PMID 27416139) found REM latency and REM duration no longer altered in comorbidity-free MDD, although REM density stayed raised. Antidepressants suppress REM and lengthen REM latency, so medicated samples show the opposite.Arıkan 2023 · Sleep Med Rev · doi
Key findings for Bipolar disorder 7 papers
- BLong sleep Between episodes, people with bipolar disorder sleep about 40 minutes longer than healthy people on wrist sensors and overnight recordings, though medication, which can lengthen sleep, was only partly documented. MD 39.69 min (95% CI 25.88-53.50); SMD 0.62 (0.44-0.81); euthymic BD, actigraphy plus PSG (only 2 PSG studies); medication effect only partly testable; sleep diaries show no difference.Marchetti 2025 · Sleep Med Rev · doi
- BRegularity Between episodes, night-to-night swings in sleep are larger than in healthy people, most clearly in time spent awake during the night. Variability SMD for WASO 0.53 (0.19-0.88), sleep latency 0.39 (0.04-0.73), total sleep 0.31 (0.01-0.60); euthymic BD vs controls, objective measures.Marchetti 2025 · Sleep Med Rev · doi
- CLong sleep In 250 young people aged 12 to 30 seeking early mental health care, longer sleep on a wrist sensor roughly doubled the risk of progressing to full bipolar or psychotic disorder over a median 2.5 years. HR 2.05 (95% CI 1.23-3.40) for time to transition; OR 2.23 (1.38-3.74); clinical cohort (Optymise, Sydney), 5-23 days of actigraphy, adjusted for medication use and baseline mania-like and psychotic-like experiences; outcome pools bipolar and psychotic disorders.Carpenter 2026 · Bipolar Disord · doi
- CShort sleep In drug-free mania, sleep on overnight recordings averaged about 312 minutes against 391 in healthy people, roughly 80 minutes less, but this rests on few small studies. Pooled means 311.88 ± 84.76 vs 391.24 ± 42.93 min (p=0.001); PSG meta-analysis of 43 drug-free case-control studies overall, mania subset small (authors call these estimates imprecise and unstable); SMDs only in tables. Marchetti 2025 (mixed PSG and actigraphy) gives -47 min in mania (SMD -0.60, -1.17 to -0.03).Barbato 2026 · Brain Sci · doi
Show 3 moreShow fewer
- CWhole night Across 17 studies of mood-disorder warning signs, sleep disturbance was the early sign of mania that patients reported most often, in a median of 77%. Median prevalence 77%; systematic review, 17 studies, n=1,191; mostly retrospective recall of prodromes, few prospective studies.Jackson 2003 · J Affect Disord · doi
- CTime to fall asleep Between episodes, people with bipolar disorder take about three minutes longer to fall asleep than healthy people on objective measures, a small difference that the two overnight-recording studies did not show. SOL MD 2.99 min (95% CI 1.28-4.70); SMD 0.25 (0.11-0.40); euthymic BD, actigraphy plus PSG; actigraphy alone SMD 0.24 (0.07-0.41); 2 PSG studies ns; sleep diaries larger (SMD 0.71); manic or mixed phase SMD 0.38 (0.08-0.67).Marchetti 2025 · Sleep Med Rev · doi
- DWhole night In a 58-person pilot trial in bipolar I disorder between episodes, bipolar-adapted CBT-I cut manic or hypomanic relapse over six months from 32% to 5% compared with psychoeducation. Relapse 4.6% vs 31.6%; any mood episode 13.6% vs 42.1% (marginal); pilot RCT (30 vs 28) designed for feasibility and effect-size estimates, added to psychiatric care.Harvey 2015 · J Consult Clin Psychol · doi
Key findings for Anxiety 6 papers
- AWhole night Across 154 experiments in 5,717 healthy people, every kind of sleep loss tested, from a full night awake to broken nights, raised anxiety symptoms afterwards by a medium amount. SMD 0.57-0.63 depending on type of sleep loss; preregistered meta-analysis of experimental studies, ages 7-79, 1,338 effect sizes; outcome is self-reported anxiety symptoms in healthy people, not anxiety disorder.Palmer 2023 · Psychol Bull · doi
- BWhole night Across 65 randomised trials in 8,608 people, treatments that improved sleep reduced anxiety by a medium amount, and across mental health outcomes as a whole, the more sleep improved, the more symptoms fell. g+ = -0.51 for anxiety (depression -0.63, composite mental health -0.53); 72 interventions, largely psychological with CBT-I prominent; mixed clinical and non-clinical samples; dose-response by meta-regression; self-reported outcomes, blinding not possible.Scott 2021 · Sleep Med Rev · doi
- BWaking in the night In anxiety disorders without other conditions, overnight recordings show only lower sleep efficiency and shorter sleep; the loss of deep sleep in the broader anxiety group comes mainly from PTSD, which made up 13 of its 21 comparisons. Hedges g by variable; PSG case-control meta-analysis across mental disorders; the anxiety category includes PTSD alongside panic, generalised anxiety and OCD; comorbidity-free subgroup analysis.Baglioni 2016 · Psychol Bull · doi
- CShort sleep In adolescents, poor sleep quality predicted later anxiety and sleep treatments lowered it, but short sleep counted as a yes-or-no category did not predict it. Longitudinal: poor sleep quality OR 1.45; categorical short sleep OR 1.07 (not significant); RCTs (k=19): sleep interventions SMD -0.47; 129 samples, n=416,426; anxiety mostly self-reported.Zhang 2026 · J Affect Disord · doi
Show 2 moreShow fewer
- CRegularity In 79,666 UK Biobank adults who wore a wrist sensor for a week, regular sleepers had a 33% lower risk of a new anxiety diagnosis over 7.5 years than irregular sleepers. HR 0.67 (95% CI 0.58-0.77); prospective cohort, accelerometer sleep regularity index 71 or more vs 51 or less; 2,097 incident cases from health records; single 7-day baseline; residual confounding possible.Li 2025 · Psychol Med · doi
- CWhole night Across six long-term studies, people with insomnia had about three times the odds of later developing an anxiety disorder, an estimate with wide uncertainty. OR 3.23 (95% CI 1.52-6.85); meta-analysis of 13 longitudinal studies with 12 or more months of follow-up; moderate risk of bias.Hertenstein 2018 · Sleep Med Rev · doi
Key findings for PTSD 7 papers
- BWhole night Gene variants linked to insomnia raised the odds of PTSD by about a third, while variants linked to PTSD raised the odds of insomnia only slightly. OR 1.32 (95% CI 1.23-1.40) insomnia to PTSD; reverse OR 1.06 (1.03-1.10); bidirectional two-sample MR. Similar estimate in Kim 2026 (OR 1.32, 1.12-1.56; PMID 42814675) from overlapping UK Biobank and PGC data, so not fully independent.Sun 2022 · Epidemiol Psychiatr Sci · doi
- BDeep sleep Across 31 overnight-recording studies, people with PTSD slept less and less efficiently, spent more time awake and had less slow-wave sleep than healthy controls; REM share was lower only in younger samples. Pooled PSG case-control meta-analysis (34 studies reviewed, 31 pooled); sleep efficiency and SWS% fall with PTSD severity; REM% lower only where mean age was under 30; heterogeneity by medication, adaptation night, scoring rules and type of controls.Zhang 2019 · Sleep Med Rev · doi
- BArousals On wrist sensors (12 studies), people with PTSD have moderately more fragmented sleep than controls and slightly lower sleep efficiency. Fragmentation SMD 0.52 (0.17-0.87, I² 0%); sleep efficiency SMD -0.26 (-0.51 to -0.004); time in bed SMD 0.41; 323 PTSD vs 416 controls; 4 veteran studies; publication bias not checked for all outcomes.Lam 2022 · J Affect Disord · doi
- BWaking in the night Across 31 overnight-recording studies, people with PTSD spent more time awake after first falling asleep than healthy controls, and wrist-sensor studies also show more broken sleep. Pooled PSG case-control meta-analysis (34 studies reviewed, 31 pooled): increased WASO, decreased TST, SWS and sleep efficiency; pooled values in full text; efficiency falls with PTSD severity; heterogeneity by medication, adaptation night and type of controls. Actigraphy: Lam 2022 (PMID 36174789) fragmentation SMD 0.52 (0.17-0.87). The earlier Kobayashi 2007 meta-analysis (PMID 17521374, 20 studies) lists differences only in N1, SWS and REM density.Zhang 2019 · Sleep Med Rev · doi
Show 3 moreShow fewer
- CShort sleep In 1,059 US soldiers followed from before deployment to more than five years after, each extra hour of self-reported sleep before deployment went with about a fifth lower odds of a worsening PTSD course. OR 0.79 (95% CI 0.63-0.98), post-deployment worsening vs asymptomatic trajectory; prospective cohort, PCL-C symptom trajectories; high-chronic (OR 0.87, 0.75-1.02) and mild (0.91, 0.78-1.06) contrasts not significant.Bottari 2026 · J Anxiety Disord · doi
- CWhole night In four small trials in people with PTSD and insomnia, CBT-I reduced PTSD symptoms by a large amount right after treatment; across all the disorders in the review, effects on comorbid symptoms were not significant at follow-up. ES 1.3 (95% CI 0.6-1.9), 4 RCTs, n=216; self-rated outcomes; controls were waitlist, treatment as usual or placebo; across all comorbidities no significant effect on comorbid symptoms at follow-up.Hertenstein 2022 · Sleep Med Rev · doi
- CREM density An earlier meta-analysis of 20 overnight-recording studies found more light N1 sleep, less slow-wave sleep and more eye movements during REM in PTSD. Meta-analysis, 20 PSG case-control studies; studies with more men or less comorbid depression found larger differences; REM density is not among the pooled changes reported in the larger Zhang 2019 analysis.Kobayashi 2007 · Psychophysiology · doi
Key findings for Schizophrenia 10 papers
- BSpindles In a new sample of 103 patients and 68 controls, the spindle and slow-oscillation deficits found in an earlier sample all replicated; medication explained some of the differences between patients but not the gap with controls. Replication of Kozhemiako 2022 (eLife, PMID 35578829; 72 patients, 58 controls; largest fast-spindle deficit 32%, -1.27 SD at FC2); combined n=301 (175 patients); high-density EEG, N2; group differences persisted after controlling for medication; olanzapine linked to exaggerated age effects. The reported AUC of 0.93 is in-sample and should not be read as clinical accuracy.Kozhemiako 2025 · Sleep · doi
