Time-series network analysis for detecting cardiac autonomic neuropathy using RR interval data

Cardiovascular autonomic neuropathy (CAN) is highly prevalent and a serious complication in patients with diabetes mellitus. In this study, we investigate the effect of changing the degree and data length on network properties (transition asymmetry and network efficiency) to differentiate negative CAN (NCAN) subjects from definite CAN (DCAN). Forty-one patients with Type 2 diabetes mellitus were included in the study: 15 patients had definite CAN (DCAN), whilst the remaining 26 were negative for CAN (NCAN), being without clinical signs and symptoms of CAN. Symbolic Aggregate approximation (SAX) was used as the discretization procedure to convert the heart rate variability (HRV) time-series signal to network. The optimal degree (m) and data length (n) were found to be m_opt = 270 and n_opt = 200 respectively with leave-one-out accuracy of 85.37% using transition asymmetry (A(G)) and network efficiency (EF) indexes. Both, A(G) and EF indexes are found to be a potential parameter for detecting CAN in diabetes.