Table 4.

Summary of receiver operating characteristics (ROC) curve metrics discrimination of non-severe and severe COVID−19 disease condition based on the metabolic profile.

		AUC		VIP score		Severe		Non-severe		Non-severe
										and severe
Performance	All	Top 5	Top 10	>1.5	≥1.0	p-value	p-value ≤0.01	p-value	p-value ≤0.01	p-value ≤0.01
	markers					<0.01	log2(FC) > |2.0|	<0.01	log2(FC) > |2.0|	log2(FC) > |2.0|
Metabolites (n)	43	5	10	5	15	10	7	18	2	9
AUC-ROC (100 CV)	0.877	0.860	0.882	0.866	0.865	0.861	0.859	0.883	0.818	0.865
CI 95% (100 CV)	0.827–0.946	0.805–0.908	0.824–0.932	0.801–0.929	0.803–0.921	0.801–0.926	0.794–0.918	0.824–0.939	0.738–0.897	0.791–0.927
AUC-ROC (25% holdout sample)	0.894	0.889	0.899	0.875	0.853	0.875	0.850	0.919	0.792	0.870
Sensitivity (%)	86.11	80.56	83.33	83.33	88.89	83.33	80.56	91.67	94.44	88.89
Specificity (%)	76.92	73.08	80.77	76.92	76.92	80.77	76.92	73.08	57.69	69.23
Accuracy (%)	81.52	76.82	82.05	80.13	82.91	82.05	78.74	82.37	76.07	79.06
Precision (%)	83.78	80.56	85.71	83.33	84.21	85.71	82.86	82.50	75.56	80.00

FC, fold change; AUC, area under curve; SEN, sensitivity; SPE, specificity; PRE, precision; ACC, accuracy; CI, confidence interval; Best score in each ROC metric (sensitivity, specificity, precision and accuracy).

The metrics were obtained using different sets of metabolite biomarkers.