(W-086) Prediction of progression free survival using tumor growth inhibition models: Effect of model uncertainty, inter-individual variability and observational noise on prediction accuracy

Arya Pourzanjani, NA – Principal Scientist, Clinical Pharmacology Modeling and Simulation, Amgen Inc., South San Francisco, CA, USA; Guido Jajamovich, NA – Scientific Associate Director, Clinical Pharmacology Modeling and Simulation, Amgen Inc., South San Francisco, CA, USA; Khamir Mehta, NA – Senior Director, Clinical Pharmacology Modeling and Simulation, Amgen Inc., South San Francisco, CA, USA

Objectives: The longitudinal changes in tumor size along with the RECIST guidelines are the basis to determine efficacy endpoints like progression free survival (PFS) and objective response rate (ORR) for solid tumor indications. Tumor growth inhibition (TGI) models aim to capture the observed dynamics of tumor size in response to therapy and are recently being used to make predictions for PFS and ORR to obtain early read on the drug efficacy. The variability associated with patient heterogeneity, measurement noise, and the potential uncertainty in model parameters can limit the accuracy of prediction of the TGI models. The objective of this work is to investigate the propagation of the model uncertainty and the variability in the TGI models w.r.t the prediction of drug efficacy metrics of ORR and PFS. Using a systematic analysis, we explore the potential biases in ORR/PFS predictions from TGI models due to parameter uncertainty, inter-individual variability (IIV) in parameters, and residual unexplained variability (RUV) in tumor size.

Methods: We explore the ability of the modeling framework to recapitulate the ground truth for several simulated scenarios for multiple TGI model structures. Time of progression and response is determined using sum of longest diameters (SLD) of target lesions, progression of nontarget and new lesions, according to RECIST guidelines [1]. We initiate our analyses of a mono-exponential tumor growth/inhibition with a analytical approach to understand the impact of the variability and uncertainty on extending the TGI models to predict ORR and PFS, and then extend the analysis for more complex TGI models that include treatment resistance and tumor heterogeneity using numerical simulations. To illustrate the implication on real world example we utilize the example simulation of a Doxorubicin TGI model with treatment resistance [1] driven by a population PK model [2].

Results: Our analysis of the mono-exponential TGI model show that while PFS predictions are sensitive to the typical value of effective tumor growth rate, ORR prediction is sensitive to both its typical value and IIV. Including uncertainty in the population parameters increases prediction intervals of PFS and ORR but does not affect median PFS. The simulation-based analysis confirms these insights.

Conclusions: Our analysis illustrates how sources of variability like IIV, RUV and parameter uncertainty may affect the prediction intervals of PFS and ORR and may bias median estimates of PFS and ORR, and can be further used for optimizing the modeling framework and/or trial design for early prediction of survival endpoints.

Citations: [1] Jiajie Yu et al., A New Method to Model and Predict Progression Free Survival Based on Tumor Growth Dynamics, CPT Pharmacometrics Syst. Pharmacol. (2020) 9
[2] Vijay S. Kumar et al., Population pharmacokinetics of doxorubicin in Indian cancer patients using NONMEM, Clinical Research and Regulatory Affairs, (2009) 26(4)