Maraviroc Exposure Response Analysis: Phase 3 Antiviral Efficacy in Treatment Experienced HIV+ Patients
L. McFadyen (1), P. Jacqmin (2), J. Wade (2), B. Weatherley (1)
(1) Clinical Pharmacology, Pfizer Global Research & Development, Sandwich, UK; (2) Exprimo NV, Lummen, Belgium
Objectives: Maraviroc, a selective CCR5 antagonist, has been shown to be active in vitro against a wide range of CCR5 tropic clinical HIV isolates and has demonstrated comparable efficacy to other effective antivirals in 10 day monotherapy studies. Phase 2b/3 studies are currently ongoing in both treatment-naïve and treatment-experienced HIV-1 infected subject with CCR5 tropic virus. The aim of the current analysis was to identify the exposure response relationship.
Methods: The effect of maraviroc exposure and various other prognostic factors on 3 viral load efficacy endpoints (binary endpoints of success or failure) at week 24 in treatment-experienced HIV-1-infected patients on optimized background therapy (OBT) has been analysed using generalized additive logistic models (GAM). This approach was used because the exposure response analysis of long term efficacy data in HIV is complicated by: left censoring of the HIV-1 RNA levels; numerous non-linear covariate effects and dropouts. Also, as viral dynamics are best described by a prey-predator model, the viral load (absolute value or change from baseline) at any given time is not necessarily indicative of the steady state inhibition-concentration relationship.
Data (up to 24 weeks of treatment) were available from 1049 subjects in two identical placebo controlled phase 2b/3 studies comparing maraviroc and optimized background therapy (OBT) to OBT alone in treatment experienced patients infected with CCR5-tropic HIV-1. Maraviroc was dosed at 150mg QD or BID when administered with Protease Inhibitors (PIs) (except tripanavir/ritonavir) or delavirdine (CYP3A4/P-gp inhibitors) otherwise 300mg QD or BID. Viral load data were transformed into binary (failure-success) endpoints: viral load higher than 50 copies/mL at 24 weeks, viral load higher than 400 copies/mL at 24 weeks. Anyone discontinuing prior to 24 weeks was regarded as a failure.
Prognostic factors used in the GAM analysis included exposure variables (dose, average concentration, Cmin and effective constant concentration), treatment variables, demographic factors, disease related factors. The concentration variables were obtained from empirical Bayes' PK parameter estimates from a population pharmacokinetic model (sparse PK sampling up to 9 samples/subject over 24 weeks). After graphical and univariate GAM analysis a step GAM procedure was implemented (automated step-wise search in S-PLUS). The uncertainty in the GAM models was quantified using a ‘bootstrapping' technique. After successful evaluation of the predictive performance of the GAM models, simulations of exposure-effect relationships including model uncertainty (but not residual error) were performed.
Results and Conclusions: The PK-PD analysis of maraviroc indicates that, in addition to the well known prognostic factors of baseline CD4+ count and viral load and the overall sensitivity score for OBT, systemic exposure to maraviroc (rather than dose) is an important prognostic factor of virological failure. The application of GAM modelling is a useful and relatively rapid means of identifying important prognostic factors for antiviral efficacy.