Title of the document:

Page. 1 of 2 Rev. Nr.: Effective date: 1.3 1 Jan 2010

Randomisation

HB Nr. : 1.1B-05 1. Aim Formulating quality requirements for randomisation. 2. Definitions Randomisation is the random assigning of a trial participant to one group in a study, often an intervention, in which two or more groups are being investigated. The groups are referred to as arms. Usually these arms consist of equal numbers of participants, but this is not a necessity. Randomisation can be undertaken at an individual level, as well as at the level of a general practice, nursing home, company or department (so-called cluster randomisation). In instances such as these the included research participants within a single practice or nursing home belong to the same arm. Block randomisation is a randomisation procedure in which the same number of research participants is allocated to smaller blocks. For instance, in a block of four there will be two participants allocated to the control group and two to the treatment group for each block. This ensures that the number of research participants is (almost) always the same for each arm of a study. Block randomisation is often applied if trial participants are successively included and if it is important to have approximately the same number of trial participants per arm over time (blocks in time). Pre-randomisation involves randomising before obtaining consent (informed consent) from trial participants. This may occasionally reduce the number of dropouts (after randomisation). A typical study involves the comparison of an experimental treatment with a standard treatment without additional measurements. Following pre-randomisation only full "informed consent" needs to be obtained from the participants in the experimental arm, reducing the number of dropouts in the standard treatment arm. Pre-stratification is the division of research participants into subgroups (strata) prior to the actual randomisation. An example here is dividing eligible participants into three subgroups for age: Young, middle, old. Randomisation is then carried out for each subgroup separately. Pre-stratification is useful if there is an expectation that the results will significantly differ in terms of outcome variable(s) between the subgroups. These differences may affect the general level of the outcomes, as well as the difference between the arms (this is an example of effect modification). It may also be the case that pre-stratification is better for practical reasons (e.g. if the study is being carried out in a number of centres or institutes). An alternative is post-stratification. Here, subgroups (strata) are created in the analysis phase. However, pre-stratification is more effective, although it is only possible for a very limited number of variables (2 or 3). 3. Keywords 4. Description Randomisation is frequently applied in randomised clinical trials (RCTs). The procedure is intended to minimalise the effect of (measured and unmeasured) confounders (see Pocock [1]), which also includes a history of the use of randomisation). Applying randomisation results in bias from confounders being avoided, in other words, each confounder is equally distributed across the study arms. However, in a concrete situation the distribution across arms may still differ by accident. This is the reason why many articles include baseline statistics (age, sex, etc.). If during the data analysis it appears that the randomisation procedure has not been sufficiently effective, this may have a significant impact on the rest of the data analysis and therefore on the validity of the research results. This will necessitate a correction in the analysis for the confounders concerned or baseline statistics. It is therefore advisable to record important potential confounders and include them as variables in the data set: This offers the opportunity for corrections to be applied during the data analysis should it appear that randomisation has unfortunately not had the desired effect.

Title of the document:

Page. 2 of 2 Rev. Nr.: Effective date: 1.3 1 Jan 2010

Randomisation

HB Nr. : 1.1B-05

5. Details It is important to ensure during the preparations for the randomisation procedure that the allocation to (both) treatment groups is truly occurring at random, that is, with the help of (computer-generated) random numbers. As patients/respondents are often successively included into a study for a number of reasons, researchers often opt for a so-called block randomisation, in which participants are equally allocated to the arms within blocks of inclusion (four, eight or sixteen). Audit questions: 1. Has the procedure followed for randomisation been described in detail, and has the researcher maintained a logbook including any potential changes in the protocols? 2. Have the most important confounders been included as variables in the data set? 3. Did randomisation work? If something went wrong with the randomization, how was this corrected for in the data analysis?

6. Appendices/references/links Pocock SJ. Clinical trials: A statistician's perspective. In: Adr HJ, Mellenbergh GJ, eds., Research Methodology in the Social, Behavioural & Life Sciences. London Thousand Oaks New Delhi: SAGE Publ., 1999; 96-109. S. Piantadosi. Clinical Trial: A methodological perspective. Wiley 2005 S. Pocock. Clinical Trial: A practical approach Wiley 1983 7. Amendments V1.3: 1 Jan 2010: Translation into English. V1.2: 11 Oct 2007: Text abbreviated, concept of pre-stratification described. V1.1: 23 Mar 2006: Minor textual modifications and references to randomisation software