Wearables can catch a convulsion. A trial shows how far they are from the rest

A seizure-detection wearable is a wrist- or body-worn sensor that watches for the physical signature of a seizure — the shaking of a convulsion, a surge in heart rate, a change in skin conductance — and raises an alarm so a carer can respond. On the seizures that shake the whole body, the technology works reasonably well; on the more common seizures that do not, a large validation study shows it is still a long way from reliable [s1][s2].

The distinction is the whole story. A tonic-clonic seizure — the stiffening-and-jerking kind — throws off signals a wrist accelerometer can catch. A focal seizure, which can be little more than a spell of altered awareness with no dramatic movement, may leave almost nothing for an external sensor to see.

What the pooled evidence shows

An updated systematic review and meta-analysis, using video-EEG monitoring as the reference standard, separated the two. Across 31 studies of tonic-clonic seizures — 2,128 patients and more than 112,000 hours of monitoring — non-invasive wearables reached a pooled sensitivity of 89.9% (95% CI 85.0–93.9), with roughly 1.43 false alarms per 24 hours (95% CI 1.22–1.64) [s2]. That is the mature end of the field: catching around nine in ten major convulsive events, at a false-alarm rate a household might tolerate.

Focal seizures were a different picture. Across 9 studies and 660 patients, pooled sensitivity was 73.5% (95% CI 57.4–87.0) — the wide interval itself a sign of how unsettled the evidence is — and the false-alarm rate roughly doubled to 2.85 per 24 hours (95% CI 1.40–5.81) [s2]. Wrist-worn devices edged out body-worn ones on sensitivity but at a higher false-alarm cost [s2]. Detection you can rely on for a convulsion becomes detection you have to second-guess for everything else.

The device built for the hard case

That gap is exactly what the SeizeIT2 study set out to probe. Most marketed wearables target the major motor seizures or rely on semi-invasive implants under the scalp; SeizeIT2 tested a non-invasive alternative aimed at the harder, more varied focal seizures [s1]. The device, called the Sensor Dot, is multimodal — it records EEG from behind the ear together with the electrical activity of the heart — on the logic that combining a brain signal with a cardiac one might catch events that either alone would miss [s1].

The validation was rigorous. Across multiple European epilepsy centres, 192 adults with drug-resistant focal epilepsy wore the Sensor Dot while undergoing hospital video-EEG monitoring, the gold standard, for a mean of five days — capturing 616 focal seizures with a definitive reference to check against [s1].

The numbers, and why they are sobering

Run on its own, the detection algorithm flagged 73% of the focal seizures — but at a precision of 0.004, meaning the overwhelming majority of its alerts were false, an alarm going off constantly for almost nothing [s1]. In that state the tool is unusable: its F1 score, which balances catching events against crying wolf, was 0.01 [s1].

Adding a human expert to review the algorithm's flags fixed the false alarms and broke the detection. After blinded review, precision rose to 0.83 — most alerts now real — but sensitivity collapsed to 0.31, so the device-plus-reviewer caught fewer than a third of seizures [s1]. The authors' honest summary is that this pairing performs about as well as patients simply keeping a seizure diary on sensitivity, while beating the diary on precision (0.83 versus 0.60) [s1].

Where it worked, it worked for a reason: in seizures that produced both a clear electrical pattern on the behind-the-ear EEG and a jump in heart rate, sensitivity reached 0.74; when the electrical activity was there but the heart rate did not move, it fell to 0.60 [s1]. The multimodal bet pays off only when both signals fire.

Why it matters

Seizure wearables are sold, and worn, on the promise of a safety net — an alert when someone seizes alone, a log more honest than memory. For convulsive seizures that promise is largely met [s2]. For the focal seizures that make up much of the burden of epilepsy, SeizeIT2 is a caution: the best non-invasive, multimodal attempt to date lands, after the false alarms are cleaned up, roughly where an unaided diary already sits [s1]. This is the same distance between a plausible signal and a dependable clinical tool that runs through the evidence on voice biomarkers, consumer EEG headbands and wearable fall detection — a demo is not a validation, and the gap only shows up when a device is measured against a real reference standard.

What to watch

Whether the multimodal approach can be tuned to lift focal sensitivity without reopening the false-alarm floodgates, and whether the algorithm can eventually do the filtering the human reviewer did here — because a device that needs an expert to read its output is not the unattended safety net it is sold as. Until then, the useful thing to know is which kind of seizure a given wearable was actually tested on, and against what.

This article is informational and is not medical advice.

Sources

• [s1] Swinnen L, Bhagubai M, Chatzichristos C, et al. "A multicenter, video-EEG-based validation of a multimodal wearable device for focal seizure detection in adults: The SeizeIT2 study." Epilepsia Open, 11(3):883–894, published online 8 April 2026. https://doi.org/10.1002/epi4.70260
• [s2] "Automated seizure detection using wearable devices: updated systematic review and meta-analysis of tonic-clonic and focal seizure detection." Frontiers in Bioengineering and Biotechnology, published online 26 June 2026. https://doi.org/10.3389/fbioe.2026.1833080