- BTime to fall asleep Across 31 overnight-recording studies, people with schizophrenia took longer to fall asleep and slept less and less efficiently; stage changes were not found in never-medicated patients, and less slow-wave sleep appeared only after about three years of illness. Hedges g; 574 patients, 515 controls, searched to January 2014; spindle findings were inconsistent at that time; the medication-naive moderator analysis rests on few studies. Morra 2025 (Sleep Med Rev, PMID 40706098, 92 studies) did find shorter REM latency in drug-naive patients.Chan 2016 · Sleep Med Rev · doi
- BWhole night In 3,755 UK students with insomnia, digital CBT-I slightly reduced paranoia and hallucinations over ten weeks, and better sleep explained part of that change. Paranoia d 0.19, hallucinations d 0.24, insomnia d 1.11; single-blind RCT vs usual care (OASIS), 26 universities; outcomes are subclinical psychotic experiences, not psychotic disorder; insomnia was a mediator.Freeman 2017 · Lancet Psychiatry · doi
- BShort sleep Across 20 overnight-recording studies of people with schizophrenia who were not taking antipsychotics, patients slept less and less efficiently and took longer to fall asleep than controls, with larger differences in those recently withdrawn from medication than in those never treated. Cohen d (values in full text); 321 unmedicated patients vs 331 controls; increased sleep latency, decreased TST and sleep efficiency; never-treated patients alone showed more total awake time and less stage 2. Chan 2016 (PMID 27061476; 31 studies, 574 patients, mixed medication status) also found shorter TST. Medicated patients show longer sleep on actigraphy (Aronica 2025).Chouinard 2004 · Schizophr Bull · doi
Show 6 moreShow fewer
- BWaking in the night Across 31 overnight-recording studies, people with schizophrenia spent more time awake after first falling asleep and slept less efficiently than healthy controls, and this continuity loss was similar however long antipsychotics had been stopped. Hedges g (values in full text); 574 patients, 515 controls, searched to January 2014; random effects; withdrawal over 8 weeks reduced architecture but not continuity abnormalities; moderator analyses rest on few studies. Chouinard 2004 (PMID 15954201) found more total awake time only in never-treated patients; Baglioni 2016 (PMID 27416139) found poor continuity in schizophrenia.Chan 2016 · Sleep Med Rev · doi
- BREM latency Across 92 overnight-recording studies, REM sleep started earlier in drug-naive and drug-free people with schizophrenia than in healthy controls, to a degree similar to depression, while antipsychotic treatment appeared to bring it back toward normal. SMD, random effects (values in full text); 1,845 patients, 1,269 controls, 262 depressed patients; studies 1955-2025. Conflict: Chan 2016 (PMID 27061476) found no architecture change in medication-naive patients, and Baglioni 2016 (PMID 27416139) found the REM latency reduction not significant in schizophrenia without comorbidity; comorbid depression may contribute.Morra 2025 · Sleep Med Rev · doi
- CLong sleep On wrist sensors, people with schizophrenia-spectrum disorders record about 106 minutes more sleep and about two hours more in bed than controls; antipsychotic use explained most of the variation between studies. MD total sleep 106.13 min (86.02-124.24); time in bed 121.58 min (88.16-155.00); 18 studies, 584 patients vs 582 controls; antipsychotics explained most heterogeneity (pseudo-R² 88%); opposite in direction to shorter sleep on PSG.Aronica 2025 · BMJ Ment Health · doi
- CWhole night In 294,856 UK Biobank adults followed almost 14 years, an unhealthy self-reported sleep pattern went with about 75% higher risk of a new schizophrenia diagnosis, highest in those with high genetic risk, but only 231 people developed it. HR 1.75 (95% CI 1.15-2.64); prospective cohort; sleep score from long sleep, evening chronotype, napping and difficulty waking; high polygenic risk plus unhealthy sleep HR 5.80 (2.55-13.20); reverse causation from a prodrome possible.Li 2025 · J Affect Disord · doi
- CWhole night In three small trials (137 people) with schizophrenia and insomnia, CBT-I improved insomnia strongly but had no clear effect on overall schizophrenia symptoms. Insomnia SMD -1.53 (-2.04 to -1.02); overall symptoms SMD -0.29 (-0.70 to 0.12); GRADE low; CBT-I vs treatment as usual.Furukawa 2025 · J Psychiatr Res · doi
- CREM density In the same 92-study analysis, eye movements during REM were more frequent in drug-free and treated schizophrenia than in controls, and unlike REM timing, they did not return toward normal with antipsychotics. SMD (values in full text); abstract reports the increase for drug-free and treated patients only; number of REM density studies not in abstract and likely small; Baglioni 2016 (PMID 27416139) found no increase in the combined REM pressure domain in schizophrenia.Morra 2025 · Sleep Med Rev · doi
Key findings for ADHD 7 papers
- BWhole night In 244 children aged 5 to 12 with ADHD, two sessions of behavioural sleep advice modestly reduced ADHD symptoms at three and six months, with about a third to a half of the gain running through better sleep. Adjusted MD -2.9 (ES -0.3) at 3 months and -3.7 (95% CI -6.1 to -1.2; ES -0.4) at 6 months on ADHD-RS-IV; RCT in 21 paediatric practices; most children on stimulants; parents not blinded, teachers also reported better behaviour.Hiscock 2015 · BMJ · doi
- BLimb movements Periodic leg movements in sleep were not more frequent in children with ADHD, in a case-control study and in a meta-analysis of 33 studies. Case-control 24 vs 22 children; meta-analysis of 33 studies; result held across PLMS scoring definitions.Fulda 2023 · Sleep · doi
- BTime to fall asleep Pooling 44 studies of children, those with ADHD took about four minutes longer to fall asleep than typically developing children, and the difference held on overnight recordings as well as wrist sensors. SMD 0.269 (p=0.003; 27.7 vs 23.4 min); 2,239 ADHD vs 57,181 typically developing; PSG, actigraphy, EEG and questionnaires; PSG subgroup significant; GRADE low to very low; no missing studies by trim-and-fill for latency. Liang 2023 (Psychiatry Res, PMID 37657199; 45 articles) found longer latency on actigraphy but not PSG; adults: Díaz-Román 2018 actigraphy SMD 0.80.Xian 2025 · Psychol Med · doi
- CWaking in the night Pooling 44 studies of children, those with ADHD spent only about three more minutes awake at night but woke noticeably more often than typically developing children. WASO SMD 0.294 (31.9 vs 28.7 min); awakening index SMD 0.781; 2,239 ADHD vs 57,181 typically developing; mixed PSG, actigraphy, EEG and questionnaires; GRADE low to very low; possible publication bias for WASO.Xian 2025 · Psychol Med · doi
Show 3 moreShow fewer
- CTime to fall asleep In adults with ADHD, wrist sensors show longer time to fall asleep and lower sleep efficiency, but overnight lab recordings show no differences. Actigraphy sleep latency SMD 0.80 (0.46-1.14); efficiency -0.68 (-1.03 to -0.34); 13 studies in total, actigraphic subset small; subjective differences larger (SMD 0.56-1.55).Díaz-Román 2018 · Neurosci Biobehav Rev · doi
- CCircadian rhythm Across 62 studies, ADHD was consistently linked to a later chronotype and a later evening rise in melatonin. Systematic review, 62 studies, 4,462 patients; no pooled effect size; melatonin evidence mainly for sleep-onset problems.Coogan 2017 · Atten Defic Hyperact Disord · doi
- CWhole night In a 2026 genetic study, variants linked to insomnia symptoms were associated with a much higher risk of ADHD, but other genetic studies disagree. OR 2.50 (95% CI 1.65-3.78), Bonferroni-significant; two-sample MR with self-reported UK Biobank insomnia instruments; conflicts with Gao 2019 (null) and Dai 2024 (ADHD leads to sleep disruption); Sun 2022 OR 1.33. DOI as indexed in PubMed (10.1159/phg/aenag005) looks malformed (Karger PII 000554501); check before linking.Kim 2026 · Public Health Genomics · doi
Key findings for Autism 6 papers
- BTime to fall asleep In medication-naive autistic children and adolescents (16 studies), objective recordings show longer time to fall asleep and less total sleep. Sleep latency g 0.59 (0.26-0.92); efficiency -0.58; total sleep -0.64; 981 ASD vs 1,220 typically developing; possible publication bias for PSG latency and total sleep.Kim 2023 · Epidemiol Psychiatr Sci · doi
- BTime to fall asleep On wrist sensors (28 studies), autistic youth take moderately longer to fall asleep and sleep slightly less than peers. Sleep latency g 0.514 (0.351-0.677); total sleep g -0.332 (-0.574 to -0.090); efficiency -0.424; 805 ASD vs 1,573 peers, ages 5-17.Liang 2023 · JAMA Netw Open · doi
- BShort sleep On overnight recordings, autistic people sleep less and less efficiently and spend more time in light N1 sleep; REM did not differ in this analysis. Total sleep d -0.52 (-0.97 to -0.08); efficiency d -0.69 (-1.27 to -0.11); N1% d 0.93 (0.37-1.48); 11 PSG case-control studies pooled.Chen 2020 · Psychiatry Res · doi
- BShort sleep In 125 autistic children whose sleep had not improved with behavioural advice, prolonged-release melatonin added about 32 minutes of sleep a night beyond placebo over 13 weeks. Adjusted difference 32.43 min (reported as -32.43, p=0.034) on caregiver sleep diaries; double-blind RCT, ages 2-17.5 (3.2% Smith-Magenis syndrome); sleep latency -25.3 min; industry-sponsored.Gringras 2017 · J Am Acad Child Adolesc Psychiatry · doi
Show 2 moreShow fewer
