Joint modeling of longitudinal tumor burden and time-to-event data to predict survival: application to aflibercept in second line metastatic colorectal cancer
Hoai-Thu Thai (1), Jean-Baptiste Fau (1), France Mentré (2), Emmanuelle Comets (2), Christine Veyrat-Follet (1)
(1) Drug Disposition, Department of Disposition, Safety and Animal Research (DSAR), Sanofi, Paris, France; (2) INSERM, IAME, UMR 1137, F-75018 Paris, France; Univ Paris Diderot, Sorbonne Paris Cité, F-75018 Paris, France
Objectives: Aflibercept (ziv-aflibercept in the US, ZALTRAP®) is a fusion protein of human vascular endothelial growth factor (VEGF) receptor domains that binds to VEGF-A, VEGF-B, and PlGF and inhibits tumor growth (1). In metastatic colorectal cancer, the VELOUR trial demonstrated significantly improved overall survival (OS) for aflibercept in combination with FOLFIRI, after failure of an oxaliplatin based regimen (2). The aim of this work was to analyze the treatment effect on tumor growth kinetics and the link to survival using a joint modeling framework accounting for informative dropouts.
Methods: Model building was conducted using 1069 evaluable patients from the VELOUR trial. Longitudinal data of tumor size (sum of target lesions) were first analyzed alone. Then, informative dropouts and OS data were both included for joint modeling using hazard frailty models (3) with shared latent random effects. Parameters were estimated by maximizing the exact joint likelihood with the SAEM algorithm implemented in MONOLIX 4.3.2. Model selection was based on log-likelihood ratio tests and BIC. VPC and Kaplan-Meier plots were used to explore the impact of dropouts and to evaluate model performance.
Results: The best longitudinal PK/PD model involved a target-mediated drug disposition (TMDD) submodel for free and bound aflibercept (4), a kinetic-pharmacodynamic (K-PD) submodel for FOLFIRI dosing data and a tumor growth inhibition (TGI) model for the tumor size data with treatment effect driven by free aflibercept. In the joint model, the hazards of dropout and death depended on time and tumor size. The Weibull distribution best described informative dropouts while the log-logistic distribution best described the OS data. The simulated median OS were in agreement with those observed in both reference (FOLFIRI alone) and treatment (aflibercept + FOLFIRI) arms.
Conclusions: This present joint model well characterized the time course of tumor size, the treatment effect and the observed OS of patients in the VELOUR trial. By linking the full time-course of tumor size to survival and taking into account the informative dropouts, this model should provide a good prediction of clinical outcomes (e.g. survival in oncology) when performing model-based simulations of new clinical trials (e.g. new dose regimen, dose intensification in subpopulations of interest). A safety component should also be taken into account in this framework to ensure an adequate efficacy/safety balance.
References:
[1] Chu QS. Aflibercept: an alternative strategy for inhibiting tumour angiogenesis by vacular endothelial growth factors. Expert Opin Biol Ther 2009; 9: 263-271.
[2] Van Cutsem E, Tabernero J, Lakomy R, Prenen H, Prausova J, Macarulla T, Ruff P, van Hazel GA, Moiseyenko V, Ferry D, McKendrick J, Polikoff J, Tellier A, Castan R, Allegra C. Addition of aflibercept to fluorouracil, leucovorin, and irinotecan improves survival in a phase III randomized trial in patients with metastatic colorectal cancer previously treated with an oxaliplatin-based regimen. J Clin Oncol 2012; 30 (28): 3499-3506.
[3] Shen Y, Anderson A, Sinha R, Li Y. Joint modeling tumor burden and time to event data in oncology trials. Pharm Stat 2014; 13(5): 286-93.
[4] Thai HT, Veyrat-Follet C, Mentré F, Comets C. Population pharmacokinetic analysis of free and bound aflibercept in patients with advanced solid tumors. Cancer Chemother Pharmacol 2013, 72(1): 167-180.
