A research team from Stanford University recently published a groundbreaking achievement in the prestigious medical journal "Nature Medicine." They developed an open-source artificial intelligence model called SleepFM, which can accurately predict an individual's health status and risk of death over the next six years by analyzing just one night of sleep monitoring data.

Deep Decoding of Physiological Signals

This study was trained on a massive clinical dataset spanning 25 years and involving 65,000 participants. Unlike simple monitoring by smartwatches, the model deeply integrates multiple physiological signals such as electroencephalogram (EEG), electrocardiogram (ECG), and respiration, enabling it to detect subtle abnormalities hidden during sleep.

SleepFM achieved an accuracy rate of 84% in predicting overall mortality, and its performance in predicting dementia reached as high as 85%. Additionally, the model's concordance indices for serious cardiovascular diseases such as heart failure and myocardial infarction are at industry-leading levels.

Promoting Universal Healthcare Early Warning

Although the technology currently mainly relies on professional polysomnography equipment, its core algorithm adopts a channel-agnostic design. This means that in the near future, the technology is expected to be adapted to portable terminals such as smartwatches, providing basic health warnings to the general public through simplified signals such as ECG or respiration alone.

The open-source nature of this research also offers new ideas for the secondary utilization of medical resources. Every year, the vast amount of raw sleep monitoring data around the world can now be transformed into valuable health management advice through this model, significantly improving the efficiency of the healthcare system's screening process.