Sleep
What the human trials show about sleep.
This page gathers EBL's sleep evidence in one place: treatment for insomnia, sleep apnea, the habits that change sleep, regular timing, and how far a sleep tracker can be trusted. Each letter belongs to a graded record that holds the trials. Findings from population studies and animal work are shown as context and carry no letter.
Insomnia
Treating insomnia without a pill first.
CBT-I. Cognitive behavioral therapy for insomnia is a structured talk treatment built around a fixed sleep window. A meta-analysis A study that pools the numbers from several trials into one combined estimate. of 20 randomized trials found people fell asleep 19 minutes faster and spent 26 fewer minutes awake at night, and the gains lasted (PMID PubMed ID: the permanent number the US National Library of Medicine gives a published paper. Each one links to that paper's record on PubMed. 26054060). The outcomes come from sleep diaries, so the record is graded B.
Against sleeping pills. Three small trials compared CBT-I directly with sleep medication: temazepam in 78 older adults (PMID 10086433), a sleeping pill in 63 adults (PMID 15451764) and zopiclone in 46 older adults (PMID 16804151). CBT-I did as well or better by the end of treatment and kept its gains longer. A review of five such trials rated that evidence low to moderate (PMID 22631616). Newer drugs, such as orexin receptor antagonists, have not been compared.
Naps. In laboratory trials, about 10 minutes of nap sleep made people more alert at once, while 30 minutes was followed by grogginess before the benefit (PMID 16796222). That use is graded A on the napping record, which grades each other nap use separately.
Sleeping longer. Adding about an hour of sleep a night in people who habitually sleep too little gave mixed results on blood pressure and blood sugar; the best-measured trial found no change in insulin sensitivity (PMID 41564347). Graded C.
A warm bath. Pooled small studies found people fell asleep faster after 10 minutes or more in water at 40 to 42.5 degrees Celsius (104 to 108.5 degrees Fahrenheit), one to two hours before bed (PMID 31102877). A warm footbath trial found no change (PMID 23669188). Graded C.
Supplements. Melatonin, magnesium, glycine, ashwagandha, L-theanine and tart cherry are graded on the sleep condition page, and the sleep supplements guide explains why products sold side by side can sit grades apart. Melatonin also carries a use for people taking beta-blockers: in 16 adults taking the beta-blockers atenolol or metoprolol, which lower the body's own nighttime melatonin, 2.5 mg of melatonin nightly increased total sleep by 36 minutes in the sleep lab (PMID 23024438). One small trial, graded C.
Sleep apnea
Sleep apnea: what each treatment was tested for.
In obstructive sleep apnea the throat closes again and again during sleep. The treatments below were tested on different outcomes, and the letters follow the outcome.
CPAP. A machine that blows air through a mask to hold the throat open. Across 75 randomized trials it lowered systolic blood pressure by about 2.5 mmHg (PMID 40254559), mostly in people whose pressure was uncontrolled (PMID 39401854), and it reduced self-rated daytime sleepiness (PMID 26497082). Those uses are graded B. On a lab test of staying awake, scored by instrument, it beat sham A fake version of a procedure or device used as the comparison, so participants cannot tell which one they received. CPAP in two independent trials of sleepy people with moderate to severe sleep apnea (PMIDs 10382693, 23204602), graded A; two small trials in mild sleep apnea found no clear change. Four large trials found it did not prevent heart attacks, strokes or cardiovascular deaths (PMIDs 27571048, 37787793), graded D on Route 2 Human randomized trials have been run and did not show benefit on the stated outcome. One of the two routes to the lowest letter on EBL's scale.. Most people in those trials were not sleepy and used CPAP about 3 hours a night, so the result does not reach sleepy patients who use it all night.
Oral appliance. A mouthpiece that holds the lower jaw forward. CPAP removed more breathing pauses, but people wore the mouthpiece longer, and daytime outcomes came out about even (PMID 23413266). Graded B.
Hypoglossal nerve stimulation. An implant that moves the tongue forward during sleep, for people who cannot use CPAP. Against a switched-off device it cut breathing pauses and sleepiness (PMID 34209581), graded B; in one trial it did not lower blood pressure (PMID 38032624), graded C.
Positional therapy. A neck-worn device that buzzes when the sleeper rolls onto the back. In one sham-controlled trial it lowered breathing pauses a little and did not change daytime sleepiness (PMID 40992934). Graded C.
Nasal valves. Small patches over each nostril that slow breathing out. In a maker-funded trial of 250 adults they cut breathing pauses and sleepiness against a sham valve (PMID 21461326), but in people with moderate to severe sleep apnea taken off CPAP they did no better than a placebo An inactive substitute made to look like the real treatment, so participants cannot tell which one they received. valve (PMID 23723343). Graded C.
Mouth and throat exercises. Myofunctional therapy is about half an hour a day of exercises for the tongue, soft palate and throat. Against sham exercises over two to four months, adults felt less sleepy and had fewer breathing pauses (PMID 33141943). Graded B, because sleepiness was self-rated and the other outcome is a breathing count; children did not benefit, mostly because few kept up the exercises (PMID 37606313).
Medicines. In two placebo-controlled trials in adults with obesity and moderate to severe sleep apnea, tirzepatide lowered the apnea-hypopnea index by 20 to 24 events an hour more than placebo over a year (PMID 38912654); both trials were run by the maker. The tirzepatide record grades this use B, with the funding stated beside the letter; its diabetes prevention use stays at C. A pill combining aroxybutynin and atomoxetine lowered the index by 4 events an hour in a phase 3 trial, and one in five people stopped it for side effects (PMID 42148495); the FDA has accepted its application, with a decision due in February 2027.
