Table 1.

Characteristics of related work.

Category	Study	Population	Sample Size	Dataset	Outcome	Models	External Validation	Performance
Traditional regression based pLOS-ICU prediction models	Zoller et al. [16]	General ICU patients	110	Single-center	pLOS-ICU	Customized SAPS II	×	AUROC: 0.70
	Houthooft et al. [18]	General ICU patients	14,480	Single-center	pLOS-ICU	Customized SOFA score	×	Sensitivity: 0.71
	Herman et al. [6]	Patients undergoing CABG	3483	Single-center	pLOS-ICU	LR	×	AUROC: 0.78
	Rotar et al. [19]	Patients following CABG	3283	Single-center	pLOS-ICU	LASSO	×	AUROC: 0.72
ML-based models in ICU	Meiring et al. [24]	General ICU patients	22,514	Multicenter	ICU mortality	AdaBoost, RF, SVM, DL, LR, and customized APACHE-II	×	AUROC: 0.88 (DL)
	Lin et al. [20]	Acute kidney injury patients	19,044	Single-center	ICU mortality	ANN, SVM, RF, and customized SAPS II	×	AUROC: 0.87 (RF)
	Viton et al. [25]	General ICU patients	13,000	Single-center	ICU mortality	DL	×	AUROC: 0.85
	Qian et al. [13]	General ICU patients	17,205	Single-center	Acute kidney injury	XGBoost, RF, SVM, GBDT, DL, and LR	×	AUROC: 0.91 (GBDT)
ML-based pLOS-ICU prediction models	Navaz et al. [26]	General ICU patients	40,426	Single-center	pLOS-ICU	Decision tree	×	Accuracy: 0.59
	Rocheteau et al. [27]	General ICU patients	168,577	Multicenter	LOS-ICU	DL	√	R2: 0.40
	Ma et al. [28]	General ICU patients	4000	Single-center	pLOS-ICU	Combining just-in-time learning and one-class extreme learning	×	AUROC: 0.85
	Our study	General ICU patients	160,238	Multicenter	pLOS-ICU	RF, SVM, DL, GBDT, and customized SAPS II	√	-