In:
Circulation, Ovid Technologies (Wolters Kluwer Health), Vol. 142, No. Suppl_3 ( 2020-11-17)
Abstract:
Background: Electrocardiogram (ECG) is a low-cost, inexpensive, non-invasive, and ubiquitous test to detect structural or electrical cardiac abnormalities. The ECG interpretation is a visual task to correlate specific patterns with diseases. Subtle ECG patterns that are invisible to the human eyes may be useful for patient risk stratification. We aimed to develop a deep neural network to predict mortality using standard 12-lead ECG. Methods: We obtained the entire clinical MUSE database of Chang Gung Memorial Hospital for standard resting 12-lead 10-sec ECG voltage-time traces, which were linked to the National Death Registry for all-cause and cause-specific mortality. Ten thousand patients remained alive at one year from the ECG examination date, and 10000 patients died within one year were randomly selected to train a one-dimensional deep neural network to predict mortality risk. Data was split into 16,000 patients for the training and 4,000 patients for the validation sets. We prepared a separate held-out dataset of 19,737 randomly selected patients whose ECG traces were fed to the model to predict their mortality risk. Results: The mean age of the held-out dataset was 60.0±16.7 years, and 50.0% were female. Among them, 3,807 died during a mean follow-up period of 6.0 years (maximum period, 14 years). The model achieved an area under the receiver operator curve of 0.85 on the held-out dataset. The age- and sex-weighted Kaplan-Meier curve for the mortality was shown in figure 1, which stratified by the model prediction for survival status (log-rank test, p 〈 0.001). Using a Cox proportional hazards model, the ECG algorithm predicted death is associated with an age- and sex-adjusted hazard ratio of 3.2 (95% confidence interval, 3.0-3.4) for all-cause mortality and 2.9 (95% CI, 2.4-3.5) for cardiovascular mortality. Conclusion: The deep learning model could extract prognostic information from standard 12-lead ECG and predict mortality with reasonable precision.
Type of Medium:
Online Resource
ISSN:
0009-7322
,
1524-4539
DOI:
10.1161/circ.142.suppl_3.15588
Language:
English
Publisher:
Ovid Technologies (Wolters Kluwer Health)
Publication Date:
2020
detail.hit.zdb_id:
1466401-X
detail.hit.zdb_id:
80099-5
Permalink