Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Sep 30.
Published in final edited form as: Vaccine. 2026 Aug 31;91:129097. doi: 10.1016/j.vaccine.2026.129097

Evaluating the influence of vaccine archetypes and epidemiological conditions on clinical trial metrics for tuberculosis vaccine candidates

Jason R Andrews 1,2, Sabine Hermans 2,3, Richard G White 4, Alberto L García-Basteiro 5, Julio Croda 6,7,8, Frank Cobelens 2
PMCID: PMC13623288  NIHMSID: NIHMS2210352  PMID: 42673713

Abstract

Background.

Conducting tuberculosis vaccine trials in high-transmission settings offers potential gains in efficiency, yet the impact of repeated exposure in the presence of incomplete (“leaky”) protection on measured vaccine efficacy is uncertain and could result in a potentially effective vaccine to be incorrectly deemed ineffective. We aimed to evaluate the effect of vaccine characteristics and epidemiological context on estimates of vaccine efficacy and requisite sample size for tuberculosis vaccine trials.

Methods.

We simulated clinical trials of hypothetical tuberculosis vaccines using a stochastic compartmental model of tuberculosis natural history. We considered vaccines that confer “leaky” (50% per-exposure reduction) versus “all-or-none” protection against infection and/or disease that were evaluated in QuantiFERON Gold Plus (QFT) positive, negative or mixed populations with 2% (medium burden; incidence 400 per 100,000), 5% (high burden; 800 per 100,000) or 50% (very high burden; 4500 per 100,000) annual risk of infection (ARI), including protection against infection alone. We compared estimates of vaccine efficacy and statistical power to detect differences according to sample size for each vaccine archetype and epidemiologic setting.

Findings.

Simulated trials of vaccines protecting against disease, alone or with protection against infection, gave comparable estimates of efficacy across epidemiological settings, whether all-or-none or leaky. The largest reduction was for a leaky vaccine protecting against both infection and disease in QFT negative participants, whose efficacy fell from 75% at medium ARI to 67% at very high ARI (relative reduction 11%). A vaccine acting only by preventing infection fell further, from 50% to 36% (28%). Trials at very high ARI could nonetheless achieve 90% power with 80-96% fewer participants than at medium ARI.

Interpretation.

While vaccines conferring leaky protection against infection and disease tested in settings with very high ARI could modestly underestimate efficacy, trials could be adequately powered with a fraction of the sample size in such settings. Further, trials could enroll a mixture of QFT negative and positive individuals, increasing generalizability.

Keywords: tuberculosis, vaccines, clinical trials, power, sample size

BACKGROUND

Tuberculosis remains the leading cause of death by an infectious disease, and incidence and mortality are declining too slowly to meet the WHO End TB Strategy milestones.1 BCG remains the only available vaccine against tuberculosis; while it protects young children against disease, it confers no protection to adults, in whom the majority of tuberculosis disease occurs globally.2 The development of a vaccine that confers protection to adolescents and adults could transform global efforts to control tuberculosis.3

More than a dozen tuberculosis vaccine candidates are in various stages of development. However, only one vaccine (M72/AS01E) has demonstrated efficacy against disease in Phase 2b trials.4 This protein adjuvant vaccine demonstrated a 50% reduction in disease among people with a positive QuantiFERON test indicating prior exposure to Mycobacterium tuberculosis (Mtb). A Phase 3 trial is now underway, at a projected cost of $550 million. This trial commenced over 5 years following the completion of the Phase 2b trial, a delay largely due to the investments required for a large, lengthy trial.5

One of the key reasons that clinical trials for tuberculosis are so time-consuming and expensive is that the incidence of disease, even in high burden communities, is typically less than 1% per year. In the Phase 2b trial of M72/AS01E, incidence was 0.6% per year in the placebo arm. For comparison, incidence of COVID-19 in the pivotal Phase 3 trial of the mRNA-1273 vaccine was 6% per year.6 Conducting tuberculosis vaccine trials in higher risk populations could potentially lead to vaccine licensure more quickly, which could be critical given the annual mortality from this disease. Further, it could reduce costs, enabling more candidates to be evaluated, increasing the chances that effective vaccines are licensed and made available.

However, conducting trials in such populations are not without concerns and risks. One of the populations that has been identified for potential inclusion in clinical trials are persons deprived of liberty (PDL). PDL consistently have among the highest rates of tuberculosis globally, often 10 times or greater than the incidence of their surrounding communities.7,8 Because of a reprehensible history of unethical experiments conducted on PDL, there have been appropriate concerns about their inclusion in vaccine trials.9 However, opinions are shifting. Ethicists have argued that excluding PDL from research has itself been harmful, both by preventing them from benefiting from research on diseases such as tuberculosis that disproportionately affect them, and by denying them the autonomy to decide whether to participate.10,11 Further, in some countries and regions, PDL constitute the largest risk group for tuberculosis;12 conducting trials that do not include them could be considered ethically and scientifically problematic.

