Back to the blog
Research

Designing a Rigorous Clinical Study

A detection model is only as credible as the data it's validated against. Here's how we're approaching that validation — deliberately, and de-identified from day one.

Good sensors and a fast pipeline aren't enough on their own — a detection system earns trust through validation against real clinical data, collected properly. We're designing our study around that principle from the start, in consultation with clinical advisors.

What We CollectBalanced CohortCommon Questions

What We Collect — De-Identified

Every participant record is tied to a study ID, never a name. Across seven categories, we capture only what's clinically relevant to detection:

Demographics
Epilepsy clinical profile
Seizure characteristics
Medication profile
Cardiovascular & autonomic baseline
Movement & lifestyle factors
Device compliance

Designing a Balanced Cohort

An unbalanced dataset produces a model that only works for one kind of person, or one kind of seizure. Our target composition is deliberately spread out:

~50/50Target sex distribution
18–50Target age range
≥70%Motor seizures — the clearest wearable signal
≥30%Sleep seizures — tests detection off-hours

These are recruitment targets for the study design — not a completed or enrolled cohort.

We deliberately track confounders too — exercise, driving, tremor, irregular sleep — the same everyday activities that can be mistaken for a seizure in raw sensor data, so the model learns to tell them apart.

Common Questions

No — LifeLyne is a monitoring and alerting system, not a diagnostic device. The study described here is what building toward clinically validated detection actually looks like.

Every record is tied to a study ID rather than a name, and stored separately from any identifying information — consistent with the encryption and access controls described on our Safety & Privacy page.

A logged, timestamped seizure event — recorded by the participant, a caregiver, or clinical staff — compared against the sensor data from the same window. It's a deliberately manual, conservative baseline before any automated confirmation is trusted.

They directly inform retraining of the detection models, and shape whether a larger, ethics-board-reviewed trial is the right next step.

Interested in the research?

We're always open to conversations with clinical partners, researchers, and investors.

Get in Touch
Continue ReadingHow Your Watch Talks to the CloudBack to the blog