1,266 tracked for 35 years, 147 nights in a sleep lab, and who sets the bar for a bad night
A 35-year cohort reports ordinary sleep while its actigraphy shows fragmentation, a cleared smartwatch's error rate turns on whom it screens, and one study lets people's own sleep history set the cutoff.
Every verdict about a night is a comparison: a reference, plus a line drawn through it. Whoever reports the number chooses both, and the choice decides which night, and which sleeper, gets called a problem.
This week one study let 462 people's own sleep history set the line, and the nights it flagged were largely different from the ones a flat six-hour rule flags, even though the share of nights called short barely moved, 35% against 37%. Around it, a 35-year cohort recorded sleep while its participants described it, a cleared smartwatch was scored in a sample chosen for its high rate of apnea, and two wearable makers compared their members against other people's published cohorts.
The report and the recording disagree, and in one direction
Cannabis is widely used and often promoted as a sleep aid, and the evidence on its long-term effect on sleep has been thin. Researchers followed 1,266 adults in the CARDIA cohort from young adulthood for 35 years, then measured their sleep twice at the year-35 examination: seven days of wrist actigraphy in 891 of them, and the Pittsburgh Sleep Quality Index in 1,128. 1
Each additional cannabis-year, counted as 365 days of use, came with higher odds of a low sleep percentage (odds ratio 1.10, 95% CI 1.06–1.14), high sleep fragmentation (1.04, 1.01–1.08) and prolonged wake after sleep onset (1.21, 1.10–1.34). Recent use looked worse than distant use: against people who had never used, those who had used in the past 30 days had 3.23 times the odds of prolonged wake after sleep onset (1.44–7.27) and 4.60 times the odds of a long night above eight hours (1.81–11.70). Self-reported sleep quality showed no association with cannabis at all. 1
The two instruments told different stories about the same nights. One reading is that a self-report is a blunt tool that misses fragmentation. The other is that a morning report answers a different question, describing sleep as it was remembered and felt, reassembled from whichever awakenings stayed in memory, and that a substance taken to help sleep moves the expectation before it moves the physiology. The study was observational: heavier lifetime use went with more fragmented recorded nights, and causation is a separate claim. Whichever reading holds, only one of the two measurements moved. 1
The sample a number was earned in
On October 5 the Journal of Clinical Sleep Medicine published a commentary on a consumer smartwatch that has crossed into the clinic's territory. Its subject is the Samsung Galaxy Watch, cleared to flag moderate-to-severe obstructive sleep apnea, and its authors, two sleep physicians at Penn State, argue the device is not yet ready to stand alone. 2 The study it accompanies, published August 27, is the part worth reading for how its accuracy figures were earned. 3
Researchers enrolled 152 adults aged 22 and over who either already carried a moderate-to-severe diagnosis or scored 3 or higher on the STOP-Bang screen, which is a sample built to contain the disease. Participants wore the watch through two nights of in-lab polysomnography; 147 completed both, covering 1,850 hours of recorded lab sleep. Against the polysomnographic apnea-hypopnea index scored at a 4% desaturation threshold, the watch's estimate reached an area under the ROC curve of 0.94 (95% CI 0.889–0.980). At the cleared cutoff of 15 events per hour, sensitivity was 94.1% and specificity 66.7%. Moving the cutoff to 25.95, a value fitted to this cohort, raised specificity to 94.9% and dropped sensitivity to 82.4%. Inside the subgroup carrying a high hypoxic burden, the accumulated depth, duration and frequency of oxygen desaturation, the watch separated low from high risk at 100% sensitivity and 100% specificity. 3
The arithmetic behind the commentary's caution is plain. A positive predictive value depends on how common the condition is among the people tested, so a figure earned where the disease is dense will not survive the move to a general population. At a specificity of 66.7%, about one in three people without moderate-to-severe apnea would test positive, and in an unselected screening population that share rises. Accuracy belongs to the sample that produced it. 2
The comparison the maker chose
Eight Sleep reported new research on October 5 and again on October 11: a retrospective study of its own members across pregnancy and the six months after birth, with a preprint posted to Research Square. 456
