Machine learning cluster analysis identifies increased 12-month mortality risk in transcatheter aortic valve...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Data Analysis

Background:

• Long-term mortality risk after transcatheter aortic valve implantation (TAVR) is often under-assessed in clinical practice.
• Unsupervised machine learning can identify patterns in complex patient data to stratify risk.
• This may help identify patients needing closer post-TAVR surveillance.

Purpose of the Study:

• To apply unsupervised machine learning to identify distinct patient clusters after TAVR.
• To assess the prognostic value of these clusters for long-term mortality.
• To compare the predictive power of machine learning clusters with conventional risk scores.

Main Methods:

• Unsupervised k-means clustering was used on demographic, biochemical, and cardiac imaging data from 200 TAVR patients.
• Feature importance for cluster assignment was determined.
• Survival analyses (Kaplan-Meier, Cox models) were performed, comparing cluster assignment to conventional risk and frailty calculators.

Main Results:

• Two distinct patient clusters were identified.
• Cluster 2 showed significantly higher 12-month all-cause mortality (HR 6.3, p < 0.01) compared to Cluster 1.
• Cluster 2 was associated with more advanced cardiac remodeling and worse multi-chamber function.
• Cluster assignment outperformed conventional scores in predicting 12-month mortality.

Conclusions:

• k-means clustering identified two prognostically distinct phenogroups post-TAVR.
• Machine learning-derived clusters demonstrated superior predictive power for mortality compared to traditional risk and frailty scores.
• These findings underscore the utility of machine learning for clinical risk prediction and enhancing patient surveillance strategies.