Valid Methods for Meta-Analyses with Few Studies and Small Sample Sizes

Research on methods for meta-analysis has received increasing attention over the past decades. In the light of a large number of results that cannot be reproduced, the rigorous and robust aggregation of potentially heterogeneous quantitative findings on one research question is of great interest.

Different clinical trials with the same aim commonly provide varying results, even if the designs are very similar. Besides natural sampling variance, differences in study populations, clinical practice and the conduct of the studies can be found in practice. A valid interpretation of research results depends therefore on adequate modelling of such heterogeneity. Small sample sizes and insufficiently validated research results can be found in many areas of science. These circumstances highlight the necessity for meta-analysis methods that provide valid results even under difficult conditions including

• between-study heterogeneity
• unbalanced designs
• small to moderate sample sizes
• few studies only

In this project we develop flexible and valid statistical methods for different meta-analytic models that do not require strong model assumptions and facilitate a clear interpretation of the results including statistical inference. In order to make recommendations for the appropriate application of the new methods, these are thoroughly examined in extensive Monte Carlo simulation studies and will be applied to current data sets from a variety of fields including medicine. The methods developed are implemented in open-source software with comprehensive documentation ensuring widespread use.

This project will be a joint venture of the Department of Medical Statistics of the University Medical Center Göttingen and the Chair Mathematical Statistics and Applications in Industry at the Technical University of Dortmund and is funded by the DFG (German Research Foundation).