Japanese AI Model Predicts Sleep Apnea Risk with High Accuracy

A collaborative research team from JMDC, Omron and the University of Tsukuba used large‑scale medical and lifestyle data to develop a machine‑learning model for predicting sleep‑apnea syndrome (SAS). The dataset comprised about 1.86 million Pep Up users (≈1.869 million health‑check, claims and personal health‑record entries collected between January 2022 and July 2024). Using LightGBM on 279 variables, the model achieved an AUROC of 0.898 (95 % CI 0.895‑0.901).

When applied to the population, the top 1 % of predicted scores included roughly 28 % of individuals currently receiving CPAP therapy, and the top 10 % captured about 10 % of such patients, substantially improving efficiency over random screening. The researchers highlight that Japan has an estimated 9.4 million undiagnosed “hidden” SAS cases, and the model could enable early referral for diagnostic testing and treatment, potentially extending healthy life expectancy and reducing productivity losses due to daytime sleepiness.

The findings were published in the international journal *Sleep and Breathing* in June 2026, and the team plans to expand the approach to other cardiovascular and lifestyle diseases.