Leveraging electronic health records for atrial fibrillation cohort generation

Figures

• 

  Fig. 1. Time-window for AF progression. AF progression (recurrence or permanence) is evaluated between 1 month and 2 years after the initial onset, the return to sinus rhythm is assessed starting from 1 day after the debut episode

• 

  Fig. 2. Manual annotation process of the dataset. The initial filtering for AF debut is performed using information from the OBI system, followed by the manual annotation of AF progression

• 

  Fig. 3. Confusion matrix showing the level of agreement between annotators based on their respective labels. Class 1 represents AF progression (recurrence or permanence), Class 0 indicates non-progression (return to sinus rhythm), and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\textit{Class -1}$$\end{document}Class-1 includes cases with insufficient information to determine progression status

• 

  Fig. 4. Overview of the rule-based approach. The pipeline includes a section identification module, a medical entity recognition (MER) and negation detection module, advanced regular expressions, and temporal management of clinical events to determine AF progression status

• 

  Fig. 5. Summarization Strategy Each discharge report is summarized in conjunction with the previous summary, and the final decision is made based on the complete accumulated summary

• 

  Fig. 6. Onset-Guided Strategy. The process is divided into two stages: first, identifying the AF onset, and second, determining the progression status

• 

  Fig. 7. Chronology-Guided Strategy. The process is divided into three stages: first, identifying the AF onset, second, determining the return to sinus rhythm in reports within 1 month after the AF onset, and third, determining the progression status for reports between 1 month and 2 years after the AF onset

• 

  Fig. 8. Total format errors across experiments by prompt language in Llama models. Each bar shows the total number of format errors made by the Llama 3.1 (left) and Aloe-Beta (right) models across all strategies, with blue (scratched) representing English prompts and red (dotted) representing Spanish prompts

• 

  Fig. 9. Mean accuracy across experiments for each prompt language in Gemma models. Each bar represents the average accuracy achieved by the Gemma-12B (left) and Gemma-27B (right) models across all strategies, with blue (scratched) indicating English prompts and red (dotted) indicating Spanish prompts

• 

  Fig. 10. Prediction outcomes analysis. Red (dotted) indicates incorrect final predictions, blue (scratched) represents correct final predictions with intermediate errors, and green (crossed) denotes fully correct predictions at every decision point

• 

  Fig. 11. Error analysis for the Incorrect category. Each bar represents a specific type of error and the number of times it was made by the model. More detailed description of the errors available in “Appendix 3”

• 

  Fig. 12. Error analysis for the Correct Final Decision category. Each bar represents a specific type of error and the number of times it was made by the model. A more detailed description of the errors is available in “Appendix 3”