Statistics from Altmetric.com
Randomisation is used within the context of therapeutic studies to try to reduce bias. It does this by using chance to spread, hopefully evenly, important prognostic factors across the groups within the study.
Randomisation can be performed in a number of ways; each variation can be used in different trial situations. To make life easier, for this article we’ll assume there are just two arms, but the principles apply to studies with greater numbers of options too. Simple randomisation is as straightforward as tossing a coin for each individual entered. In small studies, doing this may run the risk of having an uneven number of participants in the trial arms, making interpretation more difficult. The simplest variation on this is to predetermine the number of individuals in the study, and (metaphorically) put the appropriate number of “A”s and “B”s in a bag and withdraw the letters. This gives an even spread of numbers across the study arms—but is impossible to achieve for very large studies.
Block randomisation is where a block of participants (typically 6–12 in size) is randomised into an even split between “A”s and “B”s. This lets “time” be balanced between the arms too—for example, winter versus spring admissions—and balances the workload between the arms—if the treatments are not drug therapies but physiotherapy, surgery, or a multidisciplinary team intervention. It also allows a study to stop with an even spread between the arms. However, if the blocks are of the same size it may be possible for investigators to start to guess what’s coming next, upsetting the allocation concealment and jeopardising the trial.1 One way around this is taking blocks of 6, 8, and 10 participants and randomising the order of these too.
Stratified randomisation is a method where the investigator doesn’t leave the distribution of known or presumed prognostic variables entirely to chance; instead each major variable (for example, age, tumour stage, biological marker) is treated almost as a separate mini-trial, and participants within these strata are randomised independent of the other strata. (As a rule of thumb, you need at least 10 participants in each arm to make this valuable.) A similar type of process is used in minimisation allocation, which achieves similar results by a slightly different method.
Finally, cluster randomisation should be used when the unit randomised is not an individual child or family, but institution or group. For example, a trial of providing mosquito netting to prevent malaria may randomise villages, a study of a new computerised decision support system may randomise family practices.
Edited by Bob Phillips