An Intelligent Sleep Staging Framework For Sleep-Quality Characterization And Digital Interaction Design For Older Adults

Advancing age is commonly accompanied by a decline in sleep quality, while objective clinical sleep assessment often relies on polysomnography (PSG), which is costly and burdensome for older individuals. To address this issue, we develop an age-adaptive framework using single-channel EEG for automatic sleep staging, staging-derived sleep characterization, subsequent Sleep Quality Index (SQI) calibration, and digital interaction. The proposed framework consists of three successive stages. First, single-channel EEG is acquired using a low-burden sensing configuration. Second, sleep stages are predicted using a multi-scale CNN, channel-wise feature recalibration, and BiLSTM model optimized with focal loss and age-adaptive fine-tuning. Finally, sleep-architecture and continuity measures derived from the predicted hypnogram, including total sleep time (TST), sleep efficiency (SE), wake after sleep onset (WASO), N3 proportion, REM proportion, and fragmentation index (FI), are organized as candidate variables for subsequent SQI calibration. Evaluation on older-adult recordings from Sleep-EDF yielded an accuracy of 87.6% for five-stage classification and a Cohen’s kappa of 0.83, while ablation experiments further demonstrated the effectiveness of the individual design components. Correlation and multicollinearity analyses showed physiologically meaningful associations among the staging-derived variables without severe multicollinearity, with variance inflation factor (VIF) values ranging from 1.31 to 3.48. Because Sleep-EDF does not provide concurrent subjective sleep-quality references, no SQI regression weights or categorical sleep-quality cut-offs are estimated in the present study. Instead, the SQI is formulated as a calibration framework whose weights and clinically meaningful interpretation require subsequent external calibration and validation using datasets with appropriate subjective sleep-quality reference scores.