Operating characteristics of tumor growth inhibition-overall survival models to support early Phase Ib decisions: An evaluation in first-line metastatic non-small cell lung cancer patients treated with atezolizumab plus chemotherapy based on the Phase III study IMpower150

René Bruno (1), Mathilde Marchand (2), Kenta Yoshida (3), Haocheng Li (4), Wei Zhu (5), Francois Mercier (6), Pascal Chanu (7), Benjamin Wu (3), Anthony Lee (5), Chunze Li (3), Jin Y Jin (3)

(1) Clinical Pharmacology, Genentech-Roche, Marseille, France; (2) Certara Strategic Consulting, Paris, France; (3) Clinical Pharmacology, Genentech, South San Francisco, California, USA; (4) Product Development, Roche/Genentech, Mississauga, Ontario, Canada (5) Product Development, Genentech, South San Francisco, California, USA; (6) Biostatistics, Roche, Basel; (7) Clinical Pharmacology, Genentech-Roche, Lyon, France

Objectives: Tumor growth inhibition (TGI)-overall survival (OS) models have been shown to predict OS distributions and study outcomes (OS hazard ratio (HR) in a number of cancers and settings [1]. Only a few studies are evaluating tumor dynamic-based endpoints in oncology early clinical development (see e.g. [2, 3]). The goal of this work is to assess operating characteristics of TGI metrics and TGI-OS model simulations to support early Phase Ib decisions by resampling of a large Phase 3 trial in non-small-cell lung cancer (NSCLC).

Methods: IMpower150, was a randomized Phase 3 study in first-line NSCLC. Atezolizumab plus bevacizumab plus carboplatin-paclitaxel (Arm B: Atezo+Bev+CP) significantly improved progression-free survival and OS compared to the control treatment (Arm C: Bev+CP) in the intent-to-treat population with no EGFR/ALK mutations (wild type) (ITT-WT) [4]. Baseline patient’s characteristics and longitudinal tumor sum of longest diameters (SLD, RECIST 1.1) were bootstrapped to mimic a Phase 1b study of 40 patients per arm (experimental and control) with 6 months recruitment and 6 months follow-up after the last patient recruited (Nreplicate=500). Individual TGI metrics: shrinkage rate (KS), growth rate (KG), 24 weeks to baseline tumor size ratio (TR24), and time to regrowth (TTG), were estimated using a bi-exponential model [5, 6] implemented in NONMEM version 7.3.0. OS HR was simulated with a previously developed multivariate lognormal model involving baseline patient characteristics and KG estimates as covariates [7] (estimated without IMpower150 data). For OS simulations, a Phase 3 trial was simulated with 400 patients per arm obtained by bootstrapping from the 40 patients per arm of the virtual 500 replicates of the Phase 1b study to mimic real conditions of an early clinical decision. To assess the operating characteristics of the approach, TGI metrics and simulated OS distributions evaluated in two scenarios: Arm B vs. Arm C (H1) and Arm C vs. Arm C (H0) to assess ‘correct go decision’ (power) and ‘incorrect go decision’ (type 1 error) rates, respectively. As a benchmark, we also compared progression free survival (PFS) data in the subsamples. A two-sided Wilcoxon rank test (α-level 0.05) was used to compare TGI metrics, and a two sided log-rank test (α-level 0.10) was used to compare median simulated OS or observed PFS.

Results: Under H1, the power to detect a difference in KG or TTG (in a Ph1b setting), or in OS (in a Ph3 setting) was above around or 70%:

* % difference of experimental vs. control arm, median across 500 replicates

Under H0, the TGI metrics had type I error rates of 3.0 to 4.8%. The observed PFS analysis (2-sided log rank test at p=0.05) only had a 28.0% power to show a difference across Arms. 

Conclusions: This evaluation suggests that TGI-OS model-based could be used as exploratory endpoints to inform early clinical decisions as recently suggested [8]. These results compare well with conclusions from a simulation-based work in metastatic colorectal cancer [3]. Expansion of this work is planned in other settings e.g. single-arm studies commonly used in early oncology clinical development.

References:
[1] Bruno R et al. Progress and opportunities to advance clinical cancer therapeutics using tumor dynamic models. Clin Cancer Res (2020) 26, 1787-95.
[2] Claret L et al. Modeling and simulation to assess the use of change in tumor size as primary endpoint in Phase II studies in oncology. PAGE 17 (2008) Abstr 1386 [www.page-meeting.org/?abstract=1386].
[3] Seurat J et al. Comparison of various phase I combination therapy designs in oncology for evaluation of early tumor shrinkage using simulations. CPT:PSP (2020) 9, 686-94.
[4] Socinski MA et al. Atezolizumab for first-line treatment of metastatic non squamous NSCLC. N Engl J Med (2018) 378, 2288-301.
[5] Stein WD et al. Tumor regression and growth rates determined in five intramural NCI prostate cancer trials: the growth rate constant as an indicator of therapeutic efficacy. Clin Cancer Res (2011) 17, 907-17.
[6] Claret L et al. A model of overall survival predicts treatment outcomes with atezolizumab versus chemotherapy in non-small cell lung cancer based on early tumor kinetics. Clin Cancer Res (2018) 24, 3292-8.
[7] Chan P et al. Prediction of overall survival in patients across solid tumors following atezolizumab treatments: a TGI-OS modeling framework. CPT:PSP (2021) doi.org/10.1002/psp4.12686.
[8] Gong et al. An FDA analysis of the association of tumor growth rate and overall and progression-free survival in metastatic non-small cell lung cancer (NSCLC) patients. J Clin Oncol (2020) doi: 10.1200/JCO.2020.38.15_suppl.9541.

Reference: PAGE 29 (2021) Abstr 9679 [www.page-meeting.org/?abstract=9679]

Oral: Drug/Disease Modelling

PDF poster/presentation (click to open)