Background: Periodic repolarization dynamics (PRD) is an electrocardiographic biomarker that quantifies low-frequency (LF), sympathetic-activity associated instabilities of repolarization. PRD is a strong predictor of mortality in patients after myocardial infarction (MI).  Calculation of PRD using 10-second ECGs would be very advantageous, allowing the implementation of this technology in everyday clinical practice.

Purpose: We aimed to develop a novel method to calculate PRD from 10-second ECG recordings, which we called PRD_short. We finally aimed to test the prognostic value of PRD_short and the relation between PRD and PRD_short in different subgroups. 

Methods: First, the beat-to-beat change in the direction of repolarization, called dT° was measured for 30 minute-ECGs (Figure 1) and PRD was quantified as the amplitude of LF periodicities (≤0.1 Hz) within dT°. We randomly selected segments with a duration of 10-seconds. For each of these segments we calculated several parameters based on dT° and RR-interval. To overcome the issue that the wavelength of PRD is longer than 10-seconds, we performed signal-simulation and machine learning analysis. We simulated 100.000 dT°-signals using different assumptions for the level of PRD, heart rate, respiratory rate, number of premature ventricular contractions and the level of artifacts. Thereafter we used machine learning to calculate PRD from single 10-second ECG recordings (Figure 1). This method was finally validated in a cohort including 455 patients after MI. The primary endpoint was 3-year mortality. The relation between PRD and PRD_short for different subgroups was estimated using linear regression analysis. The prognostic value of PRD_short as predictor of all-cause mortality was evaluated using Cox-regression analyses.                                                                                                                   

Results: 

In the PRD-MI cohort 47 patients died within 27±11 months of follow-up. PRD_short was significantly higher in non-survivors (6.8±5.7 deg²) than survivors (4.9±3.0 deg²;p<0.001).  Dichotomization of PRD_short at the median value of ≥/< 5.0 deg² identified a high-risk group with a 3-year mortality rate of 21.0% (13.4-27.9%) compared to a mortality rate of 6.5% (2.7-10.2%;HR=3.2;1.6–6.2;p<0.001) among patients with PRD_short < 5.0 deg². The Pearson’s correlation coefficient and the beta-coefficient between PRD and PRD_short were 0.75 (0.70–0.78) and 1.25 (1.15 – 1.35), respectively. The dismatch between PRD and PRD_short significantly increased in patients with heart rate ≥ 80 bpm (beta 1.66; 95% CI 1.41 – 1.92; p-interaction < 0.001) and left-ventricular ejection fraction (LVEF) ≤ 35% (beta 1.47; 95% CI 1.23 – 1.71; p-interaction = 0.031), indicating that PRD_short significantly underestimates PRD in these subgroups (Figure 2A). There was no significant interaction for prediction of 3-year mortality within subgroups (Figure 2B). 

Conclusion: PRD_short is a novel method to calculate PRD from 10-second ECG-recordings. As normal 12-lead ECG-recordings are ubiquitous in every hospital and doctor’s office this method may allow the wide application of PRD as risk stratification tool in everyday clinical practice. In patients with resting heart-rate of ≥ 80 bpm and LVEF ≤ 35% a 30-min calculation of PRD should be considered, as PRD_short underestimates the real value of PRD.