P27.06 Multiscale NSCLC Tumor Growth Knowledge-Based Model Reproduces Tumor-Non-Progression under Gefitinib

Abstract

Non-small cell lung cancer (NSCLC) is the leading form of lung cancer and adenocarcinoma (LUAD) its most common histotype. Tumor size, part of the TNM-staging, is important for prognosis and treatment guidance. Response to Tyrosine Kinase Inhibitors, e.g. Gefitinib, is altered with certain epidermal growth factor receptor (EGFR+) gene mutations making time-to-progression (TTP) predictions difficult. We therefore developed an in silico EGFR+LUAD model to characterize tumor size and TTP in advanced-stage adenocarcinoma patients (IIIb or higher) with EGFR mutations (exon19-deletion (E19+) or exon21-L858R-substitution (E21+)). The model is a mechanistic representation of tumor evolution upon Gefitinib administration, including tumor heterogeneity, age, gender, initial clinical stage and smoking status as covariates. Three-step in silico model development: 1. Model Building using a Knowledge and a Computational Model (CM): Pathophysiology of EGFR+LUAD was characterized with seven sub-models: mutational burden, EGFR downstream pathways, tumor growth and heterogeneity, Gefitinib PK/PD, treatment-induced resistance and clinical outcome. For each sub-model, relevant biological entities and their functional relationships were extracted from scientific papers and translated into ordinary differential equations (ODEs). The CM has 43 variables, 170 parameters and 18 to 83 ODEs reflecting intra-tumor heterogeneity. 2. Calibration with information from scientific literature: Spheroids, xenografts and clinical results were used for stepwise calibration. 3. Validation against published data: Patients[1] (n=159) had E19+ or E21+ mutations and were treated with Gefitinib (250mg/day orally). A Virtual Population with equivalent baseline characteristics was simulated using the calibrated CM. Kaplan-Meier curves show tumor non-progression over time. Validation metrics: (1) “Raw-Data-Coverage” expressing the fit with the 95% prediction interval (PI) of the simulated curve (computed by bootstrapping), (2) Comparison of bootstrapped simulated survival curves (n=159) with log-rank tests (LR) for statistical significance (α=0.05). Coverage of the model simulations with raw data was 95.92%. Proportion of non-significant bootstrapped LR tests comparing the observed curve with simulated curves was 88.43% (Figure 1). We reproduced a clinical trial of advanced-stage EGFR+LUAD patients treated with Gefitinib. We created a modular, reusable, multiscale Knowledge-Based model adaptable to also test the efficacy of other treatments and drug combinations on tumor-non-progression. This enables us to identify best responders and to optimize trial designs in silico.