Single-centre trials in which randomisation is conducted on site and the randomisation method is known to the participating investigators are at high risk of this problem. If an investigator has influence over the number of patients enrolling in a trial, or the order in which they are randomised, he or she has the potential to introduce selection bias into the study by directing particular patients into preferred treatment groups while excluding others. However, even if the investigator knows the allocation sequence, the problem will not arise if all consecutive eligible patients are being enrolled in order of presentation, as is often the case in trials in emergency departments, intensive care units or any acute care setting. Ideally, a successful allocation concealment process is ensured when all investigators are ignorant of future treatment allocations and have no control over the order of patients randomised into the trial.

Patients may change their willingness to participate in a trial if they know or suspect which treatment they will receive. If patients are aware of their allocation before randomisation, they may be influenced to withdraw before randomisation or wait until their preferred treatment is available before entering the study.

Baseline imbalances as a direct result of concealment violation may not be evident across the whole trial, although patients from particular sites or patients from particular investigators may show consistent differences between treatment groups in various baseline prognostic factors. In practice, the method of allocation concealment is often not reported or is poorly described.2-7 Any trial report should provide enough detail to describe the quality of both allocation concealment and blinding strategies. The use of sequentially numbered, opaque, sealed envelopes, pharmacy-controlled allocations, coded identical containers or kits, and central randomisation systems (telephone or web based) are considered adequate concealment methods and are often implemented to protect the randomisation.8 Allocation lists that can be prepared in advance or may be guessed because of a known pattern (eg, permuted blocks of fixed size) are more susceptible to deciphering. No strategy is entirely tamper-proof, although remote systems are generally more secure.9

If patients are not blinded to their allocated treatment, then knowledge of treatment may influence their responses to the intervention and their reporting. Commonly, patients assume that the new intervention will be more beneficial than the control or standard treatment. Compared with clinical event outcomes, patient-rated outcomes (for example, quality of life, pain and discomfort) are particularly sensitive to patients’ knowledge of the intervention to which they have been allocated. Similarly, if investigators are aware of the patients’ study treatment, their knowledge may influence, first, their management of the patient and, second, their classification of responses and events.

For example, in the Aspirin Myocardial Infarction Study,11 men and women who experienced a myocardial infarction were randomised to receive 1 g aspirin a day or matching placebo. If they or their doctors had been aware that they were taking aspirin, they might have ascribed all their gastrointestinal symptoms to aspirin. Symptoms suggesting peptic ulcer, gastritis or erosion of gastric mucosa occurred in 23.7% of the aspirin group but also 14.9% of the placebo group. Because the study was well blinded, it was possible to attribute only 8.8% of the symptoms to the aspirin treatment.

Generally, drug trials involving oral, topical or intravenous administration can be set up as a double-blind study by the use of a matching placebo. Many surgical trials comparing intraoperative techniques with an identical external incision can be designed as single-blind trials, but comparisons between medical and surgical interventions are inevitably unblinded — that is, open-label trials. When a treatment has a distinctive side effect which is likely to be expressed in most patients (such as toxic effects from chemotherapy), the use of a placebo may be futile.

When routine collection of clinical data during a trial (eg, blood tests) may potentially unblind investigators, providing investigators with summary information should be considered. For example, in a long-term cardiovascular trial assessing the effect of cholesterol-lowering treatment on the incidence of cardiovascular events, knowing individual lipid profiles during follow-up may unblind the clinician and patient. If the blood tests are performed by a central laboratory, it may be possible to provide a summary report simply stating that a patient’s values do or do not fall within a prespecified range, without compromising the patient’s safety, masking the individual data and preserving the blinded status of the investigator.

Blinded assessment of outcomes is possible in most trials, regardless of whether the clinician or the patient is aware of the treatment allocation. Clinical outcomes should be assessed by people who are not aware of the patient’s treatment allocation (and preferably not involved in the patient’s clinical management) so that all patients are assessed identically. In this way, outcomes will not be influenced by beliefs about the study treatments being compared. In general, blinding becomes less important for reducing observer or information bias as the outcomes become less subjective and more objective (that is, bias is more prevalent with a subjective outcome such as degree of pain or extent of depression, but is eliminated with an outcome such as death from any cause). For outcomes such as biochemical or pathology result markers, assessment by calibrated, accurate instruments is often sufficient to ensure no bias is present. If patients are required to provide their own assessment of an outcome, it is recommended that more than one item of the same type be reported to allow reliability to be measured.

An open-label trial with successful allocation concealment and blinded assessment may provide more reliable and more valid results than a double-blind trial with unsuccessful allocation concealment or compromised outcome assessment. Blinding does not guarantee an absence of bias, although empirical evidence does endorse a reduction of bias when adequate blinding strategies have been implemented. Allocation concealment (before randomisation) is thought to have a stronger influence on the reduction of bias than blinding (after randomisation).3,5 A checklist for successful allocation concealment and blinding is provided in Box 3.