Gabriel Stillemans (1,2), Leila Belkhir (2,3), Vincent Haufroid, (2,3), Laure Elens (1,2)
(1) Louvain Drug Research Institute, Université catholique de Louvain, Belgium, (2) Institut de Recherche Clinique et Expérimentale, Université catholique de Louvain, Belgium, (3) Cliniques universitaires Saint-Luc, Belgium
Objectives: Darunavir (DRV) is a widely used protease inhibitor in de novo or pretreated HIV patients, combined with a background regimen (usually two nucleoside reverse transcriptase inhibitors). It is coadministered with a pharmacoenhancer, either ritonavir (RTV) or cobicistat (COB) to increase DRV plasma exposure. DRV is characterized by a large pharmacokinetic (PK) variability, but the reasons for that variability have yet to be fully elucidated. Additionally, fewer PK studies are available for the more recently commercialized COB-boosted DRV. There is also an interest in evaluating the feasibility and safety of reduced doses or short cycle therapies such as weekends-off regimens in order to reduce costs and toxicity and to improve patient compliance. Our objective was to explore the PK of DRV and COB in a representative cohort of HIV patients using nonlinear mixed effects modeling and to determine the effect of covariates, including single nucleotide polymorphisms (SNPs), drug-drug interactions and clinical chemistry parameters.
Methods: The study was approved by the local ethics committee and was registered at ClinicalTrials.gov (NCT03101644). HIV-positive patients were prospectively recruited. For each participant, sparse blood samples for drug quantification were drawn at random post-intake times (one sample per visit and an average of 2.5 visits per participant), plus an additional sample for genotyping (CYP3A4*22, CYP3A5*3, and ABCB1 1199G>A and 3435C>T were determined). For all participants, clinical chemistry data, DDIs and demographic parameters were recorded. Simultaneous quantification of DRV, RTV and COB in plasma was performed using a validated UPLC-UV method [1]. NONMEM was used for population PK modeling. Runs were handled through PsN and Xpose was used for additional plotting. DRV and COB PK were first modeled separately, covariates were then added in a stepwise manner, and finally, a joint interaction model was evaluated. Goodness of fit was assessed using plots of predicted versus observed concentrations, plots of weighted residuals, as well as bias, imprecision and shrinkage of population parameters. For internal validation, visual (VPC) and numeric (NPC) predictive checks were performed based on 1000 simulations from the final model. Nonparametric bootstrapping was used to evaluate model stability and generate confidence intervals.
Results: 127 patients contributed to plasma PK sampling, for a total of 249 datapoints. RTV data was too sparse to be used (n = 18 subjects) and was discarded. DRV and COB PK were both adequately described by a one-compartment model with first-order absorption. DRV clearance (CL), volume of distribution (V) and absorption rate constant (ka) were 9.6 l.h-1, 140 l and 0.38 h-1, respectively, while COB CL, V and ka were 8.7 l.h-1, 97 l and 0.86 h-1. COB concentrations were found to inhibit DRV CL in a linear fashion (r=0.47, p-value<0.05) ; however, inclusion of this relationship in the joint model did not improve model fit. Higher levels of alpha-1 acid glycoprotein (AAG) were found to correlate with lower DRV CL, in accordance with previous studies [2,3] and was added as a covariate. Despite some borderline significant associations in univariate analysis, none of the other clinical or genetic covariates were retained in the final model as they did not significantly improve the fit. No apparent relationship could be described between PK parameters and HIV viral load, the main biomarker for evaluating treatment efficacy, in line with previous results [4]. VPC and NPC results were adequate. Still, parameters suffered from high shrinkage and high relative standard errors, and bootstrap from the final model generated wide confidence intervals for parameter etas.
Conclusion: Population PK models were developed for DRV and COB. However, with such sparse data, accurate estimation of individual PK parameters proved difficult, limiting the applicability of this model (including its predictive power and ability to detect covariate relationships). The next step will be to enrich our model with dense PK data collected in a subset of participants, expand our panel of SNPs and perform extensive and rigorous internal and external validation of the model. We will also simulate the effect of alternate dose regimens on both PK and treatment outcomes.
References:
[1] L. Elens, S. Veriter, V. Di Fazio, R. Vanbinst, D. Boesmans, P. Wallemacq et al., Quantification of 8 HIV-Protease Inhibitors and 2 Nonnucleoside Reverse Transcriptase Inhibitors by Ultra-Performance Liquid Chromatography with Diode Array Detection, Clinical Chemistry 55 (2008), pp. 170–174.
[2] P. Vis, V. Sekar, E. van Schaick and R. Hoetelmans, Development and application of a population PK model of TMC114 in healthy volunteers and HIV-1-infected patients after administration of TMC114 in combination with low-dose ritonavir, PAGE 2006.
[3] J. Moltó, G. Xinarianos, C. Miranda, S. Pushpakom, S. Cedeño, B. Clotet et al., Simultaneous Pharmacogenetics-Based Population Pharmacokinetic Analysis of Darunavir and Ritonavir in HIV-Infected Patients, Clinical Pharmacokinetics 52 (2013), pp. 543–553.
[4] V. Sekar, C.V. Abeele, B.V. Baelen, P. Vis, L. Lavreys, M.D. Pauw et al., Pharmacokinetic-pharmacodynamic analyses of once-daily darunavir in the ARTEMIS study, International Workshop on Clinical Pharmacology of HIV Therapy 2008.
Reference: PAGE 28 (2019) Abstr 8893 [www.page-meeting.org/?abstract=8893]
Poster: Drug/Disease Modelling - Infection