AF/AFL Ablation Outcome Prediction Models

Overview

This repository contains the implementation of machine learning models for predicting adverse outcomes following catheter ablation treatment for atrial fibrillation (AF) and/or atrial flutter (AFL).

The models were developed using a comprehensive linked dataset from New South Wales, Australia, incorporating hospital administrative data, prescription medicine claims, emergency department presentations, and death registrations.

Research Paper

This code accompanies the research published in Heart, Lung and Circulation:

Predicting Adverse Outcomes Following Catheter Ablation Treatment for Atrial Flutter/Fibrillation

DOI: https://doi.org/10.1016/j.hlc.2023.12.016

Key Features

• Implementation of traditional and deep survival models
• Models for predicting two distinct outcomes:
  1. Major bleeding events
  2. Composite outcome (heart failure, stroke, cardiac arrest, death)
• Feature importance analysis and visualization
• Evaluation metrics including concordance index

Dataset

The study cohort included 3,285 patients who received catheter ablation for AF and/or AFL in New South Wales, Australia. Due to privacy regulations, the raw data cannot be shared publicly. However, we provide:

• Data preprocessing scripts
• Feature engineering pipelines
• Synthetic data generators for testing the models

Models

The repository implements several survival analysis models:

• Cox Proportional Hazards
• Random Survival Forest
• Gradient Boosting Survival Models
• Deep Survival Networks

Results

Our models achieved:

• Composite outcome prediction: concordance index >0.79
• Major bleeding events prediction: concordance index <0.66

Feature importance analyses identified the following as key predictors:

• Comorbidities indicating poor health
• Older age
• Therapies for heart failure and AF/AFL management

Feature Importance Analysis

The study utilized SHAP (SHapley Additive exPlanations) values to identify the most important features for predicting adverse outcomes:

Figure 1: SHAP summary plot for the composite outcome prediction model. Features are ranked by their impact on model predictions.

Citation

If you use this code in your research, please cite our paper:

@article{QUIROZ2024470,
  title = {Predicting Adverse Outcomes Following Catheter Ablation Treatment for Atrial Flutter/Fibrillation},
  journal = {Heart, Lung and Circulation},
  volume = {33},
  number = {4},
  pages = {470-478},
  year = {2024},
  issn = {1443-9506},
  doi = {https://doi.org/10.1016/j.hlc.2023.12.016},
  url = {https://www.sciencedirect.com/science/article/pii/S1443950624000039},
  author = {Juan C. Quiroz and David Brieger and Louisa R. Jorm and Raymond W. Sy and Benjumin Hsu and Blanca Gallego},
  keywords = {Atrial fibrillation, Catheter ablation, Machine learning, Survival analysis, Treatment outcome},
}

License

This project is licensed under the MIT License - see the LICENSE file for details.

Contributors

• Juan C. Quiroz
• David Brieger
• Louisa R. Jorm
• Raymond W. Sy
• Benjumin Hsu
• Blanca Gallego

Contact

For questions about the code or paper, please open an issue in this repository or contact the corresponding author.