Randomisation methods: concealment

Randomisation is the best method removing selection bias between two groups of patients.¹ However, the process of randomisation can be compromised such that the allocation results in biased groups of patients. A trial which has had its randomisation compromised may apparently show a treatment effect that is entirely due to biased allocation. The results of such a study are more damaging than an explicitly unrandomised study, as bias in the latter is acknowledged and the statistical analysis and subsequent interpretation takes this into account. Changes in clinical management based on a compromised trial may, at best, waste valuable health care resources on a useless treatment; at worst, they may also damage patients’ health. The randomisation process must therefore not be compromised.

In the past attempts were not generally made to conceal randomisation schedules from investigators who recruited patients. However, unconcealed randomisation can lead to clinicians scheduling patients such that patients with particular characteristics would receive a certain allocation, thereby biasing the allocation.² Because of this, administration of randomisation sequences was changed, and forms of concealment were introduced.

Perhaps the most common is the sealed envelope system. In this participating clinicians are given randomly generated treatment allocations within sealed opaque envelopes. Once a patient has consented to enter a trial an envelope is opened and the patient is then offered the allocated treatment regimen. However, this process is open to deliberate tampering as the investigator can open several envelopes before a clinic and then allocate patients to the desired treatment. Indeed, sometimes treatment allocation can be seen if the envelope is held against very bright light.³ Even if randomisation envelopes and allocation are sequentially numbered to detect any attempt to allocate a patient out of sequence, the process can still be compromised. For instance if the clinician knows the next allocation—for example, by opening an envelope in advance—he or she may postpone trial recruitment until a patient with certain characteristics presents, thus preferentially recruiting patients with certain characteristics into a given treatment arm. While the problem is most serious when interventions are unblinded, even when a drug trial is double blind treatment allocation has been compromised by investigators identifying drugs through poor labelling, or accessing unsecured codes which describe the codes of the active and inert drugs.³ Furthermore, treatment allocation can be guessed if blocking is used. For instance if patients are randomised in a series of blocks of four—that is, for every four patients randomised two will receive one treatment and two will receive the other—an investigator who remembers the treatments the previous three patients received will be able to predict the treatment for the fourth.

While much of the evidence on subverting randomisation is anecdotal, a recent review found that randomisation has been compromised in several controlled trials.² This review showed that trials which did not adequately conceal randomisation from the investigators demonstrated, on average, a 41% increase in effect for the active treatment compared with an adequately concealed trial.² Indeed, in a current multicentre randomised trial of a surgical procedure in the United Kingdom the median age of patients for the experimental treatment was found to be significantly lower for three groups of clinicians when an envelope system was used. This age imbalance disappeared when better concealment measures were introduced.⁴

Owing to the problems of using envelopes it is methodologically more sound to undertake “distance” randomisation (although in some instances sealed envelopes may be the only practical means of randomisation). Distance randomisation usually involves the investigator, on recruiting a patient, telephoning a central randomisation service which notes basic patient details and then issues a treatment allocation. Indeed, distance randomisation can now be performed over the internet. Such a system is being used, alongside telephone randomisation, in the Medical Research Council’s growth restricted intervention trial (GRIT). Distance randomisation is much less likely to be compromised than an envelope system.

Thus, to avoid bias it is important that randomisation is well concealed. Recent evidence has questioned the rigor of using local randomisation. Randomisation should be distant and separate from clinicians conducting the trial.