Machine Learning Identification of Modifiable Predictors of Patient Outcomes After Transcatheter Aortic Valve...

Area of Science:

• Cardiovascular Medicine
• Health Informatics
• Machine Learning in Healthcare

Background:

• Severe symptomatic aortic stenosis necessitates treatment, with transcatheter aortic valve replacement (TAVR) being a key option.
• Identifying predictors of excellent TAVR outcomes, such as good clinical results and increased home recovery time, is crucial for optimizing patient care.
• Focusing on potentially modifiable predictors can lead to improved patient-centered results after TAVR.

Purpose of the Study:

• To leverage machine learning (ML) techniques to pinpoint modifiable predictors of excellent patient-centered outcomes following TAVR.
• To analyze a large dataset of TAVR cases to identify key factors influencing treatment success.

Main Methods:

• Utilized a dataset of 8,332 TAVR cases from 21 hospitals (January 2016-December 2021).
• Trained random forest models using 57 patient characteristics and care process parameters to predict an excellent outcome composite endpoint.
• Employed recursive feature elimination and Shapley Additive Explanation (SHAP) for feature importance analysis.

Main Results:

• The final ML model included 29 predictors (15 patient characteristics, 14 care process components) with an AUC of 0.77.
• Identified four potentially modifiable predictors with high SHAP values: anesthesia type, direct transfer to a stepdown unit, time from catheterization to TAVR, and preprocedural length of stay.
• These findings highlight specific areas within the care pathway that can be adjusted to improve TAVR success.

Conclusions:

• Machine learning analysis of hospital-level data can identify modifiable predictors of excellent TAVR outcomes.
• Four key modifiable factors were identified, offering targets for improving TAVR care delivery.
• This approach can inform strategies to enhance patient recovery and reduce complications after TAVR.