[Practice of systematic reviews. VII. Pooling of results from observational studies]

Knowledge concerning aetiologic hypotheses can be obtained through systematic reviews of observational studies. In observational studies, heterogeneity between studies is expected, because of differences in study design and analysis. Data from observational studies have to be made comparable before pooling of results from component studies is possible. In this, definition of exposure and disease and adjustment for confounding have to be taken into account. Three methods to deal with heterogeneity are presented: ignoring (fixed effects model), modelling (random effects model) and exploring (meta regression). Any combination of data in meta-analysis of observational studies has to be based on statistical, methodological and clinical considerations. The same considerations are applied for the choice which method is used.