- CShort sleep In an Australian birth cohort, each extra hour or so of night sleep at 6 months went with 22% lower odds of an autism diagnosis by age 11.5. aOR 0.78 (95% CI 0.65-0.94) per 10% (about 1 hour) more night sleep; n=925 at 6 months, 64 diagnoses; parent-reported Brief Infant Sleep Questionnaire; authors note shared determinants cannot be excluded.Mohamed 2025 · Arch Dis Child · doi
- CREM amount In 193 autistic children matched to 193 children from the same clinical sleep databank, REM made up a slightly smaller share of the night (18% against 19%) and started later. REM% 17.8 ± 6.4 vs 19.1 ± 7.0 (p=0.049); REM latency median 134.5 vs 116.0 min (p=0.009); retrospective; comparison children were clinically referred, not healthy; small absolute differences.Ma 2026 · Autism Res · doi
Key findings for Alcohol use 9 papers
- BREM latency Two to four weeks into abstinence, people with alcohol use disorder reach REM about 14 minutes sooner than controls and spend a slightly larger share of the night in REM. REM latency MD -13.62 min (-15.80 to -11.44); REM share of sleep period +2.70 points (1.47-3.93); total sleep -14.32 min; WASO not different; 24 AUD studies (15 PSG), mostly inpatients and men; fixed-effect pooling where I² was under 50% narrows the intervals; some funnel asymmetry; SWS interval printed as -4.99 to 2.38.Oliva 2026 · EClinicalMedicine · doi
- BWhole night Across 8 trials in 426 adults who drink heavily or have alcohol use disorder, CBT-I cut insomnia severity by about 5.5 points, but drinking outcomes did not differ. ISI MD -5.51 (-7.13 to -3.90) post-treatment; -4.55 at 6 months (4 studies); multilevel meta-analysis; alcohol outcomes reported inconsistently, none significant; 69% men.Türkmen 2025 · Sleep Med Rev · doi
- BShort sleep On wrist sensors, people with alcohol use disorder sleep about 45 minutes less than controls, and people with opioid use disorder about 40 minutes less. AUD MD -44.67 min (-56.10 to -33.24); OUD MD -40.00 (-72.08 to -7.92); 9 studies, n=1,366; actigraphy and PSG broadly concordant.Paredes Naveda 2026 · Sleep Med Rev · doi
- BWhole night In 27 laboratory studies of healthy adults, alcohol before bed delayed the first REM period by about 18 minutes and cut REM sleep by about 11 minutes, and both effects appeared from about two drinks and grew with the dose. REM latency MD +18.0 min (8.8-27.1), k=29 effect sizes, n=212 (14 studies), 95% PI -18.8 to 54.8; +30.1 min per g/kg (6.4-53.8); REM duration MD -11.3 min (-20.8 to -1.8), k=13, n=96; -40.4 min per g/kg (-60.0 to -20.8); REM share -2.8 points (-3.9 to -1.7), k=32, n=255; dose thresholds ~0.35 g/kg for REM latency and ~0.50 g/kg for REM duration (paper converts to two to three standard drinks at 62 kg and 10 g per drink); alcohol mostly within 3 h of bedtime; 25 of 27 crossover; 12 studies men only, 4 women only; time in bed standardised or comparable in only 4 studies.Gardiner 2025 · Sleep Med Rev · doi
Show 5 moreShow fewer
- BWhole night In the same studies, alcohol shortened the time to fall asleep and to reach deep sleep only at about five or more drinks, total sleep, waking after sleep onset and sleep efficiency did not differ on average, and the few studies that split the night found more deep sleep early and lighter sleep late. Sleep onset latency MD -2.0 min (-4.3 to 0.2), k=34, n=271, NS; -6.4 min per g/kg (-9.3 to -3.5), significant from ~0.85 g/kg; N3 latency -2.8 min (-5.4 to -0.1), k=12; TST -10.1 min (-24.6 to 4.4), PI -71.5 to 51.3; WASO +4.0 min (-2.3 to 10.3); sleep efficiency -0.7% (-1.8 to 0.3); N3 duration and share not different; first-half N3 increase and second-half N1 increase from 2-3 single studies, not pooled.Gardiner 2025 · Sleep Med Rev · doi
- CApnea events Across 28 studies, alcohol users had a higher apnea-hypopnea index than non-users, and opioid users more central sleep apnea. AHI SMD 0.46 (0.21-0.71) for alcohol; central apnea SMD 0.44 (0.24-0.64) for opioids; tobacco higher but not significant; users vs non-users, not disorder diagnoses.Mauries 2026 · Sleep Med Rev · doi
- CWhole night Gene variants linked to insomnia raised the risk of alcohol dependence, starting cannabis and smoking, and starting smoking in turn raised insomnia risk. Bidirectional two-sample MR; positive effects of insomnia liability on smoking traits, alcohol dependence and cannabis initiation (not drinks per week); smoking initiation to insomnia positive; no effect of alcohol or cannabis on insomnia.Pasman 2020 · Drug Alcohol Depend · doi
- CREM amount Two to four weeks into abstinence, people with alcohol use disorder spend a slightly larger share of the sleep period in REM than controls, but no such difference was found for nicotine or cocaine use disorders. REM/SPT MD +2.70 percentage points (95% CI 1.47-3.93), AUD vs controls, mostly inpatients and men; fixed-effect pooling where I² was under 50%; NUD REM/SPT MD 0.01 (-1.66 to 1.69); CoUD REM ns. Direction depends on phase: acute alcohol reduces REM in healthy adults (Gardiner 2024, PMID 39631226) and early cannabis abstinence raises it.Oliva 2026 · EClinicalMedicine · doi
- DREM latency In 45 male veterans with alcoholism, short REM latency and more REM at admission predicted relapse within three months, correctly sorting about 80% of patients. Discriminant function 80% (36/45) on the combined sample; two consecutive cohorts (28 and 17) cross-validated at 76.5-82.1%; nondepressed primary alcoholism; 16 relapsers in total.Gillin 1994 · Arch Gen Psychiatry · doi
Key findings for High blood pressure 6 papers
- AApnea events Across 75 trials, CPAP lowered night-time systolic pressure by about 3.5 mmHg and clinic pressure by about 2.5 mmHg, with larger falls in severe apnea and in people who used it five or more hours a night. Mean difference -3.5 mmHg (95% CI -4.4 to -2.5) nocturnal SBP, 32 RCTs; office SBP -2.5 (-3.8 to -1.2), 40 RCTs; random-effects meta-analysis of 75 RCTs (10,025 patients) vs inactive control. Use of 5 h/night or more: nocturnal -4.2 vs -1.6 mmHg (subgroup comparison, not randomized by adherence).Benning 2025 · J Sleep Res · doi
- BApnea events Pooling individual patients from 36 trials, CPAP lowered blood pressure only in people whose clinic pressure was uncontrolled at the start, and the effect did not depend on how severe their night-time oxygen drops were. Individual patient data meta-analysis of 36 parallel RCTs (n=9,434); BP reduction confined to uncontrolled office SBP; no difference by severe nocturnal hypoxia. Abstract numbers not clearly assigned to a subgroup, so no estimate stored.Pengo 2025 · Eur Respir J · doi
- BShort sleep Across cohort studies, people who reported sleeping under six hours had about 17% higher risk of developing high blood pressure, and about 24% higher if they were under 60. RR 1.17 (95% CI 1.06 to 1.28) for <6 h; age <60 y RR 1.24 (1.10 to 1.39); meta-analysis of cohorts, n=173,734 (41,528 incident cases); self-reported duration.Qi 2025 · J Gen Intern Med · doi
- BWhole night Pooled genetic studies suggest that a genetic tendency to insomnia raises the risk of high blood pressure by about 16%. OR 1.16 (95% CI 1.13 to 1.18), I2 28%; meta-analysis of Mendelian randomization studies within a meta-review (9 MR studies, 4 observational meta-analyses); insomnia self-reported in source GWAS.Zhang 2024 · J Clin Sleep Med · doi
Show 2 moreShow fewer
- CShort sleep In a general-population sample with an in-lab sleep study, insomnia raised the odds of later treated high blood pressure only when the study measured under six hours of sleep, and apnea on its own roughly tripled them. Insomnia with objective short sleep OR 2.27 (95% CI 1.29 to 4.00); OSA alone (AHI 5 or more) OR 3.31 (1.85 to 5.92); OSA plus insomnia with short sleep OR 4.25 (1.52 to 11.90); insomnia with normal sleep duration ns; vs good sleepers; Penn State Adult Cohort, n=786 normotensive, 7.5 y; incident hypertension = self-reported treatment.Athanasiou 2026 · J Clin Sleep Med · doi
- CDeep sleep In the Sleep Heart Health Study, people with the least deep sleep (under 9.8% of the night) had 69% higher odds of new high blood pressure over about five years than people in the middle range, after accounting for apnea and broken sleep. OR 1.69 (95% CI 1.21 to 2.36), N3 quartile 1 (<9.8%) vs quartile 3 (17.7 to 25.2%); quartile 2 OR 1.45 (1.04 to 2.00); n=1,850, home PSG, mean follow-up 5.3 y; adjusted for AHI, oxygen desaturation, arousal index, sleep efficiency, waist.Javaheri 2018 · Sleep · doi
Key findings for Cardiovascular disease 6 papers
- AApnea events Pooling individual patients from the three main CPAP trials in people with heart disease, CPAP did not prevent repeat cardiovascular events overall. The lower risk in people who used it four or more hours a night came from a non-randomized analysis. HR 1.01 (95% CI 0.87 to 1.17) first MACCE, CPAP vs usual care; IPD meta-analysis of SAVE, ISAACC and RICCADSA, n=4,186, 82% men, mean AHI 31, mean adherence 3.1 h/day, follow-up 3.25 y; on-treatment analysis for 4 h/day or more HR 0.69 (0.52 to 0.92) by marginal structural model (not randomized).Sánchez-de-la-Torre 2023 · JAMA · doi
- BApnea events Across 18 long-term cohort studies, sleep apnea went with 82% higher cardiovascular risk, rising to about 2.5 times in severe apnea, while mild apnea showed no clear excess. HR 1.82 (95% CI 1.45 to 2.28); severe HR 2.45 (1.85 to 3.25); moderate 1.56 (1.20 to 2.03); mild 1.21 (0.98 to 1.50); 18 prospective cohorts, >25,000 participants, median 9 y. The CPAP-adherence HR 0.76 in the same abstract is observational.Craciun 2025 · Medicina (Kaunas) · doi