Inclusion of high tuberculosis burden populations such as PDL may pose scientific risks. Vaccines are often assumed to have either “all-or-none” protection, where some fraction of the population is not protected (reflecting heterogeneities in immune responses), or “leaky” protection (where protection is conferred on a per-exposure basis in the whole vaccinated population).13,14 Protection from “leaky” vaccines can be reduced through repeated exposures, as each exposure permits a chance of infection. Leakiness has been invoked to explain patterns of protection from influenza and COVID-19 vaccines.15,16 In very high transmission environments, leakiness could lead to a scenario in which a vaccine that might be effective in lower ARI settings appears ineffective or less effective.

To explore these concerns, we developed mathematical models of tuberculosis natural history and vaccination and calibrated these to data on incidence of infections and disease from settings with varying forces of infection. We simulated vaccine trials under a variety of assumptions about how tuberculosis vaccines generate immunity and examined vaccine efficacy estimates and statistical power.

METHODS

Model Overview

To simulate the efficacy of tuberculosis vaccines, we developed and calibrated compartmental models of tuberculosis natural history based on previously published models.17,18 The base model stratifies individuals into four states: susceptible, early Mtb infection, later Mtb infection, and tuberculosis disease. Susceptible individuals, upon exposure, may enter the early M. tuberculosis infection state, where they have a higher risk of progression to disease, or they may move into a later infection state, where they have a lower risk of disease (e.g. reactivation). Re-exposure can lead to re-infection, which moves an individual back into the early Mtb infection state.19 The model structure is shown in Figure 1. The model was described by a series of stochastic difference equations that govern transitions between compartments (See Appendix).

Figure 1.

Figure 1.

Open in a new tab

Tuberculosis vaccine model schemata with natural history states for Susceptible (S), early infection (E), later infection (L) and active tuberculosis (D). (A) Model with “leaky” protection against infection (VEI) and/or disease (VED); each arrow is labelled with its complete transition rate, so the transitions on which vaccine efficacy acts are explicit. (B) Model with “all-or-none” protection against infection and/or disease. Individuals with latent tuberculosis who are vaccinated with protection against infection go to early infection vaccinated (Ev) or later infection vaccinated (Lv) compartments. For susceptible individuals, protection against infection also provides full protection against disease (Pv).

For vaccination, we considered several potential forms of protection. First, we considered “all or none” (AoN) protection, whereby a fraction of individuals vaccinated receive complete protection (entering a fully protected compartment) and the remainder receive no protection. Under this model, the proportion of individuals entering the (completely) protected compartment determine vaccine efficacy. Second, we considered “leaky” protection, whereby all individuals have reduced risk of infection or disease. Under the leaky model, reduction in risk of infection or disease determines efficacy.

We then considered scenarios whereby protection was against infection (POI), against disease (POD), or against both infection and disease. For protection against disease, we assumed that protection was equally effective in protecting progression from early or later Mtb infection states, and for protection against infection, we assumed protection against reinfection was the same as protection against initial infection. These assumptions were the same for leaky and AoN vaccines. Partial immunity from prior infection is a natural history parameter applied identically in both arms and is distinct from vaccine-conferred protection (See Appendix). We did not incorporate HIV status or model its impact on natural history or vaccine protection. As a three-year trial was simulated, we did not model waning of vaccine efficacy over the trial period.

Epidemiologic Scenarios and Calibration

We considered three epidemiologic scenarios (Table 1):

Table 1.

py: person-years. The progression ratio was calibrated for each epidemiological archetype as a multiplier on the early and late progression rates. ω is a natural history parameter representing partial immunity conferred by prior Mycobacterium tuberculosis infection; it is applied identically in the vaccine and placebo arms and is used in both the base and the alternative model. Vaccine-conferred protection against infection and reinfection is represented separately by VEI.

Parameter Value Reference
Shared model parameters
   Duration of early infection 0.89 y 18
   Early progression rate 0.05 y−1 18,20
   Late progression rate 0.0015 y−1 21
   Relative risk of infection given prior infection (partial natural immunity), ω 0.21 19
Medium burden setting parameters
   Incidence 400/100,000py 1
   Annual risk of infection 0.02 y−1 22
   Infection prevalence 40% 22,23
   Progression ratio 3.58 fitted
High burden setting parameters
   Incidence 800/100,000py 1
   Annual risk of infection 0.05 y−1 24,25
   Infection prevalence 55% 23,26
   Progression ratio 4.30 fitted
Very high burden setting parameters
   Incidence 4500/100,000py 27
   Annual risk of infection 0.50 y−1 27
   Infection prevalence 70% 27
   Progression ratio 5.07 fitted
Open in a new tab

  1. A medium/high tuberculosis burden setting with an annual risk of tuberculosis infection of 2%, tuberculosis incidence of 400 per 100,000 and Mtb infection prevalence of 40%. These epidemiologic characteristics are similar to what is seen in many communities in Indonesia.22,23

