The Electrocardiogram: Are We at the Dawn of a New Era?

Brian Olshansky


Eur Heart J. 2020;41(21):2000-2002. 

The electrocardiogram (ECG) has been a foundation of cardiovascular evaluation for well over a century and for many good reasons. However, prognostic markers, present on the standard ECG, and in plain sight, in patients with coronary heart disease may be there but are being missed.

Utilizing simple ECG measures, in their study reported in this issue of the European Heart Journal, Chatterjee and colleagues[1] sought to create a risk stratification model from a large derivation cohort (PREDETERMINE), with adjustment for clinical risk factors and left ventricular ejection fraction (LVEF), to assess the differential association of ECG markers for sudden and/or arrhythmic death (SAD) and competing mortality. The population included patients with coronary artery disease and prior myocardial infarction and/or mild to moderate left ventricular dysfunction (LVEF = 35–50%); atrial fibrillation and atrial flutter were excluded. An integer scoring system assigned risks for specific ECG findings based on the results from the derivation cohort. Some of the measured parameters (such as contiguous Q waves, left ventricular hypertrophy QRS duration, and JTc prolongation) were good predictors for SAD, while others [heart rate, PR prolongation, QRS fragmentation, and left bundle branch block (LBBB)] were not.

Of 5462 individuals in PREDETERMINE, 688 deaths occurred, of which 139 were SADs and 559 were non-SADs. SADs increased progressively based on ECG risk score; in the low-risk group, SAD occurred in 1.5%, whereas in the high-risk group, SAD occurred in 6.2% over a 5.1-year median follow-up (P for Δ <0.001). Results were similar after accounting for LVEF.

The predictive value of the derived ECG score was validated in patients with coronary artery disease with or without type II diabetes (ARTEMIS Study). In ARTEMIS, the 5-year cumulative index for SAD was 0.9% in the low-risk and 5.2% in the high-risk ECG score groups (P for Δ <0.001). The high-risk score was associated with greater amounts of SAD rather than non-SAD in those at greatest risk. The score appeared robust to discriminate who was at greatest risk of SAD. The validated score enriched absolute and proportional SAD risk and significantly improved risk stratification vs. clinical risk factors alone.

While there were differences in LVEF and functional class between derivation and validation cohorts, ECG risk stratification was similar in both studies, pointing to the robustness of the score. Thus, simple ECG measures can discriminate and classify SAD risk, in lieu of other factors, including LVEF. Powerful information was gleaned from standard measures in those with only a mild to moderate decrease in LVEF. The net reclassification improvement was substantial: 25% (15–34%) P < 0.001 in the derivation cohort and 28% (7–49%) P = 0.009 in the validation cohort.

This study was well performed and carefully detailed. It considered standard and reasonable 'risk domains' involving a variety of potential mechanistic and physiological parameters (anatomic, i.e. contiguous Q waves, left ventricular hypertrophy, and left atrial enlargement; conduction, i.e. QRS duration and QRS fragmentation; autonomic, i.e. resting heart rate and PR interval; and repolarization, i.e. JTc, early repolarization, and T-wave inversion).

Now, we have a simple ECG tool that can help discriminate who is at greater risk for SAD. These data, properly applied, may target a potentially large population that could benefit from an implantable cardioverter defibrillator (ICD) but is presently being excluded. The authors assessed implications of their data indicating that the number needed to treat to save one life due to SAD is 22 in those in the intermediate risk category based on ventricular function (LVEF = 35–50%). The authors even postulated a study (sample size = 2900) that could be designed to show a 20% relative risk reduction in all-cause mortality with a therapy, such as an ICD, that could treat, and prevent, SAD. It is unclear if treatment of an arrhythmia (i.e. with an ICD or early defibrillation) would alter total mortality or just transform SAD into death from a non-SAD cause. Nevertheless, such a study is worth contemplating and may even be worth undertaking. Other attempts to study patients in this LVEF category with other risk stratification means, however, have not yet been able to show benefit of an ICD.

Take home figure.

A schematic of how the ECG score was used in the derivation and validation cohorts to predict risk of sudden/arrhythmic death. Patients at high-risk of sudden/arrhythmic death based on the ECG score who have characteristics similar to those in these cohorts may benefit from an ICD. So far, no data support ICD implants utilized in this way but future randomized trials may support ICD implants for those with high ECG scores. However, prevention of sudden/arrhythmic death by an ICD does not necessarily ensure long-term survival free from non-arrhythmic death.

Some information from the report by Chatterjee and colleagues[1] could be expected, but other information was surprising. It is not surprising that contiguous Q waves, left ventricular hypertrophy, QRS duration, and repolarization are associated with SAD. What was surprising was that heart rate,[2] LBBB,[3] PR interval,[4] and contiguous QRS fragmentation[5] were not good predictors of SAD. Why? Perhaps, too few patients had a LBBB or QRS fragmentation and thus the study was underpowered to utilize these measurements. Perhaps heart rate on an ECG does not provide an accurate measure of daily heart rate or the rate is an imprecise predictor after beta-blocker therapy. The PR interval, as stratified, was also not a predictor; perhaps it is a poor predictor of SAD based on present data. The lack of predictive value of these parameters, however, may indicate the need to choose other cut-off points to strengthen the predictive value of the ECG further. Contiguous QRS fragmentation did not seem to be a predictor. Have we been looking at the wrong things for too long? I suspect not, but it is time to delve into the reasons for this lack of prediction to see if there is any value to consider them, or anything else, further as predictors of SAD.

Is a simple integer score the best way to go? Is the score properly weighing all parameters appropriately as risk predictors? Is there a better score to discriminate, and define, the greatest majority of individuals at risk of SAD who could benefit from an ICD? These questions have yet to be answered. Further, it is uncertain that the point system used in the model of Chatterjee et al. accurately and equally reflects the ability of each parameter to distinguish best who is at risk of SAD compared with non-SAD. There may be other data from the ECG that are important. Perhaps there is value in applying artificial intelligence to the ECG to better risk-stratify these patients.[6,7] However, using information available in the ECG can better determine risk for arrhythmias and SAD in patients with mild to moderate impairment in LVEF and affect the present guidelines.[8]

A warning: the population evaluated was relatively homogeneous but ECG measurements can be race and gender specific. These data were based on a population that was 5% black, 2% Asian, and 89% white in the derivation cohort and 100% white in the validation cohort. There was a predominance of men. Perhaps further assessment of this score needs to be considered in other populations and after additions of newer therapies (such as icosapent ethyl[9]).

The ECG has been with us for so many years and has stood the test of time. Yet, we may be at the dawn of a new era of ECG analysis and risk stratification. This may be only the beginning of such a novel and important effort to understand risk and outcomes better utilizing simple and standard tools.