Subgroup Analysis in Diagnostic Test Accuracy Meta-Analysis with Trees (DamasTree)

• Applicant: Professor Dr. Alexander Hapfelmeier
• Subject Area: Epidemiology and Medical Biometry/Statistics
• Term: since 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 539422711

Findings of diagnostic test accuracy (DTA) studies and respective meta-analyses (DTA-MA) are of high relevance for patients, treating physicians and healthcare providers, as they can have a major impact on clinical decision-making and patient care. Their importance is also highlighted by a fast body of original publications, systematic reviews with DTA-MA, funding opportunities, conference talks and workshops. Own preliminary work, e.g. DTA-MA of self-report questionnaires for detecting anxiety disorders, motivated the present application as it revealed the need to perform subgroup analyses (SA) in this context. However, there is a large discrepancy between the well justified and described methods for conducting a DTA-MA, and the sparse information about respective SA. This finding was surprising, as SA are well founded and understood in other areas, such as RCTs. The respective methodology elaborated for DTA-MA, however, often only focuses on the case of comparing two pre-specified subgroups with reference to meta-regression. Other descriptions and guides on DTA-MA are essentially concerned with the investigation of sources of heterogeneity or recommend looking at subgroups separately or conducting SA through meta-regression, but without further explanation or guidance. Available methods and software often do not implement meta-regression or do not explicitly support SA. For the own DTA-MA, methods recommended for RCTs had to be transferred and adapted to perform SA in DTA-MA. Evidently, there is an urgent need for the development, implementation, guidance and dissemination of methodology for DTA-MA. There are five objectives. The first objective is a systematic review of DTA-MA to investigate the frequency, type of implementation and presentation of SA. A short ad-hoc survey already showed that often, SA are performed as results are produced per subgroup. Simultaneous modelling and estimation of interaction effects seems to be rare. Therefore, the second objective is the methodological elaboration of SA in DTA-MA. For this purpose, the well-established bivariate model will be revised and expanded appropriately. Furthermore, to enable explicit definition and hypothesis testing of subgroups with homogeneous or heterogeneous DTA conditional on potentially interacting covariates, the bivariate model will also be embedded in the framework of recursive partitioning using generalized linear mixed model trees. The third objective will address the recent development of incorporating DTA at multiple thresholds per study for a holistic analysis. The fourth objective is to provide a user-friendly implementation and to disseminate it through a freely available R-package. The fifth objective is to apply the new methods, make appropriate comparisons with other commonly used approaches identified in objective 1, and thereby eventually gain further important insights into the use case of DTA of the Hospital Anxiety and Depression Scale for Anxiety.

• DFG Programme: Research Grants