(M-009) Translation of a Preclinical Tumor Growth Inhibition Model to a Clinical Setting: Two Approaches to Incorporate Resistance in a Tumor Dynamics Model

Catherine Liu – UCSD; Micaela Reddy – Pfizer; Keagan Collins – Pfizer; Erik Hahn – Pfizer

Objectives: One key driver of acquired resistance, a major challenge in oncology and reason for treatment failure, is intratumor heterogeneity. As the tumor evolves and adapts in response to treatment, subpopulations of cancer cells emerge—variability within the tumor itself may play a crucial role in how resistance develops over time. Other factors such as acquired resistance mutations or changes to the tumor microenvironment may also contribute to acquired drug resistance. Tumor dynamics models (TDMs) aim to describe changes in tumor size. Accounting for resistance is a key aspect of various literature models.1,2 Objectives were to adapt a preclinical semi-mechanistic TDM to the clinical setting, explore the role of acquired resistance in patients with unresectable or metastatic melanoma with a BRAF V600E/K mutation, apply the model to data from patients, and compare the approach with simpler models that have been applied to clinical data.

Methods: The semi-mechanistic Tomasetti et al. model was selected for adaption and two new models were adapted from it to describe clinical data for the sum of the longest diameter over time in advanced melanoma patients. The models were applied to clinical data collected from 666 evaluable patients enrolled in the COLUMBUS study (NCT01909453).2 The resistance mutation (RM) model contains two separate compartments, one resistant to treatment and growing and another sensitive to treatment and shrinking, with the mutation rate (u) contributing to the growth of the resistant compartment. The simplified resistance mutation (SRM) model has one compartment with sensitive and resistant proportions, with u as the mutation rate and contributing to growth without explicit movement between proportions. The RM and SRM models were then compared to two biexponential models of Stein et al. and Chatterjee et al., which are TDM models with a simpler approach to resistance.3,4

Results: The observed data was well-characterized by all models tested. Both the RM and SRM models adequately described tumor shrinkage and regrowth for individual patients in our dataset despite notable heterogeneity in tumor profiles. The objective function value (OFV) was lower for the RM model than the SRM (Δ-2LL = -31); however, the models are not nested. The Stein model, the simplest model, had the lowest between-subject variability (BSV) and shrinkage with better parameter precision but with a higher OFV. There were similar trends among the Chatterjee, RM, and SRM models for OFV, BSV, shrinkage, and precision. Given resulting tumor kinetic simulations from the RM and SRM models produced different shapes than the Stein et al. and Chatterjee et al. models, these models may uniquely fit some datasets better than others. Additional model parameters were included in the RM and SRM models, which could allow for the incorporation of covariate effects beyond current literature models.

Conclusions: The RM and SRM models presented here are two new models that can be used for TDM. The modeling framework established provides confidence that more complex descriptions of resistance can be applied to clinical datasets.

Citations: Citations:
[1] Yin A, Moes DJAR, van Hasselt JGC, Swen JJ, Guchelaar HJ. CPT Pharmacometrics Syst Pharmacol. 2019;8(10):720-737
[2] Tomasetti C, Levy D. Math Biosci Eng. 2010;7(4):905-18
[3] Stein WD, Figg WD, Dahut W, et al. Oncologist. 2008;13(10):1046-1054
[4] Chatterjee MS, Elassaiss-Schaap J, Lindauer A, et al. CPT Pharmacometrics Syst Pharmacol. 2017;6(1):29-39

Keywords: Tumor Dynamics Modeling, Tumor Growth Inhibition, Oncology