Giulia Monchietto 1, Sarah Zohar 1, Karen Sinclair 2
1 Inserm, Inria, Université paris-cité (Paris, France), 2 Novartis Pharma AG (Basel, Switzerland)
Introduction: Drug development is challenging in rare and pediatric diseases, due to recruitment limitations and ethical constraints. Generation of virtual patients (VPs) is an attractive approach in a setting where patients do not exist at numbers required to conduct powered clinical trials. Model-informed drug development (MIDD [1]), facilitates the process of VP generation using computational models integrating multi-source data, however, a specific framework is yet to be developed which considers:
– how much knowledge, and of which type, is required,
– which methods are suitable in different situations,
– how reliable are the generated VPs,
A 2025 systematic review [2] highlighted this gap, underlining that only 18/76 in-silico trial publications addressed rare or pediatric populations. Our work follows six-steps to combine multi-source knowledge, available methodology and clinical studies to facilitate reliable VP cohort generation. Specifically, we focus on generating VPs through stochastic simulations of concentration-time profiles via population pharmacokinetic (popPK) models to demonstrate the purpose of this workflow.
Methods: A framework to generate VPs has been initialised, implementing 6 steps. Steps 1-4 are the focus of this work, while steps 5-6 will be developed in future work.
1. Objective statement: define the specific question of interest for the modeling activity, and where it fits in the VP generation framework.
2. Knowledge assessment: categorize available information into relevant domains to eventually understand the impact of each knowledge domain on VP generation:
drug knowledge (e.g. PK/PD models),
disease knowledge (endpoints, progression models),
knowledge coming from clinical trials (in one or more indications),
population knowledge (clinical and demographic attributes).
3. Methodological selection and procedure: identify methods and procedures applicable to various scenarios mimicking different amounts of available knowledge, and link them within an overarching framework to provide guidance on which approaches to follow for VP generation under different scenarios.
Initially we focus on extrapolation of PK to a new indication under different scenarios, considering many approaches such as external validation, integrating information from distinct indications, and model ensembling methods. [3-9]
4. Model evaluation: assess candidate models to evaluate robustness and associated prediction capability
For extrapolation of PK to a new indication, classic approaches of prediction-based diagnostics and simulation-based tools (e.g. prediction errors and visual predictive checks) are utilised to ensure accurate reproducibility of the central tendency and variability of the target populations. Published guidelines on external evaluation and on model evaluation diagnostics are leveraged [10-11].
5-6. Uncertainty quantification and VP generation: the associated uncertainty quantification of the modeling approach on the generated VPs will be a focus of future work.
Results: we present a simple case study to illustrate the implementation of the framework, whose objective is to augment cohorts of patients with psoriatic arthritis (PsA), treated with Secukinumab, using a population pharmacokinetic model (step 1). Available knowledge comes from an established popPK model for Secukinumab built on data from patients with plaque psoriasis (PsO) [12], however the clearance and absorption rate parameters are scaled to mimic a scenario where the target and reference population have differing PK profiles. Knowledge from clinical trials is fixed to N=50 PsA patients (target population)(step 2). Direct extrapolation from the scaled model to the PsA population was performed (step 3), but evaluation (step 4) revealed that the model failed to meet predefined population-based predictive criteria. Consequently, we iterate through step 3 to assess alternative approaches (investigation of sampling scheme optimisation, model re-fitting on the target population, model re-fitting on pooled indications, etc.), After re-evaluating each approach in Step 3, model evaluation (step 4) is further repeated to identify robust approaches for VP generation. VP generation may then performed with confidence, but we emphasize that the uncertainty assessment to be performed in step 5 will provide the full measure of model risk, allowing regulators to understand the total level of confidence in the evidence generated to support drug development.
Conclusions: Results show an example of applying a framework designed to provide a structured approach to VP data generation under varying amounts of knowledge volume and types. Future work will expand this framework to accommodate additional knowledge scenarios with more complex models and techniques, including VP generation in pediatric populations, and integrating pharmacodynamic modeling.
References:
[1] International Council for Harmonisation (ICH). General Principles for Model-Informed Drug Development M15.Final version 2026
[2] Chen, B., Schneider, L.C., Röver, C. et al. In Silico Clinical Trials in Drug Development: A Systematic Review. Ther Innov Regul Sci (2025). https://doi.org/10.1007/s43441-025-00893-w
[3] Nyberg J et al.. PopED: an extended, parallelized, nonlinear mixed effects models optimal design tool.. Comput Methods Programs Biomed. 2012;108(2):789–805.
[4] Gisleskog O, Karlsson MO, B SL. Use of Prior Information to Stabilize a Population Data Analysis. J. Pharmacokinet. Pharmacodyn. 2002;29(5-6):473–505.
[5] Sahasrabudhe SA, Bonate PL. Pharmacokinetic comparability between two populations using nonlinear mixed effect models: a Monte Carlo study. J. Pharmacokinet. Pharmacodyn. 2023;50:189–201.
[6] Uster DW, et al. A Model Averaging/Selection Approach Improves the Predictive Performance of Model-Informed Precision Dosing: Vancomycin as a Case Study.. Clin Pharm Therap. 2021;109(1):175–-183.
[7] Agema BC et al. Selecting the Best Pharmacokinetic Models for a Priori Model‑Informed Precision Dosing with Model Ensembling. Clin. Pharmacokinet.. 2024;63:1449–1461.
[8] Chan A et al.. Synthetic Model Combination: A new machine-learning method for pharmacometric model ensembling. Pharmacometrics Syst Pharmacol. 2023;12(7):953–962.
[9] Chan Kwong AH et al. Prior information for population pharmacokinetic and pharmacokinetic/pharmacodynamic analysis: overview and guidance with a focus on the NONMEM PRIOR subroutine. J. Pharmacokinet. Pharmacodyn. 2020;47:431–446.
[10] Nguyen, T.H.T., et al.. and for the Model Evaluation Group of the International Society of Pharmacometrics (ISoP) Best Practice Committee (2017), Model Evaluation of Continuous Data Pharmacometric Models: Metrics and Graphics. CPT Pharmacometrics Syst. Pharmacol., 6: 87-109
[11] Comets E et al. Computing normalised prediction distribution errors to evaluate nonlinear mixed-effect models: the npde add-on package for R. Comput Methods Programs Biomed. 2008 May;90(2):154-66
[12] Bruin G et al. Comparison of Pharmacokinetics, Safety and Tolerability of Secukinumab Administered Subcutaneously Using Different Delivery Systems in Healthy Volunteers and in Psoriasis Patients. Clin Pharm Therap. 2017;57(7):876–885.
Reference: PAGE 34 (2026) Abstr 12081 [www.page-meeting.org/?abstract=12081]
Poster: Drug/Disease Modelling - Other Topics