Purpose: Standard endpoints often poorly predict overall survival (OS) with immunotherapies. We investigated the predictive performance of model-based tumor growth inhibition (TGI) metrics using data from atezolizumab clinical trials in patients with non–small cell lung cancer. Patients and Methods: OS benefit with atezolizumab versus docetaxel was observed in both POPLAR (phase II) and OAK (phase III), although progression-free survival was similar between arms. A multivariate model linking baseline patient characteristics and on-treatment tumor growth rate constant (KG), estimated using time profiles of sum of longest diameters (RECIST 1.1) to OS, was developed using POPLAR data. The model was evaluated to predict OAK outcome based on estimated KG at TGI data cutoffs ranging from 10 to 122 weeks. Results: In POPLAR, TGI profiles in both arms crossed at 25 weeks, with more shrinkage with docetaxel and slower KG with atezolizumab. A log-normal OS model, with albumin and number of metastatic sites as independent prognostic factors and estimated KG, predicted OS HR in subpopulations of patients with varying baseline PD-L1 expression in both POPLAR and OAK: model-predicted OAK HR (95% prediction interval), 0.73 (0.63–0.85), versus 0.73 observed. The POPLAR OS model predicted greater than 97% chance of success of OAK (significant OS HR, P < 0.05) from the 40-week data cutoff onward with 50% of the total number of tumor assessments when a successful study was predicted from 70 weeks onward based on observed OS. Conclusions: KG has potential as a model-based early endpoint to inform decisions in cancer immunotherapy studies.
ASJC Scopus subject areas
- Cancer Research