- COxygen drops Re-analysing three CPAP trials, CPAP worked better in people whose sleep study showed large oxygen dips or heart-rate surges after each pause, and looked harmful in low-risk people, which may explain the overall null. Interaction HR 0.69 (95% CI 0.50 to 0.95), CPAP effect in high-risk vs low-risk OSA; high risk = pulse-rate response >9.4 bpm or hypoxic burden >87.1 %min/h (top tertiles); post hoc pooled analysis of RICCADSA, ISAACC and SAVE, n=3,549 (1,832 high risk); secondary-prevention trials, mostly non-sleepy. The iHR is a ratio of treatment effects, not the CPAP HR within high-risk patients.Azarbarzin 2026 · Eur Heart J · doi
- COxygen drops In two community cohorts, the hypoxic burden predicted death from cardiovascular disease while the standard apnea count did not; older men in the top fifth had about 2.7 times the risk. HR 2.73 (95% CI 1.71 to 4.36), highest vs lowest quintile of hypoxic burden, MrOS (2,743 men, mean age 76); SHHS highest quintile HR 1.96 (1.11 to 3.43), n=5,111; outcome CVD mortality; AHI not predictive.Azarbarzin 2019 · Eur Heart J · doi
Show 2 moreShow fewer
- CShort sleep In UK Biobank adults who wore a wrist device, insomnia together with below-median sleep (under 7.3 hours) raised new cardiovascular disease risk by 39%; below-median sleep without insomnia did not. HR 1.39 (95% CI 1.11 to 1.74) incident CVD, n=6,959, mean 3.6 y; cerebrovascular disease HR 1.76 (1.29 to 2.39), n=7,690; 'short' = below the sample median of 7.3 h by accelerometer, not a clinical short-sleep cut-off; insomnia self-reported.Pejovic 2026 · J Am Heart Assoc · doi
- CWhole night Pooled genetic studies suggest a genetic tendency to insomnia raises coronary artery disease risk by about 14%, though the individual studies disagree a lot. OR 1.14 (95% CI 1.10 to 1.19), I2 97%; meta-analysis of Mendelian randomization studies within a meta-review; insomnia self-reported in source GWAS.Zhang 2024 · J Clin Sleep Med · doi
Key findings for Atrial fibrillation 6 papers
- CWhole night In 1.78 million Japanese adults, a recorded insomnia diagnosis went with 14% higher risk of new atrial fibrillation, more so in people under 65 and in women. Adjusted HR 1.14 (95% CI 1.10 to 1.18); nationwide claims database (DeSC), insomnia by ICD-10 code F51.0 or G47.0, n=1,780,764 without prior CVD; no sleep measurement.Masuda 2026 · J Am Heart Assoc · doi
- CWaking in the night Among 27,232 sleep-clinic patients, each extra hour awake after falling asleep went with 11% more new atrial fibrillation and each hour less sleep with 8% more, after accounting for apnea and oxygen loss; the arousal count was not linked. HR 1.11 (95% CI 1.05 to 1.16) per hour WASO; TST HR 1.08 (1.04 to 1.11) per hour less; sleep efficiency HR 1.06 per 10-point drop; arousal index ns; Cleveland Clinic in-lab PSG, 7.8 y, AF by diagnosis code; adjusted for AHI and hypoxic impact.Heinzinger 2025 · Sleep · doi
- COxygen drops Among 43,433 sleep-clinic patients, a sleep pattern dominated by oxygen loss carried 48% more new atrial fibrillation over 7.6 years than a pattern of long sleep with plenty of REM. HR 1.48 (95% CI 1.34 to 1.64), hypoxia latent-class subtype vs long sleep plus REM reference; retrospective clinical PSG registry (STARLIT), 3,595 incident AF.Heinzinger 2024 · JACC Clin Electrophysiol · doi
- CApnea events In a trial with implanted heart monitors, five months of CPAP did not reduce time in atrial fibrillation in people with intermittent atrial fibrillation and moderate to severe apnea. Adjusted between-group difference -0.63 percentage points AF burden (95% CI -2.55 to 1.30); open-label RCT with blinded outcome, n=108; AHI 15 or more by respiratory polygraphy; 25% of the CPAP run-in did not tolerate treatment.Traaen 2021 · Am J Respir Crit Care Med · doi
Show 2 moreShow fewer
- CShort sleep Cohort studies link short sleep to slightly more atrial fibrillation, but genetic analysis found no effect of genetically short sleep, so the genetic evidence does not support a causal link. Cohort meta-analysis RR 1.06 (95% CI 1.02 to 1.11), 6 studies; MR IVW OR 0.979 (0.880 to 1.089), 16 loci.Chen 2023 · Eur J Intern Med · doi
- DWhole night In 218 pacemaker patients monitored every day, a long run of atrial fibrillation was followed by higher odds of severe apnea the next night, and a night of severe apnea by higher odds of a new episode the next day. OR 2.52 (95% CI 1.97 to 3.22) for 12 h or more of AF burden preceding next-night severe device-detected apnea; OR 1.35 (1.10 to 1.64) severe apnea preceding next-day AF onset; case-crossover, 112,242 patient-days; apnea = pacemaker respiratory disturbance index of 20 or more, not PSG.Spitaler 2026 · Heart Rhythm · doi
Key findings for Type 2 diabetes 6 papers
- BApnea events Across 14 trials in people with both apnea and type 2 diabetes, CPAP did not improve long-term blood sugar or fasting glucose, though insulin resistance fell with six months or more of use and weight crept up. HOMA-IR MD -1.67 (95% CI -2.57 to -0.77) with 24 weeks or more of CPAP; no significant effect on HbA1c, fasting or mean glucose; BMI +0.74 kg/m2 (0.20 to 1.27) long term; meta-analysis of 14 RCTs, n=1,273.Wu 2026 · BMJ Open Respir Res · doi
- BWhole night A genetic study that tested 97 possible risk factors found that a genetic tendency to insomnia raised type 2 diabetes risk by 17%, and the result held in a separate Finnish dataset. OR 1.17 (95% CI 1.11 to 1.23); two-sample MR, DIAGRAM 74,124 cases and 824,006 controls; FinnGen replication; short sleep only suggestive (p<0.05); insomnia self-reported.Yuan 2020 · Diabetologia · doi
- CRegularity In 73,630 UK Biobank adults who wore a wrist device for a week, irregular sleepers had 38% higher risk of type 2 diabetes over 8 years, even when they slept seven hours or more. HR 1.38 (95% CI 1.20 to 1.59), Sleep Regularity Index <71.6 vs >87.3; among those sleeping 7 h or more HR 1.35 (1.09 to 1.66); diabetes from self-report and health records.Chaput 2024 · Diabetes Care · doi
- CApnea events Among 8,678 sleep-clinic patients, severe apnea carried 30% higher risk of developing diabetes, and time spent below 90% oxygen also predicted it. HR 1.30, AHI >30 vs <5 (CI not given in abstract); REM AHI and time with SpO2 <90% also associated; Toronto clinical cohort, median 67 months, outcome from health administrative data; adjusted for BMI.Kendzerska 2014 · Am J Respir Crit Care Med · doi
Show 2 moreShow fewer
- CSpindles In over 600 adults, with a replication in over 1,900 more, tighter timing between slow waves and sleep spindles went with better fasting blood sugar the next morning, more than standard sleep measures did. Cross-sectional PSG cohorts; slow-oscillation and spindle coupling the strongest sleep predictor of next-day fasting glucose; pathway via insulin sensitivity; no effect size in abstract; outcome next-day glucose, not diabetes incidence.Vallat 2023 · Cell Rep Med · doi
- CDeep sleep Suppressing deep sleep with sounds all night, without shortening total sleep, sharply lowered insulin sensitivity in healthy young adults. Experimental crossover; all-night selective SWS suppression with no change in total sleep time; marked decrease in insulin sensitivity without compensatory insulin release; small sample, no effect size in abstract.Tasali 2008 · Proc Natl Acad Sci U S A · doi
Key findings for Obesity 7 papers
- BApnea events In two year-long trials in people with obesity and moderate to severe apnea, the weight-loss drug tirzepatide cut breathing pauses by about 20 to 24 an hour more than placebo. Treatment difference in AHI -20.0 events/h (95% CI -25.8 to -14.2), trial 1 (no PAP); -23.8 (-29.6 to -17.9), trial 2 (on PAP); 52-week double-blind phase 3 RCTs (SURMOUNT-OSA); baseline AHI about 50, BMI about 39; hypoxic burden, hsCRP and SBP also improved; funded by Eli Lilly.Malhotra 2024 · N Engl J Med · doi
- BApnea events A genetic study found that higher genetically predicted adult body mass index more than doubled the odds of sleep apnea, and larger childhood body size and birth weight raised them too. OR 2.419 per genetically predicted increment of adult BMI (scale and CI not in abstract), p<0.001; childhood BMI OR 1.218, birth weight OR 1.142; two-sample MR, FinnGen outcome (sleep apnoea, all types), replication dataset.Zhang 2026 · Intern Emerg Med · doi
- BShort sleep Across 11 long-term studies of adults, short sleepers were about 45% more likely to develop obesity than normal sleepers, while long sleep showed no link. Pooled OR 1.45 (95% CI 1.25-1.67), short sleep; 1.25 (1.14-1.38) after removing 3 studies driving heterogeneity; European subgroup ns (1.45, 0.79-2.64); long sleep OR 1.06 (0.98-1.15); meta-analysis of prospective studies, 197,906 participants for short sleep; observational, reverse causation possible. Children: Miller 2018 (Sleep, PMID 29401314), 42 studies, RR 1.30-2.23 by age band.Wu 2014 · Sleep Med · doi
- CApnea events In the other direction, a genetic tendency to sleep apnea raised body mass index modestly, while higher body mass index roughly doubled apnea risk. OSA to BMI OR 1.275 (95% CI 1.070 to 1.520); BMI to OSA OR 1.929 (1.808 to 2.059); bidirectional MR, FinnGen 38,998 OSA cases, GIANT BMI n=681,275. OSA genetic instruments may carry adiposity pleiotropy.Cai 2025 · Respir Med · doi
Show 3 moreShow fewer
- CShort sleep In a trial of adults with overweight who slept under 6.5 hours, sleep counselling added about 1.2 hours of sleep a night and cut energy intake by 270 kcal a day in everyday life. MD -270 kcal/day (95% CI -393 to -147); single-centre RCT, n=80, age 21 to 40; intake from doubly labelled water plus change in body energy stores; sleep by actigraphy; short intervention, surrogate outcome.Tasali 2022 · JAMA Intern Med · doi
- CWhole night Pooling 8 studies of insomnia therapy or sleep-hygiene advice, body mass index fell by about 0.6 units, while sleep-extension studies showed no clear change; most participants came from type 2 diabetes studies. MD -0.64 kg/m2 (95% CI -1.01 to -0.28), 8 studies, n=1,384, I2 16%; sleep extension 5 studies, n=138, ns; includes pre-post designs; risk of bias low in 12 studies and some concerns in 10; most participants from type 2 diabetes studies.Mostafa 2026 · Obes Rev · doi
