Designing and analysing multi-arm multi-stage clinical trials with one or more endpoints

In the early stages of drug development there is often uncertainty about the most promising among a set of different treatments. In order to ensure the best use of resources in such situations it is important to decide which, if any, of the treatments should be taken forward for further testing. An efficient solution to this problem are multi-arm clinical trials in which several active treatments are compared to a common control group. By comparing several treatments within one trial the sample size and duration required tends to be markedly smaller than if each treatment would have been evaluated separately. For added efficiency it is desirable to monitor the trial at a series of interim analyses in order to allow early stopping if efficacy is quickly established and similarly to eliminate ineffective treatments early. In confirmatory studies these designs are also useful as it has been shown that using two doses instead of a single dose can markedly improve the study's success probability.

This proposal aims to develop statistical methods to investigate how these, so called multi-arm multi-stage trials, can best be designed and analysed. The work will utilize three recently finished or currently ongoing studies in three important medical areas: HIV/AIDS, leukemia and metastatic prostate cancer. The multidisciplinary research team includes experienced clinical specialists and clinical trialists for each study from the clinical trials research unit in Leeds, the University of Liverpool and the MRC clinical trials unit as well as statisticians with experience in statistical methods for clinical trials from the University of Bremen and the Warwick medical school.

Multi-arm multi-stage designs are a broad class of designs in which several active treatment arms are compared to a common control. Their advantage over traditional 2-armed studies is that, on the one hand, a common control group is used yielding smaller expected sample sizes compared to multiple 2-armed trials and allowing direct comparisons of all active arms under the same study conditions and, on the other hand, that decisions about effective and ineffective treatments can be made at early interim analyses. Although some work on designing these trials in simple situations (e.g. play-the-winner designs) is available no adequate methods are available for many situations of practical interest (e.g. multiple endpoints). Moreover methods for adequately estimating the treatment effect and finding the corresponding confidence intervals in such trials are not available beyond two stages.

This proposal aims to
1. develop statistical designs for multi-arm multi-stage designs that use intermediate endpoints for treatment selection;
2. develop and evaluate methods for estimating the treatment effect and corresponding confidence intervals in multi-arm multi-stage designs with treatment selection and obtain their confidence intervals;
3. develop statistical methods for multi-arm multi-stage designs with multiple endpoints;
4. develop an open-source software package and associated training tailored to clinical trialists and applied statisticians working on clinical trials.

In addition to academic beneficiaries, the work proposed in this grant will have great impact on others. By extending methodology for multi-arm multi-stage clinical trials (MAMS) to allow for differing endpoints at each stage, we shall be able to apply the theory of optimal design to a broader range of MAMS trials. This is of great benefit to patients recruited to such trials - stopping boundaries can be specified that reduce the average number of patients recruited to poor treatments, but keep effective treatments in the trial. Use of such designs will mean fewer patients are exposed to poor treatments. This will also be of interest to clinicians running and supporting the trial, who are naturally interested in ensuring their patients do not receive ineffective treatments. Funders of trials would also benefit, since fewer patients would be required on average, thus reducing the average cost. Because MAMS trials are high cost trials, even relatively small improvements in the design can lead to large savings in costs. Since this is a design issue, the benefit from this work will come in the medium-term future, when actual trials using the methodology start recruiting. In addition to the immediate benefit to patients on the trial the number of patients in the trial and hence the duration of the trial is reduced resulting in effective treatments to be put into clinical practice sooner.

Reducing bias in the estimation of treatment effects after a MAMS trial will be of great interest to those using results from a clinical trial. Those planning future trials will be able to use results to inform the design of their trial with the knowledge that the bias is minimised. Regulatory organisations such as the European Medicines Agency or the National Institute for Clinical Excellence will be able to judge the cost-effectiveness of a drug using reliable data. This area of research will have a more immediate impact since it can be applied to data from a MAMS trial designed prior to this work.

Through the extension of methodology to the exact and optimal design of MAMS trials with multiple endpoints, we shall make a greater range of efficient trials possible. This would be of interest to organisations that design and run trials, such as clinical trial units (CTUs) and pharmaceutical companies, or those who support the design of trials, such as research design services (RDSs). In addition, patients recruited to such trials will benefit because there will be clear, statistically sound, rules as to when to drop a treatment which is potentially dangerous. Since this is a design issue, the impact of this work would be in the medium-term future.