Review and Limitations of Methods for Constructing a Receiver Operating Characteristic Curve in a Case-Control Design

A conditional approach for the receiver operating characteristic curve construction to evaluate diagnostic test performance in a family-matched case–control design

“…Therefore, it may not be appropriate for familial paired case-control designs. 7 Pepe presented covariate adjusted ROC curve (AROC) for data arising from a frequency matched case-control design. 10 AROC can be obtained by averaging the covariate specific ROC curves with respect to the covariate distribution among the cases.…”

“…10 These proposed methods are not useful for family matched case-control designs where groups are matched by family identification and test results among family members are likely to be correlated. 7 Liu and Wu proposed a semi parametric approach to ROC curve estimation and AUC calculation for repeated measures design. 14 In the presence of repeated measurements on the same subject over time or under various conditions, the subject's multiple observations are not independent and the intra-subject correlation should be taken into account.…”

“…However conditional approaches are unable to provide predicted probabilities needed for ROC construction. 7 To create an ROC curve illustrating a biomarker's ability to discriminate between groups in family matched study design, the appropriate approach should consider correlation between the marker measurements within a family. If the marker is informative then it is rational to expect that the ability of the marker to discriminate between groups to improve with strength of correlation between corresponding marker levels of related subjects.…”

“…However, we note that the estimation of the AUC is marginal and does not itself consider within cluster correlation, and is therefore unable to take advantage of the familial design. 7…”

“…Therefore, it may not be appropriate for familial paired case–control designs. 7…”