health π§ AI spots long-term health risks during sleep
Researchers developed a transformer-based foundation model using over 10,000 clinical polysomnography recordings. This model analyzed multimodal sleep physiology to separate patients into five stable risk groups. The highest-risk group showed a mortality hazard ratio of 2.38 compared to the lowest-risk group. Conventional apnea-hypopnea index categories were not significantly associated with mortality, proving the model's edge. This AI approach might help flag patients for early neurological or cardiovascular assessment.