Observation versus intervention in the evaluation of drugs: the story of hormone replacement therapy.

Hormone replacement therapy, which was approved for menopausal symptoms, offers an opportunity to compare clinical trials and observational studies when evaluating the risks and benefits of drugs. The differences between randomized and observational evidence relate mainly to the risks of coronary heart diseases and dementia, higher or not elevated in users in trials, and decreased in observational studies. The most likely explanation for these discrepancies is bad accounting for confounders, in particular, time-dependent confounders in classical multivariate analyses and use of prevalent user design. Marginal structural models and new user design should help to diminish strongly indication bias in future observational studies aiming at the evaluation of the risks and benefits of drugs.