Meta-analysis of environmental and occupational epidemiological studies: a method demonstrated using the carcinogenicity of PCBs as an example.

A method for a meta-analysis of several environmental or occupational epidemiological studies with small prevalences and/or incidences and long latency periods is presented. A combination of statistical evaluations of small prevalences should be done in the following way: (1) selecting relevant and comparable studies, (2) computing exact p-values of Poisson tests and aggregating them by Fisher's method, (3) estimating overall relative risks (or SMRs) by pooling the data of the individual studies and by using weighted sums of the logarithms of the individual risks, and (4) calculating confidence intervals and confidence curves for the overall risks. An example illustrates this technique by investigating the association of exposure to polychlorinated biphenyls (PCBs) at work and cancer mortality in men.