A Supervised Learning Approach Electrocardiographic Model for Differentiating Outflow Tract Premature Ventricular Complex Origins: Comparative Analysis of Seven Established Algorithms.

BACKGROUND: Premature ventricular complexes (PVCs) arising from the right and left ventricular outflow tracts (RVOT and LVOT) require accurate localization for successful ablation. Existing electrocardiographic (ECG) algorithms are limited by anatomical variability. This study aimed to develop a supervised learning approach model based on logistic regression using validated ECG parameters and to compare its diagnostic performance with 7 established algorithms. METHODS: A retrospective cohort of 116 patients with idiopathic outflow tract PVCs who underwent successful ablation between 2015 and 2020 was analyzed. Four ECG parameters were selected through backward stepwise logistic regression. The performance of the model and 7 published algorithms was assessed using receiver-operating characteristic (ROC) curve analysis, Youden index, and accuracy metrics. Subgroup analysis was performed in patients with V3 precordial transition. RESULTS: The supervised learning model achieved the highest diagnostic accuracy with an area under the ROC curve of 0.942 in the overall cohort and 0.878 in the V3 transition subgroup, significantly outperforming all comparator algorithms (P < .001). The model demonstrated a Youden index of 0.66, sensitivity of 82.7%, and specificity of 84.4%. CONCLUSION: The supervised learning approach model outperformed existing rule-based ECG algorithms in differentiating RVOT from LVOT PVCs. By integrating validated ECG features into a statistically optimized and interpretable framework, it provides a reliable noninvasive tool to support ablation planning. Larger multicenter validation studies are warranted.