Communications Medicine (Nature, 2025) followed 2,667 men aged 65+ for 12 years. Using machine learning, researchers identified three distinct sleep profiles and calculated differential risk of dementia and cardiovascular disease for each.
Bedroom of an Engadine chalet just before dawn: cold bluish light from the window, pale linen sheets, a smart ring and watch resting on the nightstand; no faces.
The three sleep profiles
Three groups identified: healthy active sleepers, sleepers with initiation difficulties, and fragmented sleepers (frequent awakenings, poor deep sleep, circadian irregularity).
Macro detail of a wearable ring and a smartwatch recording sleep phases in low light; night-blue tones, soft reflections; no faces.
The risk data
Fragmented sleepers show HR 1.35 for dementia and HR 1.32 for cardiovascular events, after multivariable adjustment.
Luminous chart of REM and deep-sleep phases projected on the dark wall of a premium clinical room; soft light, order; no faces.
In plain terms: 35% higher dementia risk and 32% higher cardiovascular risk — measured over 12 years.
The contribution of machine learning
Before this study, sleep profiles were assessed crudely.
Chalet window at dawn framing the snowy Engadine peaks seen from an unmade bed; cold natural light, stillness; no faces.
Machine learning demonstrated that complex sleep patterns — their temporal structure — are clinically significant predictors that traditional medicine systematically ignores.
How the Digital Twin applies this evidence
Silent corridor of an Alpine clinic at dawn, warm materials (wood and stone) and soft light; a mood of clinical calm; no faces.
The Digital Twin dynamically classifies your sleep profile, detecting the transition from healthy to fragmented before it consolidates. A sleep deterioration trend detected in the first weeks is worth more than any plasma biomarker emerging years later.
Bedroom of an Engadine chalet just before dawn: cold bluish light from the window, pale linen sheets, a smart ring and watch resting on the nightstand; no faces.
Cavaillès et al. — Communications Medicine / Nature (2025) — https://www.nature.com/articles/s43856-025-01019-x · www.nature.com
On the same topic
- 15 Years Before the Diagnosis, Your Sleep Already Knew. Now So Does Science.Analysis on over 1 million EHR records: sleep disorder risks emerge up to 15 years before neurodegenerative disease diagnosis.
- 52,952 People, One Clear Result: Your Sleep Predicts Atrial Fibrillation.The world's largest study on wearables and atrial fibrillation: 52,952 patients, objective sleep data as independent predictors. Discover what we measure for you.
- 9 Sleep Parameters. 4 Neurological Diagnoses. The Algorithm Redefining Neurology.A machine learning algorithm trained on 9 sleep biomarkers accurately classifies Parkinson's, Alzheimer's and Lewy Body dementia.