Predicting Disease from Sleep – A New Breakthrough Study

We sleep every day, but what sleep reveals about our health is something medicine is only beginning to fully decode. For many years, polysomnography (PSG) has been considered the gold standard for sleep assessment, recording brain activity, heart function, respiration, and muscle movement simultaneously. In practice, however, this massive volume of data has only been partially utilized, mainly to diagnose common conditions such as sleep apnea.

Vào ngày 6 tháng 1 năm 2026, các nhà khoa học tại Stanford Medicine đã công bố một thành tựu đột phá mang tên SleepFM. Đây là mô hình trí tuệ nhân tạo, đa phương thức đầu tiên có khả năng dự đoán hơn 100 tình trạng sức khỏe chỉ từ dữ liệu của một đêm ngủ duy nhất. Thành tựu này đã đưa đo đa ký giấc ngủ từ vai trò chẩn đoán các rối loạn giấc ngủ quen thuộc trở thành một hệ thống sàng lọc bệnh lý toàn thân ở quy mô lớn. SleepFM sử dụng nhiều cảm biến khác nhau để ghi lại các hoạt động não (EEG), điện nhãn đồ (EOG), điện tâm đồ (ECG), điện cơ đồ (EMG), tín hiệu hô hấp, chuyển động chân, chuyển động mắt và nhiều hơn nữa.

On January 6, 2026, scientists at Stanford Medicine announced a breakthrough achievement called SleepFM. This is the first multimodal artificial intelligence model capable of predicting more than 100 health conditions using data from just a single night of sleep. This development transforms polysomnography from a tool focused on diagnosing sleep disorders into a large-scale system for screening whole-body diseases. SleepFM integrates multiple sensors to capture brain activity (EEG), electrooculography (EOG), electrocardiography (ECG), electromyography (EMG), respiratory signals, leg movements, eye movements, and more.
 

Nguồn: Ảnh AI
 

The study was conducted to fully exploit the vast amount of physiological data embedded in sleep recordings, with the goal of clarifying the complex relationship between sleep and disease. To achieve this, the research team trained SleepFM on more than 585,000 hours of data from approximately 65,000 participants. Notably, the model demonstrated impressive performance in predicting serious conditions such as Parkinson’s disease (0.89), dementia (0.85), heart failure (0.80), and all-cause mortality risk (0.84).

These findings suggest that physiological disturbances during sleep may appear very early, serving as warning signs long before neurological or cardiovascular diseases become clinically evident. For future physicians, SleepFM opens a new era in preventive medicine, enabling early, non-invasive detection of risk signals before conditions such as Alzheimer’s disease or cardiovascular disorders manifest clinically.

SleepFM represents a promising new approach that combines multimodal physiological data with artificial intelligence to deepen our understanding of the human body at rest. In the future, the model’s analytical and predictive capabilities are expected to be integrated into compact wearable devices, supporting continuous and long-term health monitoring.
 

[Source: Nature Portfolio]