Blood-based identification of non-responders to anti-TNF therapy in rheumatoid arthritis

BMC Medical Genomics

Table 6 Performance of previously published classifiers assessed via repeated leave-one-sample-out cross validation

Classifier	AUROC(95 % CI)	Specificity(95 % CI)	Sensitivity(95 % CI)	Precision(95 % CI)	Likelihood ratio(95 % CI)
Lequerré_20	59 %(48-70 %)	58 %(47-69 %)	50 %(32-68 %)	31 %(19-46 %)	1.2(0.78-1.8)
Lequerré_8	50 %(38-62 %)	61 %(50-71 %)	47 %(29-65 %)	31 %(19-46 %)	1.19(0.76-1.9)
Julia	57 %(48-69 %)	74 %(60-86 %)	40 %(21-61 %)	45 %(24-68 %)	1.57(0.79-3.1)
Stuhlmuller_82	48 %(37-59 %)	40 %(30-52 %)	63 %(44-79 %)	29 %(18-41 %)	1.05(0.76-1.4)
Stuhlmuller_11	40 %(29-52 %)	43 %(32-54 %)	47 %(29-65 %)	24 %(14-36 %)	0.82(0.54-1.2)
Tanino	54 %(43-65 %)	51 %(40-62 %)	50 %(32-68 %)	28 %(17-42 %)	1.02(0.68-1.5)
Sekiguchi	53 %(41-66 %)	24 %(15-34 %)	75 %(57-89 %)	27 %(18-38 %)	0.98(0.78-1.2)

Each classifier was re-trained on the four whole blood data sets (GSE12051, GSE19821, GSE58795, GSE33377_infliximab) using the genes and clustering method specified in the original studies. The performance of each classifier was assessed via leave-one-sample-out cross validation across. Julia_8 was originally trained on GSE12051, so GSE12051 was omitted from cross validation for Julia_8 to avoid feature selection bias
CI: confidence interval. No AUROC p-value < 0.05

ISSN: 1755-8794