3
$\begingroup$

When we speak of the benefits of randomization, we talk about the balance of covariates so as to mitigate the possible confounding due to selection of treatment. That is, if a design were replicated again and again, each patient would be equally likely to be assigned to treatment or control. Why then do we not assign treatment in a systematic 1-0-1-0-1-... fashion? If we view the order of a participant's entry into the study as random so that their order of entry is equally likely to be even- or odd-numbered, would that effectively solve the same type(s) of biases as unconditional randomization?

$\endgroup$
4
  • 1
    $\begingroup$ Fixed assignment encourages guessing, which is often capable of discovering the actual treatment beforehand. That is why allocation concealment is as important as blinding. $\endgroup$ Commented May 28, 2018 at 8:26
  • 1
    $\begingroup$ @Joe_74 Does guessing actually bias a design? Isn't it inherent that people will try to guess or speculate at what treatment they're on? As long as "controls" and "active arm" participants are both doing guessing, they are otherwise comparable. The setting of a study introduces some issues, but guessing [which treatment you're on] isn't bad. $\endgroup$
    – AdamO
    Commented May 28, 2018 at 19:13
  • $\begingroup$ Guessing might easily bias a study. For instance, imagine that you are comparing a proton pump inhibitor (for gastritis) vs placebo. If the sequence is fixed, a patient might end up adjusting his eating pattern choosing a heavy meal for the day in which the active drug is due, and instead a light meal for the day in which placebo is due, profoundly confounding the eventual analysis. $\endgroup$ Commented May 29, 2018 at 8:33
  • $\begingroup$ @joe_74 that is definitely a consideration for open label designs. It will bias the efficacy estimate, but on terms of effectiveness this "confounding by diet" may actually reflect a drawback of the medication's function as a "crutch for bad behavior". As I said, "guessing" will be done in either arm and with equal precision (50/50) so your example wouldn't bias anything because diet couldn't correlate with actual treatment assignment. $\endgroup$
    – AdamO
    Commented May 29, 2018 at 10:58

1 Answer 1

4
$\begingroup$

If the randomization sequence is discovered, patients could be non randomly assigned to the experimental group of choice. In a truly blinded study, this is unlikely. In many studies, blinding may not be possible, which would expose a simple 1-0-1-0- process. Many investigators attempt to provide care through RCTs and hope for treatment group assignment when, theoretically, equipoise allows conduct of randomized assignment.

$\endgroup$
3
  • 1
    $\begingroup$ Good point. But in a non-blinded study, the participant knows their treatment assignment. So they need not infer it from other participants. In contaminated studies, participants infer each others' assignments even with blinding: that boils down to choice of placebo. The intent of trial coordinators to provide care is a substantial rationale for having a control (SoC) group and single-blinding at the very least. Beyond that, I would say equipoise allows for more than just randomization, it allows for treatment allocation which is inclusive of what I describe. Do you agree with that? $\endgroup$
    – AdamO
    Commented May 25, 2018 at 15:18
  • 1
    $\begingroup$ The coordinators need to be blinded to the randomization process to prevent anticipation and non-random assignment. The patients need to be blinded to normalize the treatment effect or placebo effect as much as possible. Patients do talk in waiting rooms and try to ascertain treatment or control status. Agree that equipoise allows for almost every aspect of the RCT enterprise. Without equipoise, there is no important question to answer. However, equipoise is best thought of as a consensus of ambivalence. Individuals retain strong biases, usually for new treatments. $\endgroup$
    – Todd D
    Commented May 25, 2018 at 15:30
  • $\begingroup$ Your points are excellent. This review from NEJM describes all the points you mention. I understand that the relative merits of randomization include enforcing better adherence to trial protocols. $\endgroup$
    – AdamO
    Commented May 31, 2018 at 15:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.