# Assessing REM Sleep as a Biomarker for Depression Using Consumer Wearables

## TL;DR

This study shows that Apple Watch data can detect sleep patterns linked to depression, offering a potential accessible way to monitor mental health.

## Contribution

The study demonstrates that consumer wearables can effectively capture REM sleep metrics associated with depressive symptoms.

## Key findings

- REM latency negatively correlates with depressive severity (Spearman ρ = −0.673, p < 0.001).
- REM sleep coefficient positively correlates with depressive severity (ρ = 0.678, p < 0.001).
- REM metrics together explain 62% of variance in depressive symptoms.

## Abstract

Background: Rapid-eye-movement (REM) sleep disinhibition—shorter REM latency and a larger nightly REM fraction—is a well-described laboratory correlate of major depression. Whether the same pattern can be captured efficiently with consumer wearables in everyday settings remains unclear. We therefore quantified REM latency and proportion of REM sleep out of total sleep duration (labeled “REM sleep coefficient”) from Apple Watch recordings and examined their association with depressive symptoms. Methods: 191 adults wore an Apple Watch for 15 consecutive nights while a custom iOS app streamed raw accelerometry and heart-rate data. Sleep stages were scored with a neural-network model previously validated against polysomnography. REM latency and REM sleep coefficient were averaged per participant. Depressive severity was assessed twice with the Beck Depression Inventory and averaged. Descriptive statistics, normality tests, Spearman correlations, and ordinary-least-squares regressions were performed. Results: Mean ± SD values were BDI 13.52 ± 6.79, REM sleep coefficient 24.05 ± 6.52, and REM latency 103.63 ± 15.44 min. REM latency correlated negatively with BDI (Spearman ρ = −0.673, p < 0.001), whereas REM sleep coefficient correlated positively (ρ = 0.678, p < 0.001). Combined in a bivariate model, the two REM metrics explained 62% of variance in depressive severity. Conclusions: Wearable-derived REM latency and REM proportion jointly capture a large share of depressive-symptom variability, indicating their potential utility as accessible digital biomarkers. Larger longitudinal and interventional studies are needed to determine whether modifying REM architecture can alter the course of depression.