Our statisticians and statistical programmers contribute widely to conferences and publications to help advance the knowledge base of our community. In this Q&A, we talk with Jemma Greenin, a Statistician II at Veramed, about her contribution to the 2021 PSI conference discussing Multiple Testing and Combination Testing for Treatment Selection, Adaptive Seamless Designs. Jemma was inspired to produce this work following an industry placement as part of her post-graduate degree. Adaptive and innovative trials are promising tools to save time, minimise sample size, and improve trial efficiency. ‘Seamless’ trials, where traditional phases are combined, can help to streamline the conventional clinical development pathway, and make better use of available data. However, there remain barriers to adoption as the designs may be seen as excessively elaborate, requiring complex simulations. In this interview, Jemma shares her insights on the approach’s multiple and combination testing requirements, and we offer her slides for those who may have missed the live presentation.
How did you become interested in adaptive designs?
I was working at a pharmaceutical company on placement, and they were planning an adaptive seamless design. I found the concept compelling and was intrigued by the potential to improve the way we approach clinical development. I decided to advance my knowledge by tackling some of the intricacies of adaptive seamless trials for my dissertation.
What statistical problem or issue were you trying to solve?
Traditional trial designs can be unwieldy and costly. Adaptive seamless designs are attractive as they can help sponsors reduce study durations and make better use of scarce patient resources while increasing efficiency. In a typical clinical development pathway, with separate Phase 2 and 3 trials, valuable data gained from the Phase 2 study is not used again at the confirmatory stage. By combining phases into a seamless approach, we can maximise the value of this data that would otherwise be redundant. However, there are problems that we need to resolve as Statisticians to ensure that the trials are robust. These issues encompass multiple testing, combination testing, and maintaining power.
In a seamless design, as we are looking at the same data in each stage, we need to consider the impact on the analysis and account for any bias that might arise from reusing the data. We also need to combine hypotheses from each phase without losing power and select the best treatment for the seamless transition- this is where combination testing comes in.
What methodologies did you use?
In my presentation, I compared the performance of three approaches based on how well they maintained power.
The first was a conventional method with Phase 2 and 3 designed as separate studies, the second a methodology by Thall, Simon and Ellenberg, which provides a calculation for combining trials. Then, I examined a combination and multiple testing approach by applying a reasonably straightforward Bauer and Köhne method. This latter method proposes a treatment design where you start with stage one, the Phase 2 trial, and calculate the P-value. Then multiple testing is applied at that point, followed by Phase 3 using combination testing to combine the two hypotheses. Within this approach, I also weighed up several different options for multiple testing and combination testing. I compared the overall Bauer and Köhne methodology to both the Thall, Simon and Ellenberg approach and the conventional design using simulations.
What was your conclusion?
I ultimately preferred the Thall, Simon and Ellenberg methodology as it was the most conservative and demonstrated the best power – which was my initial basis for comparison. While there was a limited difference in power between the methods, it was clear that as statisticians we must select our methodology carefully to maximise success.
I hope that my presentation can inspire others that it is possible to combine Phase 2 and Phase 3 trials successfully, without loss of power. I would like to see the industry overcome any inhibitions about adaptive seamless trials as they can be both practical and efficient while minimising patient burden.
To learn more about our statistical consultancy services, click here.
To download the slide deck, fill out the form below.