To the Editor. —
The recent article by Dupont and Page1 provides a muchneeded review of the studies of estrogens and breast cancer. However, meta-analysis of epidemiologic research faces different problems from those faced by the analysis of data from clinical trials. Evaluating the methods of observational studies, to decide which results to include, presents the first obstacle. Random assignment, placebo control subjects, and blinded assessment have become standard elements of a well-conducted clinical trial. In contrast, there is less agreement over what constitutes acceptable methods in epidemiologic research.2 The authors chose to exclude few studies, despite substantial differences among the studies in the selection of subjects and control subjects, the definitions of exposure, and the control of potential confounders. The amount and the direction of bias introduced by these differences is controversial. The authors found no "obvious" connection between methodology and risk estimates. However, bias in individual
Atkins D. Estrogens and Breast Cancer. Arch Intern Med. 1991;151(9):1882. doi:10.1001/archinte.1991.00400090148031
Customize your JAMA Network experience by selecting one or more topics from the list below.