Statistical design methods Sample Clauses

Statistical design methods. ‌ The most common statistical design method of avoiding contamination is the use of cluster randomisation, where groups of participants instead of individuals are allocated to trial arms. This strategy is often advocated by researchers and funders because it can prevent contamination, provided that treatment allocation is made at the highest level at which it is thought to take place (▇▇▇▇▇▇▇▇ and ▇▇▇▇▇▇▇▇, 1998). By ensuring that all participants within a cluster receive the same treatment, contamination of the control condition due to participants being affected by each other’s treatment receipt can be avoided. Cluster randomisation is used frequently in mental health trials, partly because of the problem of contamination, but also for logistical reasons. For example, cluster randomi- sation can reduce the number of clinicians who need to be trained in a new treatment in comparison to an individually-randomised trial. Cluster randomisation can also lead to a feeling of fairness within communities, as all participants are allocated to the same treatment. However, the use of cluster randomisation cannot always minimise conta- mination. There is at least one type where the scale of the problem is independent of the level at which treatment is allocated: control participants seeking out the active intervention outside the trial. This is a particular problem in screening trials which are typically found in other areas of medicine, for example in oncology (▇▇▇▇▇▇ et al., 2010). It is also possible that this problem may be becoming greater as patients are encouraged to take greater responsibility for disease management decisions and with the growth of online forums. There are substantial drawbacks of cluster randomisation in terms of estimator efficiency and bias. The main cost is that the correlation between participants’ outcomes within clusters means that each additional observation provides less information than it would in an individually-randomised trial. This correlation and the number of participants per cluster must be factored into a power calculation and will inflate the sample size requirement. The design factor (D) that is used for this inflation is: