Stratified randomisation, allocation concealment and minimisation

Simple randomisation, a coin toss for every patient, balances groups on average but can leave a small trial lopsided on something important, such as how many patients in each arm have a particular mutation. Stratified randomisation runs a separate randomisation list within each combination of a few pre-chosen factors (region, disease stage, biomarker status, prior therapy), so each arm is balanced on those factors by construction. Permuted blocks keep the arms level as recruitment proceeds. Minimisation is an alternative that assigns each new patient to whichever arm would best balance the factors so far, with a random element to keep it unpredictable. The stratification factors are the ones later reported in the forest plot of subgroups, which is one reason they are chosen with care.

Allocation concealment is a different safeguard: whoever enrols a patient must not be able to know or predict the next assignment, or they could, consciously or not, hold back a frail patient until the experimental arm comes up. Modern trials use a central web or telephone system that reveals the assignment only after the patient is registered. Blinding is what happens after allocation; concealment is what happens before, and even open-label trials can and should conceal allocation.

The randomisation ratio is also a design choice. Most phase 3 trials randomise 1:1, which gives the most statistical power for a given number of patients. A 2:1 ratio puts more patients on the new drug, which helps recruitment when patients want the experimental arm and builds a larger safety database, at the cost of slightly more patients overall. CheckMate 067 randomised 945 patients equally across three arms; SABR-COMET randomised 99 patients 1:2 in favour of stereotactic radiotherapy; IMvigor011 randomised only the patients whose blood test was positive for circulating tumour DNA, an example of a biomarker-defined population set before randomisation.