Predicting in-hospital mortality after transcatheter aortic valve replacement using administrative data and machine...

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Transcatheter aortic valve replacement (TAVR) is a standard treatment for symptomatic aortic stenosis.
• Current risk prediction tools for in-hospital mortality post-TAVR have limited accuracy using only pre- or post-operative data.
• This limitation hinders their use in preoperative decision-making.

Purpose of the Study:

• To evaluate the accuracy of statistical and machine learning models in predicting in-hospital outcomes (death/survival) post-TAVR.
• To determine if models trained solely on preoperative data from the National Inpatient Sample database can achieve sufficient predictive accuracy.
• To identify the optimal model for preoperative risk stratification in TAVR patients.

Main Methods:

• Utilized the National Inpatient Sample database, analyzing 54,739 TAVR procedures.
• Applied fifteen distinct binary classification methods to model in-hospital survival and death.
• Evaluated model performance using metrics including the area under the receiver operating characteristic curve (AUC) across random and time-based sampling scenarios.

Main Results:

• The top five machine learning models achieved an AUC ≥ 0.80 in both random and time-based sampling.
• This indicates strong predictive performance using only preoperative variables.
• Minimal practical differences were observed in the predictive accuracy among the top-performing models.

Conclusions:

• Machine learning models trained on preoperative data can effectively predict in-hospital mortality post-TAVR.
• L2 regularized logistic regression is recommended as the optimal model due to its computational efficiency and interpretability.
• These findings support the development of improved preoperative decision support tools for TAVR.