Incorporating MCP-Mod into ADDPLAN DF

Dose-finding remains a challenge in clinical trials. New methods are emerging, however, that aim to resolve the uncertainty over predicting which doses will prove to be effective in a dose-finding study. In a recent blog, we discussed the utility of MCP-Mod in determining the optimal dose for a Phase III trial by using model based design and analysis techniques. The EMA and FDA both issued opinions on the MCP-Mod method, promoting its use in adaptive designs and its value in improving dose-finding trials.

ICON has developed an update to its ADDPLAN DF platform that includes new features for adaptive and fixed MCP-Mod designs and simulations. We spoke with Tobias Mielke, Ph.D., a statistical consultant in ICON’s Innovation Centre, to discuss why adaptive MCP-Mod was incorporated in ADDPLAN DF and what advantages it provides to sponsors when planning and designing their studies.

Why were adaptive MCP-Mod designs the focus for this ADDPLAN DF update?

Dose-finding studies are challenging to design in part because current analytical methods, namely pairwise comparisons and modeling, can yield uncertainty regarding the selection of most informative Phase II study doses. Pairwise comparisons typically aim for the inclusion of effective doses to show positive effects, while model-based designs typically target suboptimal doses, as well, to resolve the steep part of the dose-response curve. A good dose-finding study requires a compromise between both targets. After running a trial, there should be enough evidence for phase III selection as well as enough certainty that the selected doses are close to maximal effective dose.

MCP-Mod is a model-based trial design and analysis method that, in combination with an adaptive design, can significantly enhance definition of the optimal dose for a drug in a clinical trial by reducing the uncertainty surrounding dose-finding. Primary sources of uncertainty during the initial design phase of a dose-finding study are the shape of the dose-response relationship and the effectiveness of the drug. The use of adaptive MCP-Mod designs targets both uncertainties in an efficient manner. Interim analyses make trials more efficient by flagging futile trials early, and by providing data to optimise next stage patient allocation through the inclusion of additional dose-groups, dropping doses early or by simply continuing as originally planned.

MCP-Mod is a critical piece of adaptive designs because an interim analysis using MCP-Mod takes the uncertainty on the correct dose-response model into account. This leads to more efficient futility rules and informs the allocation of remaining patients to more effective treatment groups for further data collection. This midstream change ensures an efficient study design; prevents sponsors from needing to start a new study from scratch with a new dose range; and supports early phase III preparations using predictive model-based simulations. Due to the utility of MCP-Mod, the EMA issued a qualification opinion on MCP-Mod, and it was designated as fit-for-purpose by the FDA.

Could you walk us through a trial simulation in ADDPLAN DF, and show us how the platform could help improve a study’s utility for resolving the most interesting region of the dose–response curve?

The ADDPLAN platform allows a sponsor to design, simulate, and analyse multi-stage adaptive designs, wherein a trial begins with one set of treatment arms and is modified midstream depending on the likelihood of models that are derived from the results of the previous stage.

For instance, a sponsor may want to run a two-stage adaptive seamless Proof-of-Concept Dose-Finding trial (with one interim analysis) that begins with two doses of an investigational drug and placebo (Fig 1). The purpose of this first stage is to detect an initial sign of efficacy and an indication of the dose-response relationship. For this, we suggest that patients are allocated only to the control arm, the top dose, and an intermediate dose to reduce the number of patients in case of futility. The intermediate dose provides valuable information for guiding dose selection for the second dose-finding stage.

For the second stage, the sponsor can define the allocation rules of patients to different doses depending on the results of the first stage (Fig 2). The rule displayed in Figure 2 is used to define upfront a set of dose-response scenarios. Given the interim data, the statistical algorithms will evaluate which of the scenarios is closest to the observed data. Based on these results, the algorithms will select the most informative allocation for the next stage. The scenario based-rule is a very valuable approach, both for design discussions as well as for the execution of the study design.

In ADDPLAN DF, the sponsors can choose which dose-response models to use for the actual data analysis (Fig 3) and for the generation of simulation data. The models for the data analysis will be used for interim modelling and testing and can also guide adaptive allocation rules when using adaptive optimal designs (D-optimality) or best-intention designs, which will increase patient allocation to the estimated optimal doses.

The simulation functionality of ADDPLAN DF will evaluate the proposed adaptive study designs under all specified simulation models for a range of different effect scenarios, providing an extensive output describing the quality of the design. This will support further fine-tuning of the adaptive design to improve the study design for an optimal Phase III dose selection.

Figure 1

Figure 1. Design Stage 1 will use three equal-sized treatment arms, doses 1, 5, and 9. The study will be stopped for futility if the benefit at both doses 5 and 9 is significantly below 5 units (at α=10%).

Figure 2

Figure 2. The sponsor can set up multiple scenarios that would lead to different treatment selections in stage 2 depending on the results of the first stage. In this example, the sponsor targets an allocation to the top doses, in case that first stage data is close to the specified exponential shape.

Figure 3

Figure 3. The sponsor can choose which dose-response models to use for the actual MCPMod analysis and how to weight these models when using model-averaging techniques or adaptive optimal designs.

You also introduced new features for feasibility checks when considering an adaptive design for a trial, as well as for recruitment planning. How do these new features streamline early decision-making about the relevancy and utility of adaptive designs for a given clinical trial?

Specifying the design of an adaptive trial and simulating the study is sometimes a time-consuming process. One way to optimise the process is to work in the following order:

  1. Test the feasibility of an adaptive study design.
  2. Evaluate potentially relevant adaptive designs using a small number of simulations.
  3. For the best performing candidate adaptive designs, increase the number of simulations for a more refined insight into operating characteristics and further fine-tune these designs.

Sponsors can quickly evaluate the feasibility of the adaptive approach using the recruitment functionality. In cases where the study population will be recruited quickly for indications with a long-term endpoint, the application of adaptive designs can be operationally infeasible. This is because, if all patients are already enroled in the study prior to the first interim analysis, there is no opportunity to adapt the study design following the interim analysis, and therefore it loses most of its value.

This issue could be circumvented by decelerating or interrupting the recruitment process prior to the interim analysis, what could introduce additional logistical problems introducing potentially operational bias. Short-term endpoints, which reliably predict the long-term endpoints frequently do not exist, such that it will be difficult to predict the study outcome. In this situation, an interim analysis will be of limited value. While an interim analysis may allow for predictive simulations and guide early development preparations for following Phase III studies, it will not be possible to meaningfully adapt the study at hand. Before spending time optimising adaptation rules, the sponsor should verify whether there are actually any patients available to apply these adaptations to.

The ADDPLAN DF recruitment functionality resolves this issue by estimating the number of patients enroled in each stage of the study, and defining how the interim analyses should be timed, using flexible probabilistic models on the recruitment rates and the number of sites in different regions. This functionality supports the determination of operationally feasible and efficient study designs – with and without interim analyses.

Learn more about ADDPLAN DF or download the ADDPLAN white paper.