Ayatallah Saleh (1,2,3), Ulrika Wählby Hamrén (3), Charlotte Kloft (1), Sebastian Ueckert (3, 4)
(1) Department of Clinical Pharmacy and Biochemistry, Institute of Pharmacy, Freie Universität Berlin, Germany, (2) Graduate Research Training program PharMetrX, Germany, (3) Clinical Pharmacology & Quantitative Pharmacology, R&D, AstraZeneca, Gothenburg, Sweden, (4) Ribocure Pharmaceuticals, Gothenburg, Sweden
Background: Urine is routinely collected in early clinical studies to determine the fraction excreted in urine using non-compartmental analysis (NCA), but is rarely leveraged in population pharmacokinetic (popPK) models. This study investigates the potential of urine data to strengthen the predictive performance of popPK model in renal impairment (RI), focusing on a selective endothelin A receptor antagonist known for significant renal elimination [1].
Objectives: Compare the predictive performance of two popPK models: a plasma-only model (P) and a plasma-plus-urine model (P+U) across varying RI degrees.
Methods: We developed two PopPK models sequentially using NONMEM 7.5, based on data from a single 10 mg oral dose administered to 48 individuals with varied renal function [1]. Model P was based solely on plasma data, while model P+U integrated both plasma and urine data. Model adequacy was judged by parameter plausibility, precision, and goodness-of-fit plots. Covariate significance was tested, focusing on clinically relevant factors. Model selection was based on OFV reductions and diagnostic checks. Final models were evaluated using simulation-based visual predictive checks (VPCs; n=1000). A cross-validation-based approach was implemented to compare the predictive performance of the final P and P+U models [2, 3]. This involved a stepwise exclusion of each RI group from the dataset, re-estimating model parameters, then predicting AUC and Cmax for the excluded group. Later, these predictions were compared against reference values derived from the final P model and NCA. The bias and precision of the predicted parameters were evaluated using mean prediction error (MPE) and root mean square error (RMSE), respectively.
Results:
The popPK model was first developed based on plasma data only (P). A two-compartment (CMT) model with 1st order absorption and linear elimination kinetics described the plasma data adequately. Between-subject variability (BSV) was implemented on all PK parameters except intercompartmental clearance (CL). Covariate analysis showed that estimated glomerular filtration rate (eGFR) significantly influenced CL, reducing BSV in CL by 10%, with an OFV reduction by 35 points when modelled using a power function. Body weight impacted central (V2) and peripheral (V3) volumes, decreasing OFV further by 13 points upon allometric scaling and improved the model fit. Subsequently, urine data was incorporated, starting from the final P model to construct the P+U model. A urine CMT was added to account for renal excretion, dividing the CL into renal (CLR) and nonrenal pathways, estimating the fraction excreted in urine (fe). eGFR was modelled to affect CLR.
VPCs demonstrated that both models accurately captured the central tendency and variability of the data across renal function groups. Cross-validation showed minor differences in the predictive performance between the P and P+U models. RMSE for AUC predictions were 1.00, 0.688, and 0.822 for mild, moderate and severe RI, respectively, when estimated by the P model, compared to 1.01, 0.781 and 1.07 for the corresponding groups predicted by the P+U model. Similarly, RMSE for Cmax ranged from 72.3 to 101 for P and 75 to 99.5 for P+U, demonstrating a marginal precision advantage for the P+U model. The relative prediction error analysis further confirmed these findings, with median errors closely aligned between the two models across all RI groups. The limited additional value of urine data in the P+U model could be attributed to the high proportion of BLQ values, particularly in individuals with severe and moderate RI, potentially biasing the model. The analysis also highlighted significant uncertainty associated with urine data, as evidenced by the higher residual variability in urine concentrations (σ = 72%) compared to plasma concentrations (σ = 35%).
Conclusions: While urine data provide valuable insights into drug elimination, the impact on popPK model predictions in this study was limited, partly due to the inherent challenges in urine sampling. Careful and precise collection and handling of urine is crucial, as improper sampling can introduce significant variability and affect the robustness of the model. Additionally, the substantial number of urine samples BLQ poses risks of bias and information loss. Future simulation studies are recommended to elucidate the value of urine data in popPK model predictions for drugs with varying degrees of urinary excretion.
References:
[1] Tomkinson H, Kemp J, Oliver S, et al (2011) Pharmacokinetics and tolerability of zibotentan (ZD4054) in subjects with hepatic or renal impairment: two open-label comparative studies. BMC Clin Pharmacol 11:3. https://doi.org/10.1186/1472-6904-11-3
[2] Colby E, Bair E (2013) Cross-validation for nonlinear mixed effects models. J Pharmacokinet Pharmacodyn 40:243–252. https://doi.org/10.1007/s10928-013-9313-5
[3] Hooker A, Ten Tije A, Carducci M, et al (2008) Population Pharmacokinetic Model for Docetaxel in Patients with Varying Degrees of Liver Function: Incorporating Cytochrome P450 3A Activity Measurements. Clin Pharmacol Ther 84:111–118. https://doi.org/10.1038/sj.clpt.6100476
Reference: PAGE 32 (2024) Abstr 10896 [www.page-meeting.org/?abstract=10896]
Poster: Drug/Disease Modelling - Other Topics