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

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Accurate measurement of aortic valve pressure gradient (PG_AV) is crucial for transcatheter aortic valve replacement (TAVR) outcomes.
• Echocardiography often overestimates PG_AV in patients with small aortic annuli compared to left heart catheterization (LHC).

Purpose of the Study:

• To enhance the accuracy of PG_AV measurements using echocardiography in small annuli during TAVR.
• To assess the prognostic impact of accurately measured PG_AV on patient outcomes.

Main Methods:

• Trained an extreme gradient boosting (XGBoost) algorithm on data from 273 patients with aortic stenosis and small annuli undergoing TAVR.
• Compared PG_AV measurements from transthoracic echocardiography (TTE) with LHC data.
• Evaluated the association between predicted PG_AV and a composite endpoint of mortality and heart failure readmission.

Main Results:

• Baseline TTE measurements of PG_AV were significantly overestimated compared to LHC (52.5 mmHg vs. 42.5 mmHg, P < 0.001).
• The XGBoost model significantly improved TTE PG_AV accuracy (Pearson correlation coefficient = 0.94, P < 0.001).
• Patients with a predicted mean PG_AV ≥ 68.6 mmHg had a substantially higher incidence of composite endpoints at 2 years (40.7% vs. 16.0%, P < 0.001).

Conclusions:

• The XGBoost model effectively improves the accuracy of echocardiographic PG_AV measurements during TAVR in small annuli.
• An elevated predicted PG_AV, accurately measured, is associated with a worse prognosis in TAVR patients.
• Improved PG_AV measurement accuracy can aid in better patient selection and risk stratification for TAVR.