Pith. sign in

REVIEW 3 major objections 7 minor 7 references

Project portfolio planning in the pharmaceutical industry -- strategic objectives and quantitative optimization

T0 review · 3 major / 7 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A pharmaceutical company can optimize the in-flow of new development projects—how many, when, and in which disease area—to meet revenue targets, budget limits, and portfolio-balance constraints.

desk verdict A coherent extension of the authors' own inflow-optimization framework with useful balance constraints, but the revenue calculation mixes deterministic and stochastic durations and the illustrative example is under-specified. read the letter →

arxiv 2502.09527 v1 pith:HVT4OPWC submitted 2025-02-13 stat.AP

Stig Johan Wiklund , Magnus Ytterstad , Frank Miller This is my paper · ORCID
classification stat.AP
keywords DrugdevelopmentPortfoliocompositionBudgetconstraintsRevenuetargetsProjectinflowoptimizationSimulatedannealingPharmaceuticalR&Dplanning
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper claims that the future in-flow of new projects into a pharmaceutical development portfolio can be planned quantitatively rather than merely reacted to. It frames an optimization problem in which the decision variables are the number of projects entering Phase 1 in each year and each disease area, and the goals are revenue targets, development-budget limits, and constraints that keep the portfolio balanced across phases and disease areas. Eight ways of stating the strategic objective are solved—minimizing cost or maximizing revenue, with the target measured either as a mean over time or year by year—using simulated annealing. If the approach is right, a portfolio manager gets a concrete view of how many projects to start, in which disease areas, and in which years, and can see how the recommended plan shifts as the objective is framed differently.

What carries the argument

The engine is the phase-occupancy probability $\pi_{ijt}$, defined as the probability that a project entering development in disease area $j$ is active in phase $i$ at time $t$, built from the product of phase-transition probabilities over earlier phases. Expected portfolio metrics—number of projects active in a phase, number active in a disease area, risk-adjusted development cost $\Gamma_t$, expected launches, and expected revenue—are written as sums of these probabilities weighted by the inflow counts $N_{j\tau}$. This makes the objective functions and constraints linear in the decision variables, so the yearly number and timing of project starts can be chosen by combinatorial optimization, here simulated annealing.

What would settle it

Recompute the optimal schedules with revenue $R_t^N = \sum_{j,\tau} N_{j\tau} Q_j \, \mathbb{E}_{\Delta_j}[R_{j,t-\tau-\Delta_j}]$, averaging over the lognormal duration distribution instead of using the single value $\Delta_j$; if the resulting yearly project starts shift materially, the illustrative recommendations are artifacts of that simplification.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the pipeline-replenishment problem can be formulated as a constrained optimization over the yearly inflow of new projects, $N_{j\tau}$. The current portfolio is represented by projects at known phases, each disease area has its own duration, cost, and success-probability distributions, and the expected future portfolio composition, cost, and revenue are written as linear functions of the inflow schedule. The paper then solves eight optimization problems—mean versus yearly revenue targets and mean versus yearly budget limits—under constraints requiring minimum numbers of projects per phase, per disease area, and per number of launches. The illustrating 30-year example shows that different framings recommend materially different inflows: mean-based objectives tend to front-load projects and overshoot yearly budgets, while yearly-budget objectives favor the disease area with lower cost and higher expected return.

Load-bearing premise

The plan's revenue projections assume each project in a disease area takes exactly $\Delta_j$ years to reach the market, even though the model elsewhere treats durations as random, and the paper assumes management can freely set the yearly number of new projects up to a growth cap.

Editorial extensions

If this is right

  • Portfolio managers can translate revenue targets and budget limits into a concrete year-by-year plan of how many new projects to start in each disease area.
  • The choice of objective framing matters: mean-based targets recommend front-loaded project inflow that can exceed yearly budgets, while yearly-capped objectives favor lower-cost, higher-return disease areas.
  • Portfolio-balance constraints such as minimum projects per phase, minimum presence per disease area, and minimum launches act as binding targets, sometimes forcing additional starts just to satisfy them.
  • Because the planning horizon is 30 years with the first 10 years of starts reported, the method supports long-term replenishment decisions rather than one-off project selection.
  • Comparing the eight optimal plans gives decision-makers a what-if view of how the recommended strategy changes with strategic emphasis.
  • The paper claims that the future in-flow of new projects into a pharmaceutical development portfolio can be optimized while adhering to revenue targets, budget limits, and strategic portfolio-composition constraints.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same optimization machinery could naturally accommodate in-licensing or business-development entries at later phases; the paper names this as a future enhancement, and it would make phase-balance constraints easier to satisfy in the near term.
  • Because launch revenue is computed with a single total development duration while launch probability uses the duration distribution, averaging revenue over the duration distribution is a natural correction and is the most likely place the quantitative plans would change.
  • The practical value of the approach likely lies less in a single 'optimal' schedule and more in comparing the eight objective framings; a decision-maker combining insights across solutions would learn which project-start patterns are stable across reasonable strategic goals.
  • A retrospective test is natural: apply the method to a company's portfolio state from ten years ago and compare the recommended project starts with what was actually initiated.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 7 minor

