Choice of arms

Most randomized trials can be categorized into one of two types; those that compare A with B (for example surgery versus radiotherapy for oesophageal cancer, carboplatin versus cisplatin for advanced ovarian cancer), and those which compare A with A + B (for example surgery alone versus radiotherapy followed by surgery for rectal cancer; radiotherapy with or without prophylactic anti-emetics in patients undergoing abdominal radiotherapy). The former types compare entire treatment policies, the latter, one aspect of treatment. In both cases, A is the control treatment and B the experimental treatment. Pure scientific method might suggest that only one aspect of management should ever vary between the treatment groups, as only then can differences in outcome be attributed solely to this one change in treatment. In practice absolute purity is hard to achieve and rarely necessary. There is often a good argument for making the most extreme comparison - which might involve several changes over standard therapy - first rather than attempting to justify each change through carefully controlled trials in which only one aspect of treatment changes each time. Making the most extreme comparison first means that if no worthwhile benefit is shown, there is rarely a need to go back and examine the individual components for potential efficacy. If on the other hand the most extreme comparison finds a worthwhile benefit, then one can choose whether or not to go back and examine the individual components from the happy position of having improved outcome. In general, in designing a trial one should take time to consider for each possible design what conclusions could be drawn from each of the possible trial outcomes, and where they will lead; this kind of 'options appraisal' can help determine which comparisons should be made first.

The process of treatment development means that the choice of experimental treatment will generally be clearly defined before a randomized trial is considered. However, the choice of control arm perhaps gets less consideration than it might.

0 0

Post a comment