Dynamic Sleep-Derived Heart Rate and Heart Rate Variability Features Associated with Glucose Metabolism Status: An Exploratory Feature-Selection Study Using Consumer Wearables

Impaired glucose metabolism, a known precursor to type 2 diabetes, is associated with dysregulation of the autonomic nervous system. To assess such autonomic states, consumer wearable devices provide continuous, non-invasive physiological monitoring and may capture autonomic signatures related to metabolic status. This exploratory study examined whether dynamic features of heart rate (HR) and heart rate variability (HRV) during sleep-derived from a consumer wrist-worn device (Fitbit)-are associated with glucose metabolism status in free-living adults. We analyzed 189 nights from 18 participants (7 participants in the higher-glycemic-risk group, estimated glycated hemoglobin (HbA1c) ≥ 5.5%; 11 participants in the lower-glycemic-risk group, estimated HbA1c < 5.5%). From 28 candidate HR/HRV variables, Elastic Net regression (α=0.5) was applied to identify features associated with nocturnal mean glucose. Fourteen features retained non-zero coefficients; notably, dynamic features capturing overnight trends and variability patterns showed stronger associations than conventional static mean values. The nocturnal trends of within-window standard deviation and variance of ln(RMSSD) (root mean square of successive differences between consecutive RR intervals, estimated here from PPG-derived inter-beat intervals; RMSSD) emerged as prominent candidates, alongside HR variability indices. Independent between-group comparisons further confirmed that two dynamic HRV features differed significantly between the lower- and higher-glycemic-risk groups (both p<0.05; Cohen's |d|>1.1). Specifically, the lower-glycemic-risk group exhibited decreasing overnight trends in HRV variability, consistent with progressive autonomic stabilization during sleep. In contrast, the higher-glycemic-risk group showed increasing variability trends, suggestive of persistent autonomic instability. These directional patterns are consistent with prior evidence linking autonomic dysfunction to impaired glucose metabolism. We characterize these findings as hypothesis-generating. The identified dynamic HR/HRV features represent physiologically plausible candidate correlates of glycemic status and warrant confirmatory investigation in larger, independent cohorts with laboratory-measured HbA1c. More broadly, this work highlights the potential of widely available, consumer-grade wearable devices to move beyond activity tracking and support continuous, real-world assessment of cardiometabolic health, thereby expanding their utility in everyday health monitoring and preventive medicine.