Summary. The paper develops a simulation-and-optimization framework for planning the future inflow of new drug-development projects into a pharmaceutical company's portfolio. The decision variables are the numbers of projects N_jτ entering Phase 1 in each disease area j and year τ, and the model combines the current portfolio, future additions, marketed-product forecasts, and a ramp-up/LOE revenue model. Eight objective-function variants (1A-4B) are formalized, balancing mean or year-by-year revenue targets and cost budgets, subject to constraints on phase occupancy, disease-area presence, minimum launches, and annual inflow increase. The method is illustrated on a stylized 30-year example using a lognormal duration/cost model and simulated annealing.

Significance. If the modeling inconsistencies are repaired, the paper addresses a genuinely different strategic problem from the usual project-selection literature: instead of choosing among available projects, it plans the rate and composition of future project entries needed to meet long-term revenue and portfolio-composition goals. The eight objective variants provide a useful map of how strategic framing changes the recommended inflow. The authors are to be credited for giving explicit equations for costs, phase occupancy, launch probabilities, and the optimization constraints, and for transparently presenting the illustrative results across all eight objectives. However, the central revenue equation uses a deterministic development duration in a model that is otherwise stochastic, and the example omits several numerical inputs, so the quantitative outputs as submitted are not reproducible and may be biased.

major comments (3)
  1. [Sales revenue (equation for R_t^N)] The equation R_t^N = Σ_τ Σ_j N_jτ Q_j R_{j,t-τ-Δ_j} evaluates future revenue at a single deterministic total development duration Δ_j, whereas Δ_j is earlier defined as the sum of stochastic lognormal phase durations, and the launch probability in the same paper is μ_{j,t-τ} = Pr(Δ_j < t-τ) Q_j, which uses the full duration distribution. Expected revenue from a cohort entering at τ should be Q_j times the convolution of the revenue curve with the distribution of the total development duration, i.e. Q_j ∫ R_{j,t-τ-s} dF_{Δ_j}(s) (with the appropriate truncation at launch), not the revenue evaluated at one representative duration. Because R_jt is piecewise linear with a ramp-up phase and a cliff at loss of exclusivity, replacing the duration distribution by a point mass changes both the timing and the level of projected revenue; since the revenue targets and revenue objectives are what select N_jτ, the optimized inflow inherits this bias. Please either define Δ_j explicitly as a fixed planning assumption and report sensitivity to it, or compute expected revenue by averaging over the duration distribution consistently with μ_jt.
  2. [Illustrating example - input assumptions (Tables 4-6, Figure 5)] The numerical example cannot be reproduced from the stated inputs. The revenue target S used in problem (1A), the annual revenue targets S_t used in (2A)-(4B), the budget levels B_t appearing in the constraints, and the exclusivity length Λ used in the revenue model R_jt are not reported numerically; Figure 5 shows curves but no axis values or table. Since Figures 6-29 are presented as the optimal solutions for these specific targets and budgets, please include all numerical input values used in the example so that the results are auditable and reproducible.
  3. [Optimization methods] The paper states that simulated annealing hyperparameters were 'chosen after running several cases to ensure that the solution is close to optimal', but no convergence diagnostics, multiple-restart results, or comparisons against exact solutions for smaller instances are reported. Given that the Results section repeatedly refers to the 'optimal' number of added projects, please provide evidence of solution quality, such as best-of-many-restarts values, cooling-schedule sensitivity, or a small test case with an exact integer-programming benchmark.
minor comments (7)
  1. [Current development portfolio] There is a typo in the text: 'proj0ect' should be 'project'.
  2. [Table 4] The header 'Prob. f launch' should read 'Prob. of launch'.
  3. [Sales revenue] In the piecewise definition of R_jt, the last line 'λ ∙ PYR_j Λ < t' should include the missing inequality direction or formatting, e.g. 'λ ∙ PYR_j for Λ < t'.
  4. [Sales revenue / Table 6] The exclusivity length Λ appears in the revenue model but is not listed in Table 6; please add it to the input table.
  5. [Model and assumptions] The notation Δ_j is used both as the random sum of phase durations in the launch-probability definition and as a fixed number in the revenue equation; please use distinct symbols (e.g., Δ̃_j for a fixed planning value) to avoid ambiguity.
  6. [Results] The eight cases are described qualitatively, but no table of achieved objective values, revenue shortfalls, or budget violations is provided; such a table would make the comparisons across objectives much easier to assess.
  7. [References] The reference 'Journal of Health Econonmy' contains a typo; it should be 'Journal of Health Economics'.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the optimized in-flow is the decision variable of a stated optimization with independent input forecasts; self-citations are only framing/notation.

