Evaluation of the biomarker candidate MFAP4 for non-invasive assessment of hepatic fibrosis in hepatitis C patients

Table 3 Sensitivity and specificity of the univariate and multivariate classification, respectively

	Multivariate model^a		Univariate model
	Unrestricted	Lower limit of 80 % sensitivity^b,c	Unrestricted	Lower limit of 80 % sensitivity^b,c
Sensitivity^c estimate [95 % CI]	88.8 % [82.2 %; 93.6 %]	88.8 % [82.2 %; 93.6 %]	73.1 % [64.8 %; 80.4 %]	86.6 % [79.6 %; 91.8 %]
Specificity^d estimate [95 % CI]	63.2 % [58.4 %; 67.9 %]	63.2 % [58.4 %; 67.9 %]	75.0 % [70.5 %; 79.1 %]	54.9 % [49.9 %; 59.8 %]
Leave-one-out cross validation ^e
Sensitivity^c estimate [95 % CI]	80.6 % [72.9 %; 86.9 %]	81.3 % [73.7 %; 87.5 %]	71.6 % [63.2 %; 79.1 %]	85.8 % [78.7 %; 91.2 %]
Specificity^d estimate [95 % CI]	61.5 % [56.6 %; 66.3 %]	61.5 % [56.6 %; 66.3 %]	75.0 % [70.5 %; 79.1 %]	54.9 % [49.9 %; 59.8 %]

^aA multivariate logistic regression model considering age and gender besides MFAP4 serum levels was derived and the respective Youden optimal cutoffs were determined
^bCutoff optimization was restricted to a minimum sensitivity of 80 % accounting for the importance of the identification of high fibrosis stages
^cSensitivity was defined as probability of classifying stages F3 and F4 correctly
^dSpecificity was defined as probability of classifying stages F0 to F2 correctly
^eAs sensitivity and specificity values are prone to overoptimism in the analysis of the complete data set leave-one-out cross validation was performed to obtain unbiased estimates

ISSN: 1479-5876