Abstract
Blinding is important in the conduct of clinical trials, yet it generally receives less attention and effort than other trial components (e.g., randomization, compliance) that are rigorously treated in the design and/or analysis stages. Furthermore, although the word “blind” commonly appears in the titles of publications, its use is not always well justified. We are human beings, and our behavior or decisions can change depending on whether our eyes are open or closed. That is why virtually everyone in the clinical trial community, including subjects, investigators, treating practitioners, and outcome assessors, would agree that some form of blinding is needed to reach a fair and objective conclusion uninfluenced by knowledge of intervention assignment. Despite our best efforts, however, blinding may not always be successful. There is a considerable body of literature about blinding and the need to assess whether it has been properly implemented and maintained. By discussing how the trialist and the team may handle blinding-related issues in different stages of a clinical trial, this brief article intends to offer reasonable suggestions that could improve current practice, helping researchers deal confidently with blinding-related issues and research in a manner that meets rigorous standards, but is practical and flexible.
Introduction
Blinding
Definition
Concisely speaking, blinding means that group assignments (investigational intervention or control) are concealed from subjects and members of the clinical trial team after the study begins. In contrast, allocation concealment, although often used interchangeably with blinding, means that group assignments are concealed before and until the trial begins. “Masking” and “blinding” are synonymous, although the word masking may sometimes be preferable (e.g., in an ophthalmologic trial that includes blindness as an outcome). 10 –12
Design and implementation
The team should consider the relevance, feasibility, and logistics of blinding from the concept or design stage and describe its implementation in the protocol. This includes how to make the competing interventions as indistinguishable as possible (e.g., using the same shape, color, and mode of delivery), a feature whose validity needs to be tested in a form of pilot or separate validity study with an independent group before the trial begins. Sometimes, however, blinding is unfeasible because the group assignment cannot be disguised. In dental research, for example, the two interventions may be clearly distinguishable (e.g., mechanical treatment versus ointment).
A general consensus is that complete blinding is preferable to partial blinding and partial blinding is preferable to a total absence of blinding. If a trial cannot be operated in a fully blinded manner, at least some parties (e.g., outcome assessors) should be blinded. Even if the team chooses not to collect blinding data for whatever reasons (see below “Data collection”), it is still desirable to blind relevant parties, whenever applicable and possible. It is important to note here that investigators and their communities are often a major source of bias. Even for honest investigators, cognitive bias is unavoidable. 13
Blinding methods and strategies should be reported in the main publication (in the text and/or a supplemental document), as well as discussed in the protocol for all of fully or partially blinded studies. Moreover, blinding should be treated properly in informed consent forms. 14,15 As recommended in the 2010 CONSORT statement, it is important to specify which parties (e.g., subjects, treating practitioners, study coordinators, outcome assessors, and/or statisticians) are blinded, rather than resorting to general and vague terms such as “single,” “double,” and “triple blind,” which are popularly used in practice. 16 –18
Data collection
Because “to be blinded” and “were blinded” are not the same thing, data collection to evaluate the implementation and maintenance of blinding could be generally useful. Therefore, the protocol should also describe how many times, when, and from whom data will be collected to understand the pattern and success of blinding if the team decides to collect blinding data.
There is no full consensus on when to conduct these assessments and on the optimal number, and various recommendations are available in the literature. 19 –22 The strategy will vary depending on the type and aim of the trial and on how influential blinding could be on the outcome (e.g., a hard endpoint, such as death or heart attack, versus a subjective endpoint, such as pain or taste). For instance, the team might decide to collect blinding data two times from subjects: shortly after randomization (at the clinic) and at the end of the trial (e.g., by phone). Bear in mind, however, that blinding assessment in an early stage of the trial tends to reflect a person's preference/wish, while a later assessment can be confounded with treatment effect (including side-effects).
