Performance evaluation of heart rate turbulence detection using an extended IPFM model

The well-known integral pulse frequency modulation (IPFM) model is used to generate the occurrence times of normal sinus rhythm. In a previous study, we extended the model to account for ectopic beats and subsequent variation in sinus rhythm that may accompany an ectopic beat; a phenomenon known as heart rate turbulence (HRT). Inspired by the extended model, a new approach to characterize HRT was presented, being based on Karhunen-Loeve basis functions. An HRT detection procedure was developed which involves a test statistic T(x), resulting from the generalized likelihood ratio test of a linear model. In this study, detector performance of T(x) is evaluated on both simulated and ECG data, and compared to the performance of turbulence onset (TO) and turbulence slope (TS). Two types of simulations were performed, both adding simulated heart rate variability and HRT to the input of the extended model. The first simulation evaluated HRT detection performance at different signal-to-noise-ratios (SNRs). The second simulation evaluated the influence of QRS detection inaccuracies on HRT detection. The performance was also studied on ventricular ectopic beats (VEBs) selected from 31 patients with myocardial ischemia. The relation between HRT and the degree of blood pressure reduction induced by a VEB (estimated as proportional to the sum of the coupling interval and the compensatory pause), as well as the relation between HRT and heart rate, were analysed. The simulation results at different SNRs showed that T(x) performs dramatically better than TO and TS. With a 95% sensitivity (Sn), the specificity (Sp) at 5 dB SNR was 94% for T(x), 51% for TO, and 64% for TS. The detection performance of T(x) was equally superior when assuming that QRS jitter was Gaussian with 1 ms std: with Sn=95%, Sp was 99% for T(x), 40% for TO, and 68% for TS. Based on the ECG data, it was found that the degree of blood pressure reduction is essentially proportional to the magnitude of the HRT; this finding was reflected by T(x), TO, as well as TS. Moreover, the three HRT parameters were found to be linearly related to heart rate: low heart rates was associated with large HRT and high heart rates with small HRT. This linear relation is probably due to that a VEB during low heart rates induces a larger blood pressure reduction than during high heart rates