Artificial intelligence-driven ECG biomarkers for accurate screening of large pericardial effusion

Abstract Background Pericardial effusion (PE) can progress to cardiac tamponade, a potentially life-threatening condition. Echocardiography is the standard diagnostic tool for PE; however, it is not always suitable as a initial screening due to limited resources and accessibility. In this study, we aimed to evaluate the diagnostic accuracy of a digital biomarker derived from a 12-lead electrocardiography (ECG) analysis application (ECG Buddy) for detecting large PE and cardiac tamponade, and to assess its ability to reflect clinical improvement after therapeutic pericardiocentesis (PCC). Methods Adult patients (≥19 years old) who underwent echocardiography between January 2013 and December 2023 were retrospectively analyzed. The patients confirmed large PE (with or without cardiac tamponade) were defined as positive cases, whereas negative controls were randomly selected at a ratio of 1.5 times the number of positive cases. Patient-echocardiography pairs without ECG on the same or previous day of the echocardiography were excluded. ECG data were analyzed using the ECG Buddy application, a Korean MFDS-approved Class II software medical device. The diagnostic accuracy of the ECG biomarker (PrcEff) was assessed by receiver operating characteristic (ROC) curves across three groups: normal, large PE, and tamponade. Additionally, biomarker changes were evaluated in the subgroup patients underwent PCC. Results In total, 641 patient–echocardiography pairs were included in the final analysis (normal: 344, large PE: 213, tamponade: 84). The median (IQR) ECG biomarker values clearly differed among groups (p0.001): normal, 1.0 (0.3–3.5); large PE, 37.8 (21.2–62.1); and tamponade, 61.8 (38.3–80.0). The ROC analysis demonstrated excellent diagnostic performance, with an AUC of 0.980 (95% CI: 0.972–0.988) for differentiating large PE from normal, and an AUC of 0.880 (95% CI: 0.851–0.909) for tamponade. The biomarker exhibited notably high sensitivity (94.1%) for tamponade detection. After PCC, the ECG biomarker significantly decreased from 48.95 (33.40–73.75) to 22.95 (12.75–43.08) (p0.001), reflecting clinical improvement. Conclusion The ECG-based digital biomarker derived from the ECG Buddy application demonstrated high diagnostic accuracy for identifying large PE and cardiac tamponade, particularly showing excellent sensitivity for tamponade detection. Furthermore, it effectively reflected clinical improvement after PCC procedure. Therefore, this digital ECG biomarker represents the possibility for the rapid screening and monitoring of patients with significant PE in settings where echocardiography is difficult to perform.Figure 1.Comparison of ECG Biomarker Figure 2.ROC curve for detecting PE