Why Must Nobody Know Who Got the Treatment? Removing What the Answer Depends On

A trial design is a set of defences against specific errors and knowing which defence handles which error clarifies the whole thing. Comparing against a control group removes the error of attributing to the treatment whatever would have happened anyway, including natural recovery and the tendency of extreme measurements to be less extreme on repeat. Assigning participants to groups at random removes the error of the groups differing systematically, since any difference the researchers know about and any they do not is distributed by chance. Concealing the assignment from participants removes the effect of their expectations on what they report and on what they do. Concealing it from researchers removes the effect of their expectations on how they treat, assess and record.

Unblinded assessment goes wrong in documented ways:

• •Participants who know they received treatment report more improvement in subjective measures
• •Participants who know they received a placebo drop out more often, which distorts the sample
• •Assessors who know the assignment score ambiguous findings in the expected direction
• •Clinicians who know the assignment provide different additional care
• •Researchers who know the assignment make different decisions about excluding data
• •Comparisons of blinded and unblinded studies of the same question find larger effects in the unblinded ones

Many interventions cannot be concealed and the responses are partial. Surgery, physiotherapy, psychotherapy and changes to diet or behaviour are all obvious to the participant, which rules out blinding them. Sham procedures exist and are used in some surgical trials, where a participant receives the incision and preparation without the operative step, and several such trials have found widely used procedures no better than the sham, which is why they are done despite the ethical difficulty. Where the participant cannot be blinded, the assessor usually can be, which recovers part of the benefit, and objective outcomes including death and laboratory measurements are less susceptible than subjective ones. Reporting honestly which parties were blinded is what allows a reader to judge.

The control arm is frequently misunderstood, and what it measures is not what people assume. Improvement in a placebo group combines several things that have nothing to do with any effect of the inactive treatment, including the natural course of the condition, the tendency of people to enter a trial when symptoms are at their worst and to improve afterwards regardless, and the effect of being observed and cared for. Studies comparing a placebo group against a group receiving no treatment at all find much smaller differences than the improvement within the placebo group suggests, which indicates most of that improvement is not caused by the placebo. Genuine placebo effects do exist, are largest for subjective symptoms including pain and nausea, and are far smaller than popular accounts imply.

A blinded randomised trial can still mislead and the remaining problems are the ones the field now concentrates on. Selective publication means trials with negative results are less likely to appear, so the published literature overstates effects, which registration of trials before they begin was introduced to address. Outcome switching means reporting a measure other than the one the trial was designed around, which is detectable by comparing publications against registrations and is found at substantial rates. Short trials cannot detect long-term harms. Participants may differ from the patients a treatment will actually be given to. Surrogate outcomes may not track what matters. And a single trial, however well conducted, is one result, which is why systematic review across all the evidence sits above any individual study.

A control group removes what would have happened anyway, randomisation removes systematic differences between groups, and concealment removes the effects of expectation on participants and on assessors. Comparisons find larger effects in unblinded studies of the same question. Blinding does not address selective publication, outcome switching or short follow-up, which is why review across all evidence sits above any one trial.