Bacteremia detection from complete blood count and differential leukocyte count with machine learning: complementary and competitive with C-reactive protein and procalcitonin tests

Table 2 Bacteremia prediction capability indicated with AUCs^a of biomarkers (CRP/PCT) and models (random forest/logistic regression)

Methods/Group	CBC/DC^b		CRP&CBC/DC^c			PCT&CBC/DC^d
Methods/Group	Cross-validation	Testing		Cross-validation	Testing		Cross-validation	Testing
Used biomarker	–	–	CRP	0.692 ± 0.017	0.699	PCT	0.748 ± 0.021	0.731
ML^e models
Random forest	0.792 ± 0.010	0.802	CRP excluded^f	0.797 ± 0.010	0.806	PCT excluded^h	0.759 ± 0.022	0.767
			Included^g	0.806 ± 0.011	0.814	Included	0.777 ± 0.018	0.767
Logistic regression	0.763 ± 0.009	0.772	Excluded	0.769 ± 0.009	0.775	Excluded	0.735 ± 0.030	0.734
			Included	0.784 ± 0.011	0.790	Included	0.761 ± 0.024	0.745

ISSN: 1471-2334