- CDeep sleep Over 15 years of repeated lab sleep studies, people whose deep and REM sleep fell over time gained body mass index faster, independent of sleep length. Wisconsin Sleep Cohort, 1,187 people, 4,862 PSGs, mean 14.9 y; within-person N3 and REM decreases associated with BMI gain; linear mixed models; no effect size in abstract.Reither 2021 · Sleep · doi
Key findings for Kidney disease 6 papers
- BApnea events Pooling observational studies, sleep apnea went with 77% higher risk of developing kidney disease, and kidney disease with 60% higher risk of developing apnea. Incident CKD RR 1.77 (95% CI 1.56 to 2.01); CKD to incident OSA RR 1.60 (1.35 to 1.89); meta-analysis of 63 observational studies overall (26.8 million patients), incident-outcome subsets smaller; restless legs and CKD RR 1.88 (1.48 to 2.38).Koh 2024 · Clin Kidney J · doi
- BShort sleep Genetic risk for short sleep, but not long sleep, was linked to chronic kidney disease, supporting short sleep as a cause. OR 1.80 (95% CI 1.25 to 2.60) per unit of genetic risk score (a two-fold increase in odds of short sleep); one-sample MR in UK Biobank (n=321,260) with two-sample replication in CKDGen; short sleep self-reported <6 h; CKD stages 3 to 5.Park 2020 · J Am Soc Nephrol · doi
- BShort sleep Across 42 studies and 2.6 million people, sleeping five hours or less went with 46% higher risk of developing kidney disease, and nine hours or more did too. RR 1.46 (95% CI 1.22 to 1.76) for 5 h or less vs 7 to 8 h; 9 h or more RR 1.46 (1.28 to 1.68); meta-analysis, mostly self-reported duration; long-sleep MR null (Park 2020).Koh 2024 · Clin Kidney J · doi
- BWhole night In UK Biobank, frequent insomnia went with 21% higher risk of kidney disease, and two kinds of genetic analysis pointed the same way. Observational HR 1.21 (95% CI 1.17 to 1.25); one-sample MR OR 1.35 (1.02 to 1.79); two-sample MR OR 1.06 (1.02 to 1.11); insomnia self-reported.Wang 2025 · Mayo Clin Proc · doi
Show 2 moreShow fewer
- CApnea events Across four trials, CPAP did not significantly slow the loss of kidney function; only non-randomized studies showed a small benefit. eGFR MD 1.42 mL/min/1.73 m2 (95% CI -0.99 to 3.82) in RCTs; non-randomized studies 1.95 (0.80 to 3.11); 4 RCTs and 7 non-RCTs, n=2,660; UACR ns; low to moderate certainty.Peng 2025 · Sleep Breath · doi
- CApnea events Genetic analysis found no causal effect of sleep apnea on kidney disease or filtration rate, only a small one on blood urea, while poorer genetically predicted kidney function raised apnea risk. OSA to CKD OR 1.075 (95% CI 0.916 to 1.263); OSA to eGFRcrea and eGFRcys ns; OSA to BUN beta 0.040 (0.013 to 0.067); eGFRcys to OSA OR 0.687 (0.497 to 0.950); BUN to OSA OR 1.686; two-sample bidirectional MR, FinnGen OSA.Liu 2025 · Nat Sci Sleep · doi
Key findings for Death, any cause 7 papers
- BShort sleep Across 79 cohort studies, sleeping under seven hours went with 14% higher mortality and nine hours or more with 34% higher, compared with seven to eight hours. HR 1.14 (95% CI 1.10 to 1.18) for <7 h; HR 1.34 (1.26 to 1.42) for 9 h or more; random-effects meta-analysis of 79 cohorts; mostly self-reported duration.Ungvari 2025 · Geroscience · doi
- BApnea events Pooling 10 trials and 3 observational studies, CPAP did not reduce deaths from any cause. All-cause mortality RR 0.92 (95% CI 0.72 to 1.16), I2 0%; 13,832 patients; MACE RR 0.73 (0.52 to 1.03) ns. The reported benefit with 4 h or more of use applies to MACE and CV mortality, not all-cause death, and is an adherence subgroup.Feng 2024 · Sleep Breath · doi
- BApnea events Across 16 cohort studies, people with severe sleep apnea had about 54% higher risk of death from any cause than those with the least apnea, while mild apnea showed no link. RR 1.54 (95% CI 1.21-1.97), severe vs lowest OSA category; mild ns; meta-analysis of 16 prospective cohorts, n=24,308 (11 in general or healthy samples, 5 in patient groups). Similar: Wang 2013 (Int J Cardiol, PMID 24161531) RR 1.92 (1.38-2.69), 12 cohorts; Pan 2016 (PMID 26779904) severe HR 1.60, mild and moderate ns. CPAP trials have not reduced all-cause death (Feng 2024).Xie 2017 · BMJ Open · doi
- CApnea events A 2025 review of over 1.1 million people found 37% lower mortality with PAP therapy, but 20 of its 30 studies were non-randomized and it was funded by ResMed, a CPAP maker. HR 0.63 (95% CI 0.56 to 0.72); 10 RCTs and 20 confounder-adjusted non-randomized studies, n=1,175,615; funded by ResMed, first author a ResMed employee; healthy-adherer bias likely in the non-randomized studies.Benjafield 2025 · Lancet Respir Med · doi
Show 3 moreShow fewer
- CWhole night In 451,420 UK Biobank adults, short sleep, long sleep and insomnia all predicted death, yet genetic analysis found no causal effect of any of them. Observational HR short sleep 1.246 (95% CI 1.195 to 1.298), long sleep 1.735, insomnia 1.117; MR null for duration, short, long, insomnia, chronotype and sleepiness; only napping MR OR 1.219 (1.071 to 1.387), possibly illness-related.Zhang 2025 · Rejuvenation Res · doi
- CREM amount Each 5% less REM sleep went with 13% higher mortality in older men, and the finding held in a younger group of men and women followed for 21 years. HR 1.13 (95% CI 1.08 to 1.19) per 5% REM reduction, MrOS n=2,675 men, median 12.1 y; Wisconsin Sleep Cohort HR 1.17 (1.03 to 1.34), n=1,386, median 20.8 y.Leary 2020 · JAMA Neurol · doi
- CWhole night A foundation model trained on about 65,000 people's lab sleep studies ranked who would die sooner correctly in about 84% of pairs from one night, against 79% for a model using only age, sex, body mass index and ethnicity. C-index 0.84 (0.81-0.87) all-cause mortality vs 0.79 (0.75-0.82) for a demographics-only baseline (age, sex, BMI, race/ethnicity); SleepFM itself uses age and sex; about 65,000 participants, >585,000 h PSG; not an effect estimate. Kristjansson 2025 found EEG bands lifted C-index only from 77.54% to 77.78% over confounders.Thapa 2026 · Nat Med · doi
Key findings for Inflammation 6 papers
- BShort sleep Pooling 35 experiments, three or more nights of sleep cut to about 4.3 hours raised two blood markers of inflammation, while one night of sleep loss did not. CRP d 0.76 (95% CI 0.09 to 1.43) and IL-6 d 0.42 (0.11 to 0.73), k=5 each, for 3 or more nights of partial deprivation; 35 studies, 887 healthy adults; single night of total or partial deprivation ns.Ballesio 2025 · J Sleep Res · doi
- BWhole night Across 72 studies, disturbed sleep went with slightly higher inflammation markers, and very long sleep did too. Sleep disturbance CRP ES 0.12 (95% CI 0.05 to 0.19), IL-6 ES 0.20 (0.08 to 0.31); long sleep CRP ES 0.17 (0.01 to 0.34); shorter sleep CRP ES 0.09 (0.01 to 0.17); n>50,000; its null for experimental deprivation is superseded by Ballesio 2025.Irwin 2016 · Biol Psychiatry · doi
- BApnea events People with sleep apnea had much higher inflammation markers than controls, though people with apnea also tend to carry more weight. SMD CRP 1.77, IL-6 2.16, TNF-alpha 1.03 (CIs not in abstract); meta-analysis of 51 case-control studies (30 for CRP); obesity confounding likely.Nadeem 2013 · J Clin Sleep Med · doi
- CApnea events Across 15 trials, CPAP lowered a sensitive blood marker of inflammation modestly compared with control groups. hs-CRP SMD -0.64 (95% CI -1.19 to -0.09); meta-analysis of 15 RCTs, n=1,090; control arms not restricted to sham.Ning 2019 · Sleep Breath · doi
Show 2 moreShow fewer
- CApnea events A review limited to trials with a sham CPAP device found no reliable fall in inflammatory or metabolic markers, even though CPAP clearly calmed the body's stress response. Systematic review of sham-controlled RCTs and CPAP-withdrawal studies; 'failed to demonstrate that CPAP alters metabolic or inflammatory markers'; sympathetic activity reduced; no pooled estimate.Jullian-Desayes 2014 · Sleep Med Rev · doi
- CWhole night Genetic analysis of 91 inflammatory proteins and 7 sleep traits found small causal effects running both ways, mostly on sleep timing and napping. 30 proteins with effects on sleep traits, 20 reverse effects; e.g. CD40 on chronotype OR 1.02 (95% CI 1.01 to 1.03); two-sample bidirectional MR; self-reported sleep traits, no PSG measures.Liao 2025 · Brain Behav Immun Health · doi
How I built this map
I had AI research agents count the PubMed articles for each pairing of a sleep component and a condition, and pull the strongest evidence for each: umbrella reviews, meta-analyses, genetic studies and randomized trials. I checked each number against the paper, confirmed each DOI resolves, read the key papers and graded each pairing from A (strongest) to D (weakest). The grades are my judgement, not a formal assessment, and null results are under-represented.
The evidence base
Every sleep signature: 26 conditions by 17 components
Counts. On 5 October 2026 an AI research agent counted PubMed articles for every pair of seven broad sleep measures (such as short sleep, insomnia and apnea) and 26 conditions: 110,682 distinct articles. These are search hits, not study quality, and one paper can count in several pairs.
The graded map in Figure 1. AI research agents then searched PubMed for evidence linking each of the 17 components to each condition, preferring meta-analyses of sleep-study (PSG) comparisons, then prospective cohorts, then landmark single studies. Every number was taken from an abstract or full text, and every DOI was checked in PubMed. On 10 October a second round re-checked every finding against its PubMed record, dropped those that did not hold up and regraded others. I read the key papers. 275 findings support 206 links, and 17 null results were recorded but are not drawn.
