Antoine Pitoy 1, Emmanuelle Comets 2,3, Dorothée Semiond 4, Hoai-Thu Thai 1
1 Sanofi, Translational Medicine Unit, Quantitative Pharmacology (Gentilly, France), 2 Université Paris Cité and Université Sorbonne Paris Nord, Inserm, IAME (F-75018 Paris, France), 3 Univiversité Rennes, Inserm, EHESP, Irset (Institut de recherche en santé, environnement et travail) - UMR_S 1085 (F-35000 Rennes, France), 4 Sanofi, Translational Medicine Unit, Quantitative Pharmacology (Cambridge, USA)
Introduction: Drug development in oncology faces high failure rates and significant costs, due to methodological and operational challenges. Modeling and simulation approaches increasingly support decision-making by integrating biological knowledge and existing data throughout development programs [1]. Trial simulations enable “what-if” scenario exploration, trial design optimization, and evaluation of treatment response using Virtual Populations (VPOPs). Both statistical (e.g., copula-based) and mechanistic (e.g., QSP-based) approaches have been proposed to generate realistic VPOPs [2–3]. A major challenge is ensuring that VPOPs accurately reproduce the patient characteristics and clinical outcomes observed in real studies, especially in prospective settings where predictive credibility is critical [4]. This study aimed to develop and evaluate methodologies for prospectively generating realistic clinical trial simulations, applying them to the isatuximab development program in relapsed/refractory multiple myeloma to predict the outcomes of the pivotal ICARIA (NCT02990338) Phase 3 trial [5] prior to data availability.
Methods: Baseline covariates, isatuximab concentrations, serum M-protein levels, and progression-free survival (PFS) were collected from multiple phase 1 and 1/2 studies in monotherapy (Isa) or in combination with pomalidomide and dexamethasone (Isa+Pd) or with lenalidomide and dexamethasone (Isa+Rd) (N = 468). Baseline covariates were used to develop a vine copula model. Data from 94 Isa and Isa+Pd patients, and 40 Isa+Rd patients, enrolled before ICARIA trial initiation, were used to refine a joint model of serum M-protein and PFS [6]. Additionally, we identified a subgroup of 45 patients called the ‘Eligible Ph1/2’ population, who received combination therapy (Isa+Pd or Isa+Rd) and met ICARIA inclusion/exclusion criteria.
A total of 500 Isa+Pd arm VPOPs were generated from three virtual patient simulation methods: PopDist (combining Copula-sampled covariates with population distribution parameters); CondDist (sampling individual parameters from conditional distribution); and Large Copula (LCOP) (integrating both conditional distribution parameters and covariates through Copula-based sampling). Simulated M-protein and PFS profiles were used to derive virtual patient outputs (baseline M-protein, best change from baseline, time-to-PFS). Virtual patient output distributions were compared with those of the ‘Eligible Ph1/2’ population to compute inclusion probabilities [7], which were then used to generate each VPOP by sampling virtual patients without replacement. The control arm VPOPs were obtained by simulating the same virtual patients without isatuximab exposure. Each virtual trial was constructed by combining the Isa+Pd VPOP with its corresponding control arm VPOP and applying the ICARIA protocol-defined cut-off.
We compared the predicted distributions of clinical outcomes (overall response rate (ORR), median PFS, hazard ratio (HR)) to the observed values from ICARIA (N=244).
Results: Virtual patient-based approaches slightly underestimated ORR in the Isa+Pd arm compared with ICARIA (PopDist 64% [95%PI:56–72], CondDist 63% [56–70], LCOP 63% [55–70] vs. 79% [95% CI:71–85] in ICARIA). In the control arm, PopDist and CondDist underestimated ORR (30% [22–38] and 39% [31–46], respectively, vs. 50% [41–58]), whereas the LCOP approach more accurately predicted the ICARIA observed value (43% [36–51]).
Median PFS was accurately reproduced in the Isa+Pd arm (PopDist 11.9 [9.1–NA], CondDist 10.4 [7.5–13.9], and LCOP 11.0 months [8.5–13.8] vs. 11.4 months [8.9–14.7] in ICARIA). Meanwhile, there was a trend toward overpredicting median PFS in the control arm (PopDist 6.4 [5.4–8.1], CondDist 7.4 [5.3–9.6], LCOP 7.5 [6.4–9.2] vs. 5.8 months [4.4–9.4] in ICARIA), though predictions lay within the ICARIA confidence Interval.
Regarding the hazard ratio, all methods capture the benefit of Isa+Pd arm on PFS prolongation. PopDist overpredicted the treatment effect by providing a lower HR than observed value from ICARIA (0.52 [0.42–0.61] vs. 0.61 [0.43–0.85] in ICARIA). CondDist slightly underpredicted the treatment effect with a greater HR value (0.71 [0.62–0.80]), while LCOP provided the closest prediction (0.67 [0.57–0.76]).
Conclusions: In this work, the different methods to generate VPOP provided realistic predictions for the pivotal ICARIA Phase 3 outcomes in a prospective setting. All approaches relied on an ‘Eligible Ph1/2’ subpopulation meeting Phase 3 inclusion/exclusion criteria, which closely matched the population eventually recruited in ICARIA. Further simulation studies are needed to assess the robustness of these methods under varying scenarios, including model misspecification and greater discrepancies between ‘Eligible Ph1/2’ and Phase 3 populations. The next step will be to evaluate how real-world data provide helpful information to enrich the covariate copula model and build the external control arm, ultimately facilitating prospective trial simulation.
References:
[1] Lesko LJ, van der Graaf PH. Reflections on model-informed drug development. Clin Pharmacol Ther. 2024;116:267–270.
[2] Zwep LB, Guo T, Nagler T, et al. Virtual patient simulation using copula modeling. Clin Pharmacol Ther. 2024;115:795–804.
{3] Allen RJ, Rieger TR, Musante CJ. Efficient generation and selection of virtual populations in quantitative systems pharmacology models. CPT Pharmacometrics Syst Pharmacol. 2016;5:140–146. doi:10.1002/psp4.12063
[4] Moingeon P, Chenel M, Rousseau C, Voisin E, Guedj M. Virtual patients, digital twins and causal disease models: paving the ground for in silico clinical trials. Drug Discov Today. 2023;28(7):103605.
[5] Attal M, et al. Isatuximab plus pomalidomide and low-dose dexamethasone versus pomalidomide and low-dose dexamethasone in patients with relapsed and refractory multiple myeloma (ICARIA-MM): a randomised, multicentre, open-label, phase 3 study. Lancet. 2019;394:2096–2107.
[6] Pitoy A, Desmée S, Riglet F, et al. Isatuximab–dexamethasone–pomalidomide combination effects on serum M protein and PFS in myeloma: development of a joint model using phase I/II data. CPT Pharmacometrics Syst Pharmacol. 2024. doi:10.1002/psp4.13206
[7] Braniff N, Joshi T, Cassidy T, et al. An integrated quantitative systems pharmacology virtual population approach for calibration with oncology efficacy endpoints. CPT Pharmacometrics Syst Pharmacol. 2025;14:268–278.
Reference: PAGE 34 (2026) Abstr 12109 [www.page-meeting.org/?abstract=12109]
Poster: Methodology - New Modelling Approaches