Table 3.

Overall signature classification performance stratified by dataset characteristics

Parameter	Bacterial vs. non-bacterial				Viral vs. non-viral
	Median AUC	IQR	p-value	N	Median AUC	IQR	p-value	N
All datasets	0.832	[0.796–0.849]	-	31	0.884	[0.864–0.896]	-	37
Age	-	-	-	-	-	-	-	-
Adult only	0.846	[0.826–0.870]	-	14	0.916	[0.906–0.935]	-	12
Pediatric only	0.798	[0.756–0.818]	< 0.001	15	0.860	[0.841–0.870]	< 0.001	24
# of phenotypes	-	-	-	-	-	-	-	-
2 phenotypes	0.871	[0.846–0.894]	-	12	0.911	[0.880–0.923]	-	22
>2 phenotypes	0.819	[0.788–0.835]	< 0.001	19	0.855	[0.838–0.874]	< 0.001	15
Biological source	-	-	-	-	-	-	-	-
Whole blood	0.838	[0.802–0.859]	-	28	0.884	[0.865–0.900]	-	32
PBMC	0.705	[0.668–0.753]	< 0.001	3	0.831	[0.791–0.882]	0.012	4

AUCs were calculated for each of the 29 evaluated signatures and then stratified by different dataset characteristics. Mean AUCs were first generated for each signature across the datasets in the parameter group, weighted by the number of subjects in each validation dataset. The median of the weighted AUC values and IQR were then calculated and presented here. N represents the number of datasets for the specified cohort composition.