TABLE 3.

ROC analysis of patients with AF in the training and validation cohorts.

	Training cohort (n = 46)			Validation cohort (n = 443)
		P-value			P-value#
	AUC (95% CI)	Compared to 0.5	Compared to AUC of CHA2DS2-VaSc	AUC (95% CI)	Compared to 0.5	Compared to AUC of CHA2DS2-VaSc
All patients
FD	0.746 (0.599, 0.893)	0.004*	0.811	0.602 (0.541, 0.663)	0.002*	<0.001*
CHA2DS2-VaSc	0.769 (0.643, 0.895)	0.002*	-	0.856 (0.817, 0.895)	<0.001*	-
Combined†	0.842 (0.732, 0.952)	<0.001*	0.063	0.863 (0.827, 0.899)	<0.001	0.427
Low to moderate risk
Number of patients n (%)	35 (76.1)			198 (44.7)
FD	0.762 (0.592, 0.932)	0.005*	0.545	0.696 (0.568, 0.823)	0.079	0.171
CHA2DS2-VaSc	0.696 (0.548, 0.843)	0.056	-	0.572 (0.388, 0.756)	0.519	-
Combined†	0.848 (0.733, 0.963	0.001*	0.009*	0.747 (0.608, 0.886)	0.026*	<0.001*

p-Values were obtained by z-test.

*Marked significant differences.

^†If the box-counting dimension value is greater than threshold (2.426 in the training cohort; 2.438 in the validation cohort), one point will be added to CHA₂DS₂-VaSc system, and the new risk score was assessed by ROC analysis.

^#The DeLong test was used to compare the AUCs.

AUC, area under the curve; CI, confidential interval; FD, fractal dimension (represented by box-counting dimension).