Chair: Rory Collins, Clinical Trial Service Unit, Nuffield Department of Clinical Medicine, University of Oxford, Oxford, UK

Stuart J Pocock

Jackie Brighton

Val Gebski*, Wendy Hague, Tony Keech and John Simes

Anthony Keech*, Val Gebski, Gemma Ritchie and John Simes

2.1

Stuart J Pocock

The factorial design is valuable in clinical trials research by enabling two (and occasionally more) treatment issues to be evaluated simultaneously in the same study.

For treatment factor A, all patients on A (that is, A + B and A + not-B combined) are compared with all patients not on A (that is, not-A + B and not-A + not-B combined). Similarly for factor B, all patients on B are compared with all patients not on B.

These simple analyses (and their ability to achieve two comparisons for the price of one) rest on the assumption that there is no statistical interaction: that is, the effect of A does not depend on whether B is given or not, and vice versa.

The drawback is that most factorial trials will lack sufficient statistical power to definitively clarify whether such an interaction exists.

The other benefit of a factorial trial is that the treatment combination can be investigated directly (especially important if both treatments work). However, anxieties about treatments not acting independently (for example, the combination is overkill (maybe too toxic), or the combination works but separate treatments do not), plus a lack of appreciation of their scientific merits, have inhibited the extent to which factorial trials have been used.

This presentation will present these pros and cons of factorial trials, illustrated with some recent examples, especially from cardiovascular disease trials.

2.2

Jackie Brighton

It is well known that only a small percentage of potentially eligible patients are randomised into trials. Therefore it is important that we manage limited patient resources efficiently. One option is to randomise patients to more than one trial simultaneously.

This has the potential advantages of:

• increasing recruitment
• reducing time taken to reach accrual goals (more timely evaluation of treatments)
• allowing pragmatic evaluation of new treatments in the presence of other new treatments
• gaining better patient representation (less selection of patients for trials)
• providing an opportunity to streamline consent, outcome measurements, data collection and follow-up.

Simultaneous randomisation was approved by the international steering committees of GUSTO-1 and ISIS-4, two large pragmatic trials testing optimal use of existing treatments for acute myocardial infarction.

In Australia 28 hospitals participated in both trials and allowed simultaneous randomisation (SR hospitals), and 10 sites participated in both trials but generally randomised patients to one trial or the other (non-SR hospitals).

Overall, 538 (24%) of all Australian GUSTO-1 patients were randomised to both trials. Hospitals having a policy of simultaneous randomisation randomised 517 (48%) of GUSTO-1 patients to both trials and achieved higher recruitment rates per month (8.1 patients/month versus 6.9 (non-SR hospitals)), despite randomising similar numbers of patients per month..

Some people have expressed reservations about the policy of allowing simultaneous randomisation because of potential problems:

• to do with practical issues about consent, compliance, data quality and staff
• to do with effects on the power of the studies
• for bias
• for treatment interactions.

These problems were investigated using data from the GUSTO-1 and ISIS-4 experience but found not to introduce any bias or adverse effects.

Such problems can occur in any trial situation, even in the absence of simultaneous randomisation, when nontrial treatments are administered at the clinician’s discretion. Adopting a policy of simultaneous randomisation could help to standardise nontrial treatments and allow quantification of any potential problems.

2.3

Val Gebski*, Wendy Hague, Tony Keech and John Simes

The use of a run-in period prior to randomisation can be a useful strategy to increase trial efficiency and reduce costs in chronic-treatment trials.

A placebo run-in period offers opportunities to refine baseline measures, to resolve with patients potential study issues (such as ambiguities in procedures) and to monitor initial patient compliance.

A run-in period on the active treatment allows additionally for monitoring of early toxicity, dose activity and potential response to active medication.

A major advantage of run-in phases is the increase in trial efficiency gained by screening out potentially noncomplying patients, which has a direct effect on the power of the study. The increase in sample size required to achieve the efficiency of a totally compliant study is approximately 1/(c₁+c₂–1)² where c₁ and c₂ are the compliance rates in each group. Noncompliers are assumed to have crossed over to the opposite treatment arm.

Use of a run-in phase to screen potential noncompliers may, however, alter the risk profile of the patients eventually randomised.

The objective of the LIPID study was to determine whether cholesterol reduction with pravastatin would reduce coronary mortality in patients with a history of myocardial infarction or unstable angina. The study design included an eight-week placebo run-period, which resulted in 755 of 9769 patients being eligible but not randomised. This represents a potential saving of 1580 fewer patients randomised (to achieve the same power) had the 755 patients been available for randomisation (and still discontinued treatment) in the absence of a run-in phase.

Additionally, with the incorporation of a run-in period, there was also a potential saving of 7900 patient–years of follow-up.

Comparing the baseline risk profile of these 755 with the 9014 randomised patients reveals that the randomised patients were marginally older, with lower triglycerides. The rates of the other risk factors were similar, suggesting little or no effect on the generalisability of the results by their exclusion.

2.4

Anthony Keech*, Val Gebski, Gemma Ritchie and John Simes

Frequently in clinical trials there is temptation to answer more than just the original question driving the rationale for the trial. Substudies can provide an opportunity to investigate the effects of a different intervention (for example, as in a partial factorial design), or one or more potential mechanisms of treatment action (for example, study of surrogate clinical markers of disease), or a different outcome event on the same population. Substudies should generally be approved by the study steering committee and encourage collaboration.

Potential advantages include:

Drawbacks include:

Publication of substudy results may be appropriate during conduct of the main study, provided they would not adversely affect attention to the main trial questions.

2.5

Robert M Califf

Clinical trials are undertaken for various reasons: to understand the mechanisms of treatment, to define therapeutic choices and to measure mortality.

We should no longer be developing therapies without acquiring a plausible understanding, though clinical trials, of:

• the effects, or the potential for effects, on mortality
• the effects on quality of life (such as symptoms, unpleasant experiences)
• the effects on cost.

The definitive study can be done on the basis of results for a surrogate, promise seen in a therapy, a difference or a trend in a composite endpoint (that is, an aggregation of outcomes intended to capture the total effect of a treatment), or theory alone.

In deciding which composite endpoints to include, issues are the severity of the individual endpoints and whether to separate or aggregate endpoints related to efficacy and safety.

The individual endpoints may be weighted, but an issue for weighting is that a decision must be made as to who should assign the weights: experts, doctors or healthy people (who may not reflect the patients’ perspective), patients (who may be biased), or regulatory agencies.

Treatment effects can differ between patient groups in various ways:

• the effect may be the same across the components of the composite endpoint
• the effect may be most pronounced on the most important endpoint
• the effect may be most pronounced on the least important endpoint.

Including death as a component of a combined outcome has to be planned carefully: when death is common, there is no satisfactory way of measuring a nonfatal event; when death is not common, there is a risk (if there is no real effect on death) that prevention of death appears more effective in the control group.

10 November 1999