Main page > Innovation > Flexible designs

An interview with Dr. Christian Sonesson (Senior Statistician, AstraZeneca) and Dr. Carl-Fredrik Burman (Statistical Science Director, Technical and Scientific Development, AstraZeneca) conducted by Alex Dmitrienko.

Dr. Christian Sonesson is the Sweden Implementation Coordinator for the ADDAMS project (Adaptive Design, Decision Analysis, Modelling and Simulation) and Global Lead for the Decision Analysis Team.

Alex Dmitrienko: Adaptive designs have been used in dose-finding studies in oncology for several decades (traditional up-and-down designs, continual reassessment designs, etc). What are other popular approaches to designing adaptive dose-finding studies and in which areas have they been used?
Christian Sonesson and Carl-Fredrik Burman: In cases where we a priori can predict the dose-response curve (for effect and safety) with high certainty, we don't need an adaptive approach. A fixed design study could confirm the model or we could even skip the dose-finding trial and go directly to the confirmatory phase. However, the more uncertain we are about the dose-response, the greater the value of adaptive approaches. The most efficient trial will choose the most informative dose for each patient. This is possible in some situations but usually a more simple adaptive design is more practical. One example is to start with a rather large number of doses and then stop arms for which the effect is insufficient or safety inadequate. AstraZeneca's new ADDAMS (Adaptive Designs, Decision Analysis, Modelling and Simulation) project involves the planning of two adaptive dose-finding trials in non-oncology areas.

What are the main objectives of adaptive designs in dose-finding studies? Optimize the design to maximize the amount of information about a single target dose, optimize the design to maximize the amount of information about the overall dose-response relationship for an efficacy outcome variable, etc? The answer is most likely indication-specific. For what classes of indications does each of these objectives makes most sense?
To formulate the precise objective is critical and also very hard in a practical situation. Finding the best dose is about striking the best balance between risk and benefit. In practice it is often of most interest to maximize the information on the dose response within the therapeutic window. Ultimately, we would like to find the dose that maximizes the expected Clinical Utility Index (CUI) for the patients, where the CUI weighs together the benefits and safety risks on a common scale. However, it is often difficult to agree prior to the study on which weights to use when aggregating different benefits and risks. This problem is even greater when we don't know which side effects to expect. In some cases, it may be useful to focus, for example, on the dose giving a certain effect and/or the dose where the marginal effect gain is small or more precisely, the dose for which the derivative of the effect with respect to log(dose) is a specified constant. It is also important to understand that the risks and benefits of the study patients have to be considered in the design. We should always look within the subset of ethical designs to find as efficient design as possible.

Under what conditions (or for what types of indications) do clinical trial researchers need to consider joint modelling of efficacy and safety outcomes in adaptive dose-finding studies?
It's always important to consider both effects and side effects. Thus, in principle, joint modelling of efficacy and safety should always be done. However, in some cases we don't have safety data and then safety modelling is more like trying to guess the impact of increasing the dose, possibly based on knowledge from other drugs in the same class. What's important is to get a grip on how large this uncertainty is. This is an important factor in optimising the design. If safety and/or effect are very uncertain, that means that a more careful design is needed. There is always a risk to focus the dose allocation based on a particular safety outcome. If we use an unsatisfactory safety criterion in our allocation rule, we might allocate too many subjects to high doses and thus get limited information on lower doses within the true therapeutic window. We must make sure that, when data become available, there is sufficient information on the whole dose range to assess the risk/benefit on a range of doses. The use of composite endpoints could be one alternative given that it has some clinical meaning. The risk though is that we might miss an imbalance between the dose arms on one of the variables within the composite endpoint.

Under what conditions do clinical trial researchers need to explore adaptive dose-finding studies based on early markers of efficacy (e.g., biomarkers or partial responses)?
In most cases, we only have access to data on a biomarker or on surrogate endpoints and we think this question goes beyond the discussion on adaptive designs. In many areas the lack of a validated biomarker is possibly the largest challenge in the whole clinical development and that is true whether you want to do a fixed or an adaptive trial. For seamless Phase II/III designs in oncology, an important surrogate endpoint is the progression free survival which can be used in the intermediate step whereas the final analysis is based on overall survival. In general, it is advantageous to consider variables with a fast effect in an adaptive study in order to get early information to use for the allocation of new subjects. What is deemed as "fast" is a matter of comparing the time it takes to get valuable information with the enrolment rate.

Under what conditions do clinical trial researchers need to consider continuous (each new patient is assigned to his/her dose) and group-sequential (a group of patient is assigned to a dose) allocation of patients in adaptive studies?
Continuous adaptations are more efficient in terms of how much useful information can be obtained per patient but this has to be contrasted with the greater simplicity of having only a small number of interim analyses. From the design efficiency perspective, the gain of having many interim analyses is not very large. What could motivate a continuous allocation is the potential safety concerns for the patients in the study. In general, we would say that a careful modelling and formulation of objectives is often the most important thing and more important than the adaptive design itself. It was stated at the 2006 PhRMA/FDA workshop on adaptive designs that "to have considered an adaptive design and done a good homework will leave you in a far better position even if you decide to use a traditional fixed design in the end".