How not to randomize a study

If you are designing a randomized study, make sure you actualize randomize your patients. Schiffl messed this up in his study of daily versus three days a week dialysis, for acute renal failure. Schiffl achieved randomization by alternating eligible patients to three days a week versus daily dialysis. One key aspect about randomization, especially in non-blinded studies, is that the investigators cannot know what arm of the study the patient will be in prior to enrolling the patient. With alternating patients every investigator knows which arm the next patient will end-up in and they can make subtle decisions on the appropriateness of the patient or in how they present the consent form to influence the composition of the study arms.

This comes up, because my fellow sent me a paper on prophylactic dialysis prior to CABG (PDF). From the paper comes this gem:

Repeat after me, “If you know what arm the patient will be in prior to enrolling the patient, you are not running a randomized trial.”

2 Replies to “How not to randomize a study”

  1. I see your point in the “every other one” example but you could randomly assign the medical record numbers as 3d or 6d right? Seems a pretty straight forward way to mask prior knowledge. Does the researcher even have to have access to the medical record number in order to enroll someone?

  2. Bill, I’m not sure I understand your reference to 3d and 6d but my point is there are lots of aspects of doing randomized controlled trials which are difficult. The actual randomizing of patients is not one of these aspects. Don’t let years of work be vulnerable to criticism for lack of attention to this important detail.

Comments are closed.