Separating out multi-analyte endoxifen measurements: Derivation of a conversion factor to enable assessment of analytically heterogeneous clinical trial data
Fenja Klima* (1,2), Anna M. Mc Laughlin* (1,2), Thomas Helland (3,4,5), Linda B.S. Aulin (1), Wilhelm Huisinga (6), Gerd Mikus (1), Robin Michelet (1), Daniel Hertz (3), Charlotte Kloft (1) for the CYP2D6 Endoxifen Percentage Activity Model in Breast Cancer (CEPAM) consortium
(1) Department of Clinical Pharmacy and Biochemistry, Institute of Pharmacy, Freie Universitaet Berlin, Germany, (2) Graduate Research Training Program PharMetrX, Berlin/Potsdam, Germany, (3) Deptartment of Clinical Pharmacy, University of Michigan College of Pharmacy, Ann Arbor, United States of America, (4) Hormone Laboratory, Deptartment of Medical Biochemistry and Pharmacology, Haukeland University Hospital, Bergen, Norway, (5) Deptartment of Clinical Science, University of Bergen, Norway, (6) Institute of Mathematics, University of Potsdam, Germany; * shared first authorship
Introduction: Tamoxifen (TAM) is a prodrug used in the treatment of oestrogen receptor-positive breast cancer, with Z-endoxifen (Z-EDX) being its most active metabolite. Despite its well-established use, different concentration thresholds have been reported for Z‑EDX efficacy [1-4]. An underlying factor for this observed discrepancy could be the heterogeneous chromatographic separation of Z-EDX from less active endoxifen (EDX) isomers across analytical methods [5]. In particular, many old studies have only quantified total EDX (TOT‑EDX) but not the individual isomers. To integrate all available pharmacokinetic (PK) TAM and EDX data into a quantitative framework, methods to derive Z‑EDX from measured TOT‑EDX are needed. In this work, a large clinical database incorporating studies which either quantified Z-EDX or TOT-EDX was utilised to extend and leverage a parent-metabolite nonlinear mixed-effects (NLME) PK model to include the derivation of Z-EDX from TOT-EDX and evaluate its performance.
Methods: The CEPAM database comprises 28 clinical studies with 7441 TAM patients. Male patients, patients receiving TAM for <1 month or with missing or implausible, i.e. considerably higher or lower TAM or EDX concentrations compared to the other studies were excluded from the analysis. A published parent-metabolite NLME PK model of TAM and Z-EDX [6] was extended with additional covariate relationships based on stepwise covariate modelling using the CEPAM analysis dataset [7].
To allow for the inclusion of studies with measured TOT-EDX, a conversion factor (CF) from Z-EDX to TOT-EDX with interindividual variability (IIV) was incorporated. Following a literature review [8,9], the CF was implemented based on the assumption that TOT-EDX is the sum of Z‑EDX and its inactive main isomer Z-4’-endoxifen (Z4’-EDX). The shape of the relationship between Z-EDX and TOT-EDX was investigated in function of different covariates, especially clearance (CL) from TAM to Z‑EDX. To derive the equation for individual CFs, TOT-EDX was calculated for patients with both Z-EDX and Z4’-EDX measurements (evaluation group). For these patients, both the calculated TOT‑EDX and measured Z-EDX values were included in the analysis dataset.
For internal evaluation of the CF, the model was blinded for measured Z-EDX in the evaluation group, which was then predicted either based on model-derived Empirical Bayes estimates (EBEs) or by calculation using the individual CFs and the calculated TOT-EDX (sum of observed isomers) only. CF performance was assessed via GOF plots of predicted, calculated, and measured Z-EDX, median error (ME) and median absolute error (MAE) of predicted and calculated Z-EDX, and graphical analysis of trends and distribution of individual CFs and IIV on the CF.
Results: The final analysis dataset comprised 6263 patients with measured Z-EDX only in 64.3%, measured Z-EDX and calculated TOT-EDX in 22.1%, and measured TOT-EDX in 13.6% of all patients. The CF from Z-EDX to TOT-EDX was implemented as a power function of individual CL from TAM to Z-EDX. The coefficient and exponent for the CF were estimated as 1.44 and -0.333, corresponding to a decreasing CF with increasing CL. IIV on the CF was included as an exponential function and estimated as 12.3% CV. TAM, Z‑EDX, and TOT-EDX were predicted well.
The mean CF obtained from EBEs in the internal evaluation step was 1.88 [interquartile range: 1.73-1.92] in the evaluation group. The ME and MAE of predicted Z-EDX were 8.96% and 19.5% in the evaluation group, suggesting a low bias and adequate precision of Z-EDX predictions even if measured Z‑EDX was not available. The ME and MAE of calculated Z‑EDX were very similar with 8.25% and 19.0%, respectively, supporting the hypothesis of TOT-EDX being the sum of Z‑EDX and Z4’‑EDX. Furthermore, the graphical analysis of individual CFs showed no between-study differences and IIV on the CF showed no differences between the evaluation group and patients with measured TOT-EDX only.
Conclusions: A NLME PK model of TAM and Z-EDX was extended and leveraged to successfully predict and calculate Z-EDX from TOT-EDX. The applied framework allows a joint analysis of studies with only Z-EDX or only TOT-EDX measurements in a large heterogeneous clinical database. Thus, it enables assessment of discrepancies in efficacy thresholds and application in a simulation setting to investigate exposure-response relationships in large prospective clinical trials without measured EDX.
References:
[1] L. Madlensky et al. Clin. Pharmacol. Ther. 89: 718–725 (2011).
[2] R.R. Love et al. Springerplus 2: 1–5 (2013).
[3] P. Saladores et al. Pharmacogenomics J. 15: 84–94 (2015).
[4] T. Helland et al. Breast Cancer Res. 19: (2017).
[5] N.G.L. Jager et al. Breast Cancer Res. Treat. 133: 793-798 (2012).
[6] A. Mueller-Schoell et al. Clin. Pharmacol. Ther. 108: 661–670 (2020).
[7] A.M. Mc Laughlin et al. Clin Pharmacol Ther 111: S77 (2022).
[8] Y. Chen et al. J. Pers. Med. 11: 1-12 (2021).
[9] T.E. Mürdter et al. Clin. Pharmacol. Ther. 89: 708-717 (2011).