Using novel machine learning tools to predict optimal discharge following transcatheter aortic valve replacement.

Area of Science:

• Cardiovascular Medicine
• Health Informatics
• Artificial Intelligence in Healthcare

Background:

• Transcatheter aortic valve replacement (TAVR) is a vital alternative to surgical aortic valve replacement.
• Optimizing hospital length of stay is crucial due to extensive healthcare resource utilization in TAVR.
• Predicting optimal discharge is essential for efficient patient management and resource allocation.

Purpose of the Study:

• To evaluate the predictive accuracy of novel machine learning models for optimal hospital discharge post-TAVR.
• To compare the performance of artificial neural network (ANN) and eXtreme Gradient Boost (XGBoost) models against traditional logistic regression.
• To identify key factors influencing optimal discharge timing following TAVR.

Main Methods:

• Utilized the 2016-2018 National Inpatient Sample database for TAVR patient data.
• Classified patients into optimal (0-3 days) and late (4-9 days) discharge groups.
• Employed logistic regression, ANN, and XGBoost models to predict optimal discharge, assessing performance via AUC and F1 score.

Main Results:

• Analyzed 25,874 TAVR patients; all models achieved an Area Under the Curve (AUC) of 0.80.
• Patient disposition and elective procedure status were the most significant predictors of optimal discharge.
• Coagulation disorder emerged as the strongest comorbidity predictor for optimal discharge.

Conclusions:

• ANN and XGBoost models demonstrated satisfactory performance and accuracy comparable to logistic regression in predicting optimal TAVR discharge.
• These machine learning models show potential for clinical application in optimizing hospital discharge.
• Further validation and refinement are recommended for broader clinical adoption of these predictive tools.