BCG revaccination did not prevent sustained tuberculosis infection

The trial set out to do two things: confirm that Bacille Calmette-Guerin (BCG) revaccination prevents sustained Mycobacterium tuberculosis (Mtb) infection, and identify candidate correlates of protection for that endpoint. The first failed, which removed the thing the second was supposed to correlate against. The vaccine was plainly immunogenic in the 80-participant sub-cohort, so the question that is left is narrower and answerable: does any vaccine-induced immune measure distinguish the participants who went on to sustained QuantiFERON-TB (QFT) conversion from those who did not, in a trial where the groups converted at the same rate?

1. Has anyone done this since?

Status

partly addressed

Works since

• Antigen-specific Th1 cytokine markers and protection against tuberculosis: a systematic review and meta-analysis stratified by progression to active disease and sustained IGRA conversion (Lin, Liu and Pan 2026, Frontiers in Cellular and Infection Microbiology, doi:10.3389/fcimb.2026.1780600)
  Pooled ten longitudinal human studies and found type 1 helper (Th1) cytokine markers are not correlates: odds ratios for interferon gamma, interleukin-2, tumour necrosis factor alpha and polyfunctional responses all ranged 0.97 to 1.11 for progression to disease, and for sustained interferon gamma release assay conversion the pooled odds ratios were 1.13 and 1.07 with confidence intervals spanning one, so the specific candidate this trial measured is already close to excluded at the population level.
• Efficacy, immunogenicity, and safety of BCG revaccination against Mycobacterium tuberculosis: a systematic review and meta-analysis (Ali and colleagues 2026, Human Vaccines and Immunotherapeutics, doi:10.1080/21645515.2026.2630512)
  Pooled six randomised trials and 2400 participants: no significant effect on initial (risk ratio 0.99) or sustained (risk ratio 0.80) QFT conversion, alongside increased CD8 and cytokine-producing CD4 responses, and names the gap itself as a discrepancy between immune activation and quantifiable protection. It states the problem rather than resolving it.
• To BCG or Not Two BCG (Carpenter and Boom 2025, New England Journal of Medicine, doi:10.1056/NEJMe2502491)
  The accompanying editorial, which frames the immunogenicity-without-efficacy result for the field. Europe PMC holds no abstract or full text for it, so I can confirm only that it exists and comments on this trial.

Searched

The 50 citing works in the supplied OpenAlex file, read by title; Europe PMC by digital object identifier for the four most relevant; and Europe PMC author searches for a separate correlates publication from the trial team (AUTH:"Nemes E" with BCG, revaccination or correlates in the title, 2025 to 2026), which returned only the trial paper itself, the New England Journal of Medicine correspondence reply by Schmidt, Nemes and Hatherill (doi:10.1056/NEJMc2507859), and an unrelated preprint on MR1-restricted T cells in infants.

Still open

No correlates analysis has been published on this trial's own samples, and nothing has tested immune measures outside the Th1 cytokine panel, such as antibody function, mycobacterial growth inhibition or trained innate responses, against the sustained conversion endpoint in a placebo-controlled trial.

Registry query

0 of 49 records in the open node mention any of: bcg revaccination, correlate of protection, th1 cd4 t cells, sustained igra conversion, whole blood intracellular cytokine staining, adolescents south africa. The open node holds biology and chemistry so far, so this is the expected answer; a private node of a clinical group would be asked the same question.

2. A proposed experiment

Question

In the BCG REVAX trial, does any vaccine-induced immune measure at Day 29 or Day 168 distinguish participants who later had sustained QFT conversion from those who did not?

Hypothesis

No measured immune response induced by BCG revaccination is associated with the risk of sustained QFT conversion, so Th1 CD4 frequency cannot serve as a correlate for this endpoint.

Attempt type

wet-lab
Vocabulary note: This is a laboratory correlate-of-risk analysis nested inside a completed trial. The assays are new bench work, which is why wet-lab is the closest value, but the outcomes and the allocation already exist, which makes it equally a secondary-analysis. Neither value says 'new laboratory measurement on stored specimens from a finished trial', and the honest label is both at once.

