Elsewhere in this issue Ross et al (p 1635) report on a retrospective study apparently linking menopausal use of estrogen with an excess risk of breast cancer. In fact, the overall results do not show a significant difference in estrogen use between cases and controls. The significant findings are limited to the subgroup with intact ovaries. For those with surgical menopause, there was a slight apparent (not significant) reduction in risk associated with estrogen use.
In common with other case-control study groups, Ross et al are burdened with the nonexperimental and nonprospective nature of their data. They are further burdened by high rates of refusal. (The review of pharmacy records of those subjects who declined participation raises privacy questions that are not addressed in the report.) The methodology of case-control studies is complex, and there are alternative choices possible for analysis. For example, it may be decided to eliminate from