full rationale

The paper is self-contained in its derivation. The decision variables N_{jt} are chosen by an explicit simulated-annealing optimization of stated objectives (mean or min/max revenue/cost) subject to stated constraints (revenue targets, budgets, phase/disease-area balance, launch floors, annual inflow growth cap). The revenue and cost expressions are linear convolutions of the decision variables with input probability and cash-flow curves, so the reported 'optimal inflow' is the optimizer's output for those inputs, not a parameter fitted to the same output. The self-citations to Wiklund et al. (2023) and Farid et al. (2021) supply notation, a framing convention (1A-4B), and the choice to show only the first 10 years of inflow; none of these citations carries a mathematical premise on which the optimization result depends. The only notable modeling issue is internal, not circular: the sales-revenue formula uses a fixed total duration Δ_j while the launch-probability formula uses Pr(Δ_j < t-τ), so expected revenue is not integrated over the duration distribution; this is a correctness/robustness concern, not a reduction of the conclusion to the inputs. Hence no circular step is present.

Assumptions & free parameters 10 free parameters · 6 assumptions · 0 invented entities

The central claim rests on modeling assumptions about drug development projects (independent phase successes, uniform templates within a disease area, expected-value aggregation) and on hand-chosen illustrative inputs. No parameters are fitted to real data, and several example inputs are not specified numerically. The revenue model also relies on a deterministic duration assumption that is inconsistent with the stochastic durations used elsewhere.

free parameters (10)
  • Phase success probabilities P_ij = e.g., 0.6, 0.4, 0.7, 0.95 for disease area 1
    Assumed inputs for the illustrative example (Table 4); not fitted to data. They directly determine launch counts and revenue.
  • Median phase durations delta_ij = e.g., 2, 2.5, 3.5, 1 years for disease area 1
    Assumed inputs (Table 4) with lognormal variability; affect timing of costs and revenues.
  • Median phase costs gamma_ij = e.g., 0.2, 0.3, 0.6, 0.1 $Bn for disease area 1
    Assumed inputs (Table 4); drive the cost constraints.
  • Peak year revenue PYR_j = 1.5, 5, 3 $Bn for disease areas 1-3
    Assumed inputs (Table 6); scale the revenue from future launches.
  • Ramp-up duration U_j = 3, 5, 4 years
    Assumed inputs (Table 6); shape the revenue ramp-up curve.
  • Post-LOE revenue fraction lambda_j = 0.2, 0.1, 0.15
    Assumed inputs (Table 6); set long-run revenue after exclusivity loss.
  • Exclusivity duration Lambda = not specified
    Used in the revenue model R_jt but never assigned a value in the Example; materially affects revenue from future additions.
  • Revenue targets S and S_t = not specified numerically (Figure 5 only)
    The targets that drive objectives 1-2 are shown only in Figure 5; not tabulated.
  • Budget B_t = not specified numerically (Figure 5 only)
    The yearly budget constraint appears only in Figure 5; not tabulated.
  • Maximum annual increase in project inflow delta = not specified
    The inertia constraint N_tau <= N_{tau-1} + delta is introduced in the Optimization section but delta is never given a value.
assumptions (6)
  • domain assumption Phase success probabilities are independent, so the aggregate probability of launch is the product Q_j = prod_i P_ij.
    Used in the 'Number of launched products' section to compute expected launches and revenue; assumes no correlation between phase outcomes.
  • domain assumption Projects within a disease area share identical duration, cost, and success distributions (uniform template).
    The paper states this simplification in the Discussion; it ignores project-level heterogeneity in the example.
  • domain assumption Portfolio composition can be summarized by expected numbers of projects in each phase and disease area; variances and covariances are ignored.
    The objective functions and constraints (M_it, E_jt, L_j) are all expectations; the paper notes stochastic optimization as a future enhancement.
  • ad hoc to paper Revenue from future launches is evaluated at a deterministic total duration Delta_j even though durations are stochastic.
    In the 'Sales revenue' section, R_t^N uses Q_j R_{j,t-tau-Delta_j} with fixed Delta_j, inconsistent with the stochastic duration used for launch probabilities.
  • ad hoc to paper Simulated annealing with informally tuned hyperparameters finds the global optimum.
    The Optimization methods section states hyperparameters were chosen 'after running several cases to ensure that the solution is close to optimal'; results are still labeled 'optimal'.
  • domain assumption New projects enter only at Phase 1 of clinical development.
    The model defines N_{j,tau} as entries into Phase 1; the Discussion mentions business development entries at later phases as a future enhancement.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Project portfolio planning in the pharmaceutical industry -- strategic objectives and quantitative optimization." pith.science (2026). https://pith.science/paper/HVT4OPWC

