Meta-Analysis in Epidemiology

The use of meta-analyses in order to synthesise the evidence from epidemiological studies has become more and more popular recently. It has been estimated by Egger et al. (1998) that from articles retrieved by MEDLINE with the medical subject heading (MeSH) term “meta-analysis” some 33% reported results of a meta-analysis from randomised clinical trials and nearly the same proportion (27%) were from observational studies, including 12% papers in which the aetiology of a disease was investigated. The remaining papers include methodological publications or review articles. Reasons for the popularity of meta-analyses are the growing information in the scientific literature and the need of timely decisions for risk assessment or in public health. While methods for meta-analyses in order to summarise or synthesise evidence from randomised controlled clinical trials have been continuously developed during the last years, and methods are now summarised in several text books for example Sutton et al. (2000), Whitehead (2002) and in a handbook by Egger et al. (2001), Dickersin (2002) argued that statistical methods for meta-analyses of epidemiological studies are still behind in comparison to the progress that has been made for randomised clinical trials. The use of meta-analyses for epidemiological research caused many controversial discussions, see for example Blettner et al. (1999), Berlin (1995), Greenland (1994), Feinstein (1995), Olkin (1994), Shapiro (1994a,b) or Weed (1997) for a detailed overview of the arguments.