Sleep model finds five risk groups in 10,000 studies

A sleep foundation model trained on 10,000 overnight studies sorted patients into five risk groups with sharply different long-term outcomes. The Nature Communications study reports that the highest-risk group had 2.71 times the adjusted mortality hazard of the lowest-risk group in the clinical cohort, while an independent national cohort preserved the risk gradient; conventional sleep-apnea severity categories did not. Healthcare builders should file the result as evidence that routine physiological tests may contain unused prognostic value, but keep prospective validation between this research and clinical deployment—the same boundary at issue in ChatGPT Health’s expansion into medical workflows.