Statistical Methods for Evaluation of Biomarkers

Eunice Kennedy Shriver National Institute Of Child Health & Human Development

Investigators

Linked publications & trials

Abstract

When biomarkers are used for disease diagnosis it is well known that single biomarker is usually not sufficient and combing with other biomarkers can substantially improve the diagnostic accuracy. However, combining biomarkers when only group testing data are available has not been addressed in the literature. In the past year investigators and staff on this project developed novel methods for combining biomarkers to improve the diagnostic accuracy based on group testing data. The objective is to obtain and compare the AUC estimation in group testing and individual testing, while incorporating misclassification of testing results that may vary as the group size and the number of the diseased individuals in the group changes. The team developed an innovative approach to compute the maximum likelihood estimates of the accuracy based on the pairwise bivariate fitting, assuming that the biomarkers follow multivariate distributions. The team also developed another approach through comparing Youden Index and the associated cut-off point in group testing and individual testing under Normal, Gaussian, Log-Normal, non-parametric assumption, respectively, and demonstrated that group testing not only reduces the study cost but also can improve the statistical precision for estimation of the Youden index.