Meta-analysis of observational epidemiological research; Possibilities and limitations of its application in health risk characterisation

In this report the applicability of meta-analysis is evaluated as a tool for quantitative review of observational (non-experimental) studies involving risk factors and health effects. These quantitative reviews are required for the assessment of the effectiveness of preventive policies that will be performed in the framework of next Dutch Public Health Status Forecast reports. The report describes briefly the 'art' of meta-analysis, recent methodological developments and the objectives of application. Subsequently the limitations and possibilities of meta-analysis of observational studies are evaluated. Based on this evaluation and review of the literature, guidelines are outlined for sound meta-analysis of observational studies. A statistical procedure for data aggregation is presented, providing possibilities for meta-regression analysis, taking relevant study characteristics into account, as well as possibilities for Bayesian updating of individual studies (posterior estimates) 'borrowing strength' from a set of other comparable study results. The report concludes that meta-analysis can be a useful tool for systematic, consistent, comprehensive, and reproducible quantitative review of observational studies. However, the main objective should be critical, comparative analysis, learning from heterogeneity, identifying patterns among studies, not data aggregation in itself.