Method - Design

Randomised trials are the best way to find out what treatments work, but they can be slow and expensive. We are inventing new and efficient ways to design randomised trials that can give clear answers, often answering multiple questions within the same trial. This work is supported by detailed statistical modelling to show how to analyse the new designs while strictly controlling the risk of wrongly concluding that a treatment is effective when in truth it isn’t. The new designs are implemented in our own trials, and this helps us to refine the designs and demonstrate their value for other trials.
Our novel designs are for
1. trials to evaluate whether it is safe to shorten the duration of treatment or use a treatment with fewer side effects;
2. trials to evaluate multiple treatments, where we can drop treatments that are found to be ineffective, adopt the use of treatments that are found to be effective, and start evaluating new promising treatments;
3. trials that work even when the benefit of treatment is short-term or delayed;
4. trials where groups of patients are randomised rather than individual patients;
5. trials specifically designed to evaluate treatments tailored to specific types of patient; and
6. trials comparing a number of treatments, where some patients cannot receive some of the treatments.
The results of our work are that treatments can be evaluated faster, more effectively and more cost-effectively.

The design programme aims to improve the efficiency and resilience of randomised trials by proposing new trial designs. Efficiency is the ability to identify a treatment effect (or lack of effect) quickly with best use of resources. Resilience is the ability to reach credible answers when design assumptions turn out to be inaccurate. We do this pro-actively and in response to emerging challenges in our trials, and we implement solutions first in our trials and then more widely. New design methodology usually requires new software, and we provide user-friendly software packages for internal and external use. We also improve some existing trial designs in a similar manner.
Our main areas of work are
1. Durations design and non-inferiority trials. Non-inferiority trials are widely used and are often the best way to improve outcomes, especially in tuberculosis and anti-microbial resistance. However, they are hard to implement and often lacking in resilience. We are developing ways to choose a shortened treatment regimen (the Durations Design), and ways to address particular problems that occur in sensitivity to non-adherence and to small changes in the control event rate.
2. Multi-arm multi-stage (MAMS) platform trials. To speed up the evaluation of new therapies and improve success rates in identifying effective ones, we pioneered the development and implementation of the MAMS design and wrote the nstage software suite in Stata for sample size calculation. A MAMS platform trial generally has a single master protocol in which multiple treatments are evaluated over time, and offers flexible features such as early stopping of accrual to treatments for lack of benefit, and adding new treatments to be tested during the course of a trial. We have nearly 20 years’ practical experience of designing and running such trials and we are extending the underlying methodology to increase the efficiency of the design and to broaden its application to a wide range of disease areas and types of outcome measure.
3. Non-proportional hazards. Trials with a time-to-event outcome are usually designed and analysed assuming the treatment effect remains stable over time (proportional hazards). However, non-proportional hazards (non-PH) is more common than previously thought, e.g. due to therapy ‘wearing off’ or taking time to work. If unrecognised and unaccommodated, non-PH may impact the success or failure of a trial. We are developing ways to design trials to anticipate possible non-PH.
4. Cluster randomised and stepped wedge trials. These are designs where clusters of individuals are randomised to treatment or control. We aim to optimise the design of these trials by minimising the number of clusters and/or measurements required, and also to show new ways to understand these trials.
5. Trials for stratified medicine. Our focus is on understanding how to design a trial when a treatment is suspected to work best in a particular subgroup, but this knowledge is not secure and may evolve during the trial.
6. The new “PRACTical” trial design. This is intended for situations where a number of treatments are available but typically the treatments are not appropriate for all patients, for example because of their other medical conditions. We propose to randomise each patient between the treatments that are potentially appropriate to them. We are developing suitable ways to design and analyse such a trial.