Conference Agenda

Since 2007 there are recommendations for EU regulatory assessors and sponsors seeking marketing approval on trials with design modifications based on the results of an interim analysis in the Reflection Paper on Methodological Issues in Confirmatory Clinical Trials Planned with an Adaptive Design. In 2019 a harmonisation process was initiated between worldwide regulatory agencies, Industry associations and other public health stakeholders. This ICH guideline focuses on the design, conduct, analysis, and interpretation of adaptive clinical trials and is expected to provide a transparent and harmonised set of principles for the regulatory review of these studies in a global drug development programme. This presentation will explain the opportunity given by ICH E20 to rediscuss principles for adaptations in confirmatory clinical trials and describe the key considerations for these types of clinical trial in a regulatory setting, in order to motivate comments on the draft ICH E20 during the public consultation.

Platform trials are randomized clinical trials designed to simultaneously compare multiple interventions, typically against a common control group. Arms to test experimental interventions may enter and leave the platform over time. Therefore, the number of experimental intervention arms in the trial can change over time. Determining the optimal allocation rates for assigning patients to the treatment and control arms in platform trials poses a challenge. As the treatment arms enter or exit the platform, the optimal allocation rates also need to be adjusted. Additionally, the optimal allocation strategy depends on the specific analysis method used.

In this talk, we describe optimal treatment allocation rates for platform trials with shared controls, assuming that a stratified estimation and testing procedure based on a regression model is used to adjust for time trends. We consider analysis methods using concurrent controls only as well as methods based on also non-concurrent controls. We show that to minimize the maximum of the variances of the effect estimators, the optimal solution depends on the entry time of the arms in the trial. Generally, this solution does not correspond to the square root of k allocation rule used in the classical multi-arm trials. We illustrate the optimal allocation and evaluate the power and type 1 error rate compared to trials using one-to-one and square root of k allocations. In addition, we will discuss extensions, such as allocations in trials with varying eligibility and inclusion criteria and optimal allocation to the control group, for designs where the allocation rates to the treatment arms are equal.

Background

With the growing importance of clinical trials in targeted medicine there has been recent interest in adaptive enrichment designs [1]. In these two-stage designs patients from the first stage are used to identify a biomarker-defined population in which a treatment effect is anticipated. In the second stage the trial population is ‘enriched’ by restricting recruitment to patients from the selected population. At the end of the trial a hypothesis test is conducted of the treatment effect in the selected population. The data-dependent selection leads to statistical challenges if data from both stages are used for this hypothesis test.

Methods

If the biomarker is measured on a continuous scale, with any biomarker by treatment interaction assumed to be monotonic, population selection is equivalently to identification of a cut-point for the biomarker. In this case subgroups considered are nested and the multiple testing procedure can be considered in a hierarchical fashion, enabling control of the familywise type I error rate (FWER) through a simple closed testing procedure given a valid test based on data from each possible selected subgroup.

Focussing on the case in which the outcome is normally distributed, two methods for this test are proposed. The first assumes selection of the subgroup with the largest test statistic when the distribution of this test statistic can be obtained by considering the joint distribution of the test statistics from different subgroups [2]. In the second selection is based on a fitted linear relationship between the outcome and the continuous biomarker [3].

Results

The second approach proposed is shown to be more powerful when the linear model assumptions are met, but can lead to type I error rate inflation when they are violated whereas the former method can be less powerful but provides FWER control irrespective of the relationship between the biomarker and response [3].

References

[1] Simon, N., Simon, R. Adaptive enrichment designs for clinical trials. Biostatistics, 14, 2013, 613-625.

[2] Stallard, N. Adaptive enrichment designs with a continuous biomarker. Biometrics, 79, 2023, 9-19.

[3] Stallard, N. Testing for a treatment effect in a selected subgroup. Statistical Methods in Medical Research, 33, 2024, 1967-1978.