Meta-analysis of ROC curves.

摘要: The authors present a method to combine several independent studies of the same (continuous or semiquantitative) diagnostic test, where each study reports complete ROC curve; plot true-positive rate sensitivity against false-positive one minus specificity. result analysis is pooled curve, with confidence band, as opposed earlier proposals that in area under curve. based on two-parameter model for curve can be estimated individual parameters are then bivariate random-effects meta-analytic method, and drawn from parameters. propose use specifies linear relation between logistic transformations Specifically, they define V = In(sensitivity/(1 - sensitivity)) U In((1 specificity)/specificity), D U, S + U. defined alpha betaS. beta using weighted regression bootstrapping get standard errors, maximum likelihood. show how procedure works continuous test data categorical data.