Design

A case-control study inside the trial. Cases are the 121 participants with sustained QFT conversion, 62 from the BCG group and 59 from placebo. Controls are participants who stayed QFT-negative, matched on arm, site, age stratum and sex, two or three per case. On stored specimens from the pre-vaccination visit, Day 29 and Day 168, run the same whole-blood intracellular cytokine staining panel the trial used, plus a pre-specified second tier: anti-mycobacterial antibody titre and avidity, a mycobacterial growth inhibition assay, and an unstimulated innate activation panel. The analysis is conditional logistic regression of case status on each measure, pre-specified and corrected for multiple comparisons, run separately within each arm, with the BCG arm as the correlate-of-risk analysis and the placebo arm as a check that any association is not simply a marker of exposure.

Controls

• internal-positive The Day 29 BCG-versus-placebo contrast in Th1 CD4 frequency that the trial already reported in Figure 3. The assay must recover that vaccine effect in these samples before any case-control comparison is read.
  Rules out: That a null correlate result is really assay failure or specimen degradation during storage rather than an absence of association.
• background Unstimulated wells from the same specimen, run alongside every stimulated well.
  Rules out: Background cytokine expression being read as antigen-specific response, which would blur any real difference between cases and controls.
• permutation-null Case and control labels shuffled within matching strata, a thousand times, to give the distribution of the strongest association under no effect.
  Rules out: A chance association appearing significant because several immune measures and several timepoints were tested.
• blinding-check Laboratory staff receive specimens with case status and arm masked, and record a guess for a sample of plates.
  Rules out: Measurement drift that tracks case status because the operator knew it.

Sensitivity

This cannot be set from what is published, and the reason matters. The effect sizes worth detecting are now known from Lin and colleagues 2026: pooled odds ratios of 1.13 for interferon gamma and 1.07 for interleukin-2 against sustained conversion. A design with 121 cases cannot detect an odds ratio of that size, so what this study can honestly claim is the exclusion of a large correlate, not of the published small one. Setting the detectable odds ratio requires the variance of the whole-blood intracellular cytokine staining frequencies in this population, which exists only as unlabelled points in Figure 3 and Figures S2 and S3 of the manuscript and has not been published as numbers. A pilot variance from the 80-participant sub-cohort would settle it.

Would count as

Supported → negative-no-association · opposed → positive
Inconclusive if: Stored specimens exist only for the 80-participant Worcester immunogenicity sub-cohort rather than the whole modified intention-to-treat set, which would leave too few cases to compare, or the Day 29 vaccine effect does not reproduce in the stored material.

Known before

• BCG revaccination reduced sustained QFT conversion with a vaccine efficacy of 45.4%, while both BCG and the H4:IC31 candidate were immunogenic and neither prevented initial conversion. (Nemes and colleagues 2018, New England Journal of Medicine, doi:10.1056/NEJMoa1714021)
  So: This is the result the trial was built to confirm, and its sustained conversion rates (6.7% BCG versus 11.6% placebo in 990 adolescents) are the only published numbers that describe the effect a correlate would have had to track.
• Loss of Mtb immunoreactivity, that is reversion, is common, has been systematically dismissed, and most plausibly reflects reduced or absent exposure rather than a fixed lifelong state. (Dale and colleagues 2024, European Respiratory Review, doi:10.1183/16000617.0007-2024)
  So: If reversion tracks exposure rather than immune control, an immune correlate of sustained conversion may not exist to be found, and the design should treat an exposure measure as a competing explanation rather than a nuisance covariate.
• The quest for vaccine-induced immune correlates of protection against tuberculosis was an open programme at the time this trial was designed, by authors who include this trial's immunologists. (Nemes, Fiore-Gartland, Boggiano and colleagues 2022, Vaccine Insights, doi:none given in the manuscript reference list; cited as Vaccine Insights 2022;1(3):165)
  So: Establishes that the correlates objective was deliberate rather than opportunistic, so the absence of a published correlates analysis from this trial is a gap in the record and not a sign that nobody intended one. I could not retrieve this item in Europe PMC.
• In adult Brazilian health-care workers, BCG-Denmark gave no protection against initial QFT conversion, with 34 of 996 converting in the BCG group and 32 of 989 in placebo by 12 months. (dos Santos and colleagues 2024, Lancet Infectious Diseases, doi:10.1016/S1473-3099(23)00818-6)
  So: A second population in which the same vaccine and the same assay produced immune activation without protection, which means a correlate found only in South African adolescents would need external checking before it counts.

