Antoine Pitoy (1,2,3), Solène Desmée (2), Marc Cerou (3), Hoai-Thu Thai (3), Christine Veyrat-Follet (3), Julie Bertrand (1)
(1) Université de Paris, UMR 1137 IAME, INSERM, F-75018 Paris, France; (2) Université de Tours, Université de Nantes, UMR 1246 SPHERE, INSERM, 37000 Tours, France; (3) Translational Disease Modelling Oncology, Sanofi, 91380 Chilly-Mazarin, France
Introduction: Earlier assessment of clinical trial outcomes would allow earlier decision making, especially for trials of long duration, where models developed in early phases may contribute to estimation of the primary endpoint at interim analyses. Joint modeling has been increasingly used in therapeutic evaluation, especially in oncology, and allows for the simultaneous fit of longitudinal and time-to-event data to characterize and quantify the association between biomarker dynamics and risk of event. Principally modeled with linear models[1], biomarker kinetics have been analyzed by more mechanistic models, such as tumor growth inhibition models.
At the individual level, dynamic predictions obtained using nonlinear joint models can identify the most at-risk patients in oncology clinical trials[2] and can improve patient follow-up. Evidence suggests a benefit of joint modeling at the population level to inform and support decision making; however, additional studies are needed.
Objectives: The objective of our study was to assess the ability of a nonlinear joint model selected from Phase I and II data to estimate the primary endpoint at the interim and final analyses of a Phase III oncology clinical trial. The hazard ratios (HRs) obtained with this approach were compared with estimates from Cox and classical parametric models.
Methods: We conducted a retrospective analysis of ICARIA-MM, a Phase III clinical trial comparing progression-free survival (PFS) of a standard treatment of pomalidomide and dexamethasone versus the same combination plus isatuximab in patients with relapsed and refractory multiple myeloma who have received at least 2 prior treatments[3].
HRs with 95% confidence intervals (CI) can be obtained directly from fitting the Cox and parametric models to the observed data. However, for nonlinear joint models, one needs to simulate clinical trials and then fit a Cox model to the simulated time-to-event data. From a base model, the simulation can be performed using either the population or the individual parameter conditional distribution.
We used data from interim analyses when 33% and 65% of the expected PFS events occurred, and compared the HRs and 95% CI obtained on observed data from (i) a Cox and (ii) a parametric model, and on simulated data (1000 clinical trials until the final analysis date) using estimates from (iii) a Cox, (iv) a parametric, and nonlinear joint models developed from Phase I and II data[4]. In the latter case, we sampled from (v) the population parameter distribution or (vi) the individual parameter conditional distribution.
We considered different spending functions (ie, Pocock, O’Brien and Fleming, Haybittle-Peto)[5,6] to avoid type I error inflation due to interim analyses.
Results: Among the 307 patients in ICARIA-MM, only the 256 evaluable patients with serum M protein measurements were included. At the first interim analysis, 220 days after the first patient enrolled, 226 patients were included, 46 PFS events were observed, and the median follow-up was 59 days. At the second interim and final analyses (336 and 657 days after the first patient enrolled, respectively), all 256 patients were included, 92 and 141 PFS events were observed, and the median follow-up was 129 and 231 days, respectively. At the first interim analysis, the HRs (95% CI) were 0.61 (0.34–1.10) and 0.72 (0.48–1.10) using the Cox and the parametric model on observed data; 0.61 (0.45–0.81) and 0.76 (0.58–1.03) using the Cox and the parametric model on simulated data, respectively; and 0.68 (0.51–0.81) and 0.55 (0.46–0.64) for the nonlinear joint model using simulations from the population and individual parameter conditional distribution, respectively. Significance was obtained for all alpha spending functions only with approach (vi). At the second interim and final analyses, all methods showed superiority of the isatuximab treatment arm regardless of alpha spending function. At the final analysis, HRs varied from 0.50 to 0.58 with narrower 95% CIs. The joint modeling approach using individual parameters drawn from the conditional distribution obtained the narrowest 95% CI.
Conclusions: Nonlinear joint models using simulations from the individual parameter conditional distribution may be used as a supplementary tool for treatment decision making; however, its performance must be evaluated in comparison with other methods in a simulation study.
References:
[1] Rizopoulos D Joint Models for Longitudinal and Time-to-Event Data: With Applications in R. (Routledge & CRC Press, 2012).
[2] Tardivon, C. et al. Association Between Tumor Size Kinetics and Survival in Patients With Urothelial Carcinoma Treated With Atezolizumab: Implication for Patient Follow-Up. Clin Pharmacol Ther 106, 810–820 (2019).
[3] 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 394, 2096–2107 (2019).
[4] Riglet, F., Hoai-Thu, T., Veyrat-Follet, C. & Bertrand, J. Phase I and II clinical development of isatuximab-based combination therapy in relapsed and/or refractory multiple myeloma patients: a joint model of concentrations, serum-M-protein levels and progression-free survival. CPT: Pharmacometrics & Systems Pharmacology (Submitted).
[5] DeMets, D. L. & Lan, K. K. Interim analysis: the alpha spending function approach. Stat Med 13, 1341–1352; discussion 1353-1356 (1994).
[6] Peto, R. et al. Design and analysis of randomized clinical trials requiring prolonged observation of each patient. I. Introduction and design. Br J Cancer 34, 585–612 (1976).
Reference: PAGE 30 (2022) Abstr 9993 [www.page-meeting.org/?abstract=9993]
Poster: Drug/Disease Modelling - Oncology