# Assessment of a biometric shirt for sleep body position identification in epilepsy

## TL;DR

This study evaluates a biometric shirt's ability to detect sleep positions in epilepsy patients, which could help reduce SUDEP risk by monitoring prone sleeping.

## Contribution

The study provides empirical validation of the Hexoskin biometric shirt's algorithm for sleep position detection in epilepsy patients.

## Key findings

- The Hexoskin shirt correctly classified 65-94% of sleep positions across different categories.
- Balanced accuracy was 0.76 and weighted F1-score was 0.85.
- Improving prone position detection could enhance the shirt's clinical utility for epilepsy monitoring.

## Abstract

Patients with uncontrolled epilepsy are at increased risk of sudden unexpected death in epilepsy (SUDEP). Evidence suggests that sleeping prone or being in a prone position after a seizure may increase the risk of SUDEP. A few wearable devices have the potential to track sleeping habits. These devices could eventually be used to screen patients with epilepsy with a tendency to sleep in a prone position, allowing interventions such as sleep training to influence an ideal sleep position. Additionally, they could continuously monitor body positioning, allowing for responsive alarms and/or interventions when necessary. In this study, we prospectively assessed the accuracy of the Hexoskin biometric shirt algorithm in identifying sleep body positions.

Patients were recruited at the University of Montreal Health Center (CHUM) epilepsy monitoring unit and were asked to wear the Hexoskin biometric shirt. A built-in algorithm identified prone, supine, right, left, or sitting/standing body positions using an accelerometer. Sleeping positions predicted by the algorithm were compared to “true” values collected via blind simultaneous video analysis.

Across 10 patients and 347 h of sleep analyzed, 65% of prone, 75% of supine, 94% of right lateral decubitus, 81% of left lateral decubitus, and 65% of sitting/standing positions were correctly classified by the Hexoskin algorithm. Balanced accuracy was 0.76 and weighted F1-score was 0.85.

Our results show promise in the use of the Hexoskin shirt for detecting sleep positions. Optimizing performance in identifying prone sleep could enhance its clinical utility for monitoring patients with epilepsy.