Clinical Trials

Clinical Trials: A Practical Guide ■❚❙❘subjects according to date of birth (odd or even years), hospital record number,or date of screening for the study (odd or even days), but these randomizationmethods have a level of predictability, so strictly are not acceptable methodsof randomization.Which are the common randomization methods?The generation of a randomization code can be achieved using one of a varietyof procedures. Once a code and method of allocation are decided on, their rulesmust be adhered to throughout the study. Common types of randomizationmethods are [1,2]:• simple randomization• block randomization• stratified randomization• minimization or adaptive randomizationA combination of these methods can also be used, and other special methods doexist. Let us now discuss the more common randomization methods listed.Simple randomizationThe most common form of randomization, referred to as simple or completerandomization, is a procedure that makes each new treatment allocation withoutregard to those already made. The principle of this method for a trial with twotreatments can be demonstrated by deciding treatment assignment by tossing anunbiased coin, eg, heads for treatment A and tails for treatment B. When the nextsubject is to be assigned, previous allocations are not considered.This method is easy to implement and unpredictable. However, as it is somewhatinconsiderate to previous allocations, it can often produce small inequalitiesbetween treatment groups, eg, 200 women were assigned to treatment A and 205women to treatment B. In a large trial this makes only a small difference, but insmaller trials at an early clinical stage that involve only a few dozen subjects, theseinequalities could have a substantial impact.ExampleConsider an example trial with 12 patients. While there is an equal chance ofbeing allocated treatment A or treatment B, the number of subjects randomlyassigned to each treatment ends up being 5 and 7, respectively (Table 1). Thisimbalance in the initial allocation will result in significant difficulties in thestatistics and possibly a lower power for detecting differences between the67