Recent evaluations of the methods of key studies should not change how we advise women A recently published and much publicised paper by Shapiro and colleagues, the last in a series of four, evaluated the effects of hormone therapy on the risk of breast cancer.1 The authors of the four review articles applied epidemiological principles to the findings of two randomised placebo controlled studies from the Women’s Health Initiative (WHI; 27 347 women) and two observational studies—the Collaborative Reanalysis (53 865 women) and the Million Women Study (MWS). Shapiro and colleagues concluded in their fourth paper that the MWS had design defects, that it contained multiple biases, and that its findings were thus not robust enough to show that hormone therapy increased the risk of breast cancer. All observational studies are inherently biased because subjects are not randomly assigned to treatment or control. Adjustment for confounders and careful design of observational studies help to reduce bias. However, because there is no independent variable, such studies can tell us only about association not causation. The MWS was published in the Lancet in August 2003,2 and a flurry of letters was published in a print issue later that year, many …