clinicians.build · interactive · September 21, 2026 · built on Eric Topol, Ground Truths, Sep 19 2026

Find the Trial That Doesn’t Exist

Eric Topol went looking for studies tying wearable heart rate variability to a health outcome and came back with none. Here is the same search run against the trial registry instead of the literature — 3,801 registered trials that measure HRV, one dot each. Set the filters he was looking for and watch the field go to zero.

Primary source: Eric Topol, “The Big Holes in Wearable Heart Rate Variability And Readiness Scores,” Ground Truths, September 19, 2026 — on readiness scores: “They are unvalidated against health outcomes.”
Data: ClinicalTrials.gov via the AACT registry mirror in MIMI Labs, snapshot 2026-08-17. Every study whose declared outcome measures mention heart rate variability, HRV, RMSSD, SDNN or pulse rate variability, with a stated enrollment and a start year between 1995 and 2030: 3,801 trials. Instrument and endpoint classes are derived by text matching on registry fields — see the limits at the bottom.

Read Topol’s review → Or read today’s newsletter →

the number the registry gives back

registered trials, out of 3,801 that measure HRV, that name a consumer wearable, set a hard health outcome as the primary endpoint, and have posted results.
Relax any one of those three and you still get at most a handful. The filters below let you try.

Every HRV trial ever registered

Each dot is one registered study. Horizontal axis is the year it started; vertical axis is enrollment, on a log scale, because the distribution demands one — the median HRV trial enrolls 60 people.

Instrument named in the record

Endpoint

Additional requirements

Minimum enrollment — drag to dissolve the small-n cloud 1

Named consumer wearable Generic wearable / PPG ECG or Holter Not stated

Trials matching3,801 Median enrollment60 Have posted results5.4% Hard primary endpoint102

Hover a dot for the NCT number, size and year — every point links back to its record on ClinicalTrials.gov.

The named devices, one row each

Seventy-four of the 3,801 name a consumer brand in the title, the intervention, or the outcome measure itself. Here is what those trials set out to measure.

“Hard health outcome” = the outcome text names mortality, death, survival, a cardiovascular event or MACE, hospitalisation or readmission, stroke, myocardial infarction, sudden cardiac death, or incident diabetes / hypertension / cancer / heart disease. Everything else — HRV itself, sleep scores, perceived stress, feasibility, usability, adherence — counts as soft.

80/20 If you are ingesting a readiness or HRV field from a consumer device, the registry will not tell you what it means, because nobody has registered the study that would. What it will tell you, in one query, is whether the vendor in front of you has a trial at all, what it enrolled, and whether results were ever posted. That is a five-minute check and it belongs in your vendor file before the integration does.

“To my surprise, I found none.” — Eric Topol, on searching for high-quality studies linking HRV and readiness scores to health outcomes

Where this chart is thin — read before you quote it

The instrument is usually missing, and that is the finding underneath the finding. 3,418 of 3,801 trials — 90% — never name what measured the HRV anywhere in their title, intervention or outcome text. A dot in grey is not an ECG study and not a wearable study; it is a study you cannot classify. Filter to “not stated” and you are looking at the registry’s largest blind spot.
This is keyword matching, not adjudication. Classes come from regular expressions over registry free text. A trial that used an Oura ring but wrote “wrist-worn device” lands in the generic bucket; one that mentions mortality only as a safety monitoring note counts as naming a hard outcome. Read the record before you rely on any single dot.
Registered is not published, and posted is not the same as published. Only 205 of 3,801 have results posted to ClinicalTrials.gov, which is a legal requirement for some trials and not others. A study can be published in a journal and never post here. The 5.4% is a floor on reporting, not a ceiling on evidence.
Enrollment is often a plan, not a count. Where a trial has not finished, the number is the sponsor’s target. The minimum-enrollment slider therefore filters on intention as much as on achievement.
A registry search cannot prove a negative. It shows that the trial Topol was looking for has not been registered in the United States under terms that name it. Observational cohorts, non-US registries and company-internal analyses sit outside this file entirely — including, by his account, a WHOOP study of 2,700-plus runners that is not yet published.
Small n is the norm here, not an outlier problem. 2,814 of the 3,801 trials enroll 100 people or fewer. Drag the slider to 1,000 and 55 trials remain out of 3,801 — which is the point of the control, but it also means almost any subgroup you carve out will be too small to support a trend.

What would change the picture

One prospective trial, in a population that is not exclusively young and healthy, with a readiness or HRV score as the exposure and a named health outcome as the primary endpoint, powered and registered. It would sit in the top-right of the chart above, in red, alone. As of the August 2026 snapshot the space is empty — and the companies with the users and the money to fill it are the ones who have not.