The cohort covered up to 101 pregnancies and roughly 46,000 nights. Women lowered their Pod temperature by about 2°C across pregnancy, most in the third trimester; snoring peaked 4.7% above the pre-pregnancy baseline in that trimester; heart rate rose 7.7 bpm and heart rate variability fell 17.9 ms from baseline. Against women not sleeping on the Pod, members in the third trimester averaged 12 more minutes of sleep a night, including 7 more minutes of deep sleep, a resting heart rate 3 bpm lower and heart rate variability 2 ms higher. After delivery, heart rate and heart rate variability recovered past the pre-pregnancy baseline within 12 weeks, while total sleep time, nighttime awakenings and REM sleep had not returned by 24 weeks. 6
One methodological detail decides how much to take from this. The comparison group is the mean of values reported across four published wearable cohorts that used the Apple Watch, the Oura Ring, Whoop and Fitbit. There was no concurrent control group, the design is retrospective, the authors work for Eight Sleep, and every effect is a cohort average with individual variation left unquantified. That makes this a hypothesis about overnight temperature regulation, which is a reasonable thing for a preprint to be, and not a trial of a mattress cover in pregnancy. 4
The bar you set yourself
The study that moved the line itself was announced by the National University of Singapore on October 6 and published online in Sleep Medicine on September 30. Its authors gave 462 adults sleep trackers and daily smartphone assessments for a year and asked a question a population threshold cannot answer: was last night short for this person? 7
Their personal cutoff sat an hour below each person's own 75th percentile of sleep duration, the upper part of someone's distribution being a practical stand-in for the nights when work and social life ease off. About 35% of nights fell below six hours and about 37% below the personal cutoff, and the median personal cutoff landed at roughly 6.1 hours, close to the familiar rule. The population rate held; the membership changed. A habitual six-hour sleeper with regular timing seldom crosses their own line, while someone who usually sleeps much longer can fall well short of their own range and stay above six hours.
Personalized short nights tracked how people said they felt more strongly than the fixed rule did, across alertness, sleep satisfaction, stress, motivation, sadness, sleeping heart rate and heart rate variability, and the longer a run of them ran the larger the association. A higher proportion of personal short nights also tracked body roundness, systolic blood pressure, pulse-wave velocity and arterial stiffness. Across 376,758 anonymized Oura users in seven countries the shape repeated: Australia slept earlier and longer with fewer short nights, Japan later and shorter with more. 8
The recovery numbers are the part to keep. One personal short night was followed by about 50 minutes of extra sleep on the next weekday night. Additional consecutive short nights bought less than ten minutes each. Weekend recovery reached about 1.25 hours and then flattened, so the payback is capped by the schedule around it. For 86.4% of participants the median run of short nights lasted a single night. The authors are explicit that this measures relative shortfall rather than biological sleep need, and that the health associations are observational. 8
What moved, and what did not
Two interventions ran in opposite directions this week, and each asks a different instrument for its verdict.
Lemborexant, a dual orexin receptor antagonist approved for insomnia, was tested in night shift workers whose daytime sleep is fought by their own circadian drive for wakefulness. Twenty-nine people completed the pilot trial, 15 on the drug and 14 on placebo, through a two-week baseline, a week of titration and a two-week maintenance phase. On the primary outcome, total daytime sleep duration, the two groups did not differ, by sleep diary or by wrist actigraphy, and sleep efficiency, self-rated sleep quality and insomnia severity moved no differently either. Exploratory analysis hinted that the 10 mg dose lengthened diary-recorded sleep, without reaching significance against placebo. No serious adverse events were reported. 9
Digital cognitive behavioural therapy for insomnia went the other way. A meta-analysis gathered two decades of research on one program, Sleep Healthy Using the Internet, from 25 studies and 14,166 participants. Participants given the program reported declines of at least 8 points on the insomnia severity index and at least 21 minutes of sleep-onset latency, and 64% completed a full therapeutic dose. Across 10 randomized comparisons against online education the advantage held through 6- and 12-month follow-up, with standardized mean differences of −1.03 to −0.85 on insomnia severity. 10
The hinge between the two is which instrument carries the verdict. The program with the deepest evidence base in this week's literature is scored mainly on what people say about their sleep, because for insomnia the report is the endpoint and no polysomnographic number defines a cure. The failure of the week is a drug measured both on a diary and on actigraphy, and it moved neither. A journal can report a clinical win and a clinical null in the same week without contradiction, as long as the reader can see which instrument each verdict rests on.