  2. A high tuberculosis burden community setting with an annual risk of infection of 5%, incidence of 800 per 100,000 and a Mtb infection prevalence of 55%. These characteristics are similar to that reported in some South African townships.26,28

  3. A very high tuberculosis burden prison, with annual risk of infection of 50%, incidence of 4500 per 100,000 and Mtb infection prevalence of 70%. These estimates were taken from recent studies performed in Brazilian prisons and are consistent with prior estimates from high burden prisons elsewhere.27

We held the ratio of early to late progression rates fixed at literature values and calibrated a single setting-specific multiplier applied to both, so that simulated incidence matched the target for each setting (Table 1). Because simulated incidence is stochastic, we fixed the random seed within the objective function and solved by bisection. The fitted multiplier was higher in settings with a higher force of infection (3.58 to 5.07). Progression and reactivation rates did not vary substantially across the three scenarios.

Vaccination Scenarios

We modeled hypothetical vaccines that were “all-or-none” or “leaky” and achieving POI, POD or both. For leaky vaccines we modeled POI alone, POD alone, or both; “POI alone” denotes no additional effect on progression once infection has occurred, so such a vaccine prevents disease only through the infections it averts, and is most informatively evaluated in QFT negative participants. Of note, we did not model an all-or-none vaccine with POI only, as having complete POI (at the individual level with an all-or-none vaccine) would also confer full POD. We then simulated recruitment of cohorts that were either: 1) all QuantiFERON positive (assumed to mean having current Mtb infection); 2) all QuantiFERON negative (assumed to mean not having current Mtb infection); 3) a mixture of QuantiFERON positive and negative individuals (assumed to mean have any current infection status), recruited in proportion to the assumed prevalence in the population. The proportion of individuals in the early and late infection states was taken as the split implied by a stationary population sustaining the observed force of infection and infection prevalence (See Appendix).

Model Analysis

We simulated cohorts of individuals who were recruited into a clinical trial lasting three years. For each scenario, we performed 5000 simulations for each epidemiologic scenario and vaccine archetype. For simplicity, we assumed that individuals in the cohort who developed tuberculosis were accurately diagnosed and were censored from further follow-up. We did not incorporate loss to follow up. We calculated incident tuberculosis cases and follow-up time in the placebo and vaccine groups and estimated “observed” (model produced) vaccine efficacy as 1 – incidence rate ratio of incidence in the vaccine group divided by incidence in the placebo group. We used a two sample Poisson test to compare rates in the vaccine and placebo groups and calculated statistical power as the proportion of simulations (β) in which the p-value for this comparison was less than 0.05 (α).

For the initial comparisons of vaccine efficacy, we assumed a sample size of 1500 participants per arm and a true (modeled) efficacy of 50% for each mechanism of protection, comparable to the Phase 2b trial of the M72/AS01E vaccine. Where a vaccine acted through both mechanisms these combined multiplicatively in uninfected participants, giving an expected efficacy against disease of 75%; expected efficacy therefore differs by archetype and cohort (Supplemental Table S1), and the effect of leakiness is the change across settings within an archetype rather than the departure from 50%. We repeated the efficacy simulations at 450 and 300 participants per arm. We then calculated the power to detect a difference in incidence between the arms according to epidemiologic scenario, vaccine archetype, and whether participants who were QuantiFERON positive, negative, or either were enrolled, under a range of sample sizes. Because there is some evidence that tuberculosis vaccines may have greater immunogenicity in QuantiFERON positive individuals than QuantiFERON negative individuals,29 we performed additional analyses in which efficacy was reduced by 33% or 50% in the latter group (Supplemental Figure S4).

We performed an additional sensitivity analysis around the model structural assumptions. Under the base case model, repeat exposures do not further increase the risk of disease among those in the early Mtb infection state. To relax this assumption in sensitivity analyses, we added a parameter to the model by which each Mtb exposure led to an additional risk of progression to disease among individuals with early Mtb infection (See Supplemental Figure S1); this pathway is present only in the alternative model.

The models were coded in R, and code is available at https://github.com/jasonandr/TBVaccineARI

RESULTS

In a hypothetical trial with 1500 QuantiFERON positive participants per arm, the expected number of tuberculosis outcomes over 3 years in the control arm varied by incidence setting: 31 outcomes in the medium incidence setting, 43 outcomes in the high incidence setting and 168 outcomes in the very high incidence setting. Enrolling a representative sample matching local QuantiFERON prevalence reduced the number of endpoints, with greater reduction in the medium incidence setting (18 outcomes; 42% reduction) than the high (36 outcomes; 16% reduction) or very high incidence setting (189 outcomes; no reduction).

