Is neglect of self-clearance biassing TB vaccine impact estimates?

Background. Mathematical modelling has been used extensively to estimate the potential impact of new tuberculosis vaccines, with the majority of existing models assuming that individuals with Mycobacterium tuberculosis (Mtb) infection remain at lifelong risk of tuberculosis disease. Recent research provides evidence that self-clearance of Mtb infection may be common, which may affect the potential impact of new vaccines that only take in infected or uninfected individuals. We explored how the inclusion of self-clearance in models of tuberculosis affects the estimates of vaccine impact in China and India. Methods. For both countries, we calibrated a tuberculosis model to a scenario without self-clearance and to various scenarios with self-clearance. To account for the current uncertainty in self-clearance properties, we varied the rate of self-clearance, and the level of protection against reinfection in self-cleared individuals. We introduced potential new vaccines in 2025, exploring vaccines that work in uninfected or infected individuals only, or that are effective regardless of infection status, and modelling scenarios with different levels of vaccine efficacy in self-cleared individuals. We then estimated the relative incidence reduction in 2050 for each vaccine compared to the no vaccination scenario. Findings. The inclusion of self-clearance increased the estimated relative reductions in incidence in 2050 for vaccines effective only in uninfected individuals, by a maximum of 12% in China and 8% in India. The inclusion of self-clearance increased the estimated impact of vaccines only effective in infected individuals in some scenarios and decreased it in others, by a maximum of 14% in China and 15% in India. As would be expected, the inclusion of self-clearance had minimal impact on estimated reductions in incidence for vaccines that work regardless of infection status. Interpretations. Our work suggests that the neglect of self-clearance in mathematical models of tuberculosis vaccines does not result in substantially biased estimates of tuberculosis vaccine impact. It may, however, mean that we are slightly underestimating the relative advantages of vaccines that work in uninfected individuals only compared to those that work in infected individuals.


Uninfected -Cleared Early
Individuals with previous exposure/infection who have self-cleared. Early indicates that the infection occurred at most 9 years before. No viable bacteria remaining.

Uninfected -Cleared Late
Individuals with previous exposure/infection who have self-cleared. Late indicates that the infection occurred at least 9 years before. No viable bacteria remaining.

Infection -Fast
Infected individuals who were last infected at most 2 years before

Infection -Slow Early
Individuals last infected with Mtb in between two and nine years before, who have not cleared their infection.

Infection -Slow Late
Individuals last infected with Mtb more than nine years before, who have not cleared their infection.

Subclinical Disease
Individuals with active, infectious TB disease, who do not report any of the four WHO TB symptomscreen symptoms

Clinical Disease
Individuals with active, infectious TB disease, who report any of the four WHO TB symptom-screen symptoms

Resolved
Completed treatment or naturally cured from Disease Clinical and Disease Subclinical Where Access to care dimension The access to care dimension contains 2 compartments: high-access-to-care, representing the top 3 quintiles (60% of the population in each country) and low-access-to-care, representing the bottom 2 quintiles (40% of the population in each country). We assumed that there was no transition between the high-and low-accessto-care compartments, as well as assuming random mixing between the high-access-to-care and low-access-tocare compartments.
To constrain relative burden between access-to-care compartments, we calibrated the relative tuberculosis prevalence in the high-access-to-care compartment to the low-access-to-care compartment in 2019. The calibration target, 0.674, was calculated as a weighted average from ten studies [21]-[30], with lower and upper bounds (0.575-0.801) representing the 25th and 75th percentiles of the datasets.
To incorporate access to care into our model, we assume that the differences in tuberculosis burden between compartments are due to differences in the force of infection, the rate of care-seeking (i.e., tuberculosis treatment initiation), and the rate of progression to tuberculosis following infection. We assume relative to the low-access-to-care strata, the high-access-to-care strata has a reduced force of infection per contact, an increased rate of treatment initiation, and a reduced rate of progression. Differential burden was implemented by introducing a new parameter , such that ∈ [0,1], for the high-access-to-care and = 0 for the lowaccess-to-care compartment. This new parameter was included within the model natural history structure as described in Table 3 and was fitted during calibration.  Figure B). The treatment initiation rate parameter, , represents the age specific rate of treatment initiation from the clinical disease compartment. During calibration, we varied a country-specific value for which was sampled between 0 and 1. was then multiplied by an age scaling parameter for children, 4 , also sampled between 0 and 1, to ensure that the treatment initiation rate in children was less than in adults. This was then multiplied by the value of the sigmoid curve at each year. The model was calibrated to the countryspecific notification rate in 2019 overall and by age reported by the WHO. In addition to background mortality, there are three possible exits from the on-treatment compartment: treatment completion, which progresses to the resolved compartment; treatment non-completion, which returns to the clinical disease compartment; and on treatment mortality, which counts toward tuberculosis mortality. To account for the variability in tuberculosis treatment outcomes and possible underreporting of ontreatment mortality, we used the following country-specific process: 1. For each country separately, the proportion of treatment completions was calculated and averaged over the years of available data from WHO.
2. A value for child treatment mortality 0 was sampled between 0 and 2 × . The average reported treatment mortality was multiplied by 2 to give an upper bound in the case of unreported data.
3. The age multiplier, , was sampled from (0,1), and multiplied by 0 to calculate the adult treatment mortality 15 .
4. The success and failure rates per year were calculated as in Table D. Table 1 were divided by the treatment duration to obtain the on-treatment mortality rate per year, on-treatment completion rate per year, and on-treatment non-completion rate per year. [37]

Each of the parameters in
*Implemented as 1-(proportion of overall population in the Uninfected compartment) Table E. Targets used in the calibration process.    Maximum effect when the rate of self-clearance is not within the range provided in [15] We report below the analysis we conducted to estimate the maximum effect on vaccine impact that selfclearance may have if we allowed self-clearance rates higher than those found in  India, CI POID, maximum natural protection and maximum vaccine efficacy in self-cleared individuals E. Maximum effect of self-clearance for current infection vaccines Figure H shows how the percentage reduction in tuberculosis incidence in 2050 varies for each "any infection" vaccine, when we vary the self-clearance rate and the level of natural protection in Uninfected-Cleared individuals at once (note that here we did not vary the vaccine efficacy in Uninfected-Cleared individuals, since "any infection" vaccines were assumed to take on all individuals, independently of their infection status). For each of these two characteristics, we explored the two extreme values, minimum and maximum, obtaining four possibilities for each vaccine. Here we see that self-clearance has very little effect on vaccine impact, with a maximum increase of 6% in China and 3% in India. Figure H. Effect on vaccine impact when all self-clearance characteristics are varied at once for "any infection vaccines" (China on the left, India on the right). The height of the columns show the median percentage reduction in incidence in 2050 compared to the scenario where no vaccine is implemented, with vertical bars indicating the 95% confidence interval. The numbers on top of each column correspond to the relative increase/decrease of the incidence reduction compared to the no self-clearance scenario (red). Scenarios with self-clearance are in blue, while scenarios without self-clearance are in red. NCI: no current infection, CI: current infection, POI: prevention of infection, POD: prevention of disease, POID: prevention of infection and disease.