Alexandre Marie (1,2), Prague Mélanie (1,2), Lévy Yves (2), Thiébaut Rodolphe (1,2)
(1) Inria SISTM Team, INSERM U1219, University of Bordeaux, ISPED, France, (2) Vaccine Research Institute (VRI), Créteil, France
Introduction: HIV therapeutic vaccine development is an important component in the search for long-term HIV viral control strategies in the absence of antiretroviral treatment (ART). Analytic treatment interruption (ATI) are used to evaluate their efficacy through the analysis of virological summary endpoints, such as the area under the HIV RNA load curve normalized by follow-up time (nAUC).
Objectives: We aim at developing a parametric statistical test to evaluate the difference of nAUC between two vaccine arms in clinical therapeutic trial, while considering unavoidable missing at random (MAR) monotonic/dropout data induced by early ART resumption during ATI.
Methods: To reduce potential bias induced by MAR data, we modeled individual HIV RNA load data by mixed effects model (MEM) taking into account left-censoring induced by detection limit of the HIV RNA load. Marginal dynamics have been fitted with smooth B-spline curves at group level while polynomial time-dependent random effects were used to model inter-individual variability. Afterwards, the difference of nAUC (ΔnAUC) between the two vaccine arms was estimated through trapezoid method applied only on estimated marginal dynamics. To confirm the right statistical properties of this method, this latter was tested on HIV RNA load data from two-armed randomized trials simulated by the MEM used to fit data in the method. Thereafter, the method was applied on HIV RNA load data simulated by an ordinary differential equation (ODE) -based model. We used the four compartments ODE model featuring quiescent, uninfected and productively infected CD4+ T cells, and viruses concentration studied in [1] to simulate data. Missing data were generated by the consideration of right-censoring on HIV RNA load, leading to the dropout of the individual from the trial. Finally, the statistical properties, such as Type-I error and power, and the ability of this method to deal with missing data were evaluated when ΔnAUC, the number of individual in the trial, as well as the amount of missingness were varied.
Results: We evaluated the method on simulated data where the value of ΔnAUC was varied from 0 to 0.5 log10 copies/ml and the level of censoring being 4, 4.7 or 5 log10 cp/ml. Moreover, we applied the method on real data from three HIV therapeutic vaccine trials: (1) ANRS 149 LIGHT, a phase II trial including 98 patients randomized (2:1) to receive either GTU-MultiHIVB/LIPO-5 prime-boost vaccination or placebo before 12 weeks of ATI, (2) ANRS DALIA [2] including 19 patients treated by dendritic-cell based therapy before 24 weeks of ATI and (4) ANRS 093 VAC-IL2 [3] trial including 71 patients either receiving the vaccination combining ALVAC-HIV (vCP1433) and Lipo-6T vaccines followed by interleukin-2 injections or their ART alone before 12 weeks of ATI. In accordance with these data, a significant difference of nAUC between the two vaccine arms was estimated only on the trial ANRS 093 Vac-IL2 where ΔnAUC was evaluated at 0.51 log10 cp/ml with p-value of 0.02.
Conclusions: We provided an important statistical tool for analysis of ATI studies in HIV. We showed good statistical properties for the proposed test. Moreover, we were able to demonstrate a significant difference in nAUC between treatment groups in some of the available studies.
References:
[1] Prague et al. Biometrics, 68(3): 902-911, 2012
[2] Lévy et al. European journal of immunology, 44(9): 2802-2810, 2014
[3] Lévy et al. Aids, 19(3): 279-286, 2005
Reference: PAGE () Abstr 9424 [www.page-meeting.org/?abstract=9424]
Poster: Oral: Methodology - New Tools