III-075 Antoine Pitoy

Estimation of phase III trial outcome in oncology using joint models of longitudinal biomarker and time-to-event data

Antoine Pitoy

Inserm

CONTEXT:

During drug development in oncology, a substantial amount of data is acquired to determine the safety and efficacy of novel therapies. This includes drug pharmacokinetics (PK) data, characterizing patient exposure, and of pharmacodynamics (PD) data, such as tumor size, characterizing patient response to the treatment. The latter have been linked to long-term endpoints such as progression-free survival (PFS) [1], notably using shared random effect joint models [2].

Nonlinear joint models have been shown to detect earlier the most-at-risk patients in oncology trials, paving the way for improved patient monitoring [3]. However, the benefits of these models at the population level, e.g. via decreasing study duration and/or clinical trial sample sizes, have yet to be demonstrated.

Here, we studied this question in the development of isatuximab (Isa) for the treatment of patients with relapsed/refractory multiple myeloma (RRMM) [5]. Isa is an immunoglobulin G1 monoclonal antibody inducing tumor cell killing via multiple modes of action [5]. Phase 1-2 studies highlighted good tolerability of single-agent Isa, with a greater anti-myeloma activity in RRMM patients receiving at least 10 mg/kg [6-7]. Another Phase 1b study showed significant clinical activity of Isa in combination with pomalidomide (Pom) and dexamethasone (Dex), with a manageable safety profile [8]. From these results, the randomized Phase 3 ICARIA-MM study was conducted to assess the efficacy (eg, progression-free survival [PFS]) of Isa plus Pom-Dex (IPd) compared to Pom-Dex (Pd) in RRMM patients [9]. Based on the ICARIA-MM results, Isa at a dose of 10 mg/kg in combination with Pom-Dex was approved in many countries to treat RRMM patients.

Joint modeling has been successfully applied to support the Isa clinical development [10]. The most recent application was in the selection of the Isa dosing regimen for mono- or combination therapy in Japanese patients, from early clinical trial data [11]. To go further and address the challenges raised above, we investigated the benefit of applying joint modeling to accelerate drug approval in the Isa development context.

OBJECTIVES:

  1. To assess by a simulation study the ability of a nonlinear joint model, built from Phase 1-2 clinical trial data, to estimate earlier the final outcome of a Phase 3 clinical trial, in an interim analysis context.
  2. To develop and validate a nonlinear joint model of serum M-protein and PFS, using data from Phase 1-2 studies of Isa combination therapy in RRMM patients.
  3. To estimate the outcome of the Phase 3 ICARIA-MM trial using a joint-model approach, in a retrospective interim analysis.

METHODS:

  1. A total of 200 two-arm Phase 3 clinical trials based on ICARIA-MM was simulated using a simplified joint model for longitudinal biomarker and PFS data. Final analysis was planned when 162 PFS events occurred among the 300 patients, and interim analyses were scheduled when 50% and 65% of the total number of expected events occurred. At each analysis, considering a nominal significance level defined by the O’Brien and Fleming alpha spending function (αOBF) [12], final hazard ratios (HRs) and their 1-αOBF confidence intervals (CI) were estimated using Cox (CoxObs) and parametric (ParamObs) survival models of the observed data and two joint model-based approaches. The latter require a simulation step: 1000 datasets are simulated using the population (JMpop) or the individual parameters (JMindiv) accounting for estimation uncertainty and the simulated time-to-event data are fitted with a Cox model to provide 1000 HRs. The median and percentiles are then used to derive the joint model approach HR plus its CI at the established αOBF level. We also evaluated simulation-based Cox (CoxSim) and parametric (ParamSim) survival model approaches for comparison. Two simulation scenarios were considered, no superiority of the test treatment to evaluate the type 1 error rate and significant PFS prolongation from the test treatment to assess the power.
  2. We modeled serum M-protein kinetics and PFS data from 1 Phase 1-2 (NCT01084252) [6-7] and 1 Phase 1b study (NCT02283775) [8]. We first developed a longitudinal model and a PFS model. Then, we selected the link function that best captured their association in a shared random effect joint model. We performed an external validation of the final joint model on ICARIA-MM IPd arm data. Then, through model-simulations, we compared Isa dosing regimen at 10 vs 20 mg/kg weekly for 4 weeks and then biweekly (QW/Q2W) in combination with Pom-Dex, in terms of median PFS and changes from baseline in serum M-protein levels at 8 and 12 weeks.
  3. We conducted a retrospective analysis of the ICARIA-MM trial at the final analysis date, when 162 PFS events had occurred. Two interim analyses were retrospectively planned when 50% and 65% of the total number of expected events had occurred. Available data were analyzed using a classic Cox model and the joint model built from Phase 1-2 data. JMpop and JMindiv were applied to estimate ICARIA-MM final PFS HR and its 1- αOBF CI, and then they were compared to CoxObs.

