Letter to the Editor: Anchoring composite electrocardiography–heart rate variability indices to a reference standard - the authors’ reply

Abstract

We respond to the editorial of Salimi and Hematpour on our study of mild myocardial injury during percutaneous coronary intervention detected by composite electrocardiography and heart rate variability indices. We agree that these indices must be anchored to an external reference standard, and outline a prospective design using paired high-sensitivity troponin and cardiac magnetic resonance with late gadolinium enhancement, with injury pre-specified by the Fourth Universal Definition of Myocardial Infarction. We clarify that the composite indices derive from a predefined hierarchical scaling system normalized against approximately 17000 healthy individuals, whereas the within-cohort clustering is the exploratory step requiring independent validation. We acknowledge that short peri-procedural recordings are sensitive to sedation, pain and autonomic confounders, and endorse a roadmap of larger multicentre cohorts correlating the signal with biomarker kinetics, imaging and clinical outcomes.

Key Words: Percutaneous coronary intervention; Myocardial injury; Electrocardiography; Heart rate variability; Reference standard; Cardiac troponin; Cardiac magnetic resonance; Validation

Core Tip: Composite electrocardiography and heart rate variability indices can flag small post-procedural changes that the routine 12-lead electrocardiogram misses, but they cannot separate injury from recovery unless anchored to an accepted reference standard. In this reply we set out how that anchor — paired high-sensitivity troponin and late gadolinium enhancement on cardiac magnetic resonance, with injury pre-specified by the Fourth Universal Definition of Myocardial Infarction — will be built into a larger, prospective, presentation-stratified study. We distinguish the validated construction of the composite indices from the exploratory within-cohort clustering, which still requires independent validation before any clinical use.

TO THE EDITOR

We are grateful to Salimi and Hematpour[1] in the World Journal of Cardiology for their careful and generous editorial on our recent study[2], and to the Editor for the opportunity to respond. The editorial captures the intent of our work precisely: To test whether coordinated, small-amplitude electrical and autonomic shifts can be detected shortly after percutaneous coronary intervention (PCI) when the routine 12-lead electrocardiogram (ECG) remains unremarkable. We also share, without reservation, the authors’ central framing — these findings are exploratory and hypothesis-generating, and they do not yet justify clinical decision-making. We engage below with the main points raised, agreeing with most and adding clarifications where we believe they sharpen the agenda rather than soften it.

THE NEED FOR AN EXTERNAL REFERENCE STANDARD

This is, in our view, the most important point, and we agree with it entirely. Composite ECG and heart rate variability (HRV) indices that are not anchored to an accepted definition of injury cannot, on their own, distinguish injury from recovery or from non-specific autonomic perturbation. Our planned prospective study is designed around exactly this anchor: Paired high-sensitivity troponin measured before and after the procedure, as advocated by Gustavsson et al[3], together with cardiac magnetic resonance with late gadolinium enhancement, which Selvanayagam et al[4] showed corresponds to irreversible micro-infarction even at modest biomarker rises. Injury will be pre-specified according to the Fourth Universal Definition of Myocardial Infarction[5] before any index is evaluated, so that the metric’s sensitivity, specificity, and incremental value can be quantified against an adjudicated standard rather than asserted.

CLUSTERS AS “MIXED CONSTRUCTS”

Salimi and Hematpour[1] rightly note that our three response patterns may conflate ischaemia relief, procedural injury, and autonomic perturbation. The original cohort does not support attributing the opposing directions of change to clinical presentation alone: Both the improvement and deterioration clusters consisted mainly of patients treated for acute myocardial infarction, whereas chronic coronary artery disease predominated only in the cluster with minimal change. The deterioration cluster was older and had received more stents than the other two clusters, but these descriptive differences cannot establish causation in a sample of 23 patients. Thus, age, procedural burden, clinical presentation, and other unmeasured factors remain potential sources of confounding. The next study will therefore stratify analyses by clinical presentation, adjust for age and procedural characteristics, report injury-anchored results within each stratum, and test the reproducibility and calibration of any clusters in independent data, as the editorial recommends.

DIMENSIONALITY AND INTERPRETABILITY

We agree that the analytical degrees of freedom were large relative to a cohort of 23, and that independent validation with a parsimonious, prespecified feature set and thresholds is essential. We would, however, distinguish two layers of the analysis. The composite indices themselves are not a black-box classifier trained on these 23 patients; they are computed from a pre-existing, explicitly defined hierarchical scaling system normalized against a reference distribution of approximately 17000 healthy individuals, so the per-patient scores are neither fitted to nor overfitted by the present sample. What is exploratory and sample-limited is the downstream, data-driven step — the clustering and feature selection performed within this cohort. It is that step, not the index construction, that requires external validation and correction for multiple comparisons, and we will treat it accordingly. Aggregation may improve interpretability, but it does not eliminate the analytical degrees of freedom introduced by exploratory feature selection and clustering. In future reports, we will provide the constituent variables, aggregation rules, preprocessing steps, feature-selection decisions, and thresholds for each composite index to enable independent replication. A separate large-scale analysis of 153 ECG and HRV features from 14863 individuals without known heart disorders and 8220 patients hospitalized with heart failure used Uniform Manifold Approximation and Projection and Hierarchical Density-Based Spatial Clustering of Applications with Noise to examine cluster structure across datasets and distance metrics[6]. That study demonstrates the feasibility of scalable, high-dimensional ECG-HRV pattern discovery, but its unsupervised clusters were not anchored to adjudicated myocardial injury or longitudinal outcomes and therefore do not validate the PCI-derived clusters or their clinical meaning.

ACQUISITION CONDITIONS AND CONFOUNDING

The point that short peri-procedural recordings are sensitive to analgosedation, pain, anxiety, respiratory pattern, medications, and acute coronary syndrome physiology is well taken, and it applies most forcefully to the HRV-derived autonomic and psychoemotional readouts. We will protocolize recording conditions, document the timing and agents of sedation and analgesia, and treat the autonomic indices as hypothesis-generating signals to be interpreted alongside — not in place of — the biomarker and imaging anchors above.

THE VALIDATION ROADMAP

We endorse the agenda summarized in the editorial’s Table 1: Larger prospective, multicentre cohorts with standardized acquisition; correlation with troponin kinetics, imaging, arrhythmia monitoring, and clinically meaningful outcomes; comparison with alternative assessment strategies; and evaluation of cost-effectiveness and real-world feasibility. We would emphasize one element in particular — the requirement to test whether acting on the signal improves outcomes — and we share the authors’ caution regarding over-triage and the anxiety that premature labelling could cause. A more graded physiological signal is only useful if it changes management for the better. Accordingly, any future clinical implementation should be preceded by prospective evaluation of incremental discrimination, decision impact, and net clinical benefit.

CONCLUSION

In short, we welcome this editorial not as a defence of our findings but as a clarification of how to convert a promising signal into a validated tool. We thank Drs. Salimi and Hematpour[1] for engaging so constructively with the work, and the Editor for fostering this exchange.