More than Fluid: AI-Derived Pleural Effusion Volume as a Marker of Cardiovascular Congestion in Transcatheter Aortic Valve Implantation

Background/Objectives: Pleural effusions are frequently encountered on pre-procedural computed tomography (CT) scans in patients undergoing transcatheter aortic valve implantation (TAVI). While often regarded as a marker of congestion and advanced cardiovascular disease, their clinical and prognostic significance in contemporary TAVI populations remains poorly understood. This study investigated the relationship between AI-derived pleural effusion volume, cardiovascular dysfunction, and long-term mortality after TAVI. Methods: Consecutive patients undergoing transfemoral TAVI between 2016 and 2022 who had available pre-procedural CT imaging were retrospectively included. Pleural effusion volume was quantified using an artificial intelligence-based segmentation workflow and analyzed as categorical, continuous, log-transformed, and threshold-based variables. Associations with clinical and echocardiographic characteristics were evaluated using Spearman correlation analyses. Long-term mortality was assessed using Kaplan–Meier analysis, Cox proportional hazards regression, and restricted cubic spline models. Results: A total of 470 patients were included (median age 82 years, 50.9% male). Pleural effusion was present in 124 patients (26.38%), including 73 (15.53%) with small, 26 (5.53%) with moderate, and 25 (5.32%) with larger effusions. Increasing pleural effusion volume was associated with atrial fibrillation (AF) (rho = 0.189, p < 0.001), higher systolic pulmonary artery pressure (rho = 0.224, p < 0.001), lower tricuspid annular plane systolic excursion (rho = −0.236, p < 0.001), impaired right ventricular–pulmonary arterial coupling (rho = −0.267, p < 0.001), lower stroke volume index (rho = −0.229, p < 0.001), and reduced left ventricular ejection fraction (rho = −0.221, p < 0.001). Despite these associations, pleural effusion volume was not associated with long-term mortality in univariable analyses, multivariable Cox regression models, or restricted cubic spline analyses. In the fully adjusted model, log-transformed pleural effusion volume was not independently associated with mortality (HR 1.00, 95% CI 0.93–1.08; p = 0.976). Conclusions: AI-derived pleural effusion volume is associated with markers of cardiovascular congestion and adverse hemodynamic remodeling in patients undergoing TAVI. However, pleural effusion burden does not independently predict long-term mortality, suggesting that its value lies primarily in phenotyping cardiovascular disease severity rather than risk stratification.