Abstract
Introduction:
ECG-based predictive analytics have typically focused on static assessment of beat-to-beat heart rate variability (HRV) during wakefulness. However, there is increasing evidence that quantification of time-related changes as well as non-beat-to-beat complexity indices, such as entropy, can provide important untapped insight.
Methods:
Analysis of HRV, QT variability (QTV), QT entropy (QTen), QT entropy rate (QTa) and RR entropy (RRen) in 5-min bins were performed on home polysomnographs of 6,300 asymptomatic community adults (Sleep Heart Health Study). Rate of change of ECG variables was quantified by linear regression in one- to four-hour epochs. Stepwise regression was used to identify age-, sex-, race-, comorbidity-, and sleep state-associated influences on ECG values.
Results:
Predictive multivariate models were generated for congestive heart failure, diabetes, obstructive sleep apnea, and history of myocardial infarction, each with sleep state-specific and ECG variable rate of change being of the highest predictive value in ECG-based models (Table 1).
Conclusion:
Although aggregate beat-to-beat HRV is associated with cardiovascular risk factors, unique time-varying and sleep stage-specific findings provide additional predictive power. By way of demonstrating that distinct classes of dynamic ECG features were predictive of each risk factor studied, our study suggests a potential role for patient-specific ECG phenotyping using a comprehensive analytical platform. Dynamic ECG assessment during sleep with attention to entropy-derived measures, in particular, may enhance precise risk stratification in otherwise low-risk individuals.