Observed vaccine efficacy did not differ substantially across epidemiologic settings for vaccines protecting against disease alone, in any cohort (Figure 2; Supplemental Figures S2 and S3). For vaccines also protecting against infection, efficacy in QuantiFERON-positive and mixed cohorts rose modestly with ARI, since protection against infection also prevents reinfection (52% to 62% for an all-or-none vaccine in QuantiFERON-positive participants). In a QuantiFERON-negative population, there was a modest reduction in observed vaccine efficacy when a vaccine with leaky protection against both disease and infection was evaluated in the very high incidence setting (median, 67%) compared with the high (median, 74%) and medium (median, 75%) incidence settings, a relative reduction of 11% (Supplemental Figure S3). In a trial of individuals with QuantiFERON prevalence reflecting the local population average, observed vaccine efficacy for this archetype rose modestly across settings (61.4% versus 58.5%; Figure 2). The archetype most sensitive to leakiness was a vaccine protecting against infection alone: in QuantiFERON-negative participants its efficacy against disease was 50% at medium and 50% at high ARI but fell to 36% at very high ARI, a relative reduction of 28% (Supplemental Figure S5). In QuantiFERON-positive participants the same vaccine was near-null at lower ARI (3% and 7%).

Figure 2.

Figure 2.

Open in a new tab

Observed vaccine efficacy with vaccines that confer “all-or-none” (A,C) or “leaky” (B,D) protection against tuberculosis disease (POD) or POD and prevention against infection (POI), according to annual risk of infection. Modeled efficacy for each vaccine mechanism is 50%; the dashed line indicates the expected efficacy against disease for that archetype and cohort (Supplemental Table S1). Results show observed efficacy estimates for a hypothetical study involving 1500 participants per arm, including all individuals with QFT status reflecting the infection prevalence for each setting.

While leaky vaccines had small reductions in observed efficacy at very high ARI, the power to detect differences between trial arms was much greater as ARI increased (Figure 3). For a vaccine with leaky protection against disease evaluated in QFT positive populations, the requisite sample size per arm to achieve 90% power was 3,269 in the medium ARI setting, 2,392 in the high ARI setting and 665 in the very high ARI setting; across the four archetypes these ranged from 2,941-3,269, 1,930-2,392 and 347-665 respectively. For all vaccine archetypes, power to detect vaccine efficacy, for a given sample size, was positively correlated with ARI. The differences were accentuated in settings in which the study population had a mixture of QFT negative and positive individuals. For example, to achieve 90% power, an all-or-none protection vaccine evaluated in a QFT positive population would require 548 participants per arm in a very high ARI setting and 3,032 participants per arm in a medium ARI setting. A trial of the same vaccine, if conducted in a population with QFT positivity according to local epidemiology would require 470 participants in the very high ARI setting (14% fewer) and 5,500 participants in the medium ARI setting (81% more) to achieve 90% power.

Figure 3.

Figure 3.

Open in a new tab

Power to detect a difference in incidence rates for a vaccine with an efficacy of 50% according to the sample size, annual risk of infection, vaccine mode of effects and population studied. AoN: All-or-none; QFT: QuantiFERON. POD: Prevention of disease; POI: Prevention of infection; ARI: Annual risk of infection.

Vaccines that provide protection against both infection and disease could have even greater efficacy, and trials of such vaccines could have greater power for a given sample size, particularly when studied in QFT negative individuals and in settings with very high ARI. A trial of a leaky vaccine that confers 50% protection against infection and 50% protection against disease would have 90% power to detect effects with just 202 participants per arm. Similar findings were observed for all-or-none vaccines. A vaccine protecting against infection alone was the exception, requiring 867 participants per arm in QFT negative individuals at very high ARI. At the smaller trial sizes at which such studies would realistically be conducted the central estimate changed little (median 66.3%, 66.4% and 66.6% at 1500, 450 and 300 per arm for a leaky POD/POI vaccine in QFT negative participants at very high ARI), but the 2.5th-97.5th percentile range widened from 57-74% to 44-82% and the proportion of trials returning an estimate below 50% rose from 0% to 6% (Supplemental Figure S6).

We tested a scenario where vaccines had reduced efficacy for QFT negative individuals compared with QFT positive individuals. With an all-or-nothing vaccine evaluated in a population of QFT positive and negative individuals, power was reduced as efficacy in QFT negative individuals decreased. However, even when vaccine efficacy in QFT negative individuals was 50% as great as those in QFT positive individuals, a trial conducted the very high transmission setting with a mix of QFT positive and negative individuals was still far more efficient than a trial conducted among only QFT positive individuals in a medium or high transmission setting (Supplemental Figure S4).

In structural sensitivity analysis, we tested an alternative model structure in which individuals who are in an early M. tuberculosis infection state can be re-infected and have an increased risk of fast progression to tuberculosis (Supplemental Figure S1; Supplemental Figures S7-S9). Under this model, for a vaccine with leaky protection against infection and disease, observed efficacy did not differ substantially across epidemiologic scenarios in QFT negative cohorts (75% in medium ARI, 73% in very high) (Supplemental Figure S8). In QFT positive populations, efficacy was higher (66%) in very high ARI than medium ARI (52%) due to more infections being driven by reinfections, and a vaccine with protection against infection and disease providing two opportunities for protection (Supplemental Figure S9). All-or-none and leaky archetypes gave similar results under this alternative model, as in the base model.

