On the impact of masking and blocking hypotheses for measuring the efficacy of new tuberculosis vaccines

Abstract

Over the past 60 years, the Mycobacterium bovis bacille Calmette-Guérin (BCG) has been used worldwide to prevent tuberculosis (TB). However, BCG has shown a very variable efficacy in different trials, offering a wide range of protection in adults against pulmonary TB. One of the most accepted hypotheses to explain these inconsistencies points to the existence of a pre-existing immune response to antigens that are common to environmental sources of mycobacterial antigens and Mycobacterium tuberculosis. Specifically, two different mechanisms have been hypothesized to explain this phenomenon: the masking and the blocking effects. According to masking hypothesis, previous sensitization confers some level of protection against TB that masks vaccine's effects. In turn, the blocking hypothesis postulates that previous immune response prevents vaccine taking of a new TB vaccine. In this work we introduce a series of models to discriminate between masking and blocking mechanisms and address their relative likelihood. We apply our methodology to the data reported by BCG-REVAC clinical trials, which were specifically designed for studying BCG efficacy variability. Our results yield estimates that are consistent with high levels of blocking (41% in Manaus -95% CI [14-68]- and 96% in Salvador -95% CI [52-100]-). Moreover, we also show that masking does not play any relevant role in modifying vaccine's efficacy either alone or in addition to blocking. The quantification of these effects around a plausible model constitutes a relevant step towards impact evaluation of novel anti-tuberculosis vaccines, which are susceptible of being affected by similar effects, especially if applied on individuals previously exposed to mycobacterial antigens.

Keywords: BCG; Blocking; Clinical trials; Environmental sensitization; Masking; TB vaccines; Tuberculosis.

Figure 1. Best fit estimates for each trial by models 1, 2 and 3 (yellow, blue and green dots, respectively) for the trials conducted in the BCG-REVAC study.

The colormap represents the probability of obtaining a less extreme value of the efficacy, according to the distributions considered. The probability of zero marks the central estimate (red, continuous line) while the dashed red lines mark the 95% CI reported by Barreto et al. (2014).

Figure 2. Scheme of the different contributions to the disease risk for each cohort.

Figure 3. Scheme for the temporal evolution of the level of protection for the cohorts of the three types of trials considered in the work, according to the different vaccination strategies and ES mechanisms.

Trial I: the control group is cohort one that corresponds to non-vaccinated individuals. In this cohort, a level of protection above zero can only be due to masking, which is an increasing function with age. In turn, the intervention group corresponds to cohort 2 that is the newborn vaccination group: individuals are vaccinated right after birth, which provides a protection that overcomes any possible masking effect, cannot be blocked by ES and wanes with time. Trial II: The vaccinated cohort is cohort 3, firstly immunized at school age. In this cohort individuals might be protected by masking before the vaccine is applied. Then, at the moment of vaccination, if not blocked, the vaccine will overcome masking protection up to the initial value e(0), which then will wane. Finally, if blocking takes place the protection provided by the vaccine will be reduced. The control cohort in this case is cohort 1 again. Trial III: Intervention group corresponds to cohort 4, joined by individuals firstly vaccinated at birth, and revaccinated at school age. At variance to the first dose, which cannot be blocked, the second dose might be blocked by ES or not, in which it will reset the initial protection levels provided by the vaccine. The control group for this trial is cohort 2, that corresponds to individuals only vaccinated at birth. The grey shaded area represents the age window of the individuals enrolled in the study.

Figure 4. Confidence intervals estimation scheme.

Degraded shades represent the joint probability density associated to the estimation of confidence intervals around the model best fit. The modulation coefficient c is determined so as to make the brown area within the black line of $L\left(\overrightarrow{P}\right)=0.05$ to precisely accumulate the 95% of the total joint probability distribution.

Figure 5. Distribution of the parameters which yield a maximum in the likelihood function.

(A–C) Hill climbing algorithm distributions for models 1, 2 and 3, respectively. Starting from a series of randomly distributed points in the parameter space (their coordinates distributions are represented in red), a random displacement following a uniform distribution in the parameter space within a hyper-cube of size d = 0.001 is attempted at each time step, and accepted only if it corresponds to an increasing of the likelihood function $L\left(\overrightarrow{P}\right)$. The algorithm stops when no further move is accepted after N = 10⁷ rejected displacements (i.e., the function $L\left(\overrightarrow{P}\right)$ reaches a maximum). In green, we see the peaked distribution of the end points of the algorithm around the solution of the models. (D and E) parameters cliff yielding quasi-constant values of maximum likelihood $L\left({\overrightarrow{P}}^{\ast }\right)=0.79$ for model 1 and $L\left({\overrightarrow{P}}^{\ast }\right)=0.002$ for model 3. As it can be seen in (A) and (D), the model versions that contemplate masking are unable to provide a clear univocal vaccine description yielding maximum likelihood. The reason for this behavior is the existence of a region in the parameters space, represented in (D) and (E), within which, likelihood is almost constant and close to its absolute maximum.

Figure 6. Scheme of the basis for evaluation of anti tuberculosis vaccines in absence of universally reliable protection correlates.

First stage: design of vaccine efficacy determination clinical trials: the age of the cohorts must be elected taking into account that prior exposure to mycobacteria—either environmental, M. tuberculosis after exposure or even prior TST or also BCG—may corrupt the observed vaccine efficacies. Second stage: vaccine impact evaluations: bulk, short-term and long-term impact forecasts should be equally considered, as well as age-distributed impacts in terms of cases, infections and casualties prevented.