A modelling framework for improved design and decision-making in drug development

Abstract

The development of a new drug is an extremely high-risk enterprise. The attrition rates of development projects and the average costs for each launched product are daunting, and the completion of a development program requires a very long time horizon. These facts imply that there are huge potential gains, should one be able to improve efficiency and enhance decision-making capabilities. In this paper, we argue that substantial gains can be achieved by adapting a holistic view of drug development. Historically, too much planning, design and decision-making in the pharmaceutical development has been based on locally optimising separate parts of the development program, and too often important sources of uncertainty are ignored. We propose instead a model-based approach built on two essential pillars; (1) an integrated holistic view of the development program, including post-launch marketing and sales, with all parts evaluated simultaneously; (2) an explicit appreciation of all relevant sources of uncertainty. Computer simulations are utilised to evaluate the properties of the program options at hand, and to provide valuable quantitative decision support. Applications of this modelling approach have proven to add large value to development projects in terms of better program options being generated and more value-adding decisions taken.

Fig 1. A generic example of a flow chart representation of a drug development project.

Fig 2. An example of a distribution of the true effect for subsets of simulated compounds progressing through development.

(The horizontal axis is an arbitrary scale of treatment effect. The density functions have been scaled proportional to the proportion of compounds remaining after each phase).

Fig 3. Schematic illustration of the cash flow profile from a development program.

Fig 4. Process model of the development program used for illustration.

Fig 5. The distribution of the true effect on overall response rate.

Distributions are given for all iterations, as well as for subsets of iterations progressing through the different paths of development. (The density functions have been scaled proportional to the proportion of iterations in each case).

Fig 6. The mean net present value for a range of stop/go criteria applied after the expansion trial.

Fig 7. The expected net present value as a function of the sample size of the randomized standard development trial.

Fig 8. The cost incurred throughout the development program, illustrated by box plots over time, for the subset of simulation iterations leading to a launch.

Fig 9. The cost incurred throughout the development program, illustrated by box plots over time.