Shepherd B, Gilbert P, Gurwith M, Francis D; Conference on Retroviruses and Opportunistic Infections (11th : 2004 : San Francisco, Calif.).
Program Abstr Conf Retrovir Oppor Infect 11th 2004 San Franc Calif. 2004 Feb 8-11; 11: abstract no. 284.
Univ. of Washington, Seattle, USA
BACKGROUND: An imperfect vaccine could be useful if it prevents infection for some recipients and/or reduces morbidity and mortality among individuals who would be infected whether or not they received the vaccine. In a placebo-controlled phase 2 HIV vaccine trial, it is of interest to investigate the vaccine's impact on pre-treatment set point viral load in infected participants. However, directly comparing viral loads between infected placebo and vaccine recipients can yield misleading results, because the analysis conditions on a post-randomization variable (infection), allowing for the possibility of confounding despite the fact that the trial is randomized.METHODS: Under the assumption that vaccine does not increase the risk of HIV infection, we demonstrate a procedure for comparing viral loads in a vaccine trial. We employ a causal inference method known as principle stratification and make a comparison between placebo and vaccine recipients who would have been infected whether or not they were randomized to receive the vaccine. It is not known who would have been infected regardless of treatment assignment, so we model this potential infection status using covariates and perform a sensitivity analysis. The method is applied to a completed multicenter, double-blind, randomized efficacy trial of AIDSVAX B/B. Set-point viral load was defined as the average of all pre-treatment values measured within 2 months of infection diagnosis.RESULTS: Based on 5403 randomized subjects and 347 evaluable infected subjects in the AIDSVAX B/B trial, the estimated mean difference (vaccine minus placebo) log10 viral loads of non-white, 31- to 40-year-old, high-school-or-less-educated participants with 2 high risk factors was -0.211 (95% CI: -0.72, 0.29). From the sensitivity analysis we learn that for a one-unit increase in log10 viral load for infected placebo recipients, controlling for race, age, education, and risk behavior, the odds of being infected had they been randomized to vaccine would have to increase more than 150-fold for the mean viral loads to be significantly different with p < 0.05.CONCLUSIONS: Based on the sensitivity analysis, we conclude that AIDSVAX B/B does not cause a substantial change in pre-treatment set-point viral load. When comparing endpoints (viral load, time to onset of AIDS, etc.) in subjects who become infected in a vaccine trial, it is important to perform sensitivity analyses to account for potential confounding.
Publication Types:
Keywords:
- AIDS Vaccines
- AIDSVAX
- Acquired Immunodeficiency Syndrome
- Clinical Trials as Topic
- Double-Blind Method
- HIV Infections
- Physical Examination
- Placebos
- Risk Factors
- Risk-Taking
- Treatment Outcome
- Viral Load
- methods
- organization & administration
Other ID:
UI: 102271243
From Meeting Abstracts