Grades. A: a large, consistent meta-analysis, replicated or with genetic or trial support. B: a meta-analysis with a clear, significant pooled effect. C: a small or mixed meta-analysis, or a few good studies. D: single small studies or mixed results. The grades are my judgement, not a formal GRADE assessment.
Limits. Most findings are associations, and many come from case-control comparisons, questionnaires or wrist actigraphy rather than one night of EEG. Null results are under-represented. Papers per year on sleep and dementia or cognition went from 113 in 2000 to 1,929 in 2025, and on sleep and depression from 360 to 5,558.
Papers behind Figure 1
- Akerstedt 2023, J Neurol Neurosurg Psychiatry. doi.org/10.1136/jnnp-2022-330123 short sleep and Multiple sclerosis
- Arafa 2022, EPMA J. doi.org/10.1007/s13167-022-00275-4 regularity and Atrial fibrillation
- Aronica 2025, BMJ Ment Health. doi.org/10.1136/bmjment-2024-301337 long sleep and Schizophrenia
- Arıkan 2024, Sleep Med Rev. doi.org/10.1016/j.smrv.2023.101876 rem density and Depression; rem latency and Depression
- Ayalon 2007, Neurology. doi.org/10.1212/01.wnl.0000258672.52836.30 circadian rhythm and Brain injury
- Azarbarzin 2019, Eur Heart J. doi.org/10.1093/eurheartj/ehy624 oxygen drops and Cardiovascular disease
- Baglioni 2016, PLoS One. doi.org/10.1371/journal.pone.0148496 deep sleep and Stroke; short sleep and Stroke; waking in the night and Stroke
- Baglioni 2016, Psychol Bull. doi.org/10.1037/bul0000053 deep sleep and Depression; deep sleep and PTSD; rem amount and Autism; rem latency and Depression; rem latency and PTSD; short sleep and Anxiety; waking in the night and Anxiety; waking in the night and Autism; waking in the night and Depression; waking in the night and PTSD; waking in the night and Schizophrenia
- Baldini 2026, Sleep Med Rev. doi.org/10.1016/j.smrv.2026.102245 rem without atonia and PTSD
- Barbato 2026, Brain Sci. doi.org/10.3390/brainsci16070742 deep sleep and Bipolar disorder; rem density and Bipolar disorder; rem latency and Bipolar disorder; short sleep and Bipolar disorder
- Bender 2025, Neurology. doi.org/10.1212/WNL.0000000000214459 spindles and Mild cognitive impairment
- Benning 2025, J Sleep Res. doi.org/10.1111/jsr.70072 apnea events and High blood pressure
- Berger 2022, J Am Heart Assoc. doi.org/10.1161/JAHA.121.025828 deep sleep and High blood pressure; spindles and High blood pressure
- Biancardi 2021, Sleep. doi.org/10.1093/sleep/zsab006 arousals and ADHD; eeg slowing and ADHD
- Brink-Kjaer 2022, NPJ Digit Med. doi.org/10.1038/s41746-022-00630-9 eeg slowing and Death, any cause
- Chan 2016, Sleep Med Rev. doi.org/10.1016/j.smrv.2016.03.001 deep sleep and Schizophrenia; rem latency and Schizophrenia; short sleep and Schizophrenia; time to fall asleep and Schizophrenia; waking in the night and Schizophrenia
- Chaput 2024, Diabetes Care. doi.org/10.2337/dc24-1208 regularity and Type 2 diabetes
- Chaput 2025, J Epidemiol Community Health. doi.org/10.1136/jech-2024-222795 regularity and Cardiovascular disease
- Chaput 2026, Sleep. doi.org/10.1093/sleep/zsag193 short sleep and Death, any cause
- Chen 2020, Psychiatry Res. doi.org/10.1016/j.psychres.2020.113677 short sleep and Autism
- Chen 2022, Sleep Health. doi.org/10.1016/j.sleh.2022.08.004 circadian rhythm and Kidney disease; deep sleep and Kidney disease
- Chen 2025, Headache. doi.org/10.1111/head.14904 apnea events and Migraine
- Cheung 2023, Sleep Med Rev. doi.org/10.1016/j.smrv.2023.101851 circadian rhythm and Anxiety
- Chu 2025, J Nephrol. doi.org/10.1007/s40620-025-02214-8 limb movements and Kidney disease
- Clark 1998, Biol Psychiatry. doi.org/10.1016/s0006-3223(97)00457-5 rem density and Alcohol use
- Coogan 2017, Atten Defic Hyperact Disord. doi.org/10.1007/s12402-016-0214-5 circadian rhythm and ADHD
- Craciun 2025, Medicina (Kaunas). doi.org/10.3390/medicina61111988 apnea events and Cardiovascular disease
- d'Onofrio 2025, J Sleep Res. doi.org/10.1111/jsr.70202 limb movements and Migraine
- D'Rozario 2020, Sleep Med Rev. doi.org/10.1016/j.smrv.2020.101308 eeg slowing and Mild cognitive impairment; oxygen drops and Mild cognitive impairment; rem latency and Mild cognitive impairment; spindles and Mild cognitive impairment
- Daghlas 2021, JAMA Psychiatry. doi.org/10.1001/jamapsychiatry.2021.0959 circadian rhythm and Depression
- Dai 2024, Sleep Med Rev. doi.org/10.1016/j.smrv.2024.101914 short sleep and High blood pressure
- Degenfellner 2025, J Biol Rhythms. doi.org/10.1177/07487304251391267 regularity and Obesity
- Dondé 2021, Aust N Z J Psychiatry. doi.org/10.1177/00048674211068395 time to fall asleep and Schizophrenia
- Duclos 2020, Sleep. doi.org/10.1093/sleep/zsz191 regularity and Brain injury
- Elias 2015, Sleep Med. doi.org/10.1016/j.sleep.2015.10.022 rem amount and Kidney disease
- Emamian 2016, Front Aging Neurosci. doi.org/10.3389/fnagi.2016.00078 apnea events and Alzheimer's
- Feng 2023, Lancet Healthy Longev. doi.org/10.1016/S2666-7568(23)00056-9 circadian rhythm and Cardiovascular disease; circadian rhythm and Death, any cause
- Ferman 2011, Neurology. doi.org/10.1212/WNL.0b013e31822c9148 rem without atonia and Lewy body dementia
- Fernandez-Arcos 2019, Sleep. doi.org/10.1093/sleep/zsz086 eeg slowing and Lewy body dementia; spindles and Lewy body dementia
- Galbiati 2018, Sleep Med Rev. doi.org/10.1016/j.smrv.2018.09.008 rem without atonia and Lewy body dementia
- Gao 2023, medRxiv (preprint). doi.org/10.1101/2023.05.14.23289966 circadian rhythm and Stroke; regularity and Stroke
- Geoffroy 2015, Acta Psychiatr Scand. doi.org/10.1111/acps.12367 long sleep and Bipolar disorder
- Grima 2016, J Clin Sleep Med. doi.org/10.5664/jcsm.5598 short sleep and Brain injury; waking in the night and Brain injury
- Halal 2017, Sleep Sci. doi.org/10.5935/1984-0063.20170028 time to fall asleep and Epilepsy
- Hasler 2016, J Stud Alcohol Drugs. doi.org/10.15288/jsad.2016.77.649 regularity and Alcohol use
- Heinzinger 2024, JACC Clin Electrophysiol. doi.org/10.1016/j.jacep.2024.05.027 oxygen drops and Atrial fibrillation
- Heinzinger 2025, Sleep. doi.org/10.1093/sleep/zsaf055 short sleep and Atrial fibrillation; waking in the night and Atrial fibrillation
- Henckaerts 2025, Sleep Med Rev. doi.org/10.1016/j.smrv.2025.102141 deep sleep and Depression
- Himali 2023, JAMA Neurol. doi.org/10.1001/jamaneurol.2023.3889 deep sleep and Dementia
- Ho 2024, J Affect Disord. doi.org/10.1016/j.jad.2024.05.155 arousals and Depression; circadian rhythm and Depression; regularity and Depression; waking in the night and Depression
- Huang 2020, J Am Coll Cardiol. doi.org/10.1016/j.jacc.2019.12.054 regularity and Cardiovascular disease
- Javaheri 2018, Sleep. doi.org/10.1093/sleep/zsx179 deep sleep and High blood pressure
- Johnson 2021, Sleep Med Rev. doi.org/10.1016/j.smrv.2021.101456 short sleep and Type 2 diabetes
- Juginovic 2025, Sleep Med. doi.org/10.1016/j.sleep.2025.106599 deep sleep and Cardiovascular disease; rem latency and Cardiovascular disease; spindles and Cardiovascular disease
- Kao 2025, Depress Anxiety. doi.org/10.1155/da/4329208 apnea events and Anxiety; apnea events and Bipolar disorder; apnea events and Depression
- Kendzerska 2014, Am J Respir Crit Care Med. doi.org/10.1164/rccm.201312-2209OC apnea events and Type 2 diabetes; oxygen drops and Type 2 diabetes
- Kendzerska 2018, Chest. doi.org/10.1016/j.chest.2018.08.1075 oxygen drops and Atrial fibrillation
- Kim 2023, Epidemiol Psychiatr Sci. doi.org/10.1017/S2045796023000574 time to fall asleep and Autism
- Kobayashi 2007, Psychophysiology. doi.org/10.1111/j.1469-8986.2007.537.x rem density and PTSD
- Koh 2024, Clin Kidney J. doi.org/10.1093/ckj/sfae279 apnea events and Kidney disease; limb movements and Kidney disease
- Koo 2011, Circulation. doi.org/10.1161/CIRCULATIONAHA.111.038968 limb movements and Cardiovascular disease
- Kozhemiako 2022, Elife. doi.org/10.7554/eLife.76211 spindles and Schizophrenia
- Kristjansson 2025, Sleep. doi.org/10.1093/sleep/zsae219 deep sleep and Death, any cause
- Kwon 2015, Thorax. doi.org/10.1136/thoraxjnl-2014-206655 deep sleep and Atrial fibrillation
- Lai 2021, Psychiatry Res. doi.org/10.1016/j.psychres.2021.114265 spindles and Schizophrenia
- Lam 2022, J Affect Disord. doi.org/10.1016/j.jad.2022.09.045 arousals and PTSD
- Latreille 2015, Neurobiol Aging. doi.org/10.1016/j.neurobiolaging.2014.09.009 spindles and Parkinson's
- Leary 2020, JAMA Neurol. doi.org/10.1001/jamaneurol.2020.2108 rem amount and Death, any cause
- Lehner 2022, Sleep Med Rev. doi.org/10.1016/j.smrv.2022.101689 rem latency and Epilepsy; short sleep and Epilepsy
- Leitner 2025, Sleep Med Rev. doi.org/10.1016/j.smrv.2025.102048 rem amount and Depression
- Leng 2017, JAMA Neurol. doi.org/10.1001/jamaneurol.2017.2180 apnea events and Mild cognitive impairment
- Leng 2020, JAMA Neurol. doi.org/10.1001/jamaneurol.2020.1623 circadian rhythm and Parkinson's
- Li 2024, Folia Neuropathol. doi.org/10.5114/fn.2024.145596 apnea events and Dementia
