Methods for complex trial designs with shared controls

Efficient generation of robust evidence on the safety and efficacy of novel treatments is paramount in confirmatory clinical trials. Complex trials can accelerate drug development by evaluating the efficacy of multiple experimental treatments against a shared control. These trials can also allow for the later integration of new treatment arms, which is especially useful when not all treatments are available at the start of the trial. Furthermore, the control treatment might also change during the trial. Complex trial designs pose statistical inference challenges due to potential changes in the number of treatment arms and the control treatment, which must be taken into account during both the design and analysis stages. In this project, we address two key statistical questions in the context of confirmatory complex trial designs with several treatments and shared controls: First, when newly approved treatments become available and it is no longer ethically justifiable to treat patients with the original comparator treatment, the control treatment has to be changed. This is particularly relevant for multi-regional trials or trials running over long periods of time. We will develop trial designs and testing strategies for trials that allow a change of the treatment of the control during the trial. In addition, we plan to extend the methods to adaptive designs that allow for sample size adjustment, and investigate the effects of different decision rules for control treatment change. In the second, the focus is on adaptive group sequential platform trials that allow for design modifications based on interim data, such as early termination of study arms based on meaningful interim data. Standard methods for group sequential designs fail here because they cannot guarantee control of the type 1 error and a valid estimate of the treatment effect. Therefore, in this project we will develop tailored statistical testing and estimation methods that ensure reliable control of the error rate and are suitable for both designs using only concurrent controls and designs using also non-concurrent controls. Addressing the statistical challenges of complex trial designs is particularly important in the regulatory context. Although regulatory authorities are already working on appropriate guidelines, there is still no consensus on important aspects of the analysis methods of innovative, adaptive trial designs. The project comes timely at a crucial moment to clarify the open questions addressed and thus support the implementation and analysis of confirmatory adaptive complex trials.

Efficient generation of robust evidence on the safety and efficacy of novel treatments is paramount in confirmatory clinical trials. Platform trials can accelerate drug development by evaluating the efficacy of multiple experimental treatments against a shared control. These trials can also allow for the later integration of new treatment arms, which is especially useful when not all treatments are available at the start of the trial. Furthermore, the control treatment might also change during the trial. Platform trial designs pose statistical inference challenges due to potential changes in the number of treatment arms and the control treatment, which must be taken into account during both the design and analysis stages. During this project, we worked on developing new methods and software for the analysis of platform trials utilising non-concurrent controls. In platform trials, treatment groups enter into the trial as they become available, which means that not all treatments are studied at the same time, and therefore, patients are randomised at different times to the arms available at that moment. For those treatments added later, we are working on methods that incorporate previous data from randomised patients in the control group (so-called, non-concurrent control data) but adjust the analyses to avoid bias. In recent years, researchers have developed new ways to analyse data that account for changes over time while utilising non-concurrent control data. One method involves including time as a factor in the analysis to model the effect of time in the study. In our project, we also improved and adapted these methods for different situations by introducing more advanced techniques, like mixed models and polynomial splines. A key challenge with existing methods is the assumption that changes over time affect all groups in the same way. However, this is not always true, especially in cases where different disease variants might respond differently to treatment. Therefore, we proposed new ways to relax this assumption by using a mixed model that considers the interaction between treatment and time.