Model-based prediction of outcome of the atezolizumab Phase 3 study OAK in non-small cell lung cancer based on early tumor kinetic data.

Abstract

e14517 Background: Standard early endpoints (ORR, PFS) are often poor predictors of overall survival (OS) for immunotherapies. Model-based tumor kinetic (TUK) metrics may offer an alternative (Claret, JCO 2009 and 2013). We investigated the predictive performance of TUK metrics using early data from atezolizumab clinical trials in patients (pts) with non-small cell lung cancer (NSCLC). Methods: OS benefit with atezolizumab (atezo) vs. docetaxel (doc) was observed in both Phase II POPLAR (hazard ratio 0.73, p = 0.040, n = 297, Fehrenbacher Lancet 2016) and Phase III OAK (HR = 0.73, p = 0.0003, n = 850, Rittmeyer Lancet 2016) studies although PFS was similar between arms. A multivariate model linking baseline patient characteristics and tumor growth rate (TGR, estimated using time profile of sum of longest diameters of target lesions, RECIST 1.1) to OS was developed using POPLAR data and qualified to predict outcome by PD-L1 status in both POPLAR and OAK. This model was used to predict OAK OS outcome based on patient characteristics and estimated TGR from TUK data at several OAK data cut-offs ranging from 10 to 122 weeks. Results: In POPLAR, TUK profile in atezo and doc arms crossed at about 25 weeks with more initial shrinkage in doc-treated patients and slower TGR in atezo-treated patients. A lognormal OS model with albumin and number of metastatic sites as independent baseline prognostic factors and estimated TGR predicted OS HR in sub-populations of pts based on baseline PD-L1 expression both in POPLAR as well as in OAK e.g model-predicted OAK HR: 0.73 [0.63-0.85]. Based a retrospective analysis on TUK data cuts, the POPLAR OS model predicted significant OS HR in OAK as soon as 40 week cut-off: HR: 0.67 [0.56-0.80] with success in 99% of the replicates with only 50% of the total number of tumor assessments. At this 40 week data cut, immature observed OS HR was 0.84 [0.59-1.21, p = 0.347] and HR only reached significance at 70 weeks. Conclusions: Model predictions based on TUK can be used to predict success early before maturation of OS data as shown with OAK. The model-based biomarker TUK metric TGR has strong potential as an alternative early endpoint to evaluate efficacy in cancer immunotherapy studies.