Lena Klopp-Schulze (1), Markus Joerger (2), Sebastian G. Wicha (1), Zinnia P. Parra-Guillen (1), Charlotte Kloft (1)
(1) Dept. of Clinical Pharmacy and Biochemistry, Institute of Pharmacy, Freie Universitaet Berlin, Germany, (2) Medical Oncology and Clinical Pharmacology, Dept. of Internal Medicine, Cantonal Hospital St. Gallen, Switzerland
Objectives: Two population pharmacokinetic (PK) models of tamoxifen (TAM) and its principle metabolite endoxifen (ENDX), Model 1, M1, [1] and 2, M2, [2] were built from data at steady-state (st-st); both implemented the impact of CYP2D6 and CYP3A4/5 activity but using different structural models. A previous investigation displayed considerable differences in predicted TAM (-40%) and ENDX (-20%) trough concentrations at st-st (Css,min) by M2 compared to M1 [3]. Good model predictions are crucial for model reuse as prior knowledge. The aim of this work was to explore potential reasons for the unexpected differences and their impact.
Methods: Three hypotheses were tested: (1) differences in bioavailability (F), (2) differences in medication adherence and (3) the assumption of st-st in M2. Simulations were performed with a multiple dosing regimen (20 mg/day p.o.) and considering similar CYP2D6 phenotype distributions in both models.
For the hypotheses (1) and (2) several populations of factor tiers (each n=1000) were simulated using M1 in Berkeley Madonna (v. 8.3.18) to reproduce predictions of M2. Daily drug intake was simulated as a binary outcome with a constant probability of non-adherence.
For hypothesis (3), concentration-time profiles (each n = 100) were simulated with M2 and then parameters re-estimated assuming st-st at different days (5, 10, reference: 20) in NONMEM (v. 7.3). Predicted parameters and resulting Css of TAM and ENDX were compared between the original and re-estimated parameter sets.
Results: Decreases in F or adherence (both 40%) in M1 were only able to reproduce predictions of Css,TAM in M2, but Css,ENDX were underpredicted (-20%). Hypothesis (3) showed increased CLTAM estimates with earlier assumption of st-st. Thus, simulations with the re-estimated parameter sets with early st-st assumption (day 5) displayed decreased Css,TAM (-30%) and even lower Css,ENDX (-50%) compared to the reference.
Conclusions: None of the three hypotheses per se were able to capture both the predictions of TAM and ENDX for M2. The differences in model predictions might be due to a combination of the hypothesised factors, rather than explained by one factor alone, or additional factors. This simulation exercise exemplifies that factors such as st-st assumption, differences in F or adherence have a considerable impact on model predictions of TAM and ENDX emphasising the need to account for them in clinical trials, clinical practice and data analysis.
References:
[1] Dahmane EBA. Tamoxifen Pharmacokinetics and Pharmacogenetics in Endocrine Sensitive Breast Cancer Patients. Thèse de doctorat: Univ. Genève (2013) no. Sc. 4617.
[2] Ter Heine R et al. Population pharmacokinetic modelling to assess the impact of
CYP2D6 and CYP3A metabolic phenotypes on the pharmacokinetics of tamoxifen and endoxifen. Brit J Clin Pharmacol (2014) 78(3): 572–86.
[3] Klopp-Schulze L et al. In silico simulation study : A comparison of two population pharmacokinetic models of tamoxifen and its major metabolite endoxifen. PAGE 24 (2015) Abstr. 3447 [www.page-meeting.org/default.asp?abstract=3447]
Reference: PAGE 25 (2016) Abstr 5917 [www.page-meeting.org/?abstract=5917]
Poster: Drug/Disease modeling - Oncology