Avenues for further research

In this chapter, we present an overview of the recent statistical methods for diagnostic meta-analysis and suggest a few directions for future research. We discuss two important issues regarding (a) the robustness of model misspecifications and (b) the identifiability of models and the assumption of conditional independence in the absence of a gold standard. With increasing availability of biomedical data, the individual patient-level data meta-analyses offer new insights into evidence synthesis compared to traditional aggregated data-based meta-analyses. In particular, the approaches to combine individual patient-level data with aggregated data can inform personalized medical decision based on patient-level characteristics and help to identify clinically relevant subgroups. However, such integration methods for diagnostic prediction research are limited, and hence there is a growing need for developing of novel statistical methods that can address potential issues including model validation, missing predictors, and between-studies heterogeneity while combining both types of data. Despite the perceived advantages of individual patient-level data, using individual patient-level data alone may still encounter a number of challenges, such as partial verification bias and the absence of a gold standard. We discuss these challenges by two examples.