We thank Dr. Zablah and colleagues for their thoughtful comments on our study, which examines the use of artificial intelligence-enabled electrocardiogram (AI-ECG) to inform the timing of pulmonary valve replacement (PVR) in patients with repaired tetralogy of Fallot (rTOF).

We agree that appropriate time alignment is an important consideration in observational comparisons. In our analysis, the index time point for both groups was anchored to an ECG: for PVR patients, it was the ECG obtained ≤3 months prior to intervention, and for non-PVR patients, the matched ECG was used for propensity score alignment (see Methods). The propensity score model incorporated age at ECG and key clinical and imaging variables, and postmatching balance was confirmed across measured covariates (Online Table 5), supporting comparability at the index time point. Of note, there was no statistical difference ( P =.4) in age at ECG for the PVR (median 23.6 [IQR, 16.6-36.9] years) and non-PVR groups (median 25.3 [IQR, 16.6-39.1] years).

However, we acknowledge that such alignment and covariate balance do not fully eliminate the potential for time-dependent confounding or immortal time bias. Accordingly, our findings are not definitive causal estimates of treatment effect. While we agree that alternative analytic frameworks (eg, target trial emulation or time-varying treatment models) may further refine observational estimates of treatment effect, these approaches remain subject to the same fundamental limitations inherent in nonrandomized trials, particularly in conditions with evolving disease severity and clinician-driven treatment selection. Similarly, while calibration and decision curve-based net benefit are important for evaluating clinical utility, their interpretation in this setting is constrained by the same observational structure, as both predicted risk and treatment assignment are influenced by unmeasured and time-varying factors.

In this context, we view our AI-ECG findings as hypothesis-generating. The key next step is prospective evaluation of our algorithm in real-world clinical scenarios. Ideally, validation would include multicenter studies that incorporate serial ECGs to determine whether AI-ECG can meaningfully guide PVR timing and improve outcomes in patients with rTOF.

Reference

Only gold members can continue reading. Log In or Register to continue

Share this:

• X
• Facebook

Related

Related posts:

1. Rationale and design of the vericiguat in vasospastic angina (ViVA) trial: A double-blind placebo-controlled randomized cross-over study
2. Time-alignment considerations in evaluating AI-ECG–guided pulmonary valve replacement timing in repaired tetralogy of Fallot
3. From hepatic congestion to ventricular stiffness: Mechanistic pathways of diastolic dysfunction after Fontan
4. Serial assessment of NT-proBNP and high-sensitivity cardiac troponin with glucagon-like peptide-1 receptor agonist therapy in type 2 diabetes: Insights from EXSCEL

Stay updated, free articles. Join our Telegram channel

Join