Searched: Europe PMC by digital object identifier for the four works above; a Europe PMC title search for a correlate of protection in tuberculosis published 2025 to 2026, which returned nothing; and an author search for the trial's immunologists for a separate correlates paper, which found none. I did not find any published numeric immunogenicity values from this trial, and I did not search clinical trial registries for the sample storage plan.

Feasibility

The South African Tuberculosis Vaccine Initiative at the University of Cape Town, which ran the immunogenicity assays, with the Bill and Melinda Gates Medical Research Institute as the sponsor holding the unblinded outcome data · 9 to 15 months · moderate, because the assays are established and the participants are already enrolled and followed, so the cost is batch flow cytometry and analyst time rather than recruitment
Blockers: Whether specimens were stored for the full modified intention-to-treat set or only the 80-participant sub-cohort is not stated in the manuscript; Ethics approval and participant consent for a new use of stored adolescent specimens; A correlate of risk in a trial with no efficacy is not a correlate of protection, which may make funders unwilling; Two of the listed authors have moved to Pfizer and to Third Rock Ventures, which may complicate access to trial material

Strongest objection

An immunologist would say the design is asking for something the trial cannot give. With a hazard ratio of 1.038 there is no protective effect to correlate with, so even a clean association between a Day 29 immune measure and later sustained conversion would be a correlate of infection risk, not of vaccine-induced protection, and those are different quantities with different uses. The reply is that a correlate of risk still constrains the next vaccine's go or no-go criteria, but the objection stands as stated and should be on the page.

Not derivable from what is published

• Whether specimens exist for the whole modified intention-to-treat set, or only for the 80-participant Worcester sub-cohort
• Numeric immunogenicity values and their variance, which appear only in figures
• The number of control participants available per case after matching on arm, site, age stratum and sex
• Whether exposure to an infectious household contact was recorded, which the competing explanation requires

3. Send it to a lab

The two trials differed in a way the authors noticed and could not explain. Among participants who converted, reversion in the placebo arm was 42% here against 25% in the 2018 trial, while the BCG arms were alike at 44% and 46%. Because the endpoint is conversion that does not revert, a placebo arm that reverts more has fewer events to prevent, and the earlier 45% efficacy could be partly an artefact of that difference. The authors write that whether baseline characteristics or lower exposure intensity contributed remains unknown, so the question is which of a short list of candidates moved the reversion rate.

1. Has anyone done this since?

Status

partly addressed

Works since

• Characterising progression and regression patterns across the spectrum of tuberculosis: a multistate modelling approach (Larsson and colleagues 2026, Lancet Global Health, doi:10.1016/S2214-109X(25)00454-1)
  Applied a Markov multistate model to 1789 household contacts aged 10 and over in Zimbabwe, Tanzania and Mozambique and found movement between states is common, with 13% transitioning from non-infection to infection by month 12 and 85% regressing from the non-infectious disease state by month 12. It establishes that regression is the norm rather than an anomaly, but it is a different population and does not compare the two trials.
• The role of interferon-gamma release assays in tuberculosis vaccine trials: strengths, limitations, and implications beyond trial design (Dagnew, Schmidt, Cinar, Garcia-Basteiro, Noble, Pelzer, Churchyard, Hatherill, White and Cobelens 2026, Vaccine, doi:10.1016/j.vaccine.2026.128822)
  A review of the assay as a trial endpoint written by this trial's own first and last authors with others, so it is the most likely place for an authoritative account of the reversion difference. Europe PMC holds the record with no abstract and no full text, so I can confirm the title, authors and venue but not what it concludes.
• Longitudinal dynamics and treatment-associated determinants of interferon-gamma responses in tuberculosis: a 10-year retrospective QuantiFERON-TB Gold study (2026, Frontiers in Cellular and Infection Microbiology, doi:10.3389/fcimb.2026.1823104)
  Named in the citing-works file as a ten-year retrospective study of interferon gamma response dynamics, which is the right kind of data for reversion determinants. I read the title only and did not retrieve the abstract, so I cannot say whether it stratifies by exposure or by baseline value.

