Future Wellness Briefing · Issue 004 · Friday 3 October 2026
Issue No. 004: Right on Average
Two wearable heart studies. One device is wrong in a way you can correct for. The other is right on average and still can't tell you anything about the person wearing it.
Plus: the first sham-controlled test of a cooling mattress topper, where everything the sleeper reported moved and nothing the sensors recorded did.
Measurement is bought as accuracy. What the testing establishes is how close a device sits on average, and that is a different thing.
Right about everyone, wrong about you
A watch reads your heart's electrical signal at the wrist and pulls two numbers out of it. One is the gap between beats, which is what rhythm tracking rests on. The other is how long the heart takes to reset electrically after each beat. Clinicians watch that reset, because when it runs long the heart can tip into a dangerous rhythm.
Tested on 28 adults against a laboratory machine, a consumer watch followed the beat gap well and read the reset about 43 milliseconds short. Short by a fixed amount is the easy failure, because once you know, you can correct for it.
The second study should change how you buy. A one-sensor device came out almost exactly right on average across 31 adults and 303 recordings already marked up by specialists. Its own authors still refuse to let it be used for that reset number, because person by person the readings scattered badly.
An average of nothing isn't the same as being right. It can mean one reading ran far too long, the next far too short, and the two cancelled on the way to the average.
So the number to ask for isn't how close a device sits on average. It's how far out it can be for one person on one reading. The average describes a group, and nobody wears a group.
The same gap, twice more
Four programs that score sleep stages automatically now disagree with an expert about as much as two experts disagree with each other, across 204 nights from 75 athletes. That reads as a pass, and it's really a ceiling: once a program matches the argument between two humans, there's nothing left to measure it against.
And in 1,290 adults wearing a ring for a fortnight, sleep and night-time heart rate separated serious anxiety and depression from none, while the mild group looked like neither. The signal finds the two ends and goes quiet in the middle, which is where screening would have to work.
The point: all of these devices passed a test. None passed the test a buyer cares about.
What to do this week
Ask one question of every supplier. How far out can this be for one person, on one reading? If they answer with an average, they have your number and aren't handing it over.
Buy the measurement, not the device. The same watch is sound for one of its two numbers and unfit for the other, and “clinically validated” never says which one it means.
Write down what it was tested against. A laboratory machine, a medical chest patch, or two humans reading a chart are three different bars, and the same device passes one and fails another.
Evidence Check
The dummy slept just as well
Cooling mattress toppers are sold almost entirely on how people say they slept. Someone has finally run the test that matters. Fourteen adults slept a week on a topper that lit up, made the right noises and did not cool, then a week on one that worked.
What moved. Against the dummy, people rated their sleep quality 8.9 per cent better, their mood 11.4 per cent better and their sharpness 14.0 per cent better.
What didn't. The monitor strapped to them recorded nothing that held up. The largest change was four and a half minutes quicker to fall asleep.
The limits. Fourteen people, not yet peer reviewed, and the dummy week always ran first, so the order of the two weeks could explain some of it.
Why it still counts. It's the most honest study the category has produced, because everything before it compared an obvious gadget against no gadget, which on a question about how you feel mostly measures the gadget being obvious.
Our position: a cooling topper has a real effect on how well people think they slept, and no shown effect on sleep itself. Sell it on comfort and on people sticking with it. Verdict: keep sleep-stage claims off the page.
From the Academy
Asking how far out a number can be for one person, before asking how accurate it is on average, is the measurement module of the Level 7 Diploma in Wellness Practice. Candidates get the raw data from a real device and are marked on what they would and wouldn't sign off. Applications for the next cohort open shortly. Read the prospectus.
Also this week
Position
A reading on a screen describes a population until somebody shows you how far out it can be for one person. Get that in writing before you sign.
