Performance of the 12-lead ECG in predicting short- and long-term risk of sudden cardiac death

Figures

• 

  Fig. 1. The Cumulative Incidence (CIF) of sudden cardiac deaths and deaths due to other causes during follow-up. The CIF for both sudden cardiac deaths and non-sudden deaths during the course of twelve years among patients in the MADDEC registry who underwent coronary angiography at baseline and then followed for incident deaths.

• 

  Fig. 2. Receiver operating characteristic (ROC) curves (with 95% confidence intervals) illustrating the predictive value of different ECG-based model configurations for incident sudden cardiac death. Area under the curve (AUC) values are shown for the development dataset, the unbalanced validation sample, and the risk factor–balanced validation sample with or without clinical data (top and bottom panels). ROC curves based on ECG parameters measured from baseline (a, d), from only last recording (b, e) and from last ECG combined with all previously recorded ECG information (c, f).

• 

  Fig. 3. Precision–recall (PRC) curves (with 95% confidence intervals) illustrating the predictive performance of different ECG-based model configurations for incident sudden cardiac death. PRC values are shown for the development dataset, the unbalanced validation sample, and the risk factor–balanced validation sample with or without clinical data (top and bottom panels). ROC curves based on ECG parameters measured from baseline (a, d), from only last recording (b, e) and from last ECG combine with all previously recorded ECG information (c, f).

• 

  Fig. 4. Overview of the classification pipeline, illustrating data preprocessing, feature extraction, model training, and evaluation steps. The classification pipeline shows the steps taken in constructing predictive models in the MADDEC registry using a random 80/20 split of development and validation sets. Extreme gradient boosting classifier was used to train models.