Can a Wearable's HRV Signal Actually Catch a Seizure as It Happens?

This piece covers two Phase 2 validation studies of HRV-based seizure detection using wearable ECG devices, both validated against video-EEG monitoring in patients with epilepsy. It does not cover consumer smartwatch HRV accuracy generally, and it is not medical advice for anyone managing a seizure disorder.

Two independent Phase 2 validation studies found that HRV-based algorithms, recorded with a wearable ECG device and validated against video-EEG monitoring, can detect seizures with high sensitivity, but only in patients whose seizures produce a large enough heart-rate change to begin with. In both studies, roughly half of patients qualified as reliable "responders": in one, 57.9% of patients met the criterion and the algorithm caught 87% of their seizures; in the other, 53.5% of patients were classified as responders, with 93.1% sensitivity for all seizures among them. For the other half of patients, HRV alone isn't a reliable seizure signal.

A different use for the same signal

Every other HRV article on this site treats heart rate variability as a slow-moving readout of stress, recovery, or autonomic health measured over hours or nights. Two Phase 2 validation studies used the same underlying signal for something much more acute and time-sensitive: detecting an epileptic seizure as it happens, using a wearable ECG device instead of the electrode-heavy video-EEG setup that's the actual diagnostic gold standard.

The premise works because many seizures, especially convulsive ones, produce a sharp, measurable spike in heart rate as the autonomic nervous system reacts. The question both studies tested is whether that spike is consistent and specific enough to build a real detection algorithm around.

The catch: about half of patients don't qualify

Both studies converge on the same limitation from two different patient cohorts and slightly different algorithms: HRV-based seizure detection depends entirely on whether a given patient's seizures actually produce a strong enough autonomic signature. In the first study, responders were defined by a greater-than-50-beats-per-minute heart-rate change during a seizure; in the second, by whether more than two-thirds of a patient's seizures were caught. In both, only about half of patients met that bar (57.9% and 53.5%, respectively).

For the patients who don't qualify, this isn't a story about the algorithm needing refinement so much as a documented physiological ceiling: their seizures may simply not move heart rate enough to be caught by any HRV-based approach as currently designed. Neither study reports what algorithm, if any, was more accurate for these non-responders.

What this doesn't establish

Neither study reports what happens after a detection: whether an alert reaches a caregiver in time to matter, or whether earlier detection changes clinical outcomes. Both are validation studies measuring whether the algorithm correctly flags a seizure that already happened, against a video-EEG record, not real-time-response or outcome studies.

The false alarm rate from the first study (0.9 per 24 hours) is a real, useful number, but the second study doesn't report a comparable figure in what's summarized here, so the two aren't directly comparable on that specific metric. Read the sensitivity figures as evidence that the underlying approach is real and reproducible across two independent cohorts, not as a claim that any consumer product implements it this way today.

Common questions