Abstract:
|
The receiver operating characteristic (ROC) curve is represented by a plot of the trade-off between sensitivity vs 1- specificity and is used extensively in medical research area to illustrate the performance of the biomarker in discriminating between diseased and non-diseased subjects. There have been some approaches developed to ROC curve estimation to address issues surrounding case control designs as well as designs that give rise to correlated biomarkers. Parametric, nonparametric and semiparametric approaches are available that accommodate both unmatched case-control and frequency matched designs using covariate information under some assumptions. However, these approaches do not apply to the case-control setting where subjects are matched only by family identification. In this talk, gaps and limitations of these methods will be described for family matched case-control design. An alternative conditional approach will be demonstrated, to provide appropriate ROC curves for familial paired data using information about the correlation among biomarker values. The ability and performance of conditional ROC curve will be evaluated through simulations and real dataset.
|