Drug development pipeline is composed of projects which are being evaluated in a series of clinical trials with uncertain outcomes. Advancing the pipeline is a race against time, given that most investments can only be recouped by selling the successful projects during the market exclusivity period counted from the ﬁling date. In this study, we introduce a novel optimisation methodology to manage the research and development pipeline by jointly optimising the sequence of clinical trials and the allocation of resources. Our model speciﬁcally allows to speed up clinical trials by cumulatively allocating additional resources. A discrete-time stochastic Markov decision process is used to formulate the system dynamics. The state space increases dramatically when scaling up the drug pipeline as a consequence of ﬁnding the optimal policy within a reasonable time span is not realistic. We propose an adaptive rollout algorithm that includes three computationally eﬀective innovations. First, the rolling-horizon optimisation framework constructs a base policy that estimates the expected future outcomes after taking a particular action. Second, a statistical racing procedure, which exploits the technique of common random numbers and the empirical Bernstein’s inequality, is proposed to guide the information collection process. Third, we present a hierarchical approach which priorities the actions that makes the best use of available resources so that good actions could be identiﬁed at early stage. The numerical results show that the hierarchical and normal statistical racing procedures reduce the number of required simulations, thus providing high-performance results within a reasonable computational time.
|Journal||European Journal of Operational Research|
|Publication status||Submitted - 2020|
- Dynamic Programming
- Clinical Trial Scheduling
- Resource Allocation
- Rollout Algorithm