View All Abstracts for Session 46
253 Bridging the Gap between Phase III Vaccine Trials and Vaccine Availability: A Role for Mathematical Models and New Statistical Methods
K. Desai*1, G. P. Garnett1, M.-C. Boily2, and B. R. Masse3
1Imperial Coll., London, UK; 2CDC, Atlanta, GA, USA; and 3SCHARP, Seattle, WA, USA
Background: Design aspects of phase III HIV-1 vaccine trials, such as method of analyses and sample size requirements, are chosen to detect the minimum efficacy of interest with desired power given expected number of infection endpoints. If the vaccine is efficacious, unbiased analysis should provide sufficient evidence of biological potency for licensing. However, this may not be informative enough for public health authorities who may require more precise efficacy estimates and other vaccinal information (e.g., time lag, model of action, duration of protection, cross-reactivity, population effectiveness, and costs) to decide how best to utilize the vaccine if licensed.
Methods: A mathematical model simulating HIV-1 vaccination in a sub-Saharan setting and a new method of analysis based on hazard regression (HARE) are developed to provide more precise information on vaccine properties and to help quantify population effectiveness for a range of vaccine types when used in a chosen epidemiological setting (target group, coverage rate, and timing of intervention).
Results: Time lags and other biasing vaccine properties cause theoretical power to fall by 5 to 18% (for 90% prespecified power) to detect moderate efficacy (40(50%) when employing traditional analyses in phase III trials. Increased sample sizes used with HARE improve empirical power, produce more valid and precise estimates of vaccine efficacy, and often help identify time lags and model of action. If a moderately efficacious vaccine is used in 80% of commercial sex workers in a sub-Saharan setting, HIV-1 prevalence 10 years after vaccination drops by 0 to 50% depending on model of vaccine action and duration of protection, even for vaccines equivalent in terms of efficacy (50%). Concurrently, prevalence in high risk men drops by 0 to 40%.
Conclusions: The decision to distribute a newly licensed vaccine in a given health setting should be an informed one. Mathematical modelling aids decision making by ensuring that efficacy trial results are as precise as desired and valid as possible and by determining effectiveness of different immunization strategies. HARE and other new statistical methods (sieve, frailty models) can possibly highlight model of action, time lags, and cross-reactivity.
Contact Author about this Abstract
