Using Optimal Survival Tree Model for AF Event-Free Survival Time Prediction.

Area of Science:

• Clinical data analysis
• Machine learning in healthcare
• Cardiovascular disease prediction

Background:

• Atrial fibrillation (AF) poses a significant health risk.
• Accurate prediction of long-term AF risk is crucial for patient management.
• Existing predictive models may have limitations in handling complex clinical data.

Purpose of the Study:

• To develop and evaluate a novel methodology for clinical data analysis using the Optimal Survival Tree (OST) algorithm.
• To assess the capability of the OST-based approach in predicting 10-year atrial fibrillation risk profiles.
• To compare the performance of OST against other established tree-based algorithms.

Main Methods:

• Application of the Optimal Survival Tree (OST) algorithm for data integration and analysis.
• Utilized a clinical dataset of 4114 patients with a mean follow-up of 59.0 ± 19.3 months.
• Comparative analysis with Classification and Regression Tree (CART), Conditional Inference Tree (cTree), and Random Forest (RF) algorithms.

Main Results:

• The OST-based methodology successfully predicted four distinct 10-year atrial fibrillation risk profiles.
• OST achieved an Area Under the Curve (AUC) of 0.794 and a Brier Score of 0.131.
• OST performance was comparable or superior to CART, cTree, and RF, particularly in predictive accuracy.

Conclusions:

• The proposed OST-based methodology is effective for clinical data analysis and atrial fibrillation risk prediction.
• OST offers a robust tool for identifying patient risk profiles over a 10-year period.
• This approach holds promise for improving cardiovascular risk stratification and patient care.