@inproceedings{25524d8d-0721-490b-9ad3-67f287964398,
  abstract     = {{Identification of patients at risk to develop atrial fibrillation (AF) can enable more frequent follow-up and lead to earlier detection of AF. The aim of this project was to perform long-term AF prediction based on a novel set of ECG-derived features describing the characteristics of supraventricular arrhythmias present before new onset AF. In total, 12 199 patients aged 75/76 performed repeated single-lead ECG measurements for 30-s during a period of 2–4 weeks. Arrhythmia detection was performed following quality control to ensure that only reliable detections were included. For each detected arrhythmic episode, a set of features were extracted including degree of prematurity, burden of supraventricular ectopic beats, and number of RR intervals between arrhythmic events. Following feature extraction, a 1D-convolutional neural network was employed to predict the long-term risk of developing AF. On average, the trained prediction models led to an AUC of 0.60 for the test set. At the end of the observation period, the risk stratification curves for the model showed 94% and 88% probabilities of not developing AF for the low- and high-risk groups, respectively. These results correspond to a weighted F1 score of 0.72 for the test set. The results show that ECG-derived features characterising supraventricular arrhythmias occurring before AF contribute to improved risk stratification for a future AF diagnosis}},
  author       = {{Colucci, Livia and Khan, Mashroor and Sörnmo, Leif and Svennberg, Emma and Stridh, Martin}},
  booktitle    = {{Computing in Cardiology (CinC) 2025 – Proceedings}},
  language     = {{eng}},
  title        = {{ECG-Based Long-Term Prediction of Atrial Fibrillation}},
  url          = {{http://dx.doi.org/10.22489/CinC.2025.124}},
  doi          = {{10.22489/CinC.2025.124}},
  volume       = {{52}},
  year         = {{2025}},
}