Researchers analyzing nocturnal data from 162 older adults, 111 cognitively normal and 51 with mild cognitive impairment (MCI), found that night-to-night variability in wearable-derived sleep and heart rate measures classified MCI status more accurately than models based on long-term averages alone, according to a study published in PMC. Participants wore a ring-type wearable device recording heart rate, heart rate variability (HRV), and sleep structure over an average of 70 days per person, the study reports. The researchers built variability-based features, including measures of dispersion and time-series dynamics, and compared logistic regression models using these features against models built on simple long-term averages of the same signals. The study describes this dilution problem directly: averaging nightly readings over long stretches can wash out meaningful signal, a phenomenon the authors term signal dilution from long-term averaging. Models incorporating variability metrics outperformed average-based models in distinguishing the MCI group from cognitively normal participants, the paper reports, though the study does not describe this as a diagnostic tool ready for clinical use.