Comparisons of Multiple Exposure-Response Methodologies in Oncology
Dan Polhamus (1), Jonathan French (1), Shang-Chiung Chen (2), Bei Wang (2), Chunze Li (2), Dan Lu (2), Sandhya Girish (2), Jin Jin (2), Angelica Quartino (2)
1 Metrum RG, Tariffville, CT; 2 Genentech, San Francisco, CA
Objectives: Exposure-Response assessment in oncology is complicated by many factors, e.g. alternate dosing history. In addition, for monoclonal antibodies (mAbs) exposure is confounded by key prognostic factors for the disease. M&S exposure-response (ER) analysis focuses on direct or indirect (e.g., via tumor growth inhibition (TGI)) ER for progression free survival (PFS) and overall survival (OS). Here, we compare different aspects of multiple direct ER methods, with a case example in oncology.
Methods: Direct ER methods were compared using data from an oncology Phase 3 trial: 1) stratified Kaplan-Meier (KM) estimates by exposure quartiles, 2) Cox proportional hazards (CPH) analysis with covariate adjustment by exposure quartiles or as a continuous function, 3) case matching (CM) to address confounding effects on ER by exposure quartiles, and 4) parametric survival modeling (PS) using covariate adjustment and longitudinal exposure to account for alternate dosing history and for extrapolation to other dosing regimens.
Results: KM, while easy to interpret and vizualize, don't account for confounding prognostic. Confounding factors are addressed with CPH, however it relies on assumptions about the relationship of covariates with outcome and the summary exposure metric. Case matching methods has recently been proposed by FDA to ease these assumptions [1], but ER assesment is no longer directly addressed as in the same manner as in CPH. A doubly robust ER using CPH within CM on strata of exposure is demonstrated, guarding against either poor matching or model misspecification. PS with covariate adjustments offers a direct longitudinal ER method, to complement indirect TGI methods. While TGI is a more mechanistic approach, PS may be preferred in cases where the tumor dynamics of the target leasions cannot fully predict survival due to e.g. new leasions and micrometastases. Longitudinal approaches has the advantage of directly accounting for alternate dosing history compared to ER methods that use summary exposure metrics, and allows for simulation of other dosing regimens.
Conclusions: Several direct methods for ER analysis are available. Here, we evaluate direct ER methods and propose a M&S strategy that is fit for purpose and provides a clear strategy for dose optimization if indicated, and addresses regulatory review questions.
References:
[1] Yang J, et al. J Clin Pharmacol. 53(2): 2013
