Checklist
for randomization
Randomization introduces a deliberate element of chance into the assignment of treatments to participants in a clinical trial. It tends to produce treatment groups in which the distributions of prognostic factors, known and unknown, are similar. In combination with blinding, randomization helps to avoid possible bias in the selection and allocation of participants arising from the predictability of treatment assignments. It also provides a strong statistical basis for the quantitative evaluation of the evidence relating to treatment effects.
The method of allocation generation should initially be specified pre trial (examples being a random-number table or a computerized random-number generator). The sequence may be generated by the process of simple randomization or restricted randomization. Simple randomization is based on a single sequence of random assignments and is particularly suited to large trials. It is possible that by chance alone the groups might be imbalanced with respect to important prognostic factors. Restricted randomization describes procedures used to control the randomization to achieve balance between groups in size or characteristics. Examples of restricted randomization include: blocking; stratification; minimization. This has greater potential benefit in small trials.
In multi-centre trials the randomization procedures should be organized centrally. This can entail, for example, generation of separate random schemes for each centre, the provision of telephone randomization where the collaborating centers phone up the co-ordinating office for the next allocation or a web-based system where collaborating centers access allocation over the web.
Sometimes we cannot allocate individuals to treatments, but must allocate a group (‘cluster’) of subjects together in a cluster randomised trial. The procedures for randomising clusters to trial interventions are identical to those described above.
In addition to knowing the methods used, it is also important to decide how the random sequence will be implemented: specifically, who will generate the allocation sequence, who will enrol participants, and who will assign participants to trial groups. Whatever the methodological quality of the randomization process, failure to separate the creation of the allocation sequence from assignment to trial group may introduce bias.
Illustrative
examples - Perinatal care trial
|
|
‘In this trial all units (hospitals) entered the study at the same time, and all prognostic variables were known in advance. In this situation a computer algorithm chose at random one allocation sequence among a set of sequences that minimized the imbalance between groups. The algorithm was used to allocate hospitals in the study. There were four hospital characteristics that were
selected as important prognostic variables. These were the variables
that were included in the minimization algorithm to assure balance
between groups. 1. Teaching hospital with residents (Yes-No). 2. Country (Argentina-Uruguay). 3. Hospital size (less than 2000 – 2000 or more deliveries per year). 4. Region (Montevideo, Salto/Paisandu, Rosario, Buenos Aires).
After the baseline data collection period the dataset was analyzed by an independent data center, Research Triangle Institute (RTI). RTI was then applied in the inclusion criteria to assess the eligibility of the preselected hospital according to the baseline rate of episiotomy and active management. The statistician at CLAP (the trial) elaborated a computer program implementing the minimization algorithm. The source code of the algorithm was made available in advance to RTI for audit and testing. This computer program was used by the statistician at RTI to allocate eligible hospitals to either intervention or control without the participation of others. The assignment was then communicated to CLAP. Thus, there was a clear separation between the generator of the intervention assignment and the CLAP study coordination. (CLAP Trial - go to publication) |
Illustrative
examples - WHO Pre-eclampsia trial
|
|
Simple randomization was conducted independently for each study site by the Development and Research Training Program in Human Reproduction (HRP) of the Department of Reproductive Health and Research (RHR) statistical unit at WHO headquarters and provided to the pharmaceutical manufacturer. The random allocation sequence was generated using computer-generated random numbers. Randomization was to the two arms of the trial and stratified by country. Blocking with randomly varying groups of 6-8 was used to restrict randomization within the strata (country) (SAS Software, Copyright© 1989, 1994 SAS Institute, Inc., N.C., USA). Random allocation technique consisted of allocating consecutively numbered treatment boxes for each woman, including in each box seven independent bottles, each of them with tablets for four weeks of treatment. Each bottle contained 100 tablets. Therefore each subject had seven bottles with the same randomization number. Treatment boxes were kept at the clinic. Bottles were provided consecutively as needed every month after randomization. When the woman came for antenatal care, she returned the used bottle from the previous month and was given the next month's bottle from her box. She returned the bottle after 4 weeks, regardless of whether she finished the tablets or not. Each bottle was numbered from 1 to 7 within the box and should was used sequentially. (WHO Multicentre Randomized Trial of Calcium Supplementation for the Prevention of Pre-eclampsia - go to publication) |
Illustrative
examples - Magpie trial
|
|
Hospitals with accessed the 24-hour telephone randomization service Women were entered into the trial by means of a telephone call to the 24-hour randomization service in Oxford. During this telephone call brief baseline details from the Trial Entry page in the Women’s Booklet were requested, and recorded on a computer, before the treatment allocation could be given. At the end of the call a two-digit pack number (and the four digit number of the box from which it was taken) was issued, and this number recorded immediately. Once this pack number was allocated, the woman were irrevocably entered into the trial, irrespective of whether the treatment pack was opened. Randomization was balanced for major prognostic variables; severity of pre-eclampsia, gestational age at randomization, whether delivered, whether given anticonvulsants drugs before trial entry, whether a multiple pregnancy and country. Hospitals using a local pack system This system was only for hospitals that did not have access to the 24-hour telephone randomization service. When a woman gave consent to participate in the trial the clinician completed the Trial Entry page in the Women’s Booklet. This page was completed BEFORE the treatment pack was opened, it recorded brief baseline details about the woman and the number of the next treatment pack. The treatment packs HAD to be used in the order in which they were removed through the slot in the box, which was lowest number first. Once the Trial Entry page was completed the woman were entered into the trial, irrespective of whether the treatment pack was opened. (Magpie trial - go to publication) |
Checklist
for randomization This checklist has been contributed by Dave Sackett, who prepared it for the 3rd edition of Clinical Epidemiology; A Basic Science for Answering Questions about Health Care, published by Lippincott, Williams & Wilkins, 2005.
Random
Numbers Free web based random number generator.
Pocock SJ. (1983) Clinical Trials: A Practical Approach. John
Wiley and Sons, Chichester.
Altman DG, Bland JM. How to randomize. BMJ 1999; 319: 703-704 .