PhD student: Kanella Panagiotopoulou
Title: Modelling departures from normality in network meta-analysis
Supervisor: Anna Chaimani
Doctoral school: ED 393 Epidemiology and Biomedical Information Sciences, Université Paris Cité
In the vast majority of meta-analysis and the random effects are assumed to follow a normal distribution. The main reasons for this choice are probably convenience and model simplicity, tradition, and software availability. However, under certain conditions this normality assumption can be rather implausible and potentially threaten the validity of the results.
The aim of the first project of this Thesis is to perform a large simulation study evaluating the performance of existing methods employing non-normal distributions for the random effects in the context of pairwise meta-analysis. Based on the results from the simulation study, I will extend the approach(es) with the best performance into network meta-analysis. Preliminary results suggest that mixture models that employ families of distributions may be promising for synthesizing heterogeneous data. The final goal is to develop a user-friendly R package facilitating the use of the new NMA approaches and to illustrate its features by applying the new NMA methods in a real clinical example. Appropriate NMA data can be found by the COVID-NMA database which is a living systematic review for COVID19 treatments and vaccines updated continuously.