Ana-Marija Grisic (1,2,3)*, Akash Khandelwal (3), Mauro Bertolino (3), Wilhelm Huisinga (4), Charlotte Kloft (1)**, Pascal Girard (5)**
(1) Department of Clinical Pharmacy & Biochemistry, Institute of Pharmacy, Freie Universitaet Berlin, Germany; (2) Graduate Research Training Program, PharMetrX, Berlin/Potsdam, Germany; (3) Merck KGaA, Darmstadt, Germany; (4) Institute of Mathematics, University of Potsdam, Germany; (5) Merck Institute of Pharmacometrics, Merck Serono S.A., Lausanne, Switzerland; * Author was affiliated with each of these participating institutions during the time of the analysis; ** Shared senior authorship
Objectives:
Monoclonal antibodies (mAbs) undergo nonspecific linear and target-mediated nonlinear disposition, as well as recently described time-dependent elimination [1-4]. However, the data analyzed can heavily impact the resulting final model, leading to disagreement in identified pharmacokinetic (PK) models for the same mAb, as in case of cetuximab, an anti-epidermal growth factor receptor mAb widely used in oncology [5-7].
This study aimed to (1) characterize cetuximab population PK (PPK) and compare various (semimechanistic) clearance (CL) models and (2) investigate the impact of CL model misspecification on derived exposure metrics under different study designs.
Methods:
In total, 3,821 PK samples from 2 multicenter clinical trials in patients (N=226) with metastatic colorectal cancer were used to develop the PPK model of cetuximab using the nonlinear mixed-effects modeling approach. In trial A (phase I, N=62), patients were initially treated for 6 weeks with cetuximab (doses ranged from 250 mg/m2 q1w to 700 mg/m2 q2w) and afterward received FOLFIRI co-therapy. In trial B (phase I/II, N=164), patients were initially treated for 3 weeks with cetuximab (400 mg/m2 loading dose followed by 250 mg/m2 q1w) and irinotecan. Afterward, a subset of patients underwent cetuximab dose escalation (to the maximum dose of 500 mg/m2 q1w). The sampling comprised longitudinal Cmin samples plus a Cmax sample at the end of the first infusion and dense sampling over 1 dosing interval. Six CL models were investigated: (1) linear (LCL) [7], (2) linear with exponential change over time (TVARCL) [8], (3) Michaelis-Menten (MMCL) [5], (4) linear and Michaelis-Menten (LCL+MMCL), (5) linear with exponential change over time and Michaelis-Menten (TVARCL+MMCL), and (6) linear and 0th-order (LCL+0.EL) [6].
To address the impact of CL model misspecification on accuracy (root mean square error) and bias [9] in derived exposure metrics (area under the curve [AUC] and Cmin after second dose and at assumed steady state), the stochastic simulation and estimation approach (SSE procedure in PsN) was employed to compare the reference and 5 alternative models under 6 study designs that differed in dose range (multiple dose levels vs single dose level) and sampling density (rich, semisparse, and sparse).
Results:
The PPK model that best described the data was a 2-compartment model with parallel Michaelis-Menten and linear elimination that changed exponentially over time. The baseline linear CL of 17.4 mL/h was estimated to decrease over time with a mean maximal decrease of ≈23% (38% CV), and the time to half-maximal decrease reached ≈5 months after the first dose. To address potential mechanisms [2,4], the time-varying CL in responders and nonresponders (as per Response Evaluation Criteria in Solid Tumors criteria) was compared: the CL decrease was of higher magnitude in responders than in nonresponders. These results underline the bidirectional PK-PD relationship anticipated for mAbs and its influence on unidirectional assumption of exposure-response causality. Thus, in case of time-varying CL the concept of purely drug exposure–based therapeutic drug monitoring needs to be expanded.
The second part of our analysis addressed the impact of CL model misspecification on derived exposure metrics under different study designs. Overall, for all 4 investigated exposure metrics and across all study designs, the TVARCL model resulted in the lowest inaccuracy and MMCL model resulted in highest inaccuracy and bias relative to the reference (TVARCL+MMCL) model; Cmin is least impacted by model-misspecification both in terms of accuracy and bias.
Conclusions:
This study is the first to report a combination of nonlinear and time-varying linear CL for a mAb elimination and to propose a consolidated model for cetuximab. The change of CL over time correlated with the patients’ response status. The effect of CL model misspecification on derived exposure metrics was influenced by the underlying study design. Steady-state Cmin was the most robust PK metric. The alternative model that was most likely to be identified was TVARCL. This analysis contributes to understanding of exposure-response dynamics and informing design of future clinical trials of oncology mAbs.
References:
[1] Ryman JT, Meibohm B. CPT Pharmacometrics Syst Pharmacol 2017;6:576-88.
[2] Liu C, et al. Clin Pharmacol Ther 2017;101:657-66.
[3] Zhai J, et al. Br J Clin Pharmacol 2017;83:1446-56.
[4] Wilkins J, et al. ACoP 2017:Abstract W-079.
[5] Dirks NL, et al. J Clin Pharmacol 2008;48:267-78.
[6] Azzopardi N, et al. Clin Cancer Res 2011;17:6329-37.
[7] Girard P, et al. PAGE 2013:Abstract 2902.
[8] Hongshan L, et al. J Pharmacokinet Pharmacodyn 2017;44:403-14.
[9] Walther BA, Moore JL. Ecography. 2005;28:815-29.
Reference: PAGE 28 (2019) Abstr 8994 [www.page-meeting.org/?abstract=8994]
Poster: Drug/Disease Modelling - Oncology