Table 6.

Comparison among the features used in exist state-of-the-art works and different combinations of the proposed features on ICU admission prediction on Site A dataset. One-tailed t-test is used to evaluate the statistical significance between a feature combination and the best performer.

Features	AUC			Sensitivity (PPV = 70%)			K
	Mean	95% CI	p value	Mean	95% CI	p value
Img feature (Tang 2020)	0.818	(0.796, 0.839)	$p<0.001$	51.0%	(39.3%, 62.6%)	$p<0.001$	8
Img feature (Zhu 2020)	0.776	(0.762, 0.790)	$p<0.001$	48.6%	(35.4%, 61.7%)	$p=0.001$	46
DVB	0.855	(0.844, 0.866)	$p=0.002$	76.7%	(73.2%, 80.1%)	$p=0.017$	1
HLQ	0.789	(0.781, 0.797)	$p<0.001$	51.4%	(45.3%, 57.5%)	$p<0.001$	21
WLR	0.859	(0.843, 0.873)	$p<0.001$	71.4%	(60.5%, 82.3%)	$p=0.022$	70
WLR + HLQ	0.866	(0.857, 0.875)	$p<0.001$	68.6%	(57.6%, 79.5%)	$p=0.008$	61
WLR + DVB	0.876	(0.867, 0.886)	p = 0.109	81.4%	(76.0%, 86.8%)	p = 0.152	4
HLQ + DVB	0.865	(0.844, 0.885)	p = 0.080	70.0%	(60.9%, 79.1%)	$p=0.012$	4
WLR + HLQ + DVB	0.884	(0.875, 0.893)	–	84.3%	(79.9%, 88.7%)	–	52