@misc{pith2026250209527,
  author       = {Pith},
  title        = {Pith review of: Project portfolio planning in the pharmaceutical industry -- strategic objectives and quantitative optimization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HVT4OPWC}},
  note         = {Machine review of arXiv:2502.09527}
}
read the original abstract

Many pharmaceutical companies face concerns with the maintenance of desired revenue levels. Sales forecasts for the current portfolio of products and projects may indicate a decline in revenue as the marketed products approach patent expiry. To counteract the potential downturn in revenue, and to establish revenue growth, an in-flow of new projects into the development phases is required. In this article, we devise an approach with which the in-flow of new projects could be optimized, while adhering to the objectives and constraints set on revenue targets, budget limitations and strategic considerations on the composition of the company's portfolio.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

7 extracted references · 6 canonical work pages

  1. [3]

    Computational Statistics, (2nd ed.)

    https://doi.org/10.1186/s13362-021-00099-3 Givens GH, Hoeting JA. Computational Statistics, (2nd ed.)

    work page doi:10.1186/s13362-021-00099-3

  2. [296]

    Replenishing the pipeline: A quantitative approach to optimising the sourcing of new projects

    Wiklund SJ, Önnheim M, Ytterstad M. Replenishing the pipeline: A quantitative approach to optimising the sourcing of new projects. Pharmaceutical Statistics. 2023; 22(5): 861-879. doi:10.1002/pst.2316

    work page doi:10.1002/pst.2316 2023

  3. [2013]

    Graham E, Jaki T, Harbron C

    John Wiley & Sons, Inc., Hoboken, New Jersey. Graham E, Jaki T, Harbron C. A comparison of stochastic programming methods for portfolio level decision-making. J Biopharm Stat. 2020;30(3):405-429. doi:10.1080/10543406.2019.1684307 Hassanzadeh F, Modarres M, Nemati HR, Amoako-Gyampah K. A robust R&D project portfolio optimization model for pharmaceutical co...

    arXiv 2020

  4. [2015]

    doi:10.1007/978-3-319-09075-7 Baird LG, et al

    Springer, Cham. doi:10.1007/978-3-319-09075-7 Baird LG, et al. Accelerated access to innovative medicines for patients in need, Clinical Pharmacology & Therapeutics. 2014, 96.5: 559-571. Brătianu C, and Bălănescu GV. Vision, mission and corporate values. A comparative analysis of the top 50 US companies, Management & Marketing, 2008,3(3), 19-38. Chao RO, ...

    work page doi:10.1007/978-3-319-09075-7 2014

  5. [2019]

    Innovation in the pharmaceutical industry: New estimates of R&D costs, Journal of Health Econonmy, 2016, 47, 20–33

    https://www.pharmexec.com/view/strategic-approach-pharma-rd-portfolio-planning DiMasi JA, Grabowski HG, Hansen RW. Innovation in the pharmaceutical industry: New estimates of R&D costs, Journal of Health Econonmy, 2016, 47, 20–33. Dimitri N, What Constitutes an Optimal Portfolio of Pharmaceutical Compounds?. Clinical Pharmacology and Therapeutics. 2011, 8...

    work page doi:10.1038/clpt.2010.247 2016

  6. [2022]

    Leisinger KM

    Wiley. Leisinger KM. The corporate social responsibility of the pharmaceutical industry: idealism without illusion and realism without resignation. Business Ethics Quarterly. 2005, 15(4), 577- 594 Nadar E, Akan M, Debo L, Scheller-Wolf A. Optimal Portfolio Strategies for New Product Development. SSRN Electronic Journal. 2015, 10.2139/ssrn.2675317. Patel N...

    work page doi:10.2139/ssrn.2675317 2005

  7. [2169]

    Pharmaceutical portfolio optimization under cost uncertainty via chance constrained-type method

    Farid M, Chaudhry A, Ytterstad M, Wiklund SJ. Pharmaceutical portfolio optimization under cost uncertainty via chance constrained-type method. Journal of Mathematics in Industry. 2021, 11,

    work page 2021

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.