[Limitations of meta-analysis from published data in epidemiological research]

Metaanalyses of epidemiological studies have increased during the last years and are often used to evaluate the effect of risk factors which are inconsistent in different studies, mainly for small risk factors. Very often a metaanalysis is performed from published data. In this article we discuss this form of a metaanalysis and investigate whether the requirement to get reliable information is achievable with it. We mainly ask questions whether qualitative and quantitative dose-response analysis can be performed. We point out the differences between metaanalysis from experimental data and clinical randomized studies and epidemiological studies. We discuss different arguments that were given for performing metaanalysis in clinical trials and investigate whether they are also valid in observational studies. We mainly concentrate on the problem of estimating a single pooled risk estimate. Two examples from literature are used to show problems with metaanalysis from published data.