Operating characteristics of TGI metrics to support early Phase Ib decisions in unresectable hepatocellular carcinoma patients based on an historical Phase III study (IMbrave150)
Antonio Gonçalves (1), Mathilde Marchand (1), Victor Poon (2), Phyllis Chan (2), René Bruno (3)
(1) Certara Strategic Consulting, Paris, France; (2) Clinical Pharmacology, Genentech, South San Francisco, California, USA; (3) Clinical Pharmacology, Genentech-Roche, Marseille, France
Introduction: Recently, a modeling framework has been developed to assess the operating characteristics of tumor growth inhibition (TGI) metrics to support early decision-making in non-small cell lung cancer [1]. In this indication, the tumor growth rate constant geometric mean ratio (KG GMR) had good operating characteristics with satisfactory power (>80%) and reasonable type 1 error (<20%) suggesting that it could be used to support early decisions. The objectives of this work were to further explore the operating characteristics of TGI metrics in a Phase 1b trials in another indication. In the present work we utilized IMbrave150 study, a randomized Phase III study evaluating atezolizumab plus bevacizumab (experimental arm) versus sorafenib (control arm) in unresectable hepatocellular carcinoma (HCC) patients. The study showed that atezolizumab plus bevacizumab significantly improved OS compared to the control treatment in the intent-to-treat population [2].
Methods: Longitudinal tumor and overall survival (OS) data from IMbrave150 were well characterized by two TGI-OS models [3]. To mimic a Phase Ib study, the patient’s baseline characteristics and tumor measurements from IMbrave150 were resampled. Under the alternative hypothesis (H1) that atezolizumab + bevacizumab led to benefit compared to sorafenib, we resampled 40 patients per arm from both experimental and control arms and assessed the ‘correct go decision rate’. In contrast, under the null hypothesis (H0), we resampled 40 patients per arm from the control arm twice to mimic a study with a ‘no treatment’ effect. Under both H0 and H1, we assumed a 10-month recruitment period, and 24 weeks follow-up after the last patient was recruited. Then, each replicated dataset (N=500) was re-estimated using a bi-exponential model [4, 5], and individual TGI parameters/metrics were estimated, namely, the tumor growth rate constant (KG), the tumor shrinkage rate constant (KS), the tumor ratio to baseline at week 24 (TR24) and the time-to-regrowth (TTG). For each parameter, an effect size of interest was calculated. For KG, KS, and TR24, the between-arm geometric mean ratio (GMR) was calculated, while for TTG the hazard ratio (HR) was derived. Then, the probability of the effect size being greater or lower than the desired clinical threshold was calculated across the 500 replicates under either H1 or H0, to assess ‘correct go decision’ (power) and ‘incorrect go decision’ (type 1 error) rates, respectively. The results were presented as receiver operating characteristics (ROC) curves.
Results: Previously developed TGI-OS models showed that log(KG) [3] or log(TR24) could both well predict the OS distribution of the full Phase III data (median OS HR [95%PI] = 0.722 [0.584 – 0.884] and 0.704 [0.563 – 0.870]; respectively vs. observed HR = 0.67). Interestingly, when assessing the operating characteristics, the probability of finding a KG GMR<0.9 provided insufficient power (66%) and high type 1 error (43%) while TTG HR<0.6 and TR24 GMR<0.9 had good power (79% and 91%, respectively) and low type 1 error (<20%).
Conclusions: This evaluation suggests that a TGI metrics that provides a prediction of Phase 3 outcome (OS HR) (here KG) may not have good operating characteristics and be used to support early decisions. However, TR24 or TTG could be used as relevant alternatives. Perspectives of this work include investigation of other TGI models as well as the use of machine learning approaches.
References:
[1] Bruno R et al. Tumor Dynamic Model-Based Decision Support for Phase Ib/II Combination Studies: A Retrospective Assessment Based on Resampling of the Phase III Study IMpower150. Clin Cancer Res (2023) 29, 1043-55.
[2] Finn RS et al. Atezolizumab plus bevacizumab in unresectable hepatocellular carcinoma. N Engl J Med (2020) 381, 1894-1905.
[3] Shemesh CS et al. Early Decision Making in a Randomized Phase II Trial of Atezolizumab in Biliary Tract Cancer Using a Tumor Growth Inhibition-Survival Modeling Framework. Clin Pharmacol Ther (2023) 114, 644-51.
[4] 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.
[5] 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.
