2021 - Online - In the cloud

PAGE 2021: Methodology - Other topics
David Fairman

Towards Industrialisation of Probability of Pharmacological Success (PoPS)

Tarjinder Sahota; Martin Bergstrand; Peter A. Milligan; Stefano Zamuner; Chao Chen; David Fairman

GSK

Objectives: Probability of Pharmacological Success (PoPS) integrates knowledge of pharmacokinetics, pharmacodynamics and safety together with their uncertainty to quantify confidence in the molecule and risk of inadequate human pharmacology.  The goals of PoPS assessments are to inform dose selection, explore trial design options, and support investment decision making [1,2].

An organisational goal of GlaxoSmithKline (GSK) is to deploy PoPS at scale to enhance decision making across the portfolio.  However, implementing PoPS is time consuming and computationally demanding: translational pharmacology models often involve differential equation systems describing pharmacokinetics, target engagement and pathway pharmacology in a drug-disease system; and PoPS estimation requires integrating statistics derived from large scale trial simulations over multivariate parameter uncertainty distributions. Communicating quantitative findings in non-technical language and in a consistent manner to multi-disciplinary project teams and senior stake holders is also highly important for effective application of the PoPS framework to investment decision making. There is therefore a strong need for an industrialised framework of PoPS to standardise 1) implementation, 2) outputs and 3) messaging to maximise quality and consistency of project support whilst addressing the above resourcing challenges.  The need for application across a wide range of projects and therapy areas also requires sufficient flexibility in the framework.

Methods: At GSK, we have developed a framework in collaboration with Pharmetheus using existing R packages to industrialise PoPS. The framework is based-on an input data standard, R markdown templates, and a set of dedicated R functions to collect project specific information from the user, run simulations and provide documented outputs that provide traceability and facilitate efficient communication with project teams.

Results: The R markdown templates are organised into two sections: Section 1 is templated code, designed to be modified by the user.  Section 2 requires no user modification and uses the objects defined in section 1 to conduct the PoPS calculations and produce outputs.  The templated nature of the text in these sections can aid in standardizing terminology across users as well as providing concise explanations of outputs for use in presentations.

Section 1 of R markdown: PoPS “ingredients” (templated user input)

  • – Specification of PK/PD model(s).
  • – Specification of parameter estimates (fixed and random effects) and influential covariate distributions
  • – Parameter uncertainty distributions representing both translational and estimation uncertainty
  • – Specification of dose ranges and simulation resolution settings
  • – Specification of 3 levels of summary required for PoPS:
    • Endpoint level summaries (e.g. AUC, Ctrough, area under effect curve, etc.)
    • Individual level target criteria values (specified in terms of endpoint level summaries). Flexibility for composite criteria for multivariate decisions accounting for pharmacology and safety limits, manufacturing limits, etc.
    • Population/compound level pharmacological success (PS) criteria (specified in terms individual level target criteria), including desired % of population achieving individual level success.

Section 2 of R markdown: PoPS “recipe” (problem independent code)

  • – Template samples n_sim times from uncertainty distribution of parameters. Simulate population (n_pop individuals) across prespecified range of dosing regimens in population and compute 3 levels of summary above. Compute PoPS as the probability of achieving PS.

PoPS output plots include:

  1. Dose vs PoPS: For dose selection and investment decision making
  2. Dose vs prediction interval of proportion of patients reaching individual target: for dose selection.
  3. Dose vs prediction interval (without parameter uncertainty) of summary endpoint: for communicating variability in individual endpoints
  4. Dose vs prediction interval of median summary endpoint: for communicating uncertainty in endpoints
  5. Plots illustrating the underlying dose-exposure-response relationships

 Standardised text in R markdown documents:

  • – A common language for describing the building blocks of PoPS.
  • – Description of inputs and outputs to aid project teams in 1) defining individual targets and success criteria and 2) correct interpretation of outputs

Conclusions: We have created a framework to begin industrialising PoPS within GSK.



References:
[1] Zhou X, Graff O, Chen C (2020). Quantifying the probability of pharmacological success to inform compound progression decisions. PLoS ONE 15(10): e0240234.
[2] Zhou X, Chen C, Graff O. Model-Based Estimation of Probability of Pharmacological Success for CNS Compounds. PAGE 28 (2019) Abstr 8917 [www.page-meeting.org/?abstract=8917]


Reference: PAGE 29 (2021) Abstr 9805 [www.page-meeting.org/?abstract=9805]
Poster: Methodology - Other topics
Top