Multimodal Machine Learning-Based Technical Failure Prediction in Patients Undergoing Transcatheter Aortic Valve...

Area of Science:

• Cardiovascular Medicine
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Technical failure in transcatheter aortic valve replacement (TAVR) is common and linked to poor patient outcomes.
• Predicting TAVR technical failure is complex due to interacting clinical, anatomical, and procedural factors.

Purpose of the Study:

• To develop and validate a data-driven prediction model for TAVR technical failure.
• Utilize multimodal data and machine learning algorithms for accurate failure prediction.

Main Methods:

• A prospective TAVR registry with 2,937 patients was analyzed.
• 184 parameters from diverse sources (clinical, imaging, procedural) were used to train 24 machine learning models.
• Technical success/failure was defined by Valve Academic Research Consortium (VARC)-3 criteria.

Main Results:

• Cardiac and vascular technical failure rates were 2.4% and 7.0%, respectively.
• The best models showed strong discrimination (AUC 0.769 for cardiac, 0.788 for vascular) and high negative predictive values.
• Key predictors included comorbidities, aortic/iliofemoral anatomy (CT), antithrombotic management, and procedural details.

Conclusions:

• Machine learning models integrating multimodal data can effectively predict VARC-3 technical failure in TAVR.
• These models offer potential for refining patient selection and optimizing procedural strategies.