Table 4.

Sensitivity analysis of deep learning model for external validation dataset.

Variables	Nadir90		Fall20		Fall20/MAP10
	AUROC (PC, p-value)	AUPRC (PC, p-value)	AUROC (PC, p-value)	AUPRC (PC, p-value)	AUROC (PC, p-value)	AUPRC (PC, p-value)
Vital signs	0.819 (reference)	0.106 (reference)	0.858 (reference)	0.823 (reference)	0.848 (reference)	0.835 (reference)
Vital signs + Monitored pressure	0.835 (1.9, <0.001)	0.111 (5.1, 0.028)	0.862 (0.5, <0.001)	0.829 (0.7, 0.020)	0.853 (0.6, <0.001)	0.841 (0.7, 0.016)
Vital signs + Setting measures	0.839 (2.5, <0.001)	0.112 (6.1, 0.016)	0.858 (0.1, 0.081)	0.824 (0.1, 0.624)	0.849 (0.0, 0.227)	0.835 (0.0, 1.000)
Vital signs + Time setting	0.830 (1.3, <0.001)	0.108 (2.3, 0.566)	0.856 (−0.2, <0.001)	0.822 (−0.1, 0.636)	0.848 (-0.1, 0.012)	0.835 (0.0, 1.000)

Vital signs included SBP, DBP, MAP, and pulse rate; Monitored pressure included AP, and VP; setting measures included blood flow rate, dialysate flow rate, ultrafiltration rate, total ultrafiltration volume, temperature, and dialysate sodium level. P-values were calculated compared to the models that were trained by only vital signs. The Delong test was used to calculated p-values for comparison of AUROC. The bootstrap method was used to calculated p-values for comparison of AUPRC. MAP, mean arterial pressure; AUROC, area under the receiver operating characteristic curves; AUPRC, area under the precision-recall curve; PC, percentage change.