Blinding

Bias is not only a problem at the time of treatment allocation. It can also occur during the experiment when individuals are fed or treated. If the technicians or nurses know which subjects are receiving which treatments, they may well be biased in the care they provide. As a result the control group may not be treated in exactly the same way as the treatment group. Decisions may also be biased when the outcome of the experiment is determined, and especially when the data are analysed.

These sources of bias are minimized through 'blinding'. In a fully blinded trial, the patient, the medical/veterinary staff and the statistician are all unaware of which treatment group is which until the completion of the trial - including the analysis phase. This is especially important if the assessment of the outcome of treatment is to some extent subjective.

The commonly used terminology varies somewhat between veterinary and medical trials:

• In veterinary clinical trials a single blind trial is one in which only the persons caring for the animals do not know which treatment is which; in a double blind trial both the persons caring, the veterinarian and usually the analyst are kept blind.
• In medical clinical trials a single blind trial is one in which patients do not know which treatment they are receiving, whilst in a 'double blind' trial, neither patient, doctor, nurses nor analyst know which is which.

Confusion over terminology means that one should always specify precisely who was blinded in a trial, and over what period.

Usually the only way one can conduct a blind trial is to give control groups a dummy or placebo treatment which looks identical to the real thing. In the case of drugs, the placebo must be similar in all respects to the drug except that the active ingredient is absent. It is usually recommended to use a capsule presentation for both treatment and placebo as the taste of some drugs is very distinctive and cannot be readily mimicked. Placebo injections are less common, but coloured saline injections have been used. Placebos for surgery or patient management are even more difficult to devise, although one of the examples we give below is an ingenious attempt to do just that.

Once the randomization list has been prepared by the statistician, the pharmacist needs access to the list to make up the required identical packages (drug or placebo). A batch is then made up with the each package bearing a code number which is linked to the randomization list. Usually the code number reflects the order of patients entering the trial. At the end of the trial (or better after analysis of the data), the code is broken to reveal which treatment was which.

An example

The technique of gastric freezing for the treatment of stomach ulcers was first promoted by Dr Wagensteen in the U.S.A. He had tried this method on 24 patients and all had apparently been cured. However no properly designed randomized trials had been carried out.

Ruffin et al. (1963) wanted to carry out a double blind trial to test the efficacy of the treatment. However, they had first to find an effective placebo - that is a means to make the patients think they had been given the treatment when it fact they had not. To achieve this, each control patient was given the same operation as the treatment patients, except that a bypass was put in to return the refrigerant liquid before it could freeze the stomach. Only the medical statistician knew which patients received treatment and which did not - since neither patients nor the doctor knew, it was a double blind trial.

It was found that most patients in both the treatment and control groups were improved or symptom free during the first six weeks, but as time passed their condition worsened again. At no time was there any significant difference between the two groups. The experiment therefore showed that the freezing treatment was no better than the placebo for the treatment of duodenal ulcers.

How successful is blinding

Although nowadays the majority of clinical trials are reported as being double-blind randomized controlled trials, one should perhaps be a little cautious about accepting this assertion. Fergusson et al. (2004) looked at how successful double blinding was in a sample of randomized, placebo controlled trials from leading general medicine and psychiatry journals. Unfortunately only 15 of the 191 (7.9%) trials reported on whether or not blinding had been successful in either the participants, the outcome assessors or the investigators.

Of the 15 trials, blinding was only reported as being successful in five of them, and in three of these no quantitative data were presented to support this claim. Only four trials assessed blinding in both the participants and either the outcome assessors or the investigators. Clearly we have little evidence that blinding is actually being successfully maintained in placebo controlled trials, and this should always be borne in mind when assessing results.