Nonspecific Electrocardiographic ST-T Abnormalities: Cardiac Repolarization Memory Assessment

In this issue of the journal, Shoji et al. showed that isolated nonspecific ST-segment and T-wave abnormalities (NSSTTAs) are associated with incident heart failure (HF).[1] The authors conducted a large prospective general population cohort study of more than 30,000 African-American and White participants aged ≥ 45 years, with more than 13 years of follow-up. Incident HF events were rigorously adjudicated and classified based on left ventricular ejection fraction (LVEF) measurement at or near the time of incident HF as HFrEF (LVEF < 50%) or HFpEF (LVEF ≥ 50%). Cause-specific Cox regression models were adjusted for known HF risk factors, statin use, and time-updated incident coronary heart disease (CHD) events. Remarkably, after adjustment, isolated NSSTTAs’ presence indicated a more than doubled risk of HFrEF and a 66% increased risk of HFpEF.[1]

The authors should be congratulated on this important study. In the era of artificial intelligence (AI), the authors reminded us of the importance of traditional ECG interpretation, which facilitates scientific inquiry and clinical judgment. We have to remind ourselves that it is important to actually look at the ECG in order to interpret it appropriately.

Figure 1 illustrates isolated NSSTTA, and Figure 2 shows the vectorcardiogram (VCG) of the same 12-lead ECG that is shown in Figure 1, reconstructed by applying the Kors transformation. In this example, the most prominent features of the ECG and VCG are the minuscule T-loop (T-wave). What is the mechanism behind isolated NSSTTA? Shoji et al.[1] speculated that fibrosis or ischemia are the two most likely underlying pathological processes.

Figure 1.

Representative example of a 12-lead ECG with isolated nonspecific ST-segment and T-wave abnormality, MC 5–4.

Figure 2.

The vectorcardiogram obtained by the Kors transformation of an ECG is presented in Figure 1. Red dot marks the beginning of the QRS loop. Ventricular activation along the QRS loop is shown by color transition from red to orange, then yellow. Progression of repolarization along the T loop is shown by a color transition from green to blue, then purple. Dots on VCG loops are placed every 10 seconds. Peak QRS vector (red), peak T vector (green), and peak Spatial Ventricular Gradient Vector (SVG) are shown.

Another potentially important mechanism involves the development of cardiac memory in response to intermittently abnormal ventricular activation, such as premature ventricular complexes (PVCs) or intermittent bundle branch block, in individuals with preexisting cardiac remodeling caused by repeated subclinical ischemia-reperfusion events or reduced repolarization reserve, possibly due to genetics. My group previously found that a weak repolarization memory response to right ventricular pacing indicates saturation of the already remodeled repolarization.[2] For example, different origins of PVCs produce distinct directions of abnormal activation vectors and, consequently, different directions of cardiac memory’s repolarization vectors, which can cancel each other out, resulting in a very small T-wave. Notably, we observed that women develop less cardiac memory than men.[2] Consistent with our previous findings, Shoji et al.[1] described a higher prevalence of isolated NSSTTA in women than in men. Further studies are needed to assess the burden of PVCs and other intermittent arrhythmias in individuals with isolated NSSTTA. Cardiac ablation of PVCs can reverse tachycardia-induced cardiomyopathy, and isolated NSSTTA may serve as a useful biomarker for patients who could benefit from ablation.[3]

Another important lesson highlighted by Shoji et al.[1] is about collider bias, which is frequently overlooked in observational studies. The authors draw a Directed Acyclic Graph (DAG) that illustrates two potential sources of bias in this study. I encourage readers to review Supplementary Figure 1 in the manuscript.[1] DAGs were developed as a powerful graphical syntax for causal inference.[4] Reading DAGs as nonparametric structural equation models gives them a causal interpretation. DAGs are helpful in highlighting causal assumptions of the data-generating model, deriving testable implications of a causal model, and performing causal identification analysis.[5] DAGs inform investigators about what variables they should control for, or not control for, in regression analysis.[4] When building a causal model (i.e., drawing a DAG), one must consider all factors that play a role in data generation, regardless of whether they are observed or unobserved. Careful inspection of the DAG reveals that many of its assumptions are questionable. For example, the authors assumed that unmeasured confounders in their study could affect incident coronary heart disease and incident HF but could not manifest as isolated NSSTTAs. The authors did not provide evidence supporting such an assumption. Overall, the presented DAG does not reflect the complete “true world” of incident HF development, and, therefore, does not permit causal interpretation of the study findings. Furthermore, the presented DAG shows several colliders and selection biases, indicating that the causal effect cannot be identified in this study.

The possibility of spurious, not causal, findings is especially applicable to the statistical interaction reported by the authors between isolated NSSTTAs and race. Recent AI-ECG studies showed that deep learning models can differentiate between White and African-American individuals, and that model performance differs between high- and low-income groups.[6] Therefore, any suggestions regarding which race can benefit more from NSSTTAs screening can inadvertently exacerbate biases that already exist in risk stratification and prevention of HF and other cardiovascular outcomes.[7]

In summary, isolated NSSTTAs as manifestations of cardiac repolarization memory highlight the importance of cardiac memory as a universal adaptive response of the heart, reflecting adjustment to a new activation path (whether permanent or intermittent).[8] Cardiac memory is a protective mechanism aiming to minimize dyssynchrony and optimize heart performance. Early detection of cardiac memory in the making can potentially prevent HF and other adverse cardiovascular outcomes. Both 12-lead ECG and VCG provide complementary information, helping to recognize cardiac memory early.