Model-based analysis of the relationship between pembrolizumab exposure and efficacy in patients with melanoma and NSCLC: across indication comparison
M. Chatterjee*, D.C. Turner*, M. Ahamadi*, Dinesh de Alwis*, J. Elassaiss-Schaap**, T. Freshwater*, R de Greef***, C. Li*, K. Mayawala*, J. Stone*, A. Kondic* (presenting author)
*-Merck Sharp and Dohme, 126 E Lincoln Ave, Rahway, NJ 07065; ;**PD Value; ***--Quantitative Solutions;Pivot Park Molenweg 79, 5349 AC Oss The Netherlands
Objectives: To quantify the relationship between exposure to the Programmed Cell Death 1 (PD-1) inhibitor pembrolizumab in serum (AUC over 6 weeks at steady state) and the anti-tumor response in patients with melanoma and Non-Small Cell Lung Cancer (NSCLC), measured as the sum of the longest dimension (SLD) of the tumor lesions. An additional goal is to create a modeling framework that can quantify the time course of tumor growth and shrinkage and characterize potential differences between melanoma and NSCLC indications.
Methods: Non-linear mixed effects modeling approach was used, where the structural model was parameterized with both first-order tumor growth and shrinkage rates. As pembrolizumab binds to immune cells rather than tumor cells, its effect was linked to shrinkage rate. In addition, the model assumes that some fraction of the tumor mass to be accessible for immune-mediated antitumor effect with the remaining tumor portion insensitive to treatment. Population parameter values and inter-individual variability were estimated from the available data via NONMEM 7.2, using SAEM for parameter estimation, and importance sampling (IMP) estimation method for likelihood evaluation. Simulations were conducted with final parameter estimates and compared to clinical data. .
Results: A total of 897 melanoma and 496 NSCLC patients were analyzed. Observed tumor size data showed a wide-range of longitudinal response patterns, well-characterized by the model. Model parameters were estimated with good precision. The analysis demonstrates an esseintially flat relationship between exposure and reduction in tumor size in the dose range studied. Comparing model parameters between the two indications suggest that the tumor growth characteristics are indication-specific with the growth term for melanoma predicted to be half of that for NSCLC. The model predicts that the drug effect on antitumor response is similar across indications, consistent with its pharmacological mechanism (binding to systemic T-cells that elicits a downstream tumor cell clearance).
Conclusions: The exposure-response modeling approach applied to pembrolizumab has been an important component in optimizing the current clinical dose of 2 mg/kg Q3W, demonstrating comparable efficacy to 10 mg/kg Q3W. The model presented here shows promise as a tool in providing an integrative look across indication, delineating system-specific properties from drug-specific properties. While the results are premature with two indications analyzed thus far, this approach will also be extended to other solid tumors.