Table 4 Comparison of the performances obtained for age-sex-specific groups of patients with the ResNet model.

From: Analyzing artificial intelligence systems for the prediction of atrial fibrillation from sinus-rhythm ECGs including demographics and feature visualization

 

AUC

Sensitivity

Specificity

DOR

F1 score

[18, 60)—M

0.61 (0.49–0.72)

33.33% (17.29–52.81)

80.67% (73.43–86.65)

2.09 (0.88–4.93)

0.29 (0.15–0.43)

[60, 70)—M

0.63 (0.52–0.75)

46.67% (28.34–65.67)

71.33% (63.39–78.41)

2.18 (0.98–4.84)

0.32 (0.19–0.44)

[70, 80)—M

0.78 (0.68–0.89)

70.00% (50.60–85.27)

68.00% (59.90–75.37)

4.96 (2.11–11.63)

0.42 (0.30–0.54)

[80, 90)—M

0.76 (0.66–0.87)

83.33% (65.28–94.36)

52.67% (44.36–60.87)

5.56 (2.02–15.31)

0.40 (0.29–0.50)

90 +—M

0.83 (0.72–0.93)

96.67% (82.78–99.92)

46.88% (29.09–65.26)

25.59 (3.10–211.26)

0.76 (0.65–0.86)

[18, 60)—F

0.65 (0.54–0.76)

30.00% (14.73–49.40)

88.67% (82.48–93.26)

3.35 (1.32–8.50)

0.32 (0.15–0.47)

[60, 70)—F

0.73 (0.62–0.84)

36.67% (19.93–56.14)

90.67% (84.84–94.80)

5.62 (2.23–14.17)

0.40 (0.23–0.56)

[70, 80)—F

0.69 (0.58–0.81)

43.33% (25.46–62.57)

79.33% (71.97–85.51)

2.94 (1.29–6.69)

0.35 (0.21–0.49)

[80, 90)—F

0.72 (0.61–0.83)

63.33% (43.86–80.07)

68.67% (60.59–75.98)

3.79 (1.67–8.58)

0.40 (0.27–0.51)

90 +—F

0.76 (0.66–0.87)

80.00% (61.43–92.29)

52.89% (43.61–62.03)

4.49 (1.71–11.78)

0.43 (0.31–0.54)

  1. We evaluate the same ResNet model with ten age-sex-specific test sets. The model is trained with ECG samples that fairly represent all the ten groups of patients. To compute the metrics additional to AUC, we set a unique threshold for the ResNet model, as described in “Methods”. Performances are assessed according to the aggregation rule by average. 95% confidence intervals are reported in brackets.
  2. M Male, F Female, AUC Area Under the Curve, DOC Diagnostic Odd Ratio.
Back to article page