For the sceptics
- Consumer smartwatch cardiac intervals. Cross-sectional method comparison, 28 healthy adults aged 21 to 65, four concurrent measurements per device against a research three-lead ECG and a clinically approved single-lead patch, intervals measured manually, repeated-measures Bland-Altman. Agreement for RR was strong (r 0.84 against the research ECG, 0.80 against the patch) and weak for QT (0.37) and QTc (0.27); mean error for QT was 42.9 ms (SD 30.1) against the research ECG and 51.4 ms (SD 32.3) for QTc. Noggle, N., Wing, D. and Godino, J. (2026) JMIR Formative Research, 10, e90009. https://doi.org/10.2196/90009
- Single-lead wearable ECG, multi-tier validation. Method comparison in 31 healthy adults at rest against a limb-lead reference, plus benchmarking on 200 LUDB and 103 QTDB expert-annotated records. Mean within-subject bias was negligible for heart rate, QT, QTc and RR (all p > 0.6), with QRS duration +7.3 ms (p = 0.066) and R-wave amplitude 0.15 mV (p < 0.0001); individual-record QT and QTc errors showed high dispersion, which the authors state precludes clinical QT interpretation from single traces. Resting healthy adults only; not extrapolable to arrhythmic or exercising states. Vrdoljak, J. and Racz, D.A. (2026) Sensors, 26(18), 5905. https://doi.org/10.3390/s26185905
- Automated sleep scoring in athletes. Retrospective criterion comparison, 204 polysomnography recordings from 75 athletes, four algorithms against manual scoring. Macro F1 0.76 (GSSC), 0.71 (U-Sleep), 0.70 (YASA), 0.69 (Luna); robustness to signal degradation R2 0.08 for GSSC and Luna against 0.16 and 0.29. Single laboratory. Crosbie, F., Hanif, U. and Aloulou, A. (2026) Physiological Measurement. https://doi.org/10.1088/1361-6579/aea9ea
- Consumer ring, depression and anxiety severity. Cross-sectional analysis within the Northern Finland Birth Cohort 1986, 1,290 participants aged 33 to 35, two weeks of ring wear, assessed against GAD-7 and HSCL-25. Association only; no causal or temporal inference. Azadifar, S., Sameh, A. and Nauha, L. (2026) BMC Medicine, 24(1). https://doi.org/10.1186/s12916-026-05276-y
- Sham-controlled thermoregulating mattress topper. Preprint, not peer reviewed. Sham-controlled within-subject, 14 healthy adults, two consecutive seven-night periods with the sham always first, so an order effect is not excluded. Perceived sleep quality +8.9% (p = 0.019), restedness +9.3% (p = 0.046), mood +11.4% (p = 0.049), mental sharpness +14.0% (p = 0.007); objective sleep onset latency -4.6 min (-20.1%, p = 0.110), not significant. Tucker, L. and Zwiebel, J. (2026) Research Square. https://doi.org/10.21203/rs.3.rs-11011445/v1
- Digital therapeutics for sleep. Systematic review and network meta-analysis, 96 randomised trials, 18,419 participants, six digital therapeutics. dCBT reduced ISI by 4.24 points (95% CI -4.83 to -3.65, moderate certainty) and was the only arm to reach the minimal clinically important difference; dMBT reduced ISI by 2.28 points (low certainty) without reaching it; NFB, dBBT and CRS showed no significant effect on any outcome. 79 of 96 trials rated some concerns, 15 high risk. Tong, Z., Ye, G. and Fan, S. (2026) BMC Medicine, 24(1). https://doi.org/10.1186/s12916-026-05129-8
- Cold water immersion, umbrella review. Fifteen systematic reviews with meta-analysis, 161 unique primary studies, AMSTAR 2 methodological quality low to critically low and GRADE certainty low to very low throughout. Berjisian, E. et al. (2026) Sports Medicine - Open, 12(1), 145. https://doi.org/10.1186/s40798-026-01116-8
- Supplements for lower-limb power. Systematic review and Bayesian network meta-analysis, 119 randomised trials; of 1,636 pairwise comparisons, 17 were rated moderate, 55 low and 1,564 very low certainty by CINeMA. Sun, R. et al. (2026) Journal of the International Society of Sports Nutrition, 23(1). https://doi.org/10.1080/15502783.2026.2740221
- PFAS in smartwatch straps. Trade coverage dated 28 September 2026 of analytical work published December 2024; PFAS detected in 15 of 22 bands tested. Underlying study not retrieved through PubMed or Scite, so no DOI is carried and the finding is reported as dated secondary coverage. California AB 2030 signed 29 September 2026, with age verification duties on retailers and online sellers and civil penalties to $500 per violation; New York's equivalent took effect April 2024; effective date not stated in the coverage.
- Both wearable ECG papers evaluate a named commercial device, and the mattress topper preprint tests a specific consumer product. Funding sources, author affiliations and manufacturer involvement were not retrieved on this pass for any of the three.
Verification note: verified 3 October 2026 via PubMed, Scite, publisher and Crossref records. Editorial notices were checked on every paper cited; none carries a retraction, correction or expression of concern. The mattress topper study is a preprint and is labelled as one throughout.
Written by Phil Learney. Every position we take is logged on the Register, with its evidence. Phil Learney is founder of HMN24, and the Briefing does not rule on HMN24 products.
Subscribe
Thursday mornings. One claim tested properly, what to do about it, and the position that goes on the Register.
It is the only wellness email I read where I can check the working.Estates director, five-property European hotel group
Every position lands on the Register
No brand is named in a negative finding, and where no trial exists we say never been tested rather than dressing up an absence as caution.