In many situations, randomized controlled trials are infeasible and one must draw conclusions from observational data. Certain quasi-experimental designs – for example, interrupted time series analyses – strengthen the conclusions that can be drawn from observational data. However, particularly when the intervention evaluated is important, either clinically or from a health policy perspective, implied or explicitly stated conclusions about causality often give rise to great controversy.
This month’s issue of Medical Care features an exchange between Dr. Marc Stone (and here) of the Food and Drug Administration and Professor Christine Lu and colleagues (and here) from Harvard University. The exchange focuses on the impact of the FDA boxed warnings and subsequent media attention on antidepressant treatment of youth and whether there was a resulting increase in youth suicide attempts. The exchange highlights the types of the issues that arise when attempting to draw causal conclusions from observational data:
- the appropriateness of different databases and research designs for different purposes;
- the validity of proxy measures;
- competing hypotheses to explain observed trends; and
- the strengths and weaknesses of interrupted time series designs to identify potential causal relationships.
We have created this moderated forum to provide a place for continued discussion of the issues raised by both Dr. Stone and Professor Lu and colleagues. We encourage all commenters to be civil and respectful. Our hope is that this forum will allow for meaningful exchanges among researchers and other stakeholders in this debate.