StATS: The ethics of randomization (January 14, 2005).

A recently published article

attacks the randomized trial and declares it to be

a deficient research tool both on deontologic and methodologic grounds.

 A response, published in the same issue,

is also worth reading.

Dr. Retsas tackles the difficult question of equipoise by asking

If I am genuinely uncertain about the value of a new or established treatment, does this also apply to my colleague down the road? My colleague, perhaps with greater experience or different perceptions than my own, may have a greater degree of certainty about the value or otherwise of the treatment in question. Should then this colleague be obliged to subject his or her patients to randomization to clarify my own uncertainties?

There are two differing definitions of levels of equipoise. The first states that equipoise is genuine uncertainty by the individual physician as to which therapy is better. The second states that equipoise is genuine uncertainty in the community of practicing physicians. In my opinion, both forms of equipoise need to be factored into the equation. Physicians who apply individual equipoise only place themselves in the position that they know what is best for their patients, regardless of what their colleagues might think. Such physicians need some open minded inquiry as to why others have a different viewpoint. On the other hand, physicians should not abandon their knowledge and expertise and follow the beliefs of others blindly.

But Dr. Retsas appears to be confusing two different questions: is it ethical to conduct a randomized trial versus is it ethical to compel others to conduct a randomized trial? This is actually a serious issue, because some have advocated that unproven treatments should be made available only in the context of randomized trials.

Dr. Retsas also argues that randomized trials are unnecessary because well-designed observational studies are just as good. He cites two studies

to support his case but does acknowledge that others disagree. He then poses the question

Is the information provided by a randomized trial of 1,000 patients more reliable than that from 10 observational studies, each enlisting 100 patients?

and then claims (without any apparent justification) that

If the 10 observational studies report response rates between 5% and 20% with acceptable toxicity, the true activity of the new drug or treatment lies somewhere in between.

If each of the studies replicates the same source of bias, then it is very possible that the true activity of the drug could be much less than 5% or much more than 20%. If, on the other hand, the observational studies are designed in such a way that sources of bias are deliberately and purposively varied, then perhaps you can do better. A good description of how to deliberately and purposively vary sources of bias appears in

He also questions the law of large numbers, in effect, by asking

However, how many unknown factors can randomization of a cohort of 500 patients accommodate—one, 20, or an infinite number?

It turns out that a sample size of 40 or greater will provide reasonable protection against covariate imbalance, as was shown by

Although I disagree with the general conclusions of Dr. Retsas, he does raise some very interesting issues.

This page was written by Steve Simon while working at Children's Mercy Hospital. Although I do not hold the copyright for this material, I am reproducing it here as a service, as it is no longer available on the Children's Mercy Hospital website. Need more information? I have a page with general help resources. You can also browse for pages similar to this one at Category: Equipoise in research or Category: Randomization in research.