Randomisation

Chalmers (BMJ, 1999): Why transition from alternation to randomisation in clinical trials was made?
EDITOR—It is time to put the record straight about ways of controlling selection bias when generating comparison groups in clinical trials. D'Arcy Hart has done history a great service by noting that Bradford Hill's motivation for replacing alternation with randomisation was “to better conceal the allocation schedule.”1 This is what Guy Scadding (who, with D'Arcy Hart, is the other surviving member of the team who designed the streptomycin trial) told Mike Clarke and me when we visited him on 10 June 1999, and what Bradford Hill told William Silverman and me when we visited him on 3 April 1982.

Bradford Hill's motivation for concentrating on the concealment of the allocation schedule in the streptomycin trial seems likely to have been stimulated more than a decade earlier. A Medical Research Council trial of serum treatment for lobar pneumonia had used an (unconcealed) allocation schedule based on alternation,2 and important imbalances in the characteristics of patients in the treatment and control groups had occurred.3 In an unpublished critique of the study for the council Bradford Hill noted that greater effort should be taken “that the division of cases really did ensure a random selection.”4

As Altman and Bland recently observed, treatments allocated alternately “are in principle unbiased—being unrelated to patient characteristics—[but] problems arise from the openness of the allocation system.”5 They might have gone on to note that similar problems can arise with allocation schedules based on random numbers unless they are concealed from those entering patients into trials: the fact that an allocation schedule has been based on random numbers provides no security against biased allocation to comparison groups.

It is regrettable that a certain mystique has grown up around randomisation. This seems to reflect unwarranted inferences about R A Fisher's influence on the design of medical research, even though the history of efforts to make fair treatment comparisons in medicine predates Fisher by centuries (see Controlled Trials from History at www.rcpe.ac.uk/cochrane).

The methodological advance made by the streptomycin trial was not so much the use of randomisation to generate the allocation schedule but rather the clear efforts made by the trial's designers to conceal the allocation schedule from those involved in entering patients in the trial. The results of the streptomycin trial would have been no less valid if the trial had used a system of alternation as a basis for the allocation schedule and—against the odds—had succeeded in concealing this from those taking decisions about eligibility and allocation of patients.

In summary, the only reason that allocation schedules based on random numbers are to be preferred to those based on strict alternation is because they are easier to conceal, not because they are any better at abolishing selection bias. This is the reason that a schedule based on random numbers was used for the streptomycin trial, and why the study is a methodological landmark. As Lock has suggested, Bradford Hill deserved to receive a Nobel prize for this immensely important contribution to the process of assessing the beneficial and harmful effects of medical care.2

I am indebted to Doug Altman, Joan Austoker, Mike Clarke, Philip d'Arcy Hart, Richard Doll, Stephen Lock, Irving Loudon, Guy Scadding, William Silverman and Ben Toth for checking an earlier draft of this letter.