Development and Validation of a Machine Learning Score for Readmissions After Transcatheter Aortic Valve Implantation

Area of Science:

• Cardiovascular Medicine
• Health Informatics
• Machine Learning in Healthcare

Background:

• Transcatheter aortic valve implantation (TAVI) is a critical procedure with significant readmission rates.
• Identifying patients at high risk for readmission post-TAVI is a crucial unmet clinical need.

Purpose of the Study:

• To investigate the utility of machine learning (ML) models in predicting readmissions following TAVI.
• To develop a risk prediction model for 30-day readmission after TAVI.

Main Methods:

• Utilized Nationwide Readmission Database (2016-2019) with 917 candidate predictors.
• Employed lasso regression for variable selection, K-means clustering for risk stratification, and Light Gradient Boosting Machine with SHAP for predictor impact analysis.
• Developed a parsimonious model to predict 30-day TAVI readmissions.

Main Results:

• Identified 138 and 199 informative predictors for 30- and 90-day readmissions, respectively.
• K-means clustering revealed distinct low-risk (10.1% 30-day readmission) and high-risk (23.3% 30-day readmission) groups.
• Key predictors included length of stay, frailty score, discharge diagnoses, acute kidney injury, and Elixhauser score. The TAVI readmission score showed good performance (AUC 0.74).

Conclusions:

• Machine learning models can effectively utilize administrative data to identify patients at risk of readmission after TAVI.
• These ML-driven insights can inform and enhance post-TAVI patient care strategies.
• The developed TAVI readmission score demonstrates potential for clinical application in risk stratification.