Tumor kinetic modeling and identification of predictive factors for tumor response to durvalumab in patients with non-small cell lung cancer (NSCLC).

Abstract

11555 Background: Durvalumab is a human monoclonal antibody that binds to PD-L1 and blocks its interaction with PD-1 and CD80. The primary objectives of this analysis were to describe the longitudinal tumor size profiles and identify the key factors predicting tumor growth and regression following durvalumab. Methods: Longitudinal tumor size data obtained from NSCLC patients in study 1108 (all lines of therapy) and ATLANTIC (third line and beyond) following durvalumab treatment were modeled using nonlinear mixed effect modeling. Tumor kinetics were described by four key parameters: tumor growth and killing rate constants, fraction of durvalumab-sensitive tumor cells, and delay time for tumor killing. Potential predictive factors for tumor growth and regression were evaluated in a multi-variable covariate analysis. The model was used to simulate response rates at different tumor PD-L1 expression cutoffs. Results: Tumor kinetic modeling accurately described the longitudinal tumor response profiles from NSCLC patients in both studies. The factors associated with more rapid tumor growth were liver metastases, ECOG score > 0, high neutrophil-to-lymphocyte ratio and EGFR/ALK mutation. Tumor cell PD-L1 expression, baseline tumor size and smoking history were identified as significant predictive factors for tumor killing or the fraction of sensitive tumor cells. Simulations using the tumor kinetic model showed increased response rates in patients with higher tumor cell PD-L1 expression (increased by 9-11% and 10-14% with 25% and 50% cutoff, respectively), patients receiving durvalumab as first-line therapy (increased by 12% vs. 2nd line/above), and patients with smoking histories (increased by 4-5% vs. non-smokers). Conclusions: Tumor kinetic modeling identified factors that predict tumor progression and response following durvalumab in NSCLC patients. The multivariate analysis accounts for various predictive factors within predictive biomarker strata, allowing better interpretation of different biomarker cutoffs. The modeling technique can potentially guide patient selection/enrichment, clinical trial design strategies and tumor biology. Clinical trial information: NCT02087423 and NCT01693562.