Personalized Disease Risk Scoring Using Wearable-Derived Circadian Rhythm Variability

Circadian rhythm disruption (CRD) is emerging as a risk factor for the onset of chronic disease, but has not been fully leveraged into personalized risk scoring. This paper reviews the existing studies on the use of wearable-derived circadian rhythm variability (CRV) data for generating personalized disease risk scores, discussing the potential use of actigraphy, photoplethysmography (PPG) and heart rate variability (HRV) signals. Continuous streams of wearable data are used to extract features such as interdaily stability (IS), intradaily variability (IV), relative amplitude (RA), mesor, amplitude and circadian phase, along with clinical biomarkers, within machine learning frameworks. Synthesized evidence shows that composite CRV-based risk models have area under the receiver operating characteristic curve (AUC) up to 0.92 for sleep apnea staging, up to 0.89 for mood disorder prediction and up to 0.87 for cardiovascular risk stratification – which are improvements of around 14–19% compared to traditional clinical biomarkers alone. Using cross-population validation with consumer wearable devices, the accuracy of CRV estimation ranges from 76.9 to 88.6% for sensitivity and 79–91% for specificity, depending on the type of wearable device and algorithm. The prevalence of motion artifacts (up to 38% of raw epochs) and non-compliance with wearing (22–41% of user-days) are analyzed alongside other critical challenges in the context of data quality, as are inter-device variability (±12–18%) and mitigations. The results provide a methodological strategy for incorporating the CRV obtained from wearable devices into new generation personal risk assessment tools to improve the cardio-metabolic, neuropsychiatric and sleep-related management of patients.