Statistical Methods for the Meta-Analysis of ROC Curves

• Applicant: Professorin Dr. Annika Hoyer
• Subject Area: Epidemiology and Medical Biometry/Statistics
• Term: since 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 519901253

Meta-analyses and systematic reviews are the cornerstones of evidence based medicine and inform treatment, diagnosis, or prevention of individual patients as well as policy decisions in health care. Statistical methods for the meta-analysis of intervention trials are well established today. Meta-analysis for diagnostic test accuracy studies, however, has been a vivid research area in recent years which is especially due to the increased complexity of diagnostic studies with their bivariate outcome of sensitivity and specificity. An even more increased complexity arises when single studies do not only report a single pair of sensitivity and specificity, but a full ROC curve with several pairs of sensitivity and specificity, each pair corresponding to a different threshold for the diagnostic test under evaluation. Researchers frequently ignore this information and use only one pair of sensitivity and specificity from each study to arrive at meta-analytic estimates. Meanwhile, methods to deal with the full information have been proposed, but most of them have at least one of the following disadvantages. Some of them are two-step approaches where estimation uncertainty from the first step is ignored in the second step. It also often occurs that the number of thresholds has to be identical across studies, or the specific values of thresholds are ignored thus making clinically relevant inference on sensitivity and specificity at given thresholds impossible. Recently developed models motivated from survival analysis compensate for these disadvantages. So far, however, the various models have not been systematically compared and evaluated. Furthermore, there is a growing need for methods for network meta-analysis that compare several diagnostic tests with respect to their ROC curves. This project aims to evaluate statistical methods for the meta-analysis of full ROC curves, that is, methods that can use information on sensitivity and specificity from several thresholds of a diagnostic test from the single studies. These methods will be systematically collected, implemented in one of the major statistical software packages and compared in a large simulation study that mirrors real life data. In addition, the simulation will investigate two further approaches which are to be newly developed within the project and which are free from the above mentioned disadvantages. Furthermore, the effect of the selection of individual thresholds on the resulting estimators will be investigated. Moreover, methods for network meta-analysis of diagnostic test accuracy studies will be developed, taking into account full ROC curves.

• DFG Programme: Research Grants