Researchers have found that analysing routine sleep study data using artificial intelligence can uncover hidden biological markers capable of predicting a wider range of long-term health risks than the conventional indicators currently used by physicians.
Sleep studies are typically performed to diagnose obstructive sleep apnoea, with clinicians primarily relying on the Apnoea-Hypopnoea Index (AHI), which measures the average number of breathing interruptions or reductions per hour of sleep.
In the study, an AI model analysed more than 10,000 overnight sleep studies and linked patterns in physiological data with patients’ long-term health outcomes recorded in their medical records.
The researchers identified five distinct patient subgroups with significantly different health risk profiles. According to the findings, individuals in the highest-risk group were twice as likely to die within five years as those in the lowest-risk group—a difference that was not detected by the traditional AHI measure.
The highest-risk group also showed a 65% greater likelihood of developing heart failure, an 84% higher risk of heart attacks, and a 93% increased risk of cognitive impairment. In addition, their risk of developing atrial fibrillation and epilepsy was more than three times higher than that of patients in the lowest-risk group.
The researchers reported that the AI model accurately predicted health outcomes in both men and women and maintained its performance when validated on an independent cohort of more than 6,000 patients.
