PII-124 - PREDICTION OF CLINICAL OUTCOMES IN A HETEROGENOUS NSCLC VIRTUAL PATIENT POPULATION: A QUANTITATIVE SYSTEMS PHARMACOLOGY APPROACH.

A. Mittal¹, W. Duncan¹, A. Markazi¹, N. Parish¹, R. Kowalski¹, N. Brostoff¹, D. Simpson¹, J. Woodhead², R. Suderman¹; ¹Simulations Plus, Pittsburgh, PA, USA, ²Simulations Plus, Durham, NC, USA.

Background: Immune checkpoint inhibitors (ICIs) have shown promise for treating patients with advanced and/or metastatic non-small cell lung cancer (NSCLC). However, drug developers lack a quick and efficient approach to identify efficacious combination therapies for heterogenous patient populations. We developed a quantitative systems pharmacology (QSP) model to understand how different disease features such as NSCLC histology and immune microenvironment relate to disease outcome.

Methods: We identified immune cell types and cytokines that play a key role in NSCLC pathogenesis and immune response. With this knowledge and Simulations Plus’ library of previously modeled tumor-related processes, we developed a mechanistic immune-oncology model of NSCLC in the Thales modeling platform. We used additional clinical and pre-clinical data to represent distinct tumor sub-types, heterogeneous immune environments, and inter-patient differences. We reviewed published and ongoing ICI clinical trials and collected clinical outcome data (BOR, ORR, PFS/OS) to constrain a virtual population representative of advanced/metastatic NSCLC. We also compiled data from chemotherapy-only trials to inform the virtual population’s response to chemotherapy. All together, clinical response data from over 30 clinical trials spanning 15 therapeutics were curated to fit and validate a virtual patient population that captures both clinical efficacy and inter-patient heterogeneity.

Results: The optimized virtual population demonstrates its real-world applicability by successfully predicting clinical outcomes for unseen therapeutic regimens. The virtual population captures disease heterogeneity and predicts the clinical outcomes of different patient populations, including squamous vs. non-squamous tumors and tumors with varying levels of PD-L1 expression. The mechanistic nature of the model allows for further exploration of tumor immune dynamics post ICI therapy and enables characterization of specific patient features that may enable individualized medicine.

Conclusion: Our NSCLC QSP model can serve as a computational platform to test existing and new drug regimens in the clinic. The platform will serve as a tool to accelerate drug development and can be readily extended to other NSCLC patient populations.