Searched

The 50 citing works by title, picking out the four that concern assay dynamics, reversion and regression; Europe PMC by digital object identifier for the multistate modelling paper and the Vaccine review. I did not find any work that compares reversion between these two trials at the individual level.

Still open

Nobody has put the two trials' individual participant data side by side with harmonised covariates, which is the only way to separate the three candidate explanations: a younger and more varied population, a lower force of infection during the COVID-19 school closures, and the change of assay from QuantiFERON-TB Gold In-tube to Gold Plus.

Registry query

0 of 49 records in the open node mention any of: igra reversion, quantiferon gold plus, force of infection, bcg revaccination, placebo arm, school closure. The open node holds biology and chemistry so far, so this is the expected answer; a private node of a clinical group would be asked the same question.

2. A proposed experiment

Question

Among placebo-arm QFT converters, do age, site, quantitative interferon gamma value at conversion, or calendar period explain the difference in reversion between the 2018 trial (25%) and this trial (42%)?

Hypothesis

The quantitative interferon gamma value at initial conversion, not participant age or geography, accounts for most of the between-trial difference in reversion, because conversions nearer the 0.35 IU/mL threshold revert more often.

Attempt type

secondary-analysis

Design

A pooled individual participant data analysis of the placebo arms of both trials, restricted to participants with an initial QFT conversion and at least two subsequent results. The 2018 trial contributes 49 placebo converters of 310 participants; this trial contributes 101 placebo converters with three consecutive results. Harmonise age, sex, site, calendar date of conversion and the quantitative interferon gamma value at conversion. Model reversion by six months after conversion with logistic regression, entering trial as the exposure and adding candidate explanations one at a time to see how much of the trial coefficient each absorbs. Pre-specify the order: assay generation first, because it is the hardest to separate, then quantitative value at conversion, then age stratum, then site, then calendar period as a proxy for school closure. Repeat the whole analysis restricted to the Worcester site, which both trials shared.

Controls

• reference-case The Worcester-only stratum, where both trials enrolled and where this trial reports a site-specific efficacy similar to its overall estimate, and where 92.7% of the earlier trial was enrolled.
  Rules out: Geography, community transmission intensity and clinic procedure as the explanation, since those are held nearly fixed within this stratum.
• instrument-noise-floor The repeatability of the QuantiFERON assay near its 0.35 IU/mL threshold, taken from the manufacturer's or the laboratory's own replicate data, applied as a band around the threshold.
  Rules out: That reversion is measurement noise in conversions that sat just above the cut-off, which would make the between-trial difference an artefact of where the two assays place borderline samples.
• permutation-null Trial labels shuffled within the Worcester stratum and within age and sex strata, a thousand times, to give the distribution of the trial coefficient under no real difference.
  Rules out: A difference of this size arising from the small number of converters once the data are stratified.
• within-subject Each converter's own sequence of quantitative values across visits, used as a repeated-measures check that participants classified as reverters show a genuine decline rather than one aberrant reading.
  Rules out: Single-visit misclassification of reversion status.

Sensitivity

The quantity to detect is the observed contrast itself, 42% against 25%, about 17 percentage points. The pooled sample is about 150 placebo-arm converters, 101 here and 49 in the earlier trial, so the comparison is near the edge of what that contrast needs, and every added covariate costs more. An exact detectable difference cannot be derived from the published record, because it needs the individual values and the per-site converter counts of the 2018 trial, neither of which is in either paper. What is needed is the two trials' individual participant datasets and the laboratory's replicate data near the assay threshold. Until then, this design should be presented as estimating how much of the difference each candidate absorbs, not as a test with a stated power.

Would count as

Supported → positive · opposed → negative-no-association
Inconclusive if: The individual participant data for the 2018 trial cannot be obtained, or the change of assay generation cannot be separated from calendar period because the two are perfectly confounded by design, in which case no covariate ordering will identify the cause.

