Heart rate complexity associated with diabetic cardiac neuropathy

Abstract

Cardiac autonomic neuropathy (CAN) is a serious complication of diabetes mellitus that often leads to increased morbidity and mortality. An examination of heart rate provides an opportunity to investigate the functional attributes of the autonomic nervous system (ANS) and specifically cardiac rhythm. Measuring changes in heart rate is non-invasive and may indicate increased risk of arrhythmic events associated with cardiac autonomic neuropathy (CAN) leading to sudden cardiac death, especially in diabetes mellitus. However, it is also a surrogate marker for advancement of diabetic neuropathy affecting other organ systems, and pathologies that affect neural function, such as depression, schizophrenia, and Parkinson's disease. ANS modulation of the cardiac rhythm leads to short and longterm non-stationary and nonlinear changes in heart rate. In line with this inherent complexity, computational analytics are required that are sensitive yet robust enough to adequately describe this complexity. Entropy measures are a natural candidate for this application as they are able to estimate information content or complexity of the heart rate. A number of different approaches have been used, as there are many variations of entropy measures. Empirical studies suggest that multiscale Rényi entropy allows a clear picture to emerge, that describes the advancement of diabetic neuropathy and specifically cardiac autonomic neuropathy from a preclinical, non-symptomatic stage to severe signs and symptoms of disease. This chapter describes the use of multiscale Rényi entropy to diagnose CAN progression. One hundred and forty nine ECGs were recorded at 400 samples/s of 71 controls, 67 participants with early CAN and 11 with definite CAN. All recordings were preprocessed and analysed using a number of nonlinear algorithms (Multiscale Entropy, multiscale DFA and multiscale Rényi Entropy). The results indicate that Rényi entropy is a better diagnostic tool to assess CAN progression with an area under the curve (AUC) of 0.723, 0.692 and 0.862 for differentiating no CAN from early CAN, early from definite CAN and no CAN from definite CAN respectively p < 0.05. Using machine learning algorithms to identify the best subset of measures for CAN classification, an accuracy of 71% was obtained for differentiating no CAN from early CAN. Identification of early, asymptomatic disease is of great clinical importance as early intervention is known to have the best health outcomes. The current work suggests that Rényi entropy has the best discriminatory power for identifying early CAN.