Thi Huyen Tram Nguyen (1), Jeremie Guedj (1), Jing Yu (2), Micha Levi (3) and France Mentré (1,4)
(1)University of Paris Diderot, Sorbonne Paris Cité, UMR 738, F-75018, Paris, France; INSERM, UMR 738, F-75018 Paris, France (2) Novartis Institutes for BioMedical Research, Inc, Cambridge MA 02139, USA (3) Novartis Pharmaceutical Corp., East Hanover NJ 07936, USA (4) AP-HP, Bichat Hospital, Biostatistics Service, F-75018 Paris, France
Objectives: Viral kinetics analysis based on nonlinear mixed effect models is a useful tool for individualized hepatitis C virus (HCV) treatment [1-3]. For that purpose, it is necessary to obtain precise individual parameters estimation. We evaluated the influence of a priori population parameters, sampling designs and methods handling data below detection limit (BDL) on Bayesian individual parameter estimation and prediction of response to therapy.
Methods: A viral kinetics model was used to simulate viral load profile under PegIFN/Ribavirin in 1000 HCV Genotype (G) 2/3 patients [4-6]. The estimation of four parameters was evaluated (infection rate β, death rate of infected cells δ, virion clearance c and treatment efficacy ε). Additionally, the effect of four sampling schedules was investigated (D24w with 12 measurements in 24-week (W) treatment; D4w with 6 data points within 4 weeks; D4w_sparse having 4 data points within 4 weeks but no data during W1; D4w_challenge having only 2 measurements at day 0 and W4. We used a detection limit of 45 IU/mL. Virus eradication was assumed if the infected cells reached a cure boundary during treatment [2]. Three sets of a priori information were evaluated: true model with parameters used for simulation; false model Mδε with δ and ε modified to values obtained in G1 patients; false model Mβ with a modified β. BDL data were either omitted or handled in the likelihood function. Population parameters were fixed at different a priori information to estimate individual parameters using MONOLIX 4.1.2. We used mean relative error, root mean square of relative error and shrinkage to evaluate estimation quality. We also evaluated the ability of treatment outcome prediction by comparing the response predicted using individual estimates and the simulated treatment response.
Results: Precise estimation of individual parameters and treatment outcome was obtained using only 6 data points in the first month of PegIFN/Ribavirin therapy. This result remained valid even though wrong a priori population parameters were set as long as the parameters were identifiable (δ, ε) and BDL data were properly handled. False a priori information on estimable parameters (δ, ε) could lead to severe estimation/prediction errors if BDL data were omitted and not properly accounted in the likelihood function.
Conclusions: Bayesian estimation of individual viral kinetic parameters could provide precise early prediction of treatment outcome.
References:
[1] Arends, J.E., et al., Plasma HCV-RNA decline in the first 48 h identifies hepatitis C virus mono-infected but not HCV/HIV-coinfected patients with an undetectable HCV viral load at week 4 of peginterferon-alfa-2a/ribavirin therapy. J. Viral Hepat., 2009. 16(12): p. 867-75.
[2] Snoeck, E., et al., A comprehensive hepatitis C viral kinetic model explaining cure. Clin. Pharmacol. Ther., 2010. 87(6): p. 706-13.
[3] Guedj, J., et al., A perspective on modelling hepatitis C virus infection. J. Viral Hepat., 2010. 17(12): p. 825-33.
[4] Dahari, H., R.M. Ribeiro, and A.S. Perelson, Triphasic decline of hepatitis C virus RNA during antiviral therapy. Hepatology, 2007. 46(1): p. 16-21.
[5] Bochud, P.Y., et al., IL28B polymorphisms predict reduction of HCV RNA from the first day of therapy in chronic hepatitis C. J. Hepatol., 2011. 55(5): p. 980-8.
[6] Dahari, H., et al., Modelling hepatitis C virus kinetics: the relationship between the infected cell loss rate and the final slope of viral decay. Antivir. Ther., 2009. 14(3): p. 459-64.
Reference: PAGE 22 () Abstr 2772 [www.page-meeting.org/?abstract=2772]
Poster: Other Modelling Applications