Double-Blind Procedure

Experimental Control

“Double-blinding” is intimately coupled to randomization, where participants in an experimental study are allocated to groups according to a random algorithm. Participants and experimenters are then blinded to group allocation. Hence double-blinding is an additional control element in experimental studies. If only some aspect of a study is blinded, it is a single-blind study. This is the case when the measurement of an outcome parameter is done by someone who does not know which group a participant belongs to and what hypotheses and expectations are being tested. This could, in principle, also be done in nonexperimental studies if, for instance, two naturally occurring cohorts, smokers and nonsmokers, say, are tested for some objective marker, such as intelligence or plasma level of hormones. Double-blinding presupposes that participants are allocated to the experimental procedure and control procedure at random. Hence, by definition, natural groups or cohorts cannot be subject to double-blinding. Double-blind testing is a standard for all pharmaceutical substances, such as drugs, but should be implemented whenever possible in all designs. In order for a study to succeed with double-blinding, a control intervention uses a placebo that can be manufactured in a way that makes the placebo indistinguishable from the treatment.

Allocation Concealment and Blind Analysis

There are two corollaries to double-blinding: allocation concealment and blind statistical analysis. If an allocation algorithm, that is, the process of allocating participants to experimental groups, is completely random, then, by definition, the allocation of participants to groups is concealed. If someone were to allocate participants to groups in an alternating fashion, then the allocation would not be concealed. The reason is that if someone were to be unblinded, because of an adverse event, say, then whoever knew about the allocation system could trace back and forth from this participant and find out about the group allocation of the other participants.

Double-blind studies are normally also evaluated “blind.” Here, the data are input by automatic means (Internet, scanning), or by assistants blind to group allocation of participants. Whatever procedures are done to prepare the database for analysis, such as transformations, imputations of missing values, and deletion of outliers, is done without knowledge of group assignment. Normally a study protocol stipulates the final statistical analysis in advance. This analysis is then run with a database that is still blinded in the sense that the groups are named “A” and “B.” Only after this first and definitive analysis has been conducted and documented is the blind broken.

Good clinical trials also test whether the blinding was compromised during the trial. If, for instance, a substance or intervention has many and characteristic side effects, then patients or clinicians can often guess whether someone was allocated to treatment (often also called verum, from the Latin word for true) or placebo. To test for the integrity of the blinding procedure, either all participants or a random sample of them are asked, before the blind is broken, what group they think they had been allocated to. In a good, uncompromised blinded trial, there will be a near-random answer pattern because some patients will have improved under treatment and some under control.

...