DISCUSSION

While there are many promising vaccine candidates in the development pipeline, a critical limiting factor to testing their efficacy is the size and costs of trials. Identifying ways to reduce these could accelerate testing of vaccine candidates and increase that vaccines are licensed and scaled up in populations that urgently need them. While studies can be performed more efficiently in high burden populations, one concern is that repeated exposures could underestimate vaccine efficacy, specifically if vaccine protection is ‘leaky’, meaning that it provides partial protection against each exposure. Using models to simulate infections, natural history and different archetypes of vaccine protection and epidemiological settings, we found that while leaky protection would lead to lower estimated vaccine efficacy in settings with very high ARI, this reduction was modest and more than offset by substantial gains in statistical power from conducting trials in settings with high rates of infection and disease. Further, for vaccines conferring protection against both infection and disease, conducting trials in QFT negative individuals in a setting with very high ARI would be particularly efficient, reducing sample size by >85% compared with high burden settings where current trials are being undertaken.

Enrolling QFT negative individuals in these settings raises several considerations. Although around 70% of people in high-transmission prisons are QFT positive, so that only 30% would be eligible at any one time, approximately 90% of people entering prison in Brazil are QFT negative at entry,30 and recruiting at entry would therefore enrich substantially for eligible participants. Individuals who convert during follow-up should be offered preventive treatment, although trials may not measure QFT longitudinally and serial testing is not standard practice in most prisons in low- and middle-income countries. Efficacy estimated in PDL may also raise questions of generalizability, since exposure intensity, comorbidity and demographic structure differ from the general population, and post-licensure effectiveness studies may be needed before extending a recommendation more broadly. Finally, inclusion carries a reciprocal obligation: if a vaccine is shown to be efficacious, PDL should be among the first to receive it.

Tuberculosis vaccines may prevent infection, disease, or both. We evaluated hypothetical vaccines that protected against disease or against infection and disease. If vaccines confer protection against both, they could potentially have greater overall efficacy if given to M. tuberculosis susceptible individuals. However, testing vaccines in QFT-negative individuals can require large sample sizes unless the force of infection is very high. The current Phase 3 trial of the M72 vaccine is focused predominantly on QFT positive individuals, with a small safety and immunogenicity cohort of QFT negative individuals, because incidence is anticipated to be higher in QFT positive individuals. Further, immune responses to vaccination may be more robust in individuals with positive QFT due to immune priming from prior infection;29,31 whether this will translate into differential efficacy between QFT positive and negative individuals is unknown. In settings with very high force of infection, the differences in tuberculosis incidence between QFT negative and QFT positive individuals is smaller. This could enable trials to enroll a mix of individuals, rather than needing to extrapolate efficacy from QFT positive individuals.32 Our modeling suggests that even when we assume a lower efficacy in QFT negative individuals, trials could be done efficiently in a mixed population of QFT positive and negative individuals when the ARI is very high.

For modeling and epidemiologic purposes, vaccines against any pathogen are often assumed to confer all-or-none or leaky protection. All-or-none protection reflects population heterogeneity in response to vaccines, whereas leaky protection confers partial protection against each exposure. Vaccines can have both characteristics. That measured efficacy of leaky vaccines declines as the force of infection rises is mathematically expected;13,14,16,33,34 the question for trial design is whether the magnitude of that bias is material. Many vaccines maintain high efficacy in settings with extremely high rates of exposure.35,36 For tuberculosis, it is unclear whether vaccines will confer leaky protection; leaky protection against infection could reduce vaccine efficacy in settings with high force of infection. While we found that observed efficacy was lower for a hypothetical vaccine with leaky protection against infection and disease, the reduction was fairly modest even at very high ARI. This may be in part because for tuberculosis, unlike many other diseases, individuals can be reinfected, so there is not complete, rapid depletion of individuals who are susceptible to infection in both the vaccinated and unvaccinated arms (which can bias vaccine effects towards the null in other diseases). Since leakiness classically relates to exposures, the reduction in protection is greatest in QFT negative populations, because partial immunity to reinfection mitigates the effects among already Mtb-infected individuals, as previously described.34 Nevertheless, among vaccines conferring protection against disease, and even in QFT negative populations with the archetype most sensitive to leakiness among them (leaky vaccine with POD and POI), the relative reduction in efficacy in a 1500-per-arm trial was only 11% in the very high ARI setting, and was essentially identical in trials of 450 and 300 per arm; in all other such archetypes and with QFT mixed or positive populations, efficacy was similar in very high and medium ARI settings. The exception was a vaccine protecting against infection alone, which prevents disease only by preventing the infections that cause it: in QFT negative participants its efficacy fell from 50% to 36%, a relative reduction of 28%. At an ARI of 50% nearly all control participants become infected within the trial, so halving the infection hazard reduces three-year cumulative infection by only 26%. Trials of candidates believed to act principally by preventing infection should anticipate this.

