Correlates of protection in determining vaccine efficacy is a topic of particular interest, given the current COVID-19 climate and the spotlight on vaccine development. Here at Quantics, we have a wealth of experience working with correlates of protection in vaccine development so we thought we would take a look back on our paper Estimating vaccine efficacy using animal efficacy data (written by Quantics staff Ann Yellowlees and Richard Perry) and give our newer readers a summary.
Vaccine efficacy is defined as the percentage reduction in the incidence of a disease among people who have received a vaccine compared to the incidence in unvaccinated people.
Ideally vaccine efficacy is measured directly in a randomised controlled trial with a vaccine challenge. When this is not possible, an indirect method has to be used.
The FDA document Product development under the animal rule provides some guidance for the evaluation of vaccine efficacy in the absence of human clinical efficacy data by using “an appropriate immune marker”, otherwise known as a “protective correlate”. A protective correlate is a measurable marker that is statistically related to a clinical endpoint and is reasonably likely to predict the clinical endpoint.
During vaccine development the relationship between the protective correlate marker and the clinical endpoint in animals is modelled using data from animal studies. The relationship can then be bridged to humans by the assumption that the marker is also predictive of the clinical endpoint in humans.
Deriving the vaccine efficacy
Deriving the vaccine efficacy requires two separate sets of data:
- The animal study data relating the level of protective correlate to disease protection
- The levels of the protective correlate in humans which are achieved by the proposed vaccination regimen, assessed in a healthy volunteer clinical trial.
These are combined in a 3-step process:
Step 1: model the antibody levels associated with protection in animals. Typically a logistic regression model.
Step 2: measure the distribution of antibody levels in a group of subjects vaccinated according to the planned regime.
Step 3: From the animal prediction curve, predict the probability of protection at various human antibody levels.
The vaccine efficacy (percentage reduction in the incidence of a disease) is estimated using Bayes’ Theorem (Bayes T. and Price R., 1763) to combine the probability of protection (given the antibody level is x) with the likelihood of the antibody level being x.
Precision of the estimate of protection
In order to be useful, the estimate of the probability of protection needs to be sufficiently precise, typically measured by the 95% confidence interval. If the confidence interval is too wide, the estimate of the protection may not be regarded as clinically convincing. Calculating the confidence interval is mathematically complex.
Optimising the balance between animal and human data
During early vaccine development the initial experiments to create the animal antibody vs protection curve may not result in the mean animal antibody level being around the values ultimately seen in humans, leading to imprecise estimates of vaccine efficacy.
The estimate of the vaccine efficacy can be optimised either by conducting further animal experiments to improve the primary model fit, and by adding further information on human antibody levels.
Simulation can provide a guide to the increase in precision gained by the various options
The balance can be examined using simulation of adding data in both ways. For more detail on simulation techniques see blog: using simulation in bioassay to reduce the need for laboratory work.
The left panel above illustrates adding more animal data at ED50, ED80 and ED95 to improve the primary model fit, and the right panel of this illustrates the alternative of adding more human data.
In situations where it is not possible to test a vaccine directly in humans, the indirect method using correlates of protection can be used. Early experiments use standard designs as there is usually no information about the primary model or human antibody levels.
The paper that this blog is based on shows that simulation can be used after initial data is obtained to provide some guidance on how to design further experiments. Such further experiments may be required so the resulting estimates of efficacy are adequately precise to support conclusions about the vaccine and the proposed regimen, or are adequately powered to detect differences between vaccine candidates.
Sign up to receive our blogs