Introduction

The most appropriate methods of allocating patients to treatments in trials must fulfil two major requirements; they must avoid systematic bias, and they must be unpredictable. Assigning treatment by a means that employs the play of chance - randomization - aims to ensure that patients assigned to each treatment are balanced for both known and unknown factors which could influence their response to treatment.Making sure that the treatment to be allocated cannot be known in advance, or guessed with a degree of certainty - concealment - ensures that particular types of patients cannot be chosen to receive (or avoid) a particular trial treatment. Both these aspects are equally important; a properly produced randomization schedule loses all its value if it can be found by the person responsible for deciding whether or not to enter a patient into a trial, since knowledge of which treatment that patient would be allocated could affect the decision about whether or not to enter the patient. It is therefore a means by which systematic bias could be introduced. The method used should not only meet these requirements, but it must be shown to meet these requirements. Tossing a coin is a perfectly good method of allocating one of two treatments in an unbiased and unpredictable way - but it is very hard to prove to a sceptical referee that the first toss was used, and not the second, or the best of three, or whatever it took to allocate that patient the treatment for which you or they had a preference. With respect to a principle as important as randomization, it is reasonable for a reviewer to take the position that the randomization or concealment method was inadequate unless there is a clear statement to the contrary.

The principle of unpredictability immediately calls into question some methods of 'randomization' used extensively in the past, less so now, namely the use of a patient-related characteristic such as odd or even date ofbirth, or hospital number, or alternate allocation as patients enter a clinic say, to allocate treatments. For each ofthese methods, the person responsible for randomizing a patient knows what treatment the patient will be allocated should they agree to enter the trial. They then have a choice as to whether to offer the patient the trial - and thus the treatment - or not, and that decision may be based on patient characteristics which could determine their outcome to therapy in either a blatantly biased, or possibly quite subtle, way. This can make an apparently randomized trial simply a comparison of two groups of patients chosen for different treatments, which have a high chance of differing systematically in some respects. The evaluation of treatment may therefore be confounded with these differences in patient characteristics, and impossible to interpret.

There are a number of simple methods of true randomization which can either be prepared in advance (and held by an appropriate, independent person) or which use patient characteristics dynamically. Computer-based randomization methods add an additional degree of unpredictability and concealment over pre-prepared paper lists and should therefore be used whenever possible. However, it is important to understand the underlying methodology in order that the method of randomization appropriate to a particular trial is used. The next sections illustrate 'manual' methods of randomization which can be set up quickly and simply, requiring only a table of random numbers (one can be found at http://www.rand.org/publications/MR/MR1418). All these methods can, of course, be programmed for computer-based allocations.

0 0

Post a comment