Abstract
Early detection of cardiac autonomic neuropathy (CAN) in patients with diabetes mellitus (DM) is of prime importance, as it will facilitate the prevention of its serious consequences. In the present work, the non-linear dynamics of electrocardiogram (ECG) recordings in 41 Type 2 DM patients with early CAN and 40 controls without clinical signs and symptoms of CAN were analysed with different implementations of Lempel-Ziv (LZ) complexity. LZ complexity is a non-linear analysis method that estimates the complexity of time series of finite length and reflects the arising rate of new patterns along the sequence. Results suggest that ECG traces are less complex in patients with early CAN than in those with no CAN. Differences were statistically significant (p < 0.05, Kruskal-Wallis test) when the LZ complexity was implemented with a three symbol conversion and the mean used to define the thresholds. Furthermore, the discriminative abilities of the different LZ complexity implementations in the context of CAN were evaluated with ROC curves. Accuracies over 65% were obtained when the mean was used to define the thresholds, with a sensitivity of 75.61% with a two symbol conversion. Our results suggest that LZ complexity might be a useful tool for an early detection of CAN from ECG recordings. Nevertheless, further studies are needed to address the possible usefulness of this methodology in the characterisation and early detection of CAN in type 2 DM patients.