Multivariate methods in evidence synthesis: extending their framework and applicability

• Applicant: Dr. Theodoros Evrenoglou
• Subject Area: Epidemiology and Medical Biometry/Statistics
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 554095932

Multivariate network meta-analysis (mvNMA) is the current state-of-the-art meta-analytical method, offering the potential to account for extensive information across multiple treatments and outcomes, therefore forming the basis for a well-informed medical decision making. However, the adoption of mvNMA in real-life applications remains limited due to several methodological issues that hamper its use. Common issues include the absence of suitable methods to summarize the mvNMA’s extensive output, and the complexities around the method’s modelling assumptions. Consequently, such issues create a gap between the theoretical potential and the practical adoption of mvNMA. With three work packages (WPs), this project aims to propose a novel framework upon which medical decision-making can effectively consider information based on both multiple treatments and outcomes when informing clinical practice. In WP1, based on the joint posterior distribution of the mvNMA treatment effects, I will construct a novel probabilistic ranking metric that yields an amalgamated ranking list of all treatments across multiple outcomes. This will address current challenges on summarizing and interpreting the mvNMA final results, as principally mvNMA produces an extensive output comprising several treatment effect estimates per outcome. WP2 will extend the current mvNMA toolkit from the Bayesian to the frequentist framework, therefore addressing the need for simpler methods. Additionally, by generalizing my previous work, in WP2 I will develop a novel approach that produces treatment hierarchies based on the minimal clinically important value (MCID) across multiple outcomes. Τhis will allow the incorporation of treatment benefit-risk profiles in the ranking process. In WP3, I will evaluate the methodologies proposed in WPs 1-2 by applying them into real-life clinical examples and by conducting a simulation study. Finally, the recent rise of Large Language Models (LLMs) can tempt meta-analysts to use them to enable the interpretation of the mvNMA output. In WP3, I will explore this potential, contrasting the LLM performance with the new methods from WPs 1-2, thereby highlighting whether LLMs are useful for this task. Successfully completing the proposed project can have a significant impact on meta-analysts, policymakers, and guideline developers. By addressing common issues in evidence synthesis, the project will enable a better informed medical decision-making, providing a solid foundation for clinical practice.

• DFG Programme: WBP Position