The results of this analysis should be interpreted within the context of a number of limitations. First, we used simplified models of tuberculosis natural history, based on widely published structures, but these models do not capture many complexities and heterogeneities that can affect disease risk and vaccine efficacy.37 In this model, infection is represented in two states, early and late, with reinfections returning individuals from later infection into an early infection state. We also tested an alternative model in which individuals with early infection could be reinfected, described further in the Supplement. We assumed that QuantiFERON status accurately reflects infection status, whereas in reality it has imperfect sensitivity and specificity.38 However, despite limited positive predictive value, QFT does predict progression to disease, and the qualitative results of this analysis are likely to be robust to this assumption.39,40 For modeling prevention of infection and prevention of disease, we assumed that protection against either was independent, whereas in reality immune responses are likely to be correlated; such correlation would likely reduce overall efficacy but are unlikely to do so differentially by ARI. Finally, observed efficacy of leaky vaccines may vary over time, violating proportional hazards assumptions;14 we did not model subpopulations with heterogeneity that could produce such effects, but these may be important to consider for trial planning.

With more than a million people dying of tuberculosis every year, the development of effective vaccines is a critical priority. Hastening clinical trials could increase the likelihood that new vaccines are found to be efficacious and rolled out to populations in need; however, this must be done in ways that minimize risks of type II error—that an effective vaccine is inappropriately found to be ineffective. Our results suggest this error is unlikely. We simulated clinical trials of hypothetical vaccines with different characteristics of protection in distinct epidemiologic scenarios and found that while vaccines with leaky protection against infection and disease would lead to lower observed efficacy in populations with very high force of infection, such as PDL, these reductions were modest. Further, trials in such settings could be done with 80-96% fewer participants than those being done in many high burden community settings, and they could be performed in populations with a mixture of QFT negative and QFT positive individuals, improving generalizability of results. Taken together, our findings suggest that tuberculosis vaccine trials can be conducted in high-transmission settings without materially compromising trial validity or the estimation of vaccine efficacy.

Supplementary Material

Appendix

Funding:

JRA is supported by NIH K24AI182647 and Coefficient Giving. JC is supported by Coefficient Giving. RGW is funded by the Wellcome Trust (310728/Z/24/Z, 218261/Z/19/Z), NIH (1R01AI147321-01, G-202303-69963, R-202309-71190), EDTCP (RIA208D-2505B), ESRC (ES/P008011/1), BMGF (INV-004737, INV-035506), Open Philanthropy/Good Ventures (GV673606227), and the WHO (2020/985800-0). ALGB receives funding support from: EDCTP (RIA101249120, RIA2018D-2511, RIA101190731), Marató (202510303132), ERC (101117940, 101001038), King Baudouin Foundation (2025-I4242220-0025486), Biontech (164-02 4781/0051). FC reports funding from the EDCTP2 programme supported by the EU (RIA2018D-2509, RIA2020I-3305) and the Mr. Willem Bakhuys Roozeboom Foundation (2022-2, 2022-6, 2022-7, 2022-8, 2023-8, 2024-6, 2025-2).

Footnotes

Conflicts of interest: JRA, SMH, ALGB, RW, JC, FC: none to declare.