Four more numbers
- Yoga helped the people who showed up. A four-week pilot randomized 33 adults with insomnia symptoms to recorded Vinyasa yoga sessions (n=17) or no change in activity (n=16). Insomnia severity, actigraphy and sleep diaries moved the same in both groups, while resting heart rate fell 2.91 bpm in the yoga group and rose 2.46 bpm in controls (p=0.047). Among highly adherent participants, the yoga group's insomnia severity fell further (p=0.002). A pilot, mostly female, four weeks long. 11
- Two breathing problems compounded. In the Sleep Heart Health Study, followed a median 11.45 years, people with both obstructive sleep apnea and preserved ratio impaired spirometry, a reduced FEV1 with a preserved ratio, had a 49.4% rate of major cardiovascular or cerebrovascular events against 17.0% for those with neither. Adjusted, the combination carried a hazard ratio of 1.52 (95% CI 1.08–2.12) for those events and 2.17 (1.41–3.34) for cardiovascular death, with the association appearing mainly in women (interaction p=0.009). Observational. 12
- Wearable sleep efficiency moved with grip strength. In 165 older outpatients with mainly cardiometabolic disease, average sleep efficiency across 7 to 90 nights on a wrist device was 58.0% for those with low grip strength against 62.7% for the rest (p=.006). Each percentage point of lower device-estimated sleep efficiency carried an adjusted odds ratio of 1.04 (1.01–1.08, p=.02) for low grip strength. Cross-sectional, single centre, and the association did not extend to frailty or slow walking speed. 13
- Daytime light changed what nighttime light did, in mice. Male and female mice were kept under 50-, 100- or 300-lux daylight and then exposed to dim light at night. Under the dim daytime light the sexes diverged sharply on sleep and rhythm measures; under bright daylight the differences largely disappeared. A mouse study, and a pointer that human light studies should report the daytime intensity they ran at. 14
The maker desk and the researcher desk
Eight Sleep's pregnancy study was one of two maker research releases this week. Whoop published the other on October 8: a peer-reviewed analysis, in Sports Medicine – Open, of 3,489 WNBA regular-season games across 18 seasons from 2007 to 2024. Every additional 1,000 miles travelled in the seven days before a game was associated with about 0.37 more points conceded by the away team, and home teams still estimated to be carrying eastward jet lag averaged a point differential of +0.69 against +2.55 for home teams without it. Home-court advantage fell from 64.3% of games in 2009 to 52.3% in 2024, years in which the league improved travel conditions. The study is observational and team-level, and its authors estimate circadian resynchronization at roughly an hour a day. 1516
Oura's blog carried a brand campaign on October 8 and an expansion of its Health Radar and GLP-1 features to more than 30 markets on October 6, both company announcements rather than new sleep research. 17
On the researcher side, Matthew Walker's episode 154, "Sleep and Cortisol," published October 5, takes apart cortisol's reputation on social media as a toxic villain and walks through how the hormone regulates metabolism, immunity and survival, including the healthy rise in cortisol in the hour after waking. 18
One threshold to compute this week
Open the last month of your own sleep data and find the 75th percentile of your nightly sleep duration, the level you reach on the nights your schedule lets you. Subtract one hour. That is your personal short-night cutoff, and for most people it lands somewhere near where the population rule sits.
Then count runs rather than nights. A single night below your line is ordinary, and it usually gives back about 50 minutes on the next weekday night. What matters is a streak: each extra consecutive short night buys almost no further recovery, and weekend catch-up flattens after about 1.25 hours. Track the longest run you accumulate this month, and treat three or more nights in a row as the signal worth changing something for, whether that is guarding your bedtime, fixing your wake time, or shortening one evening commitment. Compare your longest run between two months before deciding whether the change did anything.
References
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
- 12
- 13
- 14
- 15
- 16
- 17Oura blog feed
ouraring.com
- 18#154 - Sleep and Cortisol
themattwalkerpodcast.buzzsprout.com
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.