Deep neural network-estimated electrocardiographic age as a mortality predictor

Figures

• 

  Fig. 1. Chronological vs ECG-age. The scatter plots display the relation between the predicted ECG-age and chronological age. The black line is the identity line. The lateral histograms show the distributions of predicted ECG-age and chronological age among patients of the cohorts. a CODE-15% cohort, b ELSA-Brasil cohort, c SaMi-Trop cohort. The mean predicted ECG-age was 52 (s.d. 19), 47 (s.d. 11), 63 (s.d. 14), for CODE-15%, ELSA-Brasil cohort and SaMi-Trop cohort, respectively. The R2 (Pearson correlation) was 0.71 (r = 0.84) in the CODE-15%, 0.32 (r = 0.57) in ELSA-Brasil cohort and 0.35 (r = 0.59) in the SaMi-Trop cohort.

• 

  Fig. 2. Adjusted survival curves. The plots display the survival curves for the different cohorts. a CODE-15% cohort, b ELSA-Brasil cohort, c SaMi-Trop cohort. The curves are computed from the age and sex-adjusted Cox proportional model for all-cause mortality. Three groups of patients are taken into consideration: those with ECG-age more than 8 years greater than the chronological age (denoted by: “>8 years older”); those with ECG-age within a range of 8 years from their chronological age (denoted by: “±8 years”); and, those with ECG-age more than 8 years smaller than the chronological age (denoted by: “>8 years younger”).

• 

  Fig. 3. Adjusted odds ratios (ORs) for the ELSA-Brasil cohort. The figure shows the adjusted ORs of the ECG-age being more than 8 years greater than the chronological age for risk factors. a All patients; and, b only for patients with normal ECG. The dots represent the adjusted ORs (by age and sex) and the horizontal lines represent the corresponding 95% CIs. HAS hypertension, DM diabetes mellitus, DLP dyslipidemia.