CRISP: A causal relationships-guided deep learning framework for advanced ICU mortality prediction

Table 2 Performance of models trained on CMG sampling sets is presented, with mean values of corresponding 95% confidence interval for each metric. The best value for each metric is highlighted in bold

Models	Evaluation Set	AUROC	AUPRC	MCC
MIMIC-III_LR	MIMIC-IIItest	0.9249	0.7025	0.5281
MIMIC-III_SVM	MIMIC-IIItest	0.9169	0.6973	0.5689
MIMIC-III_RF	MIMIC-IIItest	0.9463	0.7335	0.6290
MIMIC-III_XGBoost	MIMIC-IIItest	0.9425	0.7447	0.6382
MIMIC-III_CRISP	MIMIC-IIItest	0.9480	0.7611	0.6678
MIMIC-III_LR	MIMIC-IVtest	0.9036	0.6231	0.5176
MIMIC-III_SVM	MIMIC-IVtest	0.9093	0.6376	0.4992
MIMIC-III_RF	MIMIC-IVtest	0.9124	0.6400	0.5560
MIMIC-III_XGBoost	MIMIC-IVtest	0.9127	0.6594	0.5666
MIMIC-III_CRISP	MIMIC-IVtest	0.9171	0.6683	0.5838
WCHSU_LR	WCHSUtest	0.8492	0.1500	0.2061
WCHSU_SVM	WCHSUtest	0.8366	0.0920	0.2543
WCHSU_RF	WCHSUtest	0.8930	0.4314	0.4211
WCHSU_XGBoost	WCHSUtest	0.8889	0.4487	0.4270
WCHSU_CRISP	WCHSUtest	0.9042	0.4771	0.4289

ISSN: 1472-6947