BlindingGuideline in PDF


Describing the purpose and methods for blinding.


Aim of blinding
The aim of blinding is to prevent information bias or bias in effect size estimates resulting from errors in measuring the variables in the study (determinants and outcomes).
For each study in which groups of participants are being compared, such as clinical trials (randomised or not), it is essential that groups are comparable, and remain comparable. It is therefore also essential that the determinants and outcomes are measured in a comparable way. Information bias may occur through knowledge of the nature of the allocated intervention. For instance: A participant may be more inclined to report no effect, if he/she is aware they are receiving a placebo treatment; this will reduce the placebo effect. The result is an underestimate of the placebo contribution in the intervention, and therefore an overestimate of the effect size. Bias may also occur when the treatment provider for instance has a strong preference for one of the treatments to be compared, which may lead to him painting a rosier picture of the condition of the participants treated. Bias may also arise when the researcher analysing the results has a strong preference for a given outcome (for instance, because a positive result would lead to a more publishable article), which may lead him to interpret the results more favourably, or decide whether or not to present certain results in the article.
A solution to all these potential problems is blinding, that is, masking which participant is receiving which treatment. Blinding can be undertaken at the level of the participant, the treatment provider, the outcome measure (blinding of the individual undertaking the measurements) or the analysis (blinding of the person analysing the data). A study is referred to as “single-blind” when the participant and/or treatment provider are unaware which intervention the participant is receiving, or when the participant and treatment provider are aware of this, but the outcome measure has been set up to be blinded. A study is referred to as “double-blind” when the participant and/or the treatment provider, as well as the outcome measure have been blinded.

N.b.: Randomisation will not prevent information bias. Randomisation is intended to reduce confounding and prevent selection bias.

When should blinding be carried out?
The need for blinding depends on the aim of the study: Blinding is extremely important in (explanatory) trials into the effectiveness of interventions. In pragmatic trials where two interventions are being applied, such as those carried out in daily practice, blinding could lead to results that can not be generalised . When an intervention is ultimately implemented, the patients would not be blind to the intervention.

Sometimes it is not possible to implement blinding fully, or at all. In cluster-randomised trails (where randomisation and intervention take place at the level of the doctor, whilst outcomes are (also) measured and analysed at the level of the participant) it is often impossible to blind the treatment provider to the treatment. The participant may often be kept blind to the treatment by not informing them that this involves an intervention study, and they therefore remain “blind”. Van der Feltz en Adèr [1] provide examples of the disastrous effect of single blinding on participant recruitment in a cluster-randomised study. At trials where there is randomisation at participants’ level, it is often difficult to keep the participant blind to the treatment (with the exception of placebo-controlled trials), particularly in the type of study in which different interventions are being compared (for instance, comparing surgery with conservative treatment, or comparing physiotherapy to treatment by the GP). In this instance it is important to ensure the outcome measure is blinded. If it is not possible to do this (for instance if the primary outcome measure is self-reported quality of life), then it is often possible to carry out the analyses blind.
When blinding is not possible, or can only be implemented partially, the expected direction of bias can usually be discerned. This should be taken into consideration in the conclusions.

How should a study be blinded?
The trick in blinding the participant to the trial lies in making the interventions look the same. In drug trials, blinding is often realised by administering drugs that look the same in terms of shape, colour and taste. In other studies attempts can be made to make the blinding procedure look the same throughout. An example of this is a comparison between injections and a placebo. In the placebo group the participants will receive an injection. However, this won’t contain an active product. Therefore the participant will not know whether or not he received the real injection or a placebo. Blinding of the treatment provider or the individual carrying out the outcome measurement is often relatively simple to implement by not informing the individual concerned which group the participant has been allocated to. The analyses can be blinded by having the grouping variables (indicating the intervention) entered by a different person and by not adding a value label (however, these will need to be thoroughly described by the other individual and stored in order to prevent the information being lost).

Monitoring the blinding procedure
It is difficult to assess how well the blinding process has worked after the study has been completed. Occasionally each participant and/or treatment provider may be asked which group they suspect they were allocated to. A cross-tabulation can then be produced of the reported against the actual group allocation; the chi-square test should not be significant. However, this method is not recommended for the following reason: If the recovery of participants is such that the participants and/or treatment providers are able to see who received the experimental and evidently superior therapy (and therefore often recognise it), then there is not a case of failed blinding, but convincing evidence for the superiority of the therapy.

[1] van der Feltz-Cornelis CM, Adèr HJ. Randomization in psychiatric intervention research in the general practice setting. International Journal of Methods in Psychiatric Research: 2000. 9(3):134-142.

  • Has a blinding procedure been applied to your study?
    1. If not, then why not?
    2. If so:

    • Were the participants and treatment providers blinded?
    • Was the outcome measurement blinded?
    • Was the data analysis blinded?