Blinding: taking a better look at the blind side
Posted on 29th March 2018 by Neelam Khan
Blinding is a common element used in rigorously designed trials. Most people are familiar with the general concept but what is its purpose and what is the best way to perform it? This blog explores both of these questions and discusses ways to tackle situations where blinding cannot be done.
What is blinding?
In randomised controlled trials (RCTs), there may be 2 or more groups receiving different treatments (possibly including a placebo group). Blinding refers to the process of concealing participants’ group allocations. This may be in the form of single-blinding, where only the participants are unaware of which group they have been assigned to, or double-blinding, where neither the participants nor the researchers know which subjects are receiving the active treatment or placebo.
Why should blinding be used?
The purpose of blinding is to minimise bias. But is this not already achieved through randomisation?
Random assignment of participants to the different groups only helps to eliminate confounding variables present at the time of randomisation, thereby reducing selection bias. It does not, however, prevent differences from developing between the groups afterwards. This is where blinding plays a role.
First of all, blinding can help to preclude the occurrence of differential co-interventions. For example, if the researchers have not been blinded, they may (subconsciously) devote extra time, attention or even apply additional diagnostic and therapeutic procedures to the group receiving the active treatment. This will, in turn, impact the results of the study. Equally, non-blinded participants who are receiving a placebo may actively seek out other treatments for themselves, which may influence the study findings. Employing the use of blinding can prevent this from happening.
Another key reason for blinding is to prevent biased assessment of outcome. When evaluating subjective outcome measures, knowledge of participants’ group allocations may influence non-blinded researchers, thereby incorrectly producing results that are more favourable of the treatment in question. Likewise, participants who are aware of their group allocations may also influence the findings, especially self-reported outcome measures. For instance, participants may selectively report information or change their behaviour to match what they believe the researchers are expecting to see. This is an experimental artefact known as demand characteristics.
In research, demand characteristics refers to an experimental artefact where participants form an interpretation of the experiment’s purpose and unconsciously change their behaviour to fit that interpretation (Wikipedia definition)
Even if objective outcome measures are used, bias may still be introduced by non-blinded data analysts selectively using and reporting statistical tests that make the treatment appear more effective than it is. If blinding is utilised throughout the whole trial, the risk of such biases affecting the results is considerably lower.
Indeed, a systematic review of 250 RCTs in the peer-reviewed Journal of the American Medical Association provides empirical evidence to support the use of blinding in RCTs. Schulz et al. report that trials that were not double-blind produced larger estimates of effects (p=0.01), with odds ratios being exaggerated by 17%, compared to blinded trials. This highlights the risk of bias present in non-blinded trials and serves a caution to readers of trial reports to be wary of such design flaws in RCTs.
What is effective blinding?
As with anything, blinding must be applied properly if it is to be maximally effective. Hence, certain criteria should be fulfilled:
- Group allocation must be concealed successfully
- The ability to accurately measure outcomes must not be hampered
- Blinding should be applied to both participants and researchers – this includes clinicians, data collectors, outcome adjudicators and data analysts (sometimes multiple roles are performed by a single group of people)
What can be done when blinding is not possible?
Of course, we do not live in an ideal world so there are situations where blinding cannot be applied because it would be unethical, or it is simply not possible for practical reasons. For example, if the intervention being investigated is a form of surgery, it would unethical to perform ‘fake’ surgery on the control group and impossible to prevent the clinician from knowing which subjects will be undergoing surgery!
Nevertheless, certain measures can be taken to minimise any potential bias. If it is impossible to blind all of the aforementioned team members, then one should blind as many of them as possible. Returning to the example of a surgical intervention, although it is not possible to blind the operator, independent outcome assessors who are unaware of participants’ group allocations may be employed. The risk of bias would be further reduced if they assessed objective outcome measures and avoided subjective measures as much as possible. Moreover, blinding should be applied until after the statistical analyses are complete. One way to ensure this is to simply assign non-identifying labels to each group e.g. group X and Y.
Summary
Blinding is a vital design feature of RCTs and is as important as randomisation in terms of minimising bias. Notwithstanding all the alternative measures applied to make up for the lack of blinding, it is imperative that researchers acknowledge this particular limitation of their study and discuss the biases that may arise as a result in the published report.
