Sequential versus fixed designs

The sample size calculations shown in this chapter assume a sample size is to be 'fixed' in advance. An alternative approach, known as sequential analysis, does not fix the sample size in advance, but accrues and analyses patients until sufficiently strong evidence in favour of the experimental or control arms (or against such a difference) emerges. In a fully sequential design, analysis takes place after each patient's response to treatment is known. This is thus most appropriate when the outcome measure can be assessed shortly after the patient enters the trial. Where this is the case, such a procedure can lead to trials being completed more quickly, compared with a fixed sample size design, if larger than expected differences occur. Perhaps because these two conditions rarely apply in cancer trials, fully sequential designs are rarely used. We therefore do not discuss them further here, but recommend a text book [1] for further details. An adaptation of this approach is much more widely used in oncology. In this, a sample size is determined according to the methods described in this chapter, but several interim analyses take place before the sample size is reached. This allows the potential for trials to be stopped early if convincing evidence for either treatment emerges. Schemes for determining the strength of evidence that must be demonstrated at an interim analysis to justify stopping are discussed in Section 9.5.

0 0

Post a comment