Document resource
Aims Early identification of reduced left ventricular ejection fraction (LVEF) after acute coronary syndrome (ACS) is important for risk stratification, escalation of care, and timely initiation of guideline-directed therapy. LVSense (PMcardio) is an AI-based ECG tool previously validated in stable cohorts. The primary objective was to compare pre-PCI versus post-PCI LVSense for identification of patients with echocardiographic LVEF ≤40%.Methods and Results In this prospective multicentre ACS cohort, consecutive patients underwent pre-PCI (admission) and post-PCI 12-lead ECGs with LVSense analysis and post-PCI transthoracic echocardiography (TTE). LVEF was stratified as ≤40%, 41–49%, and ≥50%.A total of 212 patients had complete paired datasets. On TTE, 69/212 (32.5%) had LVEF ≤40%, 66/212 (31.1%) had LVEF 41–49%, and 77/212 (36.3%) had LVEF ≥50%.For detection of TTE LVEF ≤40%, pre-PCI LVSense showed sensitivity 73.9% (51/69), specificity 60.8% (87/143), PPV 47.7% (51/107), NPV 82.9% (87/105), and accuracy 65.1% (138/212) (figure 1).Post-PCI LVSense showed sensitivity 91.3% (63/69), specificity 53.1% (76/143), PPV 48.5% (63/130), NPV 92.7% (76/82), and accuracy 65.6% (139/212).Across all three EF strata, weighted kappa improved from 0.332 (pre-PCI) to 0.358 (post-PCI), and exact agreement increased from 44.8% to 50.9% (paired p=0.124). Sensitivity for LVEF ≤40% improved significantly post-PCI (paired p=0.008), while specificity did not differ significantly (p=0.144). Post-PCI assessment substantially reduced false-negative classification of impaired LVEF (6 vs 18 pre-PCI), improving detection of patients with true LVEF ≤40%. Residual misclassification was concentrated in the intermediate 41–49% range, consistent with real-world reporting variability around borderline EF, where sonographer-estimated values are frequently reassigned to adjacent categories; this distribution is reflected in the comparative confusion matrices.Conclusions In this novel prospective evaluation, post-PCI AI-ECG (LVSense) improved detection of impaired LVEF (≤40%) versus pre-PCI assessment against echocardiographic reference, principally through higher sensitivity, higher NPV, and fewer false negatives. This likely reflects the dynamic nature of LV function during ACS, where reperfusion, peri-procedural myocardial injury (particularly in STEMI), and myocardial stunning can alter EF status over time. Pre-PCI AI-ECG remains valuable for early risk stratification at first medical contact, while post-PCI AI-ECG appears better suited to early in-hospital identification of patients with LVEF ≤40%. AI-ECG should be considered adjunctive to, not a replacement for, echocardiography. Larger prospective cohorts are needed to define optimal ECG timing and evaluate impact on guideline-directed medical therapy initiation, treatment decisions, and clinical outcomes.Panel A shows pre-PCI LVSense predictions and Panel B shows post-PCI LVSense predictions, each compared with post-PCI transthoracic echocardiography (TTE) in 212 patients. Columns represent TTE LVEF strata (≤40%, 41–49%, ≥50%) and rows represent LVSense-predicted strata. Values are patient counts per cell. Identical axis order and colour scaling were used across panels for direct visual comparison. Discordance is most prominent in the intermediate 41–49% stratum.Abstract 524 Figure 1Comparative confusion matrices for pre- and post-PCI AI-ECG LVEF classification versus echocardiographic reference