Lisa F. Amann (1), Astrid Broeker (1), Rawan Alraish (2), Magnus Kaffarnik (2), Sebastian G. Wicha (1)
(1) Dept. of Clinical Pharmacy, Institute of Pharmacy, University of Hamburg, Hamburg, Germany (2) Charité—Universitätsmedizin Berlin, Department of Surgery, Campus Charité Mitte, Berlin, Germany
Introduction/Objectives:
The hepatic panel offers many potential covariates to inform about liver function, but no single biomarker can estimate the hepatic clearance of the drug, as for renal function [1]. Some covariates are grouped in order to pursue clinical decisions. The aim of this study was to investigate and compare the predictive performance of a SCM- vs. a FREM-based modelling approach in liver failure exemplified with tigecycline.
Methods:
The clinical dataset included 40 patients receiving tigecycline with various degrees of hepatic impairment. Dense plasma sampling across up to four observed dosing occasions and data on the following covariates were included: Age, weight, sex, aspartate transaminase (AST), alanine-amino transferase, γ-glutamyl transferase (GGT), bilirubin (BILI), thrombocytes (THR), LiMax-test, prothrombin time test, De Ritis ratio (DR) , and the model for end-stage liver disease score (MELD). The data was analysed using NONMEM 7.4.3 and PSN version 4.9.0.For the SCM method, the FOCEI based likelihood ratio test with forward inclusion (α<=0.05) and backward elimination was employed (α<=0.01)[2]. FREM treats covariates as observations within the ?-matrix and the effects of all covariates were estimated simultaneously through co-variances between interindividual variability (IIV) and covariates to calculate their effect sizes [3,4]. IMPMAP estimation estimated all FREM parameters. The SCM and FREM results were compared in stratified prediction-corrected visual predictive checks (PC-VPC’s). Continuous covariates were stratified and the cut-off was set to the median of each covariate to explore the impact of them on the predictive performance of the model with regard to tigecycline pharmacokinetics in the lower and upper 50th percentile of the covariates in the dataset.
Results:
The SCM selected AST as the covariate with the highest dOFV of 29.6, followed by DR, GGT, and MELD. The MELD was rejected within the backward elimination. The final model included AST and GGT as a linear and DR as a power covariate relationship. The selected covariates were only mildly correlated: GGT-DR 7.1%, AST-DR 0.03% and AST-GGT by 15.1%. The final overall dOFV was 51.9 and the residual unexplained variability diminished by 7.3%, IIV only by 3.1%. Transferring this into a clinical perspective, the covariate with the highest significance, AST, affected the true value of clearance by only ± 13.5% (comparing values at 5th and 95th percentile: 15 u/L, 189 u/L for AST). The FREM results are presented as effect size at the 5th and 95th percentile of the respective covariate range (THR: 17-508 per/nL; MELD: 9-37; BILI: 0.22-16.75 mg/dl, LiMax: 22-510 µg/kg/h). The FREM model found the highest impact on clearance (fractional change [95% CI]) for THR (-24.8% [-32.4, -16.9] – 54.8% [32.3, 80.4]), followed by MELD (-42% [-52.7, -29.6] – 36% [21.4, 51]), BILI (-52.3% [-65.6, -37.6] – 25.7% [15.2, 37.8]) and LiMax (-19.6% [-32.8, -5.3] – 52% [10.2, 103]). The FREM quantified -7.62% [-18.6, -1.26] – 4.83% [0.72, 12.4] for AST, which is comparable to the effect calculated from the SCM model. For GGT, a less relevant effect size was found as well ranging from – 18.7% [-28.3, -1.11] to 9.68% [0.48, 15.4]. The results for DR (-25.1% [-44.4, -6.32] to 17.8% [3.53, 36.6]) indicated a moderate effect on clearance.Evaluating the predictive performance, the PC-VPC’s indicated a better alignment of predicted vs. observed PK profiles for all investigated covariate stratifications from the FREM model.
Conclusion:
Liver function covariates are challenging to implement in pharmacometric models since no single, highly predictive covariate or biomarker can be used, and their selection is complicated by partly (high) correlations. The SCM only relies on statistical criteria and does not consider effect sizes. Moreover, it suffers from selection bias and unknown ‘true’ p-values due to multiple testing. Due to the automated SCM process, these problems are shaded and partly subjective scientific or clinical interest usually guides the selection of covariates to the final model.FREM, which handles covariates as observed data points, avoids selection bias due to simultaneously estimating all covariates and can thereby handle even correlated data. FREM directly allows for assessment of the covariate-effect sizes and the results indicated that FREM might be a promising technique to implement liver function covariates.
References:
[1] Delcò F, Tchambaz L, Schlienger R, Drewe J, Krähenbühl S. Dose adjustment in patients with liver disease. Drug Saf 2005; 28:529–45.
[2] PsN. SCM user guide. PsN 2015; 4.7.0:1–23.
[3] Nyberg J, Jonsson EN, Karlsson MO, Häggström J. Properties of the full random effect modelling approach with missing covariates. BioRxiv 2019:656470.
[4] FREM userguide 2018:1–19
Reference: PAGE () Abstr 9328 [www.page-meeting.org/?abstract=9328]
Poster: Methodology - Covariate/Variability Models