Background:

Periodic repolarization dynamics (PRD) is an electrocardiographic biomarker that quantifies low-frequency (LF) instabilities of repolarization. PRD is a strong predictor of mortality in patients with ischaemic and non-ischaemic cardiomyopathy. Until recently, two methods for calculating PRD have been proposed. The wavelet analysis has been widely tested (ART, MADIT-II, PRD-MI and DANISH trials) and quantifies PRD in deg² units by application of continuous wavelet transformation (CWT). The phase rectified signal averaging method (PRSA) is an algebraic method, which doesn't require additional software tools, has been validated in the EU-CERT-ICD and MUSIC trials, and quantifies PRD in deg units. The correlation between the two methods remains as unknown yet.

Purpose:

To investigate the correlation between the CWT and PRSA methods for calculation of PRD and to propose a conversion algorithm between the two methods.

Method:

The first step for quantifying PRD is to calculate the beat-to-beat change in the direction of repolarization, called dT°. PRD is subsequently quantified as the amplitude of LF periodicities (≤0.1 Hz) within dT° by means of either CWT or PRSA-transformation. We simulated 1.000.000 dT°-signals using different assumptions for the following covariates: level and frequency of dT°, heart rate, respiratory rate, premature ventricular contractions per minute(PVCs) and signal-to-noise ratio (SNR). For each simulated signal we calculated PRD using the CWT and PRSA-methods. We used forward stepwise linear regression analysis to calculate PRD_wavelet from PRD_PRSA after adjustment for the covariates included in the simulation process (Fig. 1). For each model we calculated the predicted PRD_wavelet, based on PRD_PRSA and the model coefficients. We then calculated Pearson’s correlation coefficient (r) between the predicted and real values of PRD_wavelet. The simplest model that maximized r was finally selected to be validated in a cohort of 455 patients after myocardial infarction (MI). All patients were in sinus rhythm and underwent a high-resolution 20-min ECG recording (2kHz; TMS Porti) under resting conditions.

Results:
Figure 1 illustrates the stepwise regression models and the estimated r between the real and calculated values of PRD_wavelet in the simulation cohort. The r between PRD_wavelet and PRD_PRSA without adjustment for other covariates was 0.90 (95% CI 0.90–0.90). Addition of the mean-RRi and mean-dT° in the model maximized r to 0.97 (0.97–0.97). Inclusion of additional covariates did not show any advantage with respect to r. The median difference between the calculated and real values of PRD_wavelet was 0.07 (IQR 0.59) deg². We validated the model in the post-MI cohort. In this cohort r between PRD_wavelet and PRD_PRSA was 0.91 (0.89–0.92). Application of the conversion algorithm increased r to 0.94 (0.92–0.95; p < 0.001 for the difference; Fig. 2). The median difference between calculated and real PRD_wavelet was -0.03 (IQR 1.31) deg².

Conclusion:
This is the first analytical investigation of the different methods used to calculate PRD using simulated and clinical data. In this article we propose a novel algorithm for converting PRD_PRSA to the widely validated PRD_wavelet, by adjusting for mean-RRi and the mean-dT°. All these parameters (PRD_PRSA, mean-RRi and mean-dT°) can be easily calculated without the need for additional software, which might allow the wide application of the method in everyday clinical practice.