Servicenavigation


Sie sind hier:

 

 

 

Tutorial Evidence Synthesis/Meta-Analysis at DAGStat Tagung 2016

Instructors: Guido Knapp (TU Dortmund), Gerta Rücker, Guido Schwarzer (Uni Freiburg)

Date: Monday March 14, 2016 from 9am to 5pm.

Location: Verfügungsgebäude, VG 1.102

 

Evidence synthesis is used for decision making in various fields of applications, for instance, social sciences, medicine, psychology, ecology, and economics if multiple sources of evidence should be combined quantitatively. In this tutorial we will discuss several methods for combining information from several independent studies or experiments and show how to conduct analyses using the statistical software R. Focus of this tutorial will be model building and data analysis.

The tutorial is organised as follows: First, we will present the basics of generic fixed-effect and random-effects meta-analysis models for comparing two interventions. Especially, various estimators for the heterogeneity parameter will be presented. Then, the use of these models will be demonstrated for various effect sizes of normal, discrete, and survival outcomes like standardized mean difference, correlation coefficient, risk difference, odds ratio, or hazard ratio. Beyond the standard meta-analysis model for binary data, the use of unconditional and conditional generalized mixed linear model will be discussed if data are available as 2x2-tables. The next topic is meta-regression which can be used for exploring heterogeneity of study results or for identifying good predictor variables for the outcome variable.  Meta-regression techniques can also be used for subgroup analysis which will be discussed next. Then, we will study the so-called small study effect, which is a generic term for the phenomenon that smaller studies show larger treatment effects than larger ones. In case of more than one outcome variable, we arrive at multivariate meta-analysis. How to deal with this situation will be the next topic. Finally, methods for comparing more than two interventions, the so-called network meta-analysis also known as multiple or mixed treatment comparison, will be discussed.

All presented methods will be illustrated by real data examples from social and health sciences. For analysing the data, we will use R packages meta, netmeta, metasens, and metafor. The corresponding R code will be provided. There will be no software demonstration or practicals in this tutorial.

Further reading:

Hartung, J., Knapp, G., Sinha, B.K. (2008). Statistical Meta-Analysis with Applications. Wiley, Hoboken, NJ, USA.

Schwarzer, G., Carpenter, J.R., Rücker, G. (2015). Meta-Analysis with R. Springer International Publishing Switzerland. Web-Appendix: http://meta-analysis-with-r.org/.

 

Slides of the tutorial

Slides of the lectures in original format can be found here and in printer friendly 2x2 format here.