References

  • 1.World Health Organization. Global Tuberculosis Report 2025.; 2025. [Google Scholar]
  • 2.Martinez L, Cords O, Liu Q, et al. Infant BCG vaccination and risk of pulmonary and extrapulmonary tuberculosis throughout the life course: a systematic review and individual participant data meta-analysis. Lancet Glob Health. 2022;10(9):e1307–e1316. doi: 10.1016/S2214-109X(22)00283-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Clark RA, Mukandavire C, Portnoy A, et al. The impact of alternative delivery strategies for novel tuberculosis vaccines in low-income and middle-income countries: a modelling study. Lancet Glob Health. 2023;11(4):e546–e555. doi: 10.1016/S2214-109X(23)00045-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Tait DR, Hatherill M, Van Der Meeren O, et al. Final Analysis of a Trial of M72/AS01E Vaccine to Prevent Tuberculosis. N Engl J Med. 2019;381(25):2429–2439. doi: 10.1056/NEJMoa1909953 [DOI] [PubMed] [Google Scholar]
  • 5.Barry-Jester AM. How a Big Pharma Company Stalled a Potentially Lifesaving Vaccine in Pursuit of Bigger Profits. ProPublica. October 4, 2023. Accessed April 30, 2024. https://www.propublica.org/article/how-big-pharma-company-stalled-tuberculosis-vaccine-to-pursue-bigger-profits [Google Scholar]
  • 6.Baden LR, Sahly HME, Essink B, et al. Efficacy and Safety of the mRNA-1273 SARS-CoV-2 Vaccine. New England Journal of Medicine. 2021;384(5):403–416. doi: 10.1056/NEJMoa2035389 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Cords O, Martinez L, Warren JL, et al. Incidence and prevalence of tuberculosis in incarcerated populations: a systematic review and meta-analysis. Lancet Public Health. 2021;6(5):e300–e308. doi: 10.1016/S2468-2667(21)00025-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Martinez L, Warren JL, Harries AD, et al. Global, regional, and national estimates of tuberculosis incidence and case detection among incarcerated individuals from 2000 to 2019: a systematic analysis. Lancet Public Health. 2023;8(7):e511–e519. doi: 10.1016/S2468-2667(23)00097-X [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Strassle C, Jardas E, Ochoa J, et al. Covid-19 Vaccine Trials and Incarcerated People — The Ethics of Inclusion. New England Journal of Medicine. 2020;383(20):1897–1899. doi: 10.1056/NEJMp2025955 [DOI] [PubMed] [Google Scholar]
  • 10.Andrews JR, Charalambous S, Churchyard G, et al. The participation of people deprived of liberty in tuberculosis vaccine trials: should they be protected from research, or through research? Lancet Infect Dis. Published online June 27, 2025:S1473-3099(25)00305-6. doi: 10.1016/S1473-3099(25)00305-6 [DOI] [PubMed] [Google Scholar]
  • 11.United Nations. United Nationals Standard Minimum Rules for the Treatent of Prisoners (the Nelson Mandela Rules); 2016. Accessed August 4, 2021. https://undocs.org/A/RES/70/175 [Google Scholar]
  • 12.Liu YE, Mabene Y, Camelo S, et al. Mass incarceration as a driver of the tuberculosis epidemic in Latin America and projected effects of policy alternatives: a mathematical modelling study. Lancet Public Health. 2024;9(11):e841–e851. doi: 10.1016/S2468-2667(24)00192-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Halloran ME, Struchiner CJ. Thirty-five years of leaky vaccines. Am J Epidemiol. 2024;194(4):918–920. doi: 10.1093/aje/kwae379 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Edlefsen PT. Leaky vaccines protect highly exposed recipients at a lower rate: implications for vaccine efficacy estimation and sieve analysis. Comput Math Methods Med. 2014;2014:813789. doi: 10.1155/2014/813789 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Tokars JI, Patel MM, Foppa IM, Reed C, Fry AM, Ferdinands JM. Waning of Measured Influenza Vaccine Effectiveness Over Time: The Potential Contribution of Leaky Vaccine Effect. Clin Infect Dis. 2020;71(10):e633–e641. doi: 10.1093/cid/ciaa340 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Lind ML, Dorion M, Houde AJ, et al. Evidence of leaky protection following COVID-19 vaccination and SARS-CoV-2 infection in an incarcerated population. Nat Commun. 2023;14(1):5055. doi: 10.1038/s41467-023-40750-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Blower SM, McLean AR, Porco TC, et al. The intrinsic transmission dynamics of tuberculosis epidemics. Nat Med. 1995;1(8):815–821. [DOI] [PubMed] [Google Scholar]
  • 18.Menzies NA, Wolf E, Connors D, et al. Progression from latent infection to active disease in dynamic tuberculosis transmission models: a systematic review of the validity of modelling assumptions. Lancet Infect Dis. 2018;18(8):e228–e238. doi: 10.1016/S1473-3099(18)30134-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Andrews JR, Noubary F, Walensky RP, Cerda R, Losina E, Horsburgh CR. Risk of progression to active tuberculosis following reinfection with Mycobacterium tuberculosis. Clin Infect Dis. 2012;54(6):784–791. doi: 10.1093/cid/cir951 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Sutherland I, Svandová E, Radhakrishna S. The development of clinical tuberculosis following infection with tubercle bacilli. 1. A theoretical model for the development of clinical tuberculosis following infection, linking from data on the risk of tuberculous infection and the incidence of clinical tuberculosis in the Netherlands. Tubercle. 1982;63(4):255–268. [DOI] [PubMed] [Google Scholar]
