To illustrate some of the issues that can arise in determining the appropriate difference a trial should target, we consider two further trials as case studies.

Example 1: Design ofTE18. TE18 was a randomized trial in patients with stage I testicular seminoma conducted by the MRC Testicular Tumour Working Party [11]. At the time of its design, the following pertinent information was known:

♦ the disease is rare; approximately 500 men are diagnosed with stage I testicular seminoma every year in the UK (approximately 1 per 100,000 population);

♦ the average patient age is approximately thirty-seven years;

♦ if untreated following surgery (orchidectomy) alone, approximately 20 per cent of patients will have a recurrence of their disease, with the most common site being the abdominal lymph nodes;

♦ treatment of recurrence depends on the extent of disease at relapse, but is likely to need several months of chemotherapy;

♦ currently, we are not able to predict reliably which patients will relapse, and follow-up is intensive and prolonged (up to ten years) in order that relapses are detected at an early stage;

♦ only one randomized trial had ever been attempted.

For these reasons, to prevent recurrence, standard treatment in the UK and much of Europe was radiotherapy for all patients following surgery. Concerning radiotherapy, the following additional points of relevance were known:

♦ the 'standard' dose has evolved over time, and differs across Europe - in the UK a dose of 30 Gy given in fifteen treatment fractions over three weeks was fairly standard;

♦ this is very effective - around 97 per cent of patients will be relapse-free at three years;

♦ it is also associated with some morbidity - nausea/vomiting and diarrhoea during treatment, peptic ulceration in some patients, prolonged lethargy which can delay return to work for several months, reduced fertility - sperm counts can be reduced to minimal levels for up to two years after treatment and may never recover - and finally a small but definite risk of treatment-induced second cancers.

A policy of adjuvant treatment for all, though effective, would therefore mean that around 80 per cent of patients receive unnecessary treatment, as they were never destined to relapse. This implies that minimizing treatment (and thus side effects) without compromising efficacy is a worthwhile target for a randomized trial. A dose of 20 Gy in ten fractions was chosen as the comparison - a dose that was considered low enough to impact on morbidity, but for which there was reasonable (though non-randomized) data to suggest it maybe 'safe' with respect to relapse rates.

The question to address was, would a small increase in relapse rates be an acceptable trade-off for all patients having fewer side effects? To ascertain clinical opinion the questionnaire described above was employed to identify what differences in relapse rate would lead potential participants in the trial (and 'consumers' of the results) to switch from one treatment to another. The results of the questionnaires are shown in Fig. 5.8. Here, points lying to the left of the left-hand side arrows indicate the point at which each clinician would definitely use 30 Gy as standard treatment; points to the right of the right-hand side arrows indicate the point at which each clinician would definitely use 20 Gy. In between are the individual ranges of equivalence. This is an equivalence, or more specifically a non-inferiority trial, where the aim was to show that the difference in relapse rates associated with these two schedules lay within an 'acceptable range.' The data shown in Fig. 5.8 were used to deduce what was an acceptable range. It was hoped that the trial result would show a confidence interval centered close to zero (no difference) and with sufficient patients to ensure, ideally, that the whole confidence interval would lie to the right of the majority of right-hand side arrows (so all plausible differences would still lead clinicians to use 20 Gy). Such an outcome is represented by confidence interval A in Fig. 5.8. This though required several thousand patients, an impossible target for a disease this rare. A reasonable fall-back position is a confidence interval lying to the right of the majority of the left-hand side arrows - so all plausible differences are less than that which would lead clinicians to use 30 Gy as standard (confidence interval B).

Ranges of clinical equivalence

Ranges of clinical equivalence

0 12 3 Advantage to 20 Gy -

0 12 3 Advantage to 20 Gy -

Absolute differences in relapse rate (per cent)

Fig. 5.8 Proposed stage I seminoma trial

These considerations led to the need to exclude differences of 3-4 per cent, which in turn required a sample size of between 600 and 1000 patients. This was the aim of the trial.

Example 2: Design of AXIS (Adjuvant Xray and 5FU Infusion Study). In contrast, AXIS was a trial in colorectal cancer, which is second in incidence only to lung cancer in the UK. It was designed in the late 1980s when many trials of adjuvant radiotherapy and of chemotherapy had been performed, but few trials had involved more than a few hundred patients and results were conflicting.

As previous, relevant trials were available, the likely treatment effect in AXIS was estimated by a meta-analysis of previous trials of radiotherapy, and of different chemotherapy regimens [12]. From these the following was concluded:

♦ Portal vein infusion of fluorouracil (PVI) appeared promising (estimated absolute survival improvement 10-15 per cent) but many previous trials were flawed in their analysis and may have over-estimated the treatment effect. In addition, only 1500 patients had been treated in randomized trials.

♦ PVI involves one week of chemotherapy given when patients would usually remain in hospital recovering from surgery - it is fairly easy and cheap to administer and the incidence of serious side effects appeared very low. It was therefore potentially a widely applicable treatment and one for which small survival benefits could be very important.

♦ PVI could be given to both colon and rectal cancer patients.

Based on the results of the previous trials, therefore, the 'likely' difference was large, if flaws in previous trials were ignored, but the size of difference that could be considered clinically worthwhile was much smaller at around 5 per cent. The nature of the disease and treatment meant that it was feasible to design a trial powered to detect this smaller difference; if unequivocal evidence of larger benefit became apparent during the course of the trial, it would be possible to stop the trial early. For radiotherapy, there were the following considerations.

♦ There was much more data on radiotherapy (~9000 patients), but the schedules used varied widely and many would be considered inadequate or unsafe in the present day.

♦ The overall estimate of absolute survival benefit was approximately 4-5 per cent. To detect this reliably required 4000-6000 patients.

♦ Radiotherapy can be given only to rectal cancer patients; the incidence ratio of colon to rectal cancer is approximately 60 : 40.

♦ Radiotherapy was already becoming routine treatment because there were reasonably strong data to suggest that local relapse rates were decreased by its use. Although it was not clear if deferred radiotherapy would be any less effective than immediate radiotherapy with respect to survival, it was clear that the number of surgeons willing to randomize patients to a no-radiotherapy control group was limited, and a 4000+ patient target was not felt to be feasible.

Instead therefore, the overall sample size was based on that required for the PVI comparison, namely 4000 patients, with the aim that this should provide definitive evidence for or against the treatment. A radiotherapy randomization would be included in a factorial design for rectal cancer patients; as we could not produce definitive evidence from this trial alone, the aim was to randomize a minimum of 1000 patients (sufficient to detect a 10 per cent survival improvement) thereby contributing a reasonably substantial amount of additional data to the worldwide evidence, and with the aim of contributing to a formal meta-analysis at the end of the trial.

0 0

Post a comment