Different methods of data collection have been used. 23 –26 A common method is to ask subjects “Which treatment do you think you received (or were assigned to)?” and to give them the following choices: (1) A; (2) B; and (3) Don't know, or (1) Strongly A; (2) Somewhat A; (3) Don't know; (4) Somewhat B; and (5) Strongly B. In any situation, the objective is to elicit an honest response, but not stimulate undue curiosity about the treatment identity; at times, asking too often or via a face-to-face interview may make respondents less comfortable about telling the truth or prompt efforts to break the blindness. A sample blinding assessment protocol has been proposed, but each team may develop a protocol better suited to the individual trial. 4
Data analysis
Data analysis itself, or the use of a particular analytic method, is not mandated. Let it be assumed that the team collected the blinding data as described above. The data may then be summarized in a 2×3 or 2×5 table format. Among various alternatives, specialized methods, such as a blinding index, have been proposed and increasingly employed. 24,27 –30
Like many other statistical operations, using a method correctly and interpreting the results in a valid manner are key tasks. Also, whether the team decides to use a blinding index or other analytic method, or to report the data without any analysis, some qualitative considerations about the context and underlying reasons for a breach of blinding (e.g., how/why this happened, is this problematic, alarming or just a random occurrence?) should be made. In addition, secondary analyses taking blinding into account could be conducted. These may include exploratory subgroup analyses or stratified analyses based on correctness of guess, regression adjustment, or sensitivity analyses 31,32 or more advanced and rigorous analyses that have been recently developed. 33
Reporting and Take-Home Messages
It is critical to understand that unsuccessful blinding does not necessarily invalidate the results of a trial. For example, loss of blinding may have no or minimal influence on the outcomes analyses, particularly for hard endpoints, 9 and correct guess due to clear therapeutic effect may not be a bad thing at all. Moreover, different analytic/statistical methods tend to address slightly different problems, entailing different assumptions. Accepting these facts would make trialists less hesitant to collect, analyze, and report data on blinding or less tempted to do selective reporting. At the same time, however, a loss of blinding, or a correct guess about treatment allocation (e.g., due to a dramatic or no therapeutic effect), may bias a trial by influencing the participants' attitudes and behaviors, systematically affecting compliance, retention, and/or ascertainment of endpoints or other factors. To illustrate, in a multiple sclerosis trial of active treatment versus placebo, all subjects were examined by a blinded and an unblinded neurologist. In interim assessments, the unblinded—but not the blinded—neurologists reported a benefit with active treatment over an identically designed placebo regimen, an erroneous observation given the outcome of the trial. Thus, unblinding appeared to introduce ascertainment bias—a systematic distortion of the results that can occur when the person assessing the outcome (whether an investigator or the subject receiving the intervention) knows the group assignment. 33
Because unblinding can lead to varying forms of bias, it would be wise to do something on this issue rather than ignoring or being quiet, wishing this specific aspect of the trial unnoticed. 1,2 But blinding research, itself, is inherently subjective and could be complicated. When querying participants about treatment assignment, for example, it generally cannot be known if a response is truthful or if it represents a personal wish. This means that the data generated would usually reflect correct versus incorrect guesses, but not blinding versus unblinding per se.
As with other scientific disciplines, the role of “common sense” should also be highlighted in blinding research. For instance, if I (as trialist) test a treatment I developed, I would be greatly tempted to show that my treatment is better than the competing one. If I (as patient) know that I am on placebo and I am busy, I think I would be more likely to drop out if I know I am on placebo. Also I am not quite sure if I will pick my favorite wine with my eyes closed.
Conclusions
In summary, collecting and reporting data or other forms of empirical evidence are important if one wants to call a trial “blinded.” Also, investigators who learn that a specific treatment is difficult to blind, or who identify a good or better way to blind it, should be willing to share this information with others in the field. 7,34 While quantitative data and statistics on blinding are crucial, they can also create doubt, confusion, and reluctance to engage in further research unless they are interpreted in a valid manner and meaningful ways are sought to explain how or why unblinding occurred. If many subjects or team members are able to discern an intervention assignment easily, there must be reason(s)—good or bad—and researchers should try to explain this phenomenon.
Blinding research, while complex and imperfect, is an essential part of the clinical trial enterprise. Investigators as well as readers must find ways of addressing its practical challenges so that blinding research will be given its rightful place in the field of clinical investigation, and must try to understand the published results more objectively.
Footnotes
Disclosure Statement
The authors declare no competing interests. This article was not financially supported by any institution or commercial entity.