Known before

• Reversion of Mtb immunoreactivity is common, has been dismissed rather than disproved, and most plausibly indicates reduced or absent exposure, so immunoreactivity is dynamic in individuals and in populations. (Dale and colleagues 2024, European Respiratory Review, doi:10.1183/16000617.0007-2024)
  So: Makes exposure intensity the leading candidate rather than an afterthought, and supplies the argument for entering calendar period as a proxy for the COVID-19 school closures.
• In the 2018 trial, 49 of 310 placebo participants had an initial QFT conversion and the sustained conversion rate was 11.6%, against 6.7% in the BCG group. (Nemes and colleagues 2018, New England Journal of Medicine, doi:10.1056/NEJMoa1714021)
  So: These are the only published counts available for the comparison arm of the pooled analysis, and they set its size: about 49 converters, which is why the design leans on the Worcester stratum rather than on full adjustment.
• The tuberculin skin test and QuantiFERON differ in how they predict tuberculosis disease in a South African adolescent cohort, in the same Western Cape setting. (Mahomed and colleagues 2011, PLoS One, doi:10.1371/journal.pone.0017984)
  So: Shows that assay choice changes who counts as infected in exactly this population, which supports putting assay generation first in the covariate order rather than treating it as a nuisance.
• In the Brazilian health-care worker trial, QFT Plus conversion occurred in only 66 of 1985 participants over 12 months, with no significant difference by study site. (dos Santos and colleagues 2024, Lancet Infectious Diseases, doi:10.1016/S1473-3099(23)00818-6)
  So: An independent demonstration that conversion rates with the same assay vary by an order of magnitude with population and exposure, which is the effect the design is trying to isolate.

Searched: Europe PMC by digital object identifier for the 2018 trial, the reversion review, the Brazilian nested trial and the Malawi follow-up; the manuscript's own reference list for the adolescent cohort work in the same setting. I looked for a published head-to-head comparison of QuantiFERON-TB Gold In-tube against Gold Plus for reversion specifically and did not find one.

Feasibility

A group holding both datasets, in practice the South African Tuberculosis Vaccine Initiative with the Bill and Melinda Gates Medical Research Institute, or an independent statistical group under a data-sharing agreement · 4 to 8 months · low, because no new specimens, participants or assays are required and the work is statistical
Blockers: Data-sharing agreements across two sponsors and two trial eras; The 2018 trial's consent may not cover pooled reuse; Assay generation and calendar period may be inseparably confounded; Per-site converter counts for the earlier trial are not published

Strongest objection

A trialist would say the comparison is confounded beyond repair. The assay changed, the age range changed, the sites changed and the calendar period changed all at once, and with about 150 converters no covariate ordering will separate four changes that moved together. On that view the design can only describe, not explain, and presenting it as a test of competing explanations oversells it. The design's answer is the Worcester-only stratum, which holds site fixed, but that stratum is small and still carries the assay change.

Not derivable from what is published

• Per-site and per-age converter counts for the 2018 trial
• Quantitative interferon gamma values at conversion for either trial
• Assay repeatability near the 0.35 IU/mL threshold for either assay generation
• Any individual-level measure of Mtb exposure, such as a household contact, in either trial
• The exact dates of school closure relative to each participant's conversion

3. Send it to a lab

Send to a lab

This design is a reanalysis of existing data. There is nothing for a bench to run; it needs the data holder.
No bench job exists: this is a pooled reanalysis of individual participant data from two completed trials, and the only new measurement it might need is the laboratory's existing replicate data near the assay threshold.

This is the authors' own next step, not a new idea. They state plainly that their data do not allow conclusions about prevention of disease, that supporting evidence for it is limited, and that a revaccination trial for disease is under way in India. What they leave unspecified is everything that makes such a trial decidable: the population to enrol, the number of events needed, what establishes that the trial could have detected protection at all, and how to keep an endpoint committee blind when the vaccine leaves a visible scar.

