Allocating sufficiently large numbers of patients at random to create similar treatment comparison groups

Comparing like with like

Random allocation of patients to treatment comparison groups is one of the ways to ensure that tests of treatments are fair. Although random allocation can never ensure that treatment comparison groups are exactly comparable in all respects, it ensures that the composition of the comparison groups has not been biased by clinician or patient choices. Unless allocation bias is reduced in this way, any differences in the progress of patients in the comparison groups may simply reflect the fact that they were composed of patients who were different to begin with, and nothing to do with relative effects of different treatments.

Recorded, known patient characteristics

Patients in the treatment comparison groups that have been generated using random allocation need to be sufficiently alike in characteristics (age, for example) which are known to influence their health problems. Reports of treatment comparisons, therefore, need to show that 'like patients' have been compared with 'other like patients'.

Unrecorded and, as yet, undiscovered patient characteristics

Patients in treatment comparison groups who differ in respect of unrecorded and as yet undiscovered characteristics of importance obviously cannot be shown to be comparable. However, because random allocation ensures that any unrecorded differences have been due to chance, these differences can be taken into account in statistical analyses, usually by calculating confidence intervals.

Achieving balance in important patient characteristics

The larger the number of patients allocated at random to treatment comparison groups, the more likely it is that the comparison groups will be made up of patients with comparable characteristics, and the less likely that large imbalances will remain. You can see this effect of increasing the numbers of patients using the random allocation program below. Ask the program to randomise successively larger numbers of patients (up to 100,000) and watch the way that the characteristics of the patients in the treatment comparison groups become increasingly balanced.

Enter the number of patients to be randomised, and hit the Go button....