- Li 2025, Psychol Med. doi.org/10.1017/S0033291725101281 regularity and Anxiety; regularity and Depression
- Liang 2023, Psychiatry Res. doi.org/10.1016/j.psychres.2023.115447 time to fall asleep and ADHD
- Liang 2023, JAMA Netw Open. doi.org/10.1001/jamanetworkopen.2023.36129 time to fall asleep and Autism
- Lim 2013, Sleep. doi.org/10.5665/sleep.2802 arousals and Alzheimer's
- Lin 2016, Sleep Breath. doi.org/10.1007/s11325-016-1391-3 apnea events and Epilepsy
- Liu 2025, Ann Med. doi.org/10.1080/07853890.2024.2447422 circadian rhythm and Type 2 diabetes
- Lucey 2019, Sci Transl Med. doi.org/10.1126/scitranslmed.aau6550 deep sleep and Alzheimer's
- Luo 2025, Sleep Med. doi.org/10.1016/j.sleep.2025.106832 limb movements and Stroke
- Lyall 2018, Lancet Psychiatry. doi.org/10.1016/S2215-0366(18)30139-1 circadian rhythm and Bipolar disorder
- Maggi 2023, Sleep Med Rev. doi.org/10.1016/j.smrv.2023.101759 rem without atonia and Parkinson's
- Maggi 2024, Eur J Neurol. doi.org/10.1111/ene.16109 apnea events and Parkinson's
- Maggi 2024, Sleep Med. doi.org/10.1016/j.sleep.2024.07.015 limb movements and Parkinson's
- Maia 2025, Can J Neurol Sci. doi.org/10.1017/cjn.2025.10114 long sleep and Brain injury
- Mantua 2018, Sleep Med Rev. doi.org/10.1016/j.smrv.2018.01.004 deep sleep and Brain injury
- Marchetti 2025, Sleep Med Rev. doi.org/10.1016/j.smrv.2025.102137 long sleep and Bipolar disorder; regularity and Bipolar disorder; rem amount and Bipolar disorder; short sleep and Bipolar disorder; time to fall asleep and Bipolar disorder
- Mathias 2012, Sleep Med. doi.org/10.1016/j.sleep.2012.04.006 waking in the night and Brain injury
- Mellman 2002, Am J Psychiatry. doi.org/10.1176/appi.ajp.159.10.1696 rem amount and PTSD
- Mokhlesi 2014, Am J Respir Crit Care Med. doi.org/10.1164/rccm.201406-1136OC apnea events and High blood pressure
- Morra 2025, Sleep Med Rev. doi.org/10.1016/j.smrv.2025.102134 rem amount and Schizophrenia; rem density and Schizophrenia
- Nadeem 2013, J Clin Sleep Med. doi.org/10.5664/jcsm.3070 apnea events and Inflammation
- Ng 2015, Sleep Med Rev. doi.org/10.1016/j.smrv.2014.06.006 regularity and Bipolar disorder
- Ning 2019, Sleep Breath. doi.org/10.1007/s11325-018-1662-2 apnea events and Inflammation
- Nota 2015, Neurosci Biobehav Rev. doi.org/10.1016/j.neubiorev.2015.01.002 circadian rhythm and Anxiety; short sleep and Anxiety
- Oliva 2026, EClinicalMedicine. doi.org/10.1016/j.eclinm.2025.103723 deep sleep and Alcohol use; rem amount and Alcohol use; rem latency and Alcohol use; short sleep and Alcohol use
- Oye-Somefun 2026, PLoS One. doi.org/10.1371/journal.pone.0343621 long sleep and Dementia; short sleep and Dementia
- Ozdogar 2021, Mult Scler Relat Disord. doi.org/10.1016/j.msard.2021.103275 limb movements and Multiple sclerosis
- Paredes Naveda 2026, Sleep Med Rev. doi.org/10.1016/j.smrv.2026.102297 short sleep and Alcohol use
- Pase 2017, Neurology. doi.org/10.1212/WNL.0000000000004373 rem amount and Dementia; rem latency and Dementia
- Pejovic 2024, J Clin Sleep Med. doi.org/10.5664/jcsm.11046 short sleep and Cardiovascular disease
- Pejovic 2026, J Am Heart Assoc. doi.org/10.1161/JAHA.125.045898 short sleep and Cardiovascular disease
- Peppard 2000, JAMA. doi.org/10.1001/jama.284.23.3015 apnea events and Obesity
- Postuma 2019, Brain. doi.org/10.1093/brain/awz030 rem without atonia and Parkinson's
- Punjabi 2009, PLoS Med. doi.org/10.1371/journal.pmed.1000132 apnea events and Death, any cause
- Qian 2021, Sleep. doi.org/10.1093/sleep/zsaa256 circadian rhythm and Obesity
- Ravenhall 2026, Sleep Med Rev. doi.org/10.1016/j.smrv.2026.102345 regularity and Anxiety
- Reither 2021, Sleep. doi.org/10.1093/sleep/zsab031 deep sleep and Obesity; rem amount and Obesity
- Ren 2025, Sleep Med. doi.org/10.1016/j.sleep.2025.106562 apnea events and Mild cognitive impairment
- Ricardo 2017, J Am Soc Nephrol. doi.org/10.1681/ASN.2016121288 arousals and Kidney disease; short sleep and Kidney disease
- Ritter 2018, Acta Psychiatr Scand. doi.org/10.1111/acps.12924 spindles and Bipolar disorder
- Rossignol 2011, Dev Med Child Neurol. doi.org/10.1111/j.1469-8749.2011.03980.x circadian rhythm and Autism
- Sabia 2021, Nat Commun. doi.org/10.1038/s41467-021-22354-2 short sleep and Dementia
- Sakaguchi 2013, Clin J Am Soc Nephrol. doi.org/10.2215/CJN.11931112 oxygen drops and Kidney disease
- Sanders 2024, Sleep Med. doi.org/10.1016/j.sleep.2024.09.008 waking in the night and Kidney disease
- Sedky 2013, Sleep Med Rev. doi.org/10.1016/j.smrv.2013.12.003 apnea events and ADHD
- Seiler 2019, Neurology. doi.org/10.1212/WNL.0000000000006904 apnea events and Stroke
- Shahrbabaki 2021, Eur Heart J. doi.org/10.1093/eurheartj/ehab151 arousals and Cardiovascular disease
- Simou 2018, Sleep Med. doi.org/10.1016/j.sleep.2017.12.005 apnea events and Alcohol use
- Smagula 2016, Psychosom Med. doi.org/10.1097/PSY.0000000000000312 arousals and Death, any cause; oxygen drops and Death, any cause; oxygen drops and Inflammation; short sleep and Inflammation
- Srivali 2023, Sleep Med. doi.org/10.1016/j.sleep.2023.01.008 limb movements and High blood pressure
- Stamatakis 2009, Chest. doi.org/10.1378/chest.09-0791 arousals and Type 2 diabetes
- Stanyer 2021, Neurology. doi.org/10.1212/WNL.0000000000012701 rem amount and Migraine; short sleep and Migraine; time to fall asleep and Migraine; waking in the night and Migraine
- Stubbs 2016, J Affect Disord. doi.org/10.1016/j.jad.2016.02.060 apnea events and Bipolar disorder; apnea events and Depression; apnea events and Schizophrenia
- Sun 2026, JAMA Netw Open. doi.org/10.1001/jamanetworkopen.2026.1521 eeg slowing and Dementia
- Tasali 2008, Proc Natl Acad Sci U S A. doi.org/10.1073/pnas.0706446105 deep sleep and Type 2 diabetes
- Tasali 2022, JAMA Intern Med. doi.org/10.1001/jamainternmed.2021.8098 short sleep and Obesity
- Terzaghi 2013, Mov Disord. doi.org/10.1002/mds.25523 apnea events and Lewy body dementia; limb movements and Lewy body dementia
- Toubasi 2026, J Cent Nerv Syst Dis. doi.org/10.1177/11795735261485680 apnea events and Multiple sclerosis; short sleep and Multiple sclerosis; waking in the night and Multiple sclerosis
- Tranah 2011, Ann Neurol. doi.org/10.1002/ana.22468 circadian rhythm and Mild cognitive impairment
- Ungvari 2025, Geroscience. doi.org/10.1007/s11357-025-01637-2 apnea events and Alzheimer's
- Ungvari 2025, Geroscience. doi.org/10.1007/s11357-025-01592-y short sleep and Death, any cause
- Van Veen 2010, Biol Psychiatry. doi.org/10.1016/j.biopsych.2009.12.032 circadian rhythm and ADHD
- Velzeboer 2025, Sleep Med Rev. doi.org/10.1016/j.smrv.2025.102164 rem amount and Alcohol use
- Wang 2013, Int J Cardiol. doi.org/10.1016/j.ijcard.2013.08.088 apnea events and Cardiovascular disease
- Wang 2019, Sleep Med. doi.org/10.1016/j.sleep.2019.01.027 limb movements and Migraine
- Wang 2021, Sleep Med. doi.org/10.1016/j.sleep.2021.11.001 long sleep and Stroke; short sleep and Stroke
- Wang 2025, Neurology. doi.org/10.1212/WNL.0000000000214513 circadian rhythm and Dementia
- Wang 2025, Nat Sci Sleep. doi.org/10.2147/NSS.S492630 apnea events and Migraine
- Wang 2025, Front Med (Lausanne). doi.org/10.3389/fmed.2025.1497703 apnea events and Type 2 diabetes
- Wei 2024, Sleep Med. doi.org/10.1016/j.sleep.2023.12.010 rem amount and Mild cognitive impairment; short sleep and Mild cognitive impairment; time to fall asleep and Mild cognitive impairment; waking in the night and Mild cognitive impairment
- Windred 2024, Sleep. doi.org/10.1093/sleep/zsad253 regularity and Death, any cause
- Winer 2019, J Neurosci. doi.org/10.1523/JNEUROSCI.0503-19.2019 spindles and Alzheimer's
- Winsor 2021, Sleep Med Rev. doi.org/10.1016/j.smrv.2021.101416 short sleep and Epilepsy
- Wu 2022, Gen Hosp Psychiatry. doi.org/10.1016/j.genhosppsych.2022.10.009 rem without atonia and Depression
- Xia 2018, Sleep Med. doi.org/10.1016/j.sleep.2018.03.016 apnea events and High blood pressure
- Xian 2025, Psychol Med. doi.org/10.1017/S0033291725102158 arousals and ADHD; short sleep and ADHD; time to fall asleep and ADHD; waking in the night and ADHD
- Xiao 2022, J Am Geriatr Soc. doi.org/10.1111/jgs.17926 circadian rhythm and Mild cognitive impairment
- Xiao 2022, J Gerontol A Biol Sci Med Sci. doi.org/10.1093/gerona/glab095 circadian rhythm and Inflammation
- Yaffe 2011, JAMA. doi.org/10.1001/jama.2011.1115 apnea events and Mild cognitive impairment; oxygen drops and Mild cognitive impairment
- Yang 2026, Alzheimers Res Ther. doi.org/10.1186/s13195-026-02088-3 circadian rhythm and Alzheimer's; circadian rhythm and Dementia
- Yeh 2022, Seizure. doi.org/10.1016/j.seizure.2022.01.014 deep sleep and Epilepsy; rem amount and Epilepsy; waking in the night and Epilepsy
- Yeh 2022, Sleep. doi.org/10.1093/sleep/zsac041 arousals and Epilepsy
- Yiallourou 2023, Neurology. doi.org/10.1212/WNL.0000000000208029 regularity and Dementia
- Zalai 2020, CMAJ Open. doi.org/10.9778/cmajo.20190137 circadian rhythm and Brain injury
- Zali 2023, Neurol Sci. doi.org/10.1007/s10072-022-06364-6 limb movements and Multiple sclerosis