Screening. The US Preventive Services Task Force found the evidence insufficient to weigh the benefits and harms of screening adults in general for sleep apnea (PMID 36378202). That statement concerns people without symptoms; it does not address people with snoring, witnessed pauses or daytime sleepiness.
Habits
Caffeine, alcohol and a hot bedroom.
These are shown as context and carry no letter. The trials gave people caffeine or alcohol, or a hotter room, and measured what happened to sleep. None tested cutting back as a treatment, which is what a letter here would grade.
Caffeine. Pooling 24 studies, caffeine shortened total sleep by about 45 minutes and lengthened the time to fall asleep by about 9 minutes (PMID 36870101). In a crossover A trial design in which each participant receives both the treatment and the comparison, one after the other, and serves as their own control. of 23 young men, 400 mg changed sleep even when taken 12 hours before bed, while 100 mg had no measurable effect at any time tested (PMID 39377163). A second pooled analysis of 22 sleep-lab trials found total sleep about 35 minutes shorter (PMID 41124973).
Alcohol. Across 27 studies in healthy adults, alcohol before sleep delayed and reduced REM sleep from about two standard drinks, more so at higher doses. People fell asleep faster only at about five drinks, and the effects on total sleep were uncertain (PMID 39631226). Sleep is often deeper in the first half of the night and more broken in the second (PMID 23347102).
Heat. In small sleep-lab studies, rooms well above body-comfort temperature cost sleep. Adults over 65 slept 26 minutes less at 30 degrees Celsius (86 degrees Fahrenheit) than at 27 (PMID 36437666), and older men lost REM sleep at 32 degrees Celsius (PMID 15173935). In 40 people with untreated sleep apnea, a 16-degree room gave 30 minutes more sleep than a 24-degree room but made the apnea worse (PMID 22467989).
Timing
Regular sleep, long sleep and the population studies.
These findings come from people followed over years, not from trials, so they show what travels together, not what causes what. They carry no letter.
Regularity. In 60,977 UK Biobank adults who wore activity trackers for a week, the most regular sleepers had a 20 to 48 percent lower risk of death over about 6 years than the least regular, and regularity predicted death better than sleep length did (PMID 37738616). In 3,231 Finnish adults aged 46, irregular bedtimes went with about twice the rate of heart attacks and strokes over 10 years, but only among those who slept less than about 8 hours, and with a confidence interval The range the true effect probably falls in. A wide interval means the estimate is uncertain. that touched no effect (PMID 41877022).
More is not always better. In 431,924 UK Biobank adults, dementia risk was lowest at about 7 hours of sleep and rose on both sides (PMID 35701596). Pooling 76 cohorts, short sleep went with a 27 percent higher risk of dementia or cognitive decline, and long sleep with up to 66 percent higher risk of Alzheimer's disease (PMID 41081870). People who are already becoming unwell may sleep longer, which these studies cannot separate.
Light. Daylight and night light are covered on the sleep condition page, and bright light therapy has its own graded record.
An open question
Does sleep clean the brain? Two mouse studies disagree.
A widely quoted idea holds that the brain flushes out waste, including amyloid, during sleep. It comes from a 2013 study in mice, in which the space between brain cells widened by about 60 percent during sleep and amyloid cleared faster (PMID 24136970). A 2024 study in mice, using a different method, found clearance was lower during sleep and anesthesia, not higher (PMID 38741022). Neither was done in people, and no human trial has tested whether sleeping more clears the brain. EBL shows the question and grades nothing on it.
Trackers
What a sleep tracker gets right and wrong.
Rings and watches are good at telling that someone is asleep and poor at telling that someone is awake. Against sleep-lab recordings, seven devices detected 93 percent or more of sleep, but only 18 to 54 percent of time awake (PMID 33378539). In 62 adults, six newer wrist devices detected 29 to 52 percent of wake, and agreement on sleep stages was fair to moderate (PMID 40303381). Pooling 24 studies, wrist devices underestimated total sleep by about 17 minutes and overestimated time awake by about 13 (PMID 39484805).
A 2026 review of 29 real-world studies found trackers reasonable for sleep duration and timing and unreliable for deep sleep, REM and fragmentation, with few studies in older adults or people with insomnia (PMID 41811282). The research consensus statement reaches the same view and notes that skin color and body size can affect the sensors (PMID 38149978). Some of the best published results for one ring come from a study the maker paid for (PMID 38382312) and one with an author on its advisory board (PMID 39460013). Device by device, see wearables.
Questions
What readers ask.
Why is CPAP graded D if it helps people?
Each use has its own letter. CPAP is graded A for measured wakefulness in sleepy people with moderate to severe sleep apnea, and B for blood pressure and self-rated sleepiness. The D is for heart attacks, strokes and cardiovascular deaths, where four large trials found no reduction. Those trials enrolled people who were mostly not sleepy and used CPAP about 3 hours a night.
Why do the regularity and dementia findings carry no letter?
EBL grades on randomized trials. The regularity and dementia findings come from people followed over time, which shows that two things go together but not that one causes the other. No trial has yet made people's sleep more regular and measured death or dementia.
Is CBT-I better than sleeping pills?
In three small head-to-head trials it did as well or better by the end of treatment and its gains lasted longer. The trials had 18 to 20 people per group and did not test newer drug classes, so the comparison is shown on the record and does not set the letter.
Can a sleep tracker diagnose sleep apnea?
The consumer trackers studied here were tested for sleep and wake, not for apnea. Sleep apnea is diagnosed with a home sleep apnea test or a sleep-lab study.
Go further
From the overview to the records.
The same treatment can carry different letters for different outcomes. Each letter follows the outcome it was tested on.
Educational use only, not medical advice.