Model based support to biosimilarity assessment planning – A case study of pegfilgrastim
Ari Brekkan(1,2), Luis Lopez-Lazaro(3), Elodie L. Plan(1), Chayan Acharya(1), Gunnar Yngman(1,2), Joakim Nyberg(1), Andrew C. Hooker(1,2), Suresh Kankanwadi(3), Mats O. Karlsson(1,2)
1) Pharmetheus AB, Uppsala, Sweden. 2) Pharmacometrics Research Group, Department of Pharmaceutical Biosciences, Uppsala University, Uppsala, Sweden. 3) Dr. Reddy’s Laboratories, Basel, Switzerland
Introduction/Objectives
Pegfilgrastim (PG) is a recombinant pegylated granulocyte colony stimulating factor (GCSF) used in the treatment of chemotherapy induced neutropenia and febrile neutropenia (FN) [1]. PG induces the maturation, proliferation and survival of neutrophil precursors resulting in an increase in absolute neutrophil count (ANC). Administration of PG is associated with a high treatment cost which can be mitigated by the approval of biosimilar versions of the drug. However, the first approvals of biosimilar PG are very recent and reasons for the difficulties related to development of biosimilar PG were explored using model-based simulation in this work. The aim of this work was twofold: to develop a population PK/PD model for PG and ANC using the data from three PG formulations tested in a clinical trial and to perform sensitivity simulations with the model to elucidate exposure sensitivity of PG and ANC to differences in delivered dose, EC50 and baseline ANC levels.
Methods
Data from a three way cross over clinical study (N=174) of a potential biosimilar and 2 batches of Reference product (Neulasta®) was used for model building. An integrated bidirectional PK/PD model coupling PG concentrations and ANC was developed. The main modelling focus was to describe absorption and elimination mechanisms in the model, both of which were believed to be relatively complex. The PD model was a neutrophil kinetic model based on previous publications with all system parameters apart from ANC baseline fixed to literature values [2]. PG induced neutrophil proliferation, maturation and expansion of central volume (as a margination effect) through Emax effects. Covariate influence was assessed using FREM [3].
A biosimilarity trial was simulated using the model comparing hypothetical reference and test PG products. The system was evaluated by adding dose and potency (by perturbation of EC50) differences between the two administered products. Further, the influence of ANC baseline was evaluated. The power to conclude PK and PD similarity based on areas under the PG concentration and ANC curves from 0 to 312 hours (AUC and AUEC, respectively) was calculated for the simulated scenarios by comparing the geometric mean ratios of AUC and AUEC between the two products [4]. The expected statistical power to conclude PK and PD similarity was calculated as the percentage of simulated studies that demonstrated equivalence according to traditional bioequivalence criteria.
Results
The final PK model was a one-compartment model with sequential zero- and first-order absorption and parallel ANC-dependent and non-specific saturable elimination. Non-specific saturable elimination was the primary elimination pathway identified based on the data at hand (single dose data). ANC mediated elimination accounted for 50% of the elimination rate at the highest PG concentrations. FREM revealed that tested covariates could explain only a small degree (2%) of the variability in either AUC or Cmax.
Simulations of a two-way biosimilarity trial with the model indicated PK sensitivity and PD insensitivity to differences in delivered doses between the reference and test PG products. With a 2% delivered dose difference the difference in AUC was approximately 8% while a 10% dose difference resulted in an AUC difference of >50%. AUECs were less impacted by differences in delivered doses. A potency difference of up to 50% did not impact AUC or AUEC to a large degree.
The power to conclude PK similarity was impacted by differences in delivered doses between the products. A sample size of ~200 individuals was needed to conclude PK similarity with a 2% dose difference between the test and reference products. The power to conclude PD similarity was unaffected by differences in delivered dose amounts between a reference and test product. The power to conclude PK and PD similarity was relatively unaffected with EC50 differences up to 50%, but for larger EC50 differences between the reference and test products the power to conclude PD similarity was ~0%.
Conclusions
A well performing semi-mechanistic population PKPD model was developed to describe PG and ANC disposition. The model was used, by means of simulations, to determine the sensitivity of PK and PD to differences in delivered doses and potency and the results show a very high sensitivity of the PK parameters to changes in the amount of PG delivered to the circulation.
References:
[1] Quartino AL, Karlsson MO, Lindman H and Friberg LE, 2014, Characterization of endogenous G-CSF and the inverse correlation to chemotherapy-induced neutropenia in patients with breast cancer using population modeling. Pharmaceutical Research vol. 31: 3390–3403.
[2] Roskos LK, Lum P, Lockbaum P, Schwab G and Yang BB, 2006, Pharmacokinetic/pharmacodynamic modeling of pegfilgrastim in healthy subjects. J Clin Pharmacol vol. 46: 747–757
[3] Yngman G, Nyberg J, Jonsson EN, Karlsson MO. Practical considerations for using the full random effects modeling (FREM) approach to covariate modeling [Internet]. Available from: https://www.page-meeting.org/default.asp?abstract=7365.
[4] Vong C, Bergstrand M, Nyberg J, Karlsson MO. Rapid sample size calculations for a defined likelihood ratio test-based power in mixed
effects models. 2012. AAPS J. Jun;14(2):176–86.