- Zhai 2015, Depress Anxiety. doi.org/10.1002/da.22386 long sleep and Depression; short sleep and Depression
- Zhang 2017, Sleep Med. doi.org/10.1016/j.sleep.2017.04.020 apnea events and PTSD
- Zhang 2019, Sleep Med Rev. doi.org/10.1016/j.smrv.2019.08.004 deep sleep and PTSD; rem amount and PTSD; short sleep and PTSD
- Zhang 2020, Sleep Med Rev. doi.org/10.1016/j.smrv.2020.101281 apnea events and Parkinson's; deep sleep and Parkinson's; limb movements and Parkinson's; rem amount and Parkinson's; rem latency and Parkinson's; short sleep and Parkinson's; waking in the night and Parkinson's
- Zhang 2022, Transl Psychiatry. doi.org/10.1038/s41398-022-01897-y arousals and Alzheimer's; deep sleep and Alzheimer's; eeg slowing and Alzheimer's; rem amount and Alzheimer's; rem density and Alzheimer's; rem latency and Alzheimer's; short sleep and Alzheimer's; spindles and Alzheimer's; time to fall asleep and Alzheimer's; waking in the night and Alzheimer's
- Zhang 2022, Psychol Med. doi.org/10.1017/S0033291722001581 time to fall asleep and Depression
- Zhang 2022, Medicine (Baltimore). doi.org/10.1097/MD.0000000000029443 apnea events and Atrial fibrillation
- Zhang 2022, Front Endocrinol (Lausanne). doi.org/10.3389/fendo.2022.1008820 circadian rhythm and Obesity
- Zhang 2023, J Clin Sleep Med. doi.org/10.5664/jcsm.10304 waking in the night and Multiple sclerosis
- Zhang 2025, J Neurol. doi.org/10.1007/s00415-025-13372-x long sleep and Mild cognitive impairment; short sleep and Mild cognitive impairment
- Zheng 2024, Nat Med. doi.org/10.1038/s41591-024-03155-8 circadian rhythm and High blood pressure; deep sleep and Atrial fibrillation; long sleep and High blood pressure; regularity and High blood pressure; regularity and Obesity; rem amount and Atrial fibrillation; short sleep and High blood pressure; short sleep and Obesity
- Zhu 2025, Sleep Breath. doi.org/10.1007/s11325-025-03348-6 apnea events and Inflammation
- Zong 2025, Sleep. doi.org/10.1093/sleep/zsaf175 regularity and Brain injury
Reading the map
Two things stand out to me. The first is a shared core. Almost every condition is linked to the same few disturbances. Short sleep shows up in 25 of the 26 conditions, apnea in 24, changes in deep (slow-wave) sleep and circadian rhythm in 18 each, and waking in the night in 16. Part of the reason is that these are the most studied measurements, and the links vary a lot in strength. The core works a bit like a fever for the brain. It tells you something is wrong, but not what.
The second is that each condition has its own fingerprint. A meta-analysis of sleep studies across mental disorders found that no single measure was altered in only one disorder, yet no two disorders shared the same profile8. The specific signs mostly sit in the finer detail of the recording. Schizophrenia, for example, shows far fewer sleep spindles, the brief bursts of activity that mark stage 2 sleep9.
This is what got me excited. If each condition leaves its own fingerprint, one night could be a readout for many conditions at once. These profiles are group averages, though, and whether a model can read them in one person's night is the question the next part turns to.
Cause or signal
A fingerprint tells you something is there, but not whether sleep drives the disease or just reflects it. If it drives it, fixing sleep could help. If it only reflects it, sleep is still a window worth watching. Three kinds of evidence separate the two: cohorts show what predicts what, genetic studies (Mendelian randomization) use inherited variants as a natural experiment, and trials test whether fixing sleep changes the outcome. Depression and neurodegeneration tell two different stories.
Depression: sleep as part of the cause. All three point the same way. Insomnia roughly doubles the risk of later depression10, and genetic studies find a smaller effect in both directions11. Treating insomnia with cognitive behavioural therapy for insomnia (CBT-I), a short course that retrains sleep habits and thoughts about sleep, halved new or recurring depression in a single-site trial of 291 older adults with insomnia disorder, from 26% to 12% over three years12, though not every prevention trial has worked13. During my depression I couldn't tell whether bad sleep was dragging my mood down or the other way round. The research says both, and that matches what I went through. Once my sleep improved, my mood lifted, and a better mood made it easier to sleep.
Neurodegeneration: sleep as an early warning. RBD is the sharpest case. Acting out dreams doesn't cause Parkinson's, but it is one of the disease's earliest signs6. With dementia it helps to ask two separate questions: whether sleep predicts it, and whether it causes it. It predicts it, if you read the night closely. When five cohorts pooled single-night sleep studies, none of the standard sleep scores predicted dementia14, but read as a brain age, largely the same recordings did: each 10 years of extra brain age carried a 39% higher risk15. Whether sleep causes it is less clear. Genetic studies mostly find no causal effect on Alzheimer's16,17. One night of disrupted deep sleep does raise one form of amyloid-β, the protein that builds up in Alzheimer's, the next morning18, but the disease also disturbs sleep for years before diagnosis19. For now, sleep looks more like a warning light than a cause, and a warning light is still worth having.
One night, or many
So why isn't a sleep study as routine as a blood panel at your GP? Most people never get one. In a US cohort of working adults studied in the 1990s, 82% of men and 93% of women with moderate to severe apnea had never been diagnosed20, and apnea is the condition sleep clinics are built for.
Even the few people who do get a sleep study get a single night, in a bed that isn't theirs. Across 53 studies of 1,422 healthy people, sleep was shorter and more broken on the first night in a lab21. When 66 patients had one sleep study in hospital and one at home, they slept 47 minutes longer at home, fell asleep 17 minutes faster and spent more time in REM sleep22. A night in the lab can reveal a lot, but it is a snapshot taken in an unfamiliar place.
But nights at home vary too. When 94 men wore an EEG headband for five weeknights, the time each spent in deep sleep varied by 44% to 58% from one night to the next23.
No two nights are the same, just as no two days are. Researchers split any repeated measurement into three parts24. The trait is the level your nights keep returning to, the part that makes your sleep yours. A state is a departure from that level that passes: one bad night, a stressful month, a new medicine, a depressive episode that lifts25. A change that lasts becomes part of the trait. Noise is error in the measurement itself, such as two experts scoring the same night differently26. One night mixes all three. Over many nights, short states and noise average out, leaving the trait and any state that lasts (Figure 2a).
a Three simulated people, four weeks of total sleep
b How alike two nights are (1 = identical)
c How far brain age moves, in years
The traits appear strongest in the brain's rhythms. Sleep spindles and slow waves come back almost unchanged from one night to the next, while total sleep barely repeats (Figure 2b)29. The pattern is so personal that in 40 twins it was 96% heritable31.
States show up as a change against that baseline, and some only show in the brain. Across the menstrual cycle, a ring picked up the change in skin temperature but not in sleep32, while the sleep EEG showed spindle power rising and falling33. A new medicine is a state too. In Figure 2a, its 40 extra minutes of sleep are lost among single nights and stand out in the weekly average28.
Averaging helps a lot. A brain age read from the sleep EEG moved by several years between single nights, and averaging four nights cut that spread by more than half (Figure 2c)30.
Watches and rings give you many nights, but they only read movement and heart rate34, so they can't see the spindles and slow waves where the traits are strongest. EEG headbands now record them at home, and their sleep staging agrees with a full sleep study almost as well as two human scorers agree with each other35,26.
I've slept with most of these, more than 1,000 nights with watches, rings and EEG headbands, and my own nights show the same layers. They wobble from day to day and week to week, and I think I've started to see a state that comes with the seasons. I live in Copenhagen, where summer nights barely get dark and winter days are short. I'm prone to seasonal affective disorder, the low mood that comes with winter, and I wonder how much of it runs through changes in my sleep and circadian rhythm. And if I protect my sleep through the winter, will the low mood stay away? I'll share the data in a later post.
In part 2, we look at how AI is unlocking new applications in sleep and what is holding it back. In part 3, we look at where the technologies are heading and what it could mean for the future of brain medicine.