Clinical significance of machine learning algorithm in predicting PPM during TAVR in small annuli

Compared with left heart catheterization (LHC), the pressure gradient of an aortic valve (PG_AV) measured by echocardiography during transcatheter aortic valve replacement (TAVR) in small annuli is overestimated. The purpose of this study was to improve the accuracy of PG_AV measurements by echocardiography in small annuli and to evaluate the influence of PG_AV on prognosis. The internal derivation cohort included 273 consecutive patients with aortic stenosis and a small annulus (computed tomographic scan showing an annulus circumference < 72 mm or area < 400 mm²) who underwent TAVR. Patients completed transthoracic echocardiography (TTE) and LHC measurements during TAVR, and an extreme gradient boosting (XGBoost) algorithm was trained. The primary outcome was a composite end point of all-cause mortality and readmission for heart failure. The mean PG_AV level measured by TTE was overestimated compared to the LHC measurement {52.5 [interquartile range: 47.5-57.0] mmHg vs. 42.5 (interquartile range: 38.0-46.0) mmHg, P < 0.001}. After adjustments to the XGBoost, the mean PG_AV measured by TTE could be significantly improved [Pearson correlation coefficient = 0.94, P < 0.001]. Importantly, patients with a predicted mean PG_AV ≥ 68.6 mmHg showed a significantly increased incidence of composite end points at 2 years after the procedures (40.7% vs. 16.0%, P < 0.001). The XGBoost model could effectively improve the accuracy of the mean PG_AV measured by TTE during TAVR, and the predicted mean increase in the PG_AV level may lead to a worse prognosis.Clinical Trial Registration: ClinicalTrials.gov Protocol Registration System (NCT05044338).