  • 21.Horsburgh CR Jr. Priorities for the treatment of latent tuberculosis infection in the United States. N Engl J Med. 2004;350(20):2060–2067. doi: 10.1056/NEJMsa031667 [DOI] [PubMed] [Google Scholar]
  • 22.Schwalb A, Dodd P, Rickman HM, Ugarte-Gil C, Horton KC, Houben RMGJ. Estimating the Global Burden of Viable Mycobacterium Tuberculosis Infection. Social Science Research Network. Preprint posted online November 13, 2024. doi: 10.2139/ssrn.5017943 [DOI] [Google Scholar]
  • 23.Ding C, Hu M, Guo W, et al. Prevalence trends of latent tuberculosis infection at the global, regional, and country levels from 1990–2019. International Journal of Infectious Diseases. 2022;122:46–62. doi: 10.1016/j.ijid.2022.05.029 [DOI] [PubMed] [Google Scholar]
  • 24.Middelkoop K, Bekker LG, Liang H, et al. Force of tuberculosis infection among adolescents in a high HIV and TB prevalence community: a cross-sectional observation study. BMC Infect Dis. 2011;11:156. doi: 10.1186/1471-2334-11-156 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Wood R, Liang H, Wu H, et al. Changing prevalence of tuberculosis infection with increasing age in high-burden townships in South Africa. Int J Tuberc Lung Dis. 2010;14(4):406–412. [PMC free article] [PubMed] [Google Scholar]
  • 26.Mahomed H, Hawkridge T, Verver S, et al. Predictive factors for latent tuberculosis infection among adolescents in a high-burden area in South Africa. Int J Tuberc Lung Dis. 2011;15(3):331–336. [PubMed] [Google Scholar]
  • 27.Bampi J, Jung E, Salindri A, et al. Intensive screening reduces tuberculosis transmission over time in prisons. In: 2025. [Google Scholar]
  • 28.Moyo S, Ismail F, Walt MV der, et al. Prevalence of bacteriologically confirmed pulmonary tuberculosis in South Africa, 2017–19: a multistage, cluster-based, cross-sectional survey. The Lancet Infectious Diseases. 2022;22(8):1172–1180. doi: 10.1016/S1473-3099(22)00149-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Penn-Nicholson A, Tameris M, Smit E, et al. Safety and immunogenicity of the novel tuberculosis vaccine ID93 + GLA-SE in BCG-vaccinated healthy adults in South Africa: a randomised, double-blind, placebo-controlled phase 1 trial. Lancet Respir Med. 2018;6(4):287–298. doi: 10.1016/S2213-2600(18)30077-8 [DOI] [PubMed] [Google Scholar]
  • 30.Carbone AS, Paião DSG, Sgarbi RVE, et al. Active and latent tuberculosis in Brazilian correctional facilities: a cross-sectional study. BMC Infect Dis. 2015;15:24. doi: 10.1186/s12879-015-0764-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Penn-Nicholson A, Geldenhuys H, Burny W, et al. Safety and immunogenicity of candidate vaccine M72/AS01E in adolescents in a TB endemic setting. Vaccine. 2015;33(32):4025–4034. doi: 10.1016/j.vaccine.2015.05.088 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Cobelens F, Pelzer PT, Churchyard GJ, et al. Sample size efficiency of restricting participation in tuberculosis vaccine trials to interferon-gamma release assay-positive participants. Vaccine. 2025;61:127301. doi: 10.1016/j.vaccine.2025.127301 [DOI] [PubMed] [Google Scholar]
  • 33.Shim E, Galvani AP. Distinguishing vaccine efficacy and effectiveness. Vaccine. 2012;30(47):6700–6705. doi: 10.1016/j.vaccine.2012.08.045 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Ragonnet R, Trauer JM, Denholm JT, Geard NL, Hellard M, McBryde ES. Vaccination Programs for Endemic Infections: Modelling Real versus Apparent Impacts of Vaccine and Infection Characteristics. Sci Rep. 2015;5(1):15468. doi: 10.1038/srep15468 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Qadri F, Khanam F, Liu X, et al. Protection by vaccination of children against typhoid fever with a Vi-tetanus toxoid conjugate vaccine in urban Bangladesh: a cluster-randomised trial. Lancet. 2021;398(10301):675–684. doi: 10.1016/S0140-6736(21)01124-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Aiemjoy K, Seidman JC, Saha S, et al. Estimating typhoid incidence from community-based serosurveys: a multicohort study. Lancet Microbe. 2022;3(8):e578–e587. doi: 10.1016/S2666-5247(22)00114-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Basu S, Andrews J. Complexity in mathematical models of public health policies: a guide for consumers of models. PLoS Med. 2013;10(10):e1001540. doi: 10.1371/journal.pmed.1001540 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Diel R, Goletti D, Ferrara G, et al. Interferon-γ release assays for the diagnosis of latent Mycobacterium tuberculosis infection: a systematic review and meta-analysis. Eur Respir J. 2011;37(1):88–99. doi: 10.1183/09031936.00115110 [DOI] [PubMed] [Google Scholar]
  • 39.Zhou G, Luo Q, Luo S, et al. Interferon-γ release assays or tuberculin skin test for detection and management of latent tuberculosis infection: a systematic review and meta-analysis. Lancet Infect Dis. 2020;20(12):1457–1469. doi: 10.1016/S1473-3099(20)30276-0 [DOI] [PubMed] [Google Scholar]
  • 40.Hamada Y, Gupta RK, Quartagno M, et al. Predictive performance of interferon-gamma release assays and the tuberculin skin test for incident tuberculosis: an individual participant data meta-analysis. EClinicalMedicine. 2023;56:101815. doi: 10.1016/j.eclinm.2022.101815 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Appendix
NIHMS2210352-supplement-Appendix.pdf (1.4MB, pdf)

RESOURCES