Integrating Sleep Foundation Model Representations and Digital Biomarkers for Age-Aware Cognitive Risk Prediction

Leyu Zhou1, Donglin Xie2, Guangkun Nie3, Jinshuai Gu4, Qingshuo Zhao5, Shijie Chen6, Deyun Zhang7, Yuxi Zhou3, Shenda Hong8
1School of Intelligence Science and Technology University of Science and Technology Beijing, 2National Institute of Health Data Science, Peking University, 3Peking University, 4National Institute of Health Data Science, Peking University, Beijing, China Department of Computer Science, Tianjin University of Technology, Tianjin, China., 5School of Computer Science and Engineering, Tianjin University of Technology, 6Anhui Heartvoice Medical Technology Co., 7Heartvoice Medical Technology, 8Georgia Institute of Technology

Abstract

Predicting future cognitive impairment from polysomnography (PSG) is challenging because sleep physiology varies strongly with age and across clinical centers. We integrated pretrained SleepFM representations with compact and extended digital sleep biomarkers, demographic information, and age-matched pairwise ranking in a unified prediction framework. The submitted system achieved an official validation age-conditioned AUROC of 0.626. Among candidate sleep biomarkers, transition instability showed consistent associations with cognitive impairment across all three development centers and a positive adjusted association after accounting for age, sex, BMI, and site, whereas the respiratory effort-related arousal (RERA) index showed marked center dependence. These results support combining foundation-model representations with robust digital biomarkers for age-aware multicenter cognitive-risk prediction.