An Efficient Oncology Phase II/III Development Strategy Using Response and Overall Survival Information

Cancer has a big impact on public health. The years of life and productivity lost to diseases and death, and the economic burden to the community make cancer a focus of public health. Efficiency in drug development in oncology area is critical as it impacts millions of families. In clinical development, there is a trade-off between investment and level of confidence in the potential of the drug before going into phase III. Contemporary approaches of decision making with reduced investment often require use of short-term endpoints or limited information on long-term endpoints. The probability of success for a future phase III trial based on long-term endpoints is the measure of confidence. Both aspects can be integrated within a framework that models the probability of success using both short-term and long-term endpoints data. This dissertation compares current methods of determining the probability of success and proposes a mixed approach that combines the Bayesian and frequentist ideas to utilize individual-level information from a multi-categorical short-term endpoint (response status) and a long-term endpoint (overall survival). The Bayesian paradigm is used to obtain predicted overall survival of censored or not-enrolled-yet patients, while the analysis for estimating the treatment effects is based on frequentist methods to comply with regulatory requirements of controlling type I error. During the development of a new medication, usually there is limited reliable prior information about the relationship between response and overall survival. Therefore, the proposed approach utilizes the ‘phase 2+’ design concept that further follow-up data of the long-term endpoint from phase II, after a go-to-phase-III decision (1st look), are considered for further decision making (2nd look) during phase III. The proposed method was evaluated and compared with current approaches in a simulation study. The results show that the proposed method with a planned 2nd look benefits estimating the probability of success and could provide a customized development planning strategy with respect to quality of decision making and program duration. Furthermore, various scenarios such as informative prior on control arm, heterogeneity in response status between phase II and phase III trials, different sample sizes of phase II trials and side conditions for decision making and one-trial design were assessed.