Observational studies analyzed like randomized trials, and vice versa
*Miguel Hernan, Harvard University
Keywords: comparative effectiveness research, randomized trials, observational studies, time-varying confounding
Randomized experiments and observational studies are often used to answer the same causal questions in comparative effectiveness and safety research. For example, the effects of statins and of postmenopausal hormone therapy on the risk of coronary heart disease have been estimated both using randomized clinical trials and observational studies. Often the design of open-label randomized trials and of observational follow-up studies is similar, except that baseline treatment assignment is not randomly assigned in the latter. Because treatment choices and participation decisions after baseline are not randomized in neither randomized trials nor observational follow-up studies, time-varying confounding and selection bias may arise under both designs. Therefore randomized trials are just follow-up studies with baseline randomization. Leaving aside the adjustment for baseline confounders, which is generally necessary in observational follow-up studies, there are no reasons why the analysis of follow-up studies with and without randomization should differ. This talk reviews a framework for the analysis of both randomized trials and observational studies in the presence of time-varying confounding, and presents several applications.
Important Dates & Deadlines
- October 9 - 11, 2013