Related Experiment Video
Updated: Aug 6, 2026
Multi-Modal Home Sleep Monitoring in Older Adults
Published on: January 26, 2019
Longitudinal Variability of Wearable-Derived Sleep and Heart Rate as a Digital Biomarker for Early Detection of Mild
Hyunsuk Kim1, Jin Young Park2,3,4,5
1Department of Integrative Medicine, Yonsei University College of Medicine, Seoul, Korea.
Purpose:
This study evaluated the potential of wearable-derived longitudinal data for early detection of mild cognitive impairment (MCI). We examined whether variability in nocturnal sleep and heart rate (HR) patterns reflect autonomic nervous system instability associated with cognitive decline.
Materials And Methods:
Nocturnal data from 162 participants (111 cognitively normal; 51 MCI) were collected via smart rings over an average of 70 days. To mitigate signal dilution from long-term averaging (LTA), we derived variability-based features (including measures of dispersion, distributional properties, and time-series indices) and compared them to mean-based models. Logistic regression models were constructed through stepwise integration and validated with cross-validation. Key predictors were identified using SHapley Additive exPlanations (SHAP).
Results:
Mean-only models showed limited discrimination. In contrast, variability-focused models improved performance, with the model that integrated extended mean, distributional, and time-series features achieving an area under the curve of 0.861 and sensitivity of 0.820. LTA analyses revealed instability of mean effects (signed Cohen's d changed in magnitude/direction across 35-70-day windows), supporting a shift away from long-window means. SHAP identified HR_drop variability (short-term variability, mean absolute deviation-based distributional instability, time-bin fluctuation) as top contributors, while sleep-behavior rhythm (e.g., median/trimmed mean of daily sleep count) and circadian-timing features (e.g., mode of sleep midpoint, variability in the timing of the lowest HR) provided complementary information.
Conclusion:
Longitudinal wearable monitoring suggests that variability in sleep and HR patterns can capture subtle signs of cognitive decline more effectively than conventional, mean-based measures. These variability-based digital biomarkers may support early MCI screening and personalized intervention strategies.
