Invited commentary: how far can epidemiologists get with statistical adjustment?

In 2002, the Women’s Health Initiative (WHI) clinical trial reported that combined estrogen-plus-progestin hormone therapy did not prevent coronary heart disease in women (1). Combined estrogen-plus-progestin therapy increased the risk of stroke by a factor of 1.4 on average and doubled the risk of venous thromboembolism. Observational research up to the time of the WHI suggested that the relative risk of coronary heart disease was 0.50–0.65 in hormone users compared with nonusers (2, 3). Based on observational research, the relative risk of stroke was considered less than or near 1.0 for hormone therapy (2). Both observational studies (4) and a randomized trial, the Heart Estrogen/progestin Replacement Study (HERS) (5), showed a relative risk of 1–2 for venous thromboembolism in hormone users. The WHI clinical trial joined a body of experimental evidence showing no effect of postmenopausal hormone therapy on coronary heart disease clinical endpoints or measures of subclinical coronary atherosclerosis (6–16). In 1998, the Heart Estrogen/progestin Replacement Study reported no beneficial effect of hormone therapy on morbidity or mortality from coronary heart disease in women with established coronary disease (6). Seven other randomized trials of the effect of hormone therapy on arterial disease endpoints published in 2000–2002 found no effect of hormone therapy (7–13). A 1997 meta-analysis of small randomized trials of hormone replacement found that hormone therapy did not prevent cardiovascular disease (14). Follow-up of participants in the Heart Estrogen/progestin Replacement Study (15) showed no benefit of hormone therapy for any cardiovascular endpoint. The estrogen-only arm of the WHI clinical trial was terminated early after showing that hormones did not prevent coronary heart disease (16). In this issue, Prentice et al. (17) report an analysis of data from the WHI clinical trial and the WHI observational study. The WHI clinical trial and observational study were conducted by the same group of researchers, in the same time frame, and had similar procedures for ascertaining and monitoring events. Prentice et al. explore the reasons for the differences in estimates of coronary heart disease, stroke, and venous thromboembolism risk for hormones between the WHI clinical trial and the observational study. They attempt to resolve the discrepancies by statistical adjustment.