Adaptive Informational Design of Confirmatory Phase III Trials With an Uncertain Biomarker Effect to Improve the Probability of Success

ABSTRACT

Oncology drug developers sometimes decide to initiate Phase III randomized confirmatory trials at risk after significant preliminary anti-tumor activities are observed in small Phase I/II single arm studies. There are two clear challenges. First, these investigational drugs may have a greater benefit in a biomarker enriched population. But the limited data from Phase I/II can hardly provide the much-needed information for selecting a biomarker cutpoint or prioritizing a biomarker hypothesis for Phase III testing. Second, the data seldom provide any insight on how the treatment benefit evolves over time. Risk-mitigation strategies such as conventional adaptive-designs that rely on interim analyses for modifying the study design are less reliable because the treatment effect observed at an interim analysis may not be the same as in the final analysis. The use of an intermediate endpoint for interim decision makes it even more unreliable because the predictive value of an intermediate endpoint is often unknown for drugs with a new mechanism of action. In this article, we present an alternative design strategy to mitigate the risks. The idea is to add an analysis of the primary endpoint at the end of the Phase III trial in a subgroup of patients representing the overall study population. We call it informational analysis and the corresponding design informational design to emphasize its difference from the conventional event-time or calendar-time-driven interim analysis. From a high-level perspective, the subgroup analysis is equivalent to a Phase II trial conducted under the same study design at the same time in the same population at the same sites as the Phase III trial. It provides a more reliable resource of information for inference than a separate Phase II trial or a conventional interim analysis. The strategy is applied to address a wide range of statistical issues encountered in expedited development of personalized medicines, including alpha splitting between a biomarker subpopulation and the overall population and de-selection of nonperforming biomarker subpopulations. Applications to hypothetical Phase III trials are illustrated. Although the strategy is motivated by oncology studies, it may be applied to drug development in other therapeutic areas with similar concerns. Supplementary materials for this article are available online.

Keywords:

• Adaptive design
• Bayesian decision analysis
• Informational analysis
• Seamless Phase II/III
• Sub-study

Supplementary Materials

The online supplementary materials contain the Branchburg Seminar, presented by authors Cong Chen and Nicole Li.

Acknowledgment

This research is part of the collaborative work with the DIA Small Population Work Stream under the DIA Adaptive Design Scientific Working Group. During the preparation of the article, the authors have received helpful comments from Sue-Jane Wang. The authors have also received valuable comments from two anonymous referees, which have helped greatly improve the presentation of the article.