Dynamically Optimizing Budget Allocation for Phase 3 Drug Development Portfolios Incorporating Uncertainty in the Pipeline
In this chapter we describe a mathematical approach to maximize the Expected Net Present Value (ENPV) of a portfolio over a planning horizon by determining the optimal designs of Phase 3 (Ph3) trials for a given budget. The model is formulated as a Stochastic Integer Programming (SIP) model that incorporates uncertainty regarding availability of drugs in the pipeline. The SIP model provides an optimal policy that specifies the optimal design for each drug for every possible scenario of availability of future drugs for Ph3 trials. It optimizes the trade-off between committing budget to drugs available for Ph3 funding at any point in time and preserving budget for drugs in the development pipeline that will need funding in the future. Optimizing this trade-off is challenging because it is uncertain which drugs will need funding in the future as they may fail to progress to Ph3. This important trade-off is not handled in a consistent, quantitative way in portfolio budgeting models used in practice today. We have also developed a simulation model to assess the technical, regulatory, and commercial risk of the optimal budget allocation policy. We show how our models can be used for dynamic re-optimization of the portfolio when changes in the internal and external environment occur and as new information becomes available. This capability will enable rapid, frequent, and consistent realignment of the strategy to optimize future use of the budget available for reallocation.
We illustrate our approach using an example that shows how our models can be used to decide on the best budget level to meet a target Return on Investment (ROI) and to evaluate risk of the optimal allocation strategy associated with this budget level. We have shown how these models can be used to answer important what-if questions such as those that arise when in-licensing or out-licensing drug development.
KeywordsPortfolio optimization Budget allocation Decision analysis Stochastic integer programming Risk Re-optimizing budget allocation In-licensing Out-licensing
We are indebted to Kraig F. Schulz of Ernst & Young, LLC for help in constructing a realistic dataset for our example and to Jaydeep Bhattacharyya of Cytel Inc. for assistance in testing our computer programs.
The SIP model for our example has been implemented in the mathematical programming language, AMPL . An Excel front-end, SolverStudio , was used to run the model using CPLEX  and Gurobi  solvers on a PC. The model was also tested on the publicly available Neos server  on the Web using the “AMPL on Neos” feature of SolverStudio with the Cbc solver . We sincerely thank the providers of these software tools.
- 2.Sharpe P, Keelin T (1998) How SmithKline Beecham makes better resource-allocation decisions. Harv Bus Rev 76(2):45–46Google Scholar
- 3.Martello S, Toth P (1990) Knapsack problems. Wiley, ChichesterGoogle Scholar
- 9.Patel N, Bolognese J, Chuang-Stein C, Hewitt D, Gammaitoni A, Pinheiro J (2012) Designing phase 2 trials based on program-level considerations: a case study of neuropathic pain. Drug Inf J 46:439–454Google Scholar
- 11.Bodie Z, Kane A, Marcus A (2012) Investments, 10th edn. McGraw-Hill-Irwin, New York City, NYGoogle Scholar
- 13.http://ampl.com/products/ampl/ Accessed on May 13, 2014
- 14.http://solverstudio.org/ Accessed on May 13, 2014
- 15.http://ampl.com/products/solvers/solvers-we-sell/cplex/ Accessed on May 13, 2014
- 16.http://ampl.com/products/solvers/solvers-we-sell/gurobi/ Accessed on May 13, 2014
- 17.http://neos-server.org/neos/ Accessed on May 13, 2014
- 18.https://projects.coin-or.org/Cbc Accessed on May 13, 2014