In prevention, I’ve noticed something pretty predictable happens when a new study comes out noting the health harms of certain lifestyle factors (alcohol, obesity, and red meat intake are particularly prone to this) – the internet lights up with someone posting (and plenty of others linking) about why we can’t trust anything that comes out of observational studies.
I am not going to sit here and pretend that there aren’t limitations to observational studies – there are and I know them well – that’s what the foundation courses in epidemiology teach you – the limits of each study design.
But let’s be perfectly clear about something else, randomized trials aren’t perfect either. They aren’t perfect when perfectly conducted and almost none are perfectly conducted. They can’t be. Randomized trials are done on people, who are imperfect. So that means people don’t comply with randomization and they quit the study.
This isn’t news. If we thought the (THE) randomized trial was perfect, we’d only do one. So we wouldn’t have need for meta-analyses of randomized trials, which proliferate in the literature.
Randomized trials are subject to interpretation. One need only to compare how the same set of studies can be interpreted quite differently by the thoracic surgeons and gastroenterologists when each group issues treatment guidelines.
Does this mean randomized trials are no good? Absolutely not! In fact I recently wrote a paper with my colleague Graham Colditz about the importance of intervention trials specifically in cancer survivors, though we also noted the limits to what such trials can show.
We need to remember that we can’t (and never will) base all of our health and medical decision making on trial data. I will continue to trust in the value of parachutes, despite the alarming findings of this analysis of trial data from the esteemed British Medical Journal.
My colleagues at Cancer News In Context & Washington University School of Medicine have put together a nice graphic on the pros and cons of observational versus randomized trial data. I think it is a nice summary of how BOTH study designs add value to our understanding of health. Which one is “best” is VERY dependent on the specific question of interest!