Na IBMI bo 25. novembra 2008 predavanje dr. Michela Baccini iz Univerze v Firencah (Department of Statistics "G. Parenti", University of Florence and Biostatistics Unit, ISPO) .
Meta-analysis in environmental epidemiology
Statistical methods for meta-analysis can be used to combine and compare results from independent studies designed to make inference on the same scientific issue. In environmental epidemiology research, meta-analysis is commonly employed to estimate overall measures of effect or to identify inconsistency among results arising from different locations within multi-centre studies. Examples concern the study of the short term effect of urban air pollution on mortality and morbidity and the effect of warm and cold temperatures on health.
In this context, because of regional, demographic and environmental differences among locations, the assumption of homogeneity of effect measures across trials is usually inappropriate. In the presence of heterogeneity, random effects meta-analysis can be used, but the interpretation of the combined effect estimate is not straightforward and its internal and external validity falls.
This is a relevant problem when meta-analytic results are used for health impact assessment purposes. Use of shrunken estimates, which are a compromise between overall and location-specific estimates, is recommended in these situations.