|
Abstract:
|
The evaluation of the predictive performance of biomarkers is a vital and growing area of research in precision medicine. However, statistical methods for meta-analysis of the predictive accuracy of tests, as measured by the positive and negative predictive value (PPV and NPV respectively), have received limited attention in the literature, in contrast to methods for meta-analysis of diagnostic accuracy. In this paper, we propose a hierarchical summary predictive ROC (HSPROC) curve model to summarize estimates of PPV, NPV and disease prevalence jointly. The model accounts for the relation between PPV and NPV stemming from the dependence on the threshold for test positivity, and also addresses the monotonicity of the summary predictive ROC curve. The HSPROC curves generated from the model can be used for comparison of different biomarkers. We applied the proposed method to a meta-analysis of prognostic capabilities of biomarkers for rapid rule-out of acute myocardial infarction, and compared the prognostic capabilities of these biomarkers.
|
