Model-based estimates of tumor growth inhibition (TGI) metrics to predict for overall survival (OS) in first-line non-small cell lung cancer (NSCLC).

Abstract

e19049 Background: A survival model using change in tumor size at week 8 from baseline (CTS) (Clin Pharmacol Ther 86:167-174, 2009) was used to predict OS of the motesanib phase III study in NSCLC (Clin Pharmacol Ther 92:631-634, 2012). However CTS may not fully capture treatment effect on OS. We investigated other TGI metrics to predict OS using historical phase III data. Methods: Various TGI metrics, CTS, time to growth (TTG), and tumor growth rate (G), were estimated using two longitudinal tumor size models (Clin Cancer Res 17:907-917, 2011, PAGE 21 [www.page-meeting.org/?abstract=2328], 2012, JCO in press, 2013) developed from phase III data (1218 1st-line patients with NSCLC treated with doxetaxel (D) plus cisplatin (Cs) or carboplatin (Cb) or vinorelbine (V) plus Cs) (J Clin Oncol, 21:2016-3024, 2003). TGI metrics and baseline prognostic factors were assessed in univariate Cox and multivariate parametric survival models. Alternative models were qualified by simulating OS distributions in 1000 replicates of the phase III study and checking that observed OS distribution belongs to the 95% prediction intervals. Results: In a univariate analysis all TS metrics were strong predictors of OS (p<0.0001). TTG and G were similar and much better than CTS (likelihood ratio test). OS was best described by a log-normal distribution and the influence of the covariates was assessed based on linear regression of the logarithm of OS (days). Parameter estimates of the best multivariate model are shown in the Table. The model was qualified in predicting OS distributions in the 3 arms (D+Cs, D+Cb, V+Cs). Conclusions: With further external validation, this model could be used to analyze longitudinal NSCLC tumor size data and their relationship to OS, in phase II studies of new agents and support end-of-phase-II decisions. [Table: see text]