The Limits of Consumer Health Data

The last chapter showed how much processing sits between a raw sensor reading and the number on your screen — noise removed, samples averaged, a pattern inferred rather than measured directly. A consumer wearable's “irregular rhythm detected” alert is the output of exactly that kind of estimate, tuned to flag anything unusual rather than to confirm a specific condition.

Flagging unusual and diagnosing a condition are not the same claim, even though a notification on your wrist can make them feel identical in the moment.

A marketing page saying a feature is “highly accurate” is describing performance under the specific conditions the validation study actually tested — and that scope is almost always narrower than how the device gets used afterward. The largest study of a wearable's irregular-rhythm alert enrolled more than 419,000 participants, but the population skewed young and healthy, and atrial fibrillation is rare in exactly that group. Of the small number who got a notification and then wore a real clinical-grade ECG patch to check it, only about a third had atrial fibrillation show up on the patch — partly because the condition can come and go, and partly because some alerts were simply false alarms.

That is not a reason to dismiss the feature — it caught real cases that would have gone unnoticed. It is a reason to read “accurate” as a claim with a population and a test protocol behind it, not a blanket guarantee that travels with the device wherever it goes.

Wellness data is built to be directionally useful across millions of different bodies cheaply and continuously — good enough to notice a trend, not built to survive a courtroom or a treatment decision. Medical data is validated against a clinical gold-standard measurement, on a specific device, cleared for a specific claim. A hospital ECG and a watch's single-lead reading can describe the same heartbeat and still not carry the same evidentiary weight.

State the boundary rather than treating “it has limits” as a vague disclaimer. A consumer wearable can only ever flag something that shows up as a pattern in the exact signal it measures — and most of what can go wrong in a body never touches that signal at all.

None of this is a flaw in a specific product. It is a description of what a sensor on your wrist physically can and cannot see, no matter how good the software behind it gets.

A device that used to give you one number a year now gives you thousands. Somewhere in that many readings, an unusual one is inevitable — not because anything is wrong, but because that is what noisy data does across a large enough sample. Sleep researchers have a name for what happens next in one specific domain: orthosomnia, an unhealthy preoccupation with hitting a perfect sleep-tracker score that can make a person's actual sleep worse, not better.

1. 1

   A continuous stream produces an unusual reading, purely by chance

   With thousands of data points a year, a statistical outlier is expected, not exceptional.

2. 2

   The reading triggers real worry, whether or not anything is actually wrong

   A number with no context reads as a warning, because that is exactly how the app presents it.

3. 3

   That worry leads to an appointment, or a repeat test, to rule something out

   Reasonable on its own — but multiplied across millions of users, a meaningful new load on a healthcare system.

4. 4

   Most of those follow-ups find nothing, at a real cost in money and time

   The confirmation gap from the last section compounds here: most flagged readings don't turn out to be the condition.

None of this argues for ignoring the data. It argues for treating a single unusual reading the way you would treat a single unusual comment from a stranger. Note it, but do not rearrange your week over it.

A single unusual reading is weak evidence — noise from the last two chapters can produce one on its own. A pattern that repeats across multiple days, especially paired with a symptom you can actually feel, is worth a real appointment where a clinician can order the medical-grade version of the same measurement. The wearable's real job is not to replace that appointment — it is to give you a reason to book it sooner than you otherwise would have.

That is the honest summary of this entire chapter: not that the data is worthless, and not that it is trustworthy either, but that it earns a specific, narrower kind of trust than the confident number on the screen implies.