RESULTS:

  1. All approaches control the type 1 error rate. JMindiv offers similar power estimates to detect treatment effect compared to the classic approach CoxObs at the final analysis, but slightly lower power estimates at interim analysis. The other simulation-based approaches show much lower power estimates.
  2. The selected nonlinear joint model on Phase 1-2 trial data successfully captured drug effects on serum M-protein kinetics and its association with PFS was best captured by the instantaneous change in serum M-protein level. Model-based simulations shown the model abilities to predict the ICARIA-MM study IPd arm and confirmed the Isa 10 mg/kg QW/Q2W dosing with Pom-Dex used in the ICARIA-MM trial.
  3. Using JMindiv or JMpop did not allow to detect the benefit of IPd on PFS prolongation earlier than the classic approach CoxObs. All approaches were still inconclusive at the 2nd interim analysis.

CONCLUSION:

Our results show the ability of a joint modeling-based approach based on phase 1-2 trial to support the choice of dose and predict accurately the median PFS of the experimental treatment arm in a pivotal phase 3 trial. In our simulation study, we observed that accounting for the estimation uncertainty around the several parameters of the joint model led to a loss of power in the interim analysis context, to a lesser extent when using the individual parameters conditional distribution.  Finally, our findings encourage systematic development of a nonlinear joint model, using data from early-phase clinical trials, to guide dosing regimen selection and inform trial design for a Phase 3 pivotal study. This modeling approach would possibly support decision making in the context of interim analyses if the event of interest is less contemporary to the biomarker dynamics (e.g. considering overall survival).

  1. Thai H, et al. Joint modelling and simulation of M-protein dynamics and progression-free survival for alternative isatuximab dosing with pomalidomide/dexamethasone. Br. J. Clin. Pharmacol. 88, 2052–2064 (2022).
  2. Rizopoulos D. Joint models for longitudinal and time-to-event data with applications in R. 20121833, i–xiii (2012).
  3. 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).
  4. Sarclisa® (isatuximab-irfc) injection, for intravenous use [prescribing information]. Sanofi, Bridgewater, NJ. (2021).
  5. Deckert J, et al. SAR650984, A novel humanized CD38-targeting antibody, demonstrates potent antitumor activity in models of multiple myeloma and other CD38+ hematologic malignancies. Clin. Cancer Res. 20, 4574–4583 (2014).
  6. Martin T, et al. Phase I trial of isatuximab monotherapy in the treatment of refractory multiple myeloma. Blood Cancer J. 9, 41 (2019).
  7. Mikhael J, et al. A dose-finding Phase 2 study of single agent isatuximab (anti-CD38 mAb) in relapsed/refractory multiple myeloma. Leukemia 34, 3298–3309 (2020).
  8. Mikhael J, et al. A phase 1b study of isatuximab plus pomalidomide/ dexamethasone in relapsed/refractory multiple myeloma. Blood. 134, 123–133 (2019).
  9. 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).
  10. Thai H, et al. Joint modelling and simulation of M-protein dynamics and progression-free survival for alternative isatuximab dosing with pomalidomide/dexamethasone. Br. J. Clin. Pharmacol.88, 2052–2064 (2022).
  11. Thai H, et al. Model-based simulation to support the approval of isatuximab alone or with dexamethasone for the treatment of relapsed/refractory multiple myeloma in Japanese patients. CPT: Pharmacomet. Syst. Pharmacol.12, 1846–1858 (2023).
  12. Demets DL, Lan KKG. Interim analysis: The alpha spending function approach. Stat. Med.13, 1341–1352 (1994).

Reference: PAGE 32 (2024) Abstr 11152 [www.page-meeting.org/?abstract=11152]

Poster: Drug/Disease Modelling - Oncology

PDF poster / presentation (click to open)