FDA draft guidance on “Co development of two or more unmarketed investigational drugs for use in combination” notes that:
“Combination therapy is an important treatment modality in many disease settings, including cancer, cardio-vascular disease, and infectious diseases. Recent scientific advances have increased our understanding of the pathophysiological processes that underlie these and other complex diseases. This increased understanding has provided further impetus for new therapeutic approaches using combinations of drugs directed at multiple therapeutic targets to improve treatment response or minimize development of resistance.” In this setting, it’s important to be able to design dose escalation studies which can identify the synergistic activity of compounds, and less toxic combinations.
Historically, the vast majority of Phase 1 dose escalation studies have used rule based approaches. Clearly, rule based designs such as 3+3 have advantages in that they are easy to deploy. However, there are also well documented limitations including the lack of ability to incorporate prior knowledge- only considering the outcome of the last cohort to guide escalation. It has also been previously noted that 3 + 3 in fact has a low probability of selecting the true MTD. We may therefore consider alternative statistical designs which have been developed to reflect the changing needs of the dose-finding paradigm.
To streamline the process for statisticians looking to explore and implement novel dual agent designs, within the ESCALATE module in this Spring's upcoming EAST 6.4 release Cytel has incorporated two published Bayesian based methodologies. These are the Bayesian logistic regression model (BLRM) for two agents, and the Product of Independent beta Probabilities dose Escalation (PIPE) method.
PIPE and BLRM dual agent approaches in EAST
The two approaches were discussed in depth at the recent East User Group Meeting. We were delighted to have the MRC Biostatistics Unit’s Adrian Mander presenting on the PIPE method and the ORCA trial and Cytel’s Pantelis Vlachos presented and delivered a talk and workshop on Single and Dual Agent Dose Escalation designs including discussion of the BLRM two agent approach (Neuenschwander et al.). Pantelis' slides are available for download below.
In his talk, Pantelis outlined the factors which can be seen in common in the two approaches. The common approach starts by specifying a dose-combination for first cohort and recording the observed number of toxicities. Given a parametric dose-toxicity model, π(x, θ), with priors on the parameter vector θ, the method proceeds by obtaining the posterior distribution of θ and using it to choose the next dose combination out of a set of admissible dose combinations. The recruitment of patients continues until either a fixed sample size is obtained or a stopping rule is satisfied.
A product of Independent beta probabilities dose escalation design for dual-agent Phase I trials.
Stat. Med. 34 (8): 1261-1276 Mander, A.P. and Sweeting M.J (2015)
Adaptive designs for dual-agent phase I dose-escalation studies. Nat Rev Clin Oncol. 2013 May;10(5):277-88. Harrington JA1, Wheeler GM, Sweeting MJ, Mander AP, Jodrell DI.
A Bayesian Industry Approach to Phase I Combination Trials in Oncology (In book Statistical Methods in Drug Combination Studies) Beat Neuenschwander, Alessandro Matano, Zhongwen Tang, Stuart Bailey