The medical literature contains a vast number of studies of the accuracy of diagnostic tests. Increasingly, meta-analysis is being used as a tool to synthesize the results of individual tests. Such analyses are characterized typically by large amounts of aggregated data, but with substantial uncertainties about the validity of the data. Moreover, they are frequently conducted by clinical epidemiologists and subject matter experts. Consequently the use of complex or highly sophisticated statistical tools to obtain maximum statistical efficiency is neither necessary nor desirable. We propose to develop and study the properties of relatively simple statistical methods for synthesizing the data using ROC analysis. Our goal is to develop methods that, for the most part, can be applied by hand calculation, but which nonetheless enjoy high efficiency, relative to full maximum likelihood estimation, a technique which is computationally burdensome and may exhibit convergence problems in the meta-analytic setting. Our work builds on the ideas of other researchers who have recently proposed techniques for this purpose.
Begg, C B; Mazumdar, M (1994) Operating characteristics of a rank correlation test for publication bias. Biometrics 50:1088-101 |