Table 1.

Summary of machine learning models developed for disease prediction.

Readout	Parameters	Algorithm	Sensitivity (Recall)	Specificity	Precision (PPV)	F1-score	Accuracy	AUROC	Test Cohort	Ref.
Mortality	33 clinical parameters	Random forest	85.71 %	92.45%	–	–	89.47%	0.921	No	[54]
Mortality	45 proteins	Bayesian network	92.68%	86%	–	–	89.01%	0.953	No	[54]
Mortality	CRP, BUN, serum calcium, serum albumin, lactic acid	SVM	91%	91%	62.5%	–	–	0.93	No	[55]
In-hospital mortality	Age, lymphocyte, D-dimer, CRP, creatinine (ALDCC)	Logistic regression	0.91 ± 0.03	0.78 ± 0.04	0.92 ± 0.03	0.92 ± 0.03	0.91 ± 0.03	0.992	Yes	[56]
In-hospital mortality	Age, hs-CRP, lymphocyte, d-dimer	Logistic regression	0.839	0.794	–	–	–	0.881	Yes	[57]
In-hospital mortality	LDH, neutrophils, lymphocyte, hs-CRP, age (LNLCA)	Logistic regression	92 ± 2.6%	92 ± 3%	–	–	–	0.991	Yes	[58]
In-hospital mortality	PTA, urea, WBC, IL-2r, indirect bilirubin, myoglobin, FgDP	LASSO logistic regression	98%	91%	–	–	–	0.997	No	[59]
In-hospital mortality	Disease severity, age, hs-CRP, LDH, ferritin, IL-10	Simple-tree XGBoost	>85%	–	>90%	>0.90	>0.90	1.000	Yes	[60]
Disease severity	28 blood and urine parameters	SVM	–	–	–	–	0.8148	–	Yes	[61]
Disease severity	Different biomarker combinations	Penalized logistic regression	>82%	>71%	>87%	–	>85%	–	Yes	[62]

BUN, blood urea nitrogen; CRP, c-reactive protein; FgDP, fibrinogen degradation products; hs-CRP, high-sensitivity C-reactive protein; IL-2r, interleukin-2 receptor; IL-10, interleukin-10; LASSO, least absolute shrinkage and selection operator; LDH, lactate dehydrogenase; MCHC, mean corpuscular hemoglobin concentration; PPV, positive predictive value; PTA, prothrombin; SVM, support vector machine; WBC, white blood cell activity; XGBoost, eXtreme Gradient Boosting.