We recently had a speaker come talk about adaptive clinical trials. It was a good seminar with lots of clear, real-life examples and it was well-attended by RDS advisers and researchers based both in universities and in the NHS. I won’t go into detail about the content of the seminar – people far better qualified than me have already done so in abundance – but it did get me thinking about how such designs could be used by the researchers with whom I and my colleagues work.
A large proportion of the health research funded by the NIHR, through such programmes as HTA, EME and RfPB, are clinical trials. And there are many issues to consider when designing an RCT, some of which I’ve discussed before. Most researchers engaged in such projects have input from one of the many Clinical Trials Units (CTUs) and have experienced statisticians, data managers and trial managers on their research teams. And, of course, many RDS advisers who support such studies ourselves have this experience.
What is particularly relevant about adaptive trials, not least in the context of providing advice for health researchers, is the opportunity for flexibility they offer. The goal of an adaptive design is to enable researchers to learn from the accumulating data and make key design changes accordingly as things progress. Usually, while a trial is still in progress, we do not look at any of the data. However, the accumulating material may be more informative than that available before the trial had started. And it is this previously available data – things like effect sizes, recruitment strategies and randomisation, dosages – on which the design of the trial was based. The idea behind adaptive designs is that a trial can be improved by making use of interim data to refine certain aspects of it.
Improvement can mean a number of things, usually to do with making a clinical trial more efficient. It can, for example, lead to doing away with a treatment arm if a particular dosage or intervention is demonstrated to be inappropriate. It can also mean identifying with greater accuracy the number of patients needed or lead to refinements in the recruitment strategy, target population or treatment randomisation. It can even lead to decision-making about, amongst others, key objectives, end-points, test statistics, or subsequent phases of the research.
The trick to it all is in the pre-specification: making it very clear in the trial protocol what purpose will be served by carrying out an interim analysis, when it will be done, on what measures, by whom, and to what end. You need to put procedures in place to ensure that the blind holds and limit the people who see the data at this pre-specified interim stage.
As an RDS adviser, I can see merit in this approach. The ability to carry out a sample size re-estimation, for example, is something that I think could benefit many projects. So too is the ability to drop inferior treatment groups or use preferential randomisation. There are, of course, many other options for adaptive designs, and these are just two examples that I can think of which could be applied to many projects.
The drive for efficiency in health research is nothing new. I’ve written about it before and there are examples of it in the NIHR funding programmes themselves – a recent example being the call the HTA programme issued in the summer of 2014 for ‘efficient study designs’. And, as always, an integral part of any NIHR funding application is the demonstration of the value for money of the research itself. This is, of course, not to say that adaptive trials are the answer or even appropriate. Such designs come with their own risks – errors due to their greater complexity, more time needed in the planning stages, uncertainties around the ethical implications, and the need for greater regulatory review, to name just a few.
As always, the design of a study needs to fit its research question. Adaptive trials offer an intriguing option when uncertainties mean refinements are required during the trial itself in order to optimise its design.