Table 10.

Multivariate ablation outcome classification performance and NRI assessment.

	FCLIN						FCLIN+SIG
	Training			Validation			Training			Validation
	AUC [%]	Sensitivity [%]	Specificity [%]	AUC [%]	Sensitivity [%]	Specificity [%]	AUC [%]	Sensitivity [%]	Specificity [%]	AUC [%]	Sensitivity [%]	Specificity [%]	NRI [a. u.]	p-value
NDI [a.u.]	58	19	97	67	67	90	60	25	92	70	100	50	0	N.A
AF duration [months]	19	0	100	43	33	100	56	98	19	67	67	90	0	N.A
ABSPM [mV]	59	15	97	67	67	90	66	15	97	79	67	100	0.25	0.32

ROC analysis of the multivariate predictors of ablation outcome obtained by using clinical data only (F_CLIN) and by integrating an AF complexity descriptor (F_CLIN+SIG) in the training and validation sets. Training and validation of classification models were first performed on multivariate features depending on patient's clinical information only (F_CLIN). The same procedure was applied again to multivariate classifiers obtained by integrating clinical data with the parameter of signal complexity under exam (F_CLIN+SIG). Classification models based on F_CLIN were trained, tested and re-evaluated each time a signal complexity feature was examined. The variable F_CLIN+SIG including AF duration was compared to F_CLIN^*. Differences in classification accuracy were quantified by the NRI. Results for the parameters with the highest classification performance (AUC≥70%) are highlighted in boldface; AUC, area under curve; a.u., arbitrary units; N.A., not applicable.