Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Early Diagnosis of Atrial Fibrillation Recurrence: A Large Tabular Model Approach with Structured and Unstructured Clinical Data

Ane G. Domingo-Aldama, Marcos Merino Prado, Alain García Olea, Koldo Gojenola Galletebeitia, Josu Goikoetxea Salutregi, Aitziber Atutxa Salazar

TL;DR

This work addresses predicting atrial fibrillation (AF) recurrence within a 1-month to 2-year window after onset, a task where traditional scores underperform. It introduces a Large Tabular Model (LTM) framework that fuses structured EHR data with unstructured discharge reports via a three-step NLP pipeline to produce high-quality tabular vectors (silver data). The study demonstrates that LTMs, particularly TabPFN with minimal preprocessing, outperform traditional clinical scores and standard ML models in AF recurrence prediction, while revealing gender and age biases that merit fairness considerations. The dataset-generation approach is generalizable to other conditions, enabling scalable, data-rich predictive modeling in healthcare with reduced annotation burden and better handling of missing data. Practically, this work supports earlier, more accurate identification of patients at risk of AF recurrence, informing personalized treatment strategies and resource allocation.

Abstract

BACKGROUND: Atrial fibrillation (AF), the most common arrhythmia, is linked to high morbidity and mortality. In a fast-evolving AF rhythm control treatment era, predicting AF recurrence after its onset may be crucial to achieve the optimal therapeutic approach, yet traditional scores like CHADS2-VASc, HATCH, and APPLE show limited predictive accuracy. Moreover, early diagnosis studies often rely on codified electronic health record (EHR) data, which may contain errors and missing information. OBJECTIVE: This study aims to predict AF recurrence between one month and two years after onset by evaluating traditional clinical scores, ML models, and our LTM approach. Moreover, another objective is to develop a methodology for integrating structured and unstructured data to enhance tabular dataset quality. METHODS: A tabular dataset was generated by combining structured clinical data with free-text discharge reports processed through natural language processing techniques, reducing errors and annotation effort. A total of 1,508 patients with documented AF onset were identified, and models were evaluated on a manually annotated test set. The proposed approach includes a LTM compared against traditional clinical scores and ML models. RESULTS: The proposed LTM approach achieved the highest predictive performance, surpassing both traditional clinical scores and ML models. Additionally, the gender and age bias analyses revealed demographic disparities. CONCLUSION: The integration of structured data and free-text sources resulted in a high-quality dataset. The findings emphasize the limitations of traditional clinical scores in predicting AF recurrence and highlight the potential of ML-based approaches, particularly our LTM model.

Early Diagnosis of Atrial Fibrillation Recurrence: A Large Tabular Model Approach with Structured and Unstructured Clinical Data

TL;DR

This work addresses predicting atrial fibrillation (AF) recurrence within a 1-month to 2-year window after onset, a task where traditional scores underperform. It introduces a Large Tabular Model (LTM) framework that fuses structured EHR data with unstructured discharge reports via a three-step NLP pipeline to produce high-quality tabular vectors (silver data). The study demonstrates that LTMs, particularly TabPFN with minimal preprocessing, outperform traditional clinical scores and standard ML models in AF recurrence prediction, while revealing gender and age biases that merit fairness considerations. The dataset-generation approach is generalizable to other conditions, enabling scalable, data-rich predictive modeling in healthcare with reduced annotation burden and better handling of missing data. Practically, this work supports earlier, more accurate identification of patients at risk of AF recurrence, informing personalized treatment strategies and resource allocation.

Abstract

BACKGROUND: Atrial fibrillation (AF), the most common arrhythmia, is linked to high morbidity and mortality. In a fast-evolving AF rhythm control treatment era, predicting AF recurrence after its onset may be crucial to achieve the optimal therapeutic approach, yet traditional scores like CHADS2-VASc, HATCH, and APPLE show limited predictive accuracy. Moreover, early diagnosis studies often rely on codified electronic health record (EHR) data, which may contain errors and missing information. OBJECTIVE: This study aims to predict AF recurrence between one month and two years after onset by evaluating traditional clinical scores, ML models, and our LTM approach. Moreover, another objective is to develop a methodology for integrating structured and unstructured data to enhance tabular dataset quality. METHODS: A tabular dataset was generated by combining structured clinical data with free-text discharge reports processed through natural language processing techniques, reducing errors and annotation effort. A total of 1,508 patients with documented AF onset were identified, and models were evaluated on a manually annotated test set. The proposed approach includes a LTM compared against traditional clinical scores and ML models. RESULTS: The proposed LTM approach achieved the highest predictive performance, surpassing both traditional clinical scores and ML models. Additionally, the gender and age bias analyses revealed demographic disparities. CONCLUSION: The integration of structured data and free-text sources resulted in a high-quality dataset. The findings emphasize the limitations of traditional clinical scores in predicting AF recurrence and highlight the potential of ML-based approaches, particularly our LTM model.

Paper Structure

This paper contains 37 sections, 3 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Dataset
  4. Cohort Generation: AF onset detection
  5. Silver training dataset Generation
  6. Tabular Data Generation
  7. Automatic labelling process: AF Recurrence Detection
  8. Resulting AF recurrence dataset
  9. Experimental setup
  10. Clinical scores
  11. ML models
  12. Handling Missing Values
  13. Feature Selection and Dimensionality Reduction
  14. Sampling Strategies
  15. Model Optimization and Validation
  16. ...and 22 more sections

Figures (3)

  • Figure 1: Overview of the generated dataset and the experimental setup for AF recurrence prediction.
  • Figure 2: Summary of the AF recurrence cohort generation process. First, patients with an AF onset are identified, followed by the assignment of the AF recurrence label.
  • Figure 3: Summary of the vector generation process. Information from both free-text discharge reports and structured codified data from OBI system are merged in a single vector per patient describing its health status around the time of AF onset.