Parallel groups

The simplest form of randomized trial is known as the parallel group trial. This follows the basic design illustrated in Fig. 4.3. Eligible patients are randomized to two or more groups, treated according to their assigned groups, subsequently assessed for their response to treatment and the groups are compared in a manner appropriate to the data.

There are numerous examples, one being the first MRC trial of radiotherapy in rectal cancer [19]. In this trial, eligible patients (those with potentially operable rectal cancer) were randomized to one of three treatment groups: surgery alone, surgery preceded by low-dose radiotherapy or surgery preceded by higher-dose radiotherapy. Following treatment, patients were followed up until death, and the overall survival rates were compared.

The parallel group design is simple and widely applicable. In theory, the number of treatment arms is unlimited. In practice, however, the number of groups may be limited either because there are only a small number of treatments which people are

Treat according to allocation

Assess patients and compare groups

Fig. 4.3 Parallel group trial interested in comparing, or because the number of patients available limits the number of treatments which can be compared while achieving a sample size sufficient to give useful results over a reasonable timeframe. In the latter case, attempting to answer several questions at once may lead to none being adequately powered to provide convincing evidence. Where neither of these constraints apply there remain two other issues worth considering; the quantity of information for a patient considering the trial to absorb, and the practical difficulties in the clinic of having many possible treatments in the same group of patients at one time. Although these are reasonable concerns, there is now empirical evidence to suggest that neither need be a bar to embarking on a multi-arm trial [20].

For many trials with more than two arms, questions may arise concerning the necessity for all patients to be randomized with respect to all arms. There may, within the medical community, be sufficient uncertainty about a number of treatments to warrant their study in a trial, but often practical difficulties if not frank disagreement mean that some clinicians would be unwilling to randomize patients with respect to all the arms. Indeed, multi-arm trials sometimes evolve for this very reason. One possibility is to allow participants to opt out of some of the arms. Although this maybe a practical solution in some cases, it needs careful thought, making sample size planning difficult. In general, it is often best to begin with a requirement that patients should be randomized with respect to all arms, and only relax this requirement if accrual proves particularly difficult. Given choice from the start, it is perhaps inevitable that patients and clinicians may avoid the 'difficult' arms (for example, 'no treatment' control arms) if they can still contribute to a trial. In the 1980s, the EORTC brain tumour group planned a trial to evaluate the role of post-operative radiotherapy in the treatment of low grade gliomas. There was no compelling evidence of benefit to radiotherapy; nevertheless, sufficient clinicians argued that they could not participate in a trial with a 'no-treatment' control arm that two trials were launched: the 'believers' trial randomized patients between two radiotherapy doses [21]; the 'non-believers' trial randomized patients between immediate post-operative radiotherapy (at the higher of the two doses) and radiotherapy deferred until disease progression [22]. The former recruited at three times the rate of the latter while both were open to accrual; it found no survival benefit to the higher dose. The non-believers trial finally completed its planned accrual after 11 years, the rate of accrual picking up notably on closure of the competing trial; it found no survival benefit to immediate radiotherapy over deferred radiotherapy. One can only speculate at the rate of accrual which might have been achieved in the non-believers trial had this been the only trial available.

If the reasons for comparing several groups are compelling, then there are designs which are more efficient than a simple parallel group trial in that they enable a reliable result to be obtained with fewer patients than would be required in a multi-arm parallel group trial. Sections 4.5.3 and 4.5.4 describe two such designs.

0 0

Post a comment