1. Has anyone done this since?

Status

partly addressed

Works since

• Design and feasibility considerations for a phase 3 efficacy trial of the M72/AS01E-4 tuberculosis vaccine (Dagnew and colleagues 2026, Vaccine, doi:10.1016/j.vaccine.2026.128690)
  Simulated the design of a disease-endpoint trial and supplies the numbers this design needs: 110 events rule out a vaccine efficacy confidence bound at or below 10% and 185 events rule out 25%, assuming 90% power and a true efficacy of 55%; 18,000 interferon gamma release assay positive participants give a 90% chance of 110 events in 3.5 to 4 years; and a standalone disease-endpoint evaluation in assay-negative participants would need about 134,800 participants, which the authors call infeasible. That last figure settles the shape of any revaccination disease trial, because this trial's population was assay-negative.
• Efficacy, immunogenicity, and safety of BCG revaccination against Mycobacterium tuberculosis: a systematic review and meta-analysis (Ali and colleagues 2026, Human Vaccines and Immunotherapeutics, doi:10.1080/21645515.2026.2630512)
  Pools six randomised trials and 2400 participants on infection endpoints and concludes the benefit of revaccination in high-risk groups is limited. It covers infection, not disease, so it narrows the prior rather than answering the issue.
• Lessons from another negative tuberculosis vaccine trial (McShane 2026, Lancet Infectious Diseases, doi:10.1016/S1473-3099(26)00412-3)
  A commentary on this class of negative result by a tuberculosis vaccine trialist. Europe PMC holds no abstract, so I can confirm only its existence, authorship and venue.

Searched

The citing-works file by title for anything reporting a disease endpoint; Europe PMC by digital object identifier for the M72 feasibility paper, the revaccination meta-analysis and two commentaries; and Europe PMC title searches for a BCG revaccination trial in India and for revaccination trials published 2024 to 2026. The India disease-endpoint trial has no published results or protocol in Europe PMC that I could find; the only Indian revaccination study I retrieved is a safety report in elderly participants for a COVID-19 indication (doi:10.1016/j.ijtb.2024.05.003). The manuscript itself cites a newspaper report for the India trial, not a registry entry or a protocol.

Still open

Everything the issue asks. No disease-endpoint result for BCG revaccination in adolescents has been published since this trial, and the one trial said to be under way is documented in the record only by a press report.

Registry query

0 of 49 records in the open node mention any of: bcg revaccination, prevention of disease, pulmonary tuberculosis, adolescents, event-driven trial, assay sensitivity control. The open node holds biology and chemistry so far, so this is the expected answer; a private node of a clinical group would be asked the same question.

2. A proposed experiment

Question

Does BCG revaccination of adolescents in a high-incidence setting reduce the incidence of laboratory-confirmed pulmonary tuberculosis disease?

Hypothesis

BCG revaccination reduces laboratory-confirmed pulmonary tuberculosis incidence in adolescents by enough to exclude a vaccine efficacy confidence bound at or below 10%.

Attempt type

clinical-trial

Design

An event-driven, individually randomised, placebo-controlled trial in adolescents and young adults in high-incidence settings, enrolling irrespective of baseline QFT status and stratifying randomisation by it, because the M72 simulations show a disease-endpoint trial restricted to assay-negative participants would need about 134,800 people. Participants receive BCG-Denmark 1331 at 0.1 mL intradermally or intradermal 0.9% saline, as in the Brazilian nested trial. The primary endpoint is laboratory-confirmed pulmonary tuberculosis, adjudicated by a committee blind to allocation, with the analysis triggered at 110 events and a pre-specified interim at a lower count. Enrol a third arm receiving a comparator already shown to move the disease endpoint, so that a null BCG result can be read as absence of effect rather than a trial that could not detect anything. Follow for at least four years, with annual QFT to let infection and disease endpoints be compared in the same people.

Controls

• placebo Intradermal 0.9% saline, the comparator used in the Brazilian nested revaccination trial.
  Rules out: Background tuberculosis incidence and ascertainment effects of being in a trial being read as vaccine effect.
• active-comparator An arm receiving a vaccine with demonstrated efficacy against the disease endpoint, for which the published anchor is M72/AS01E-4 at 50% efficacy against laboratory-confirmed pulmonary tuberculosis in assay-positive adults in a phase 2b trial.
  Rules out: That a null BCG result means the trial could not detect protection in this population and era. This is the assay sensitivity control that neither this trial nor its predecessor had on the disease endpoint.
• blinding-check Participants and site staff are asked to guess allocation at 6 and 12 months, and agreement with the true allocation is reported, because the BCG injection-site lesion unblinded participants in this trial.
  Rules out: Differential care-seeking and differential investigation driving the endpoint once participants work out their arm.
• active-placebo A comparator that would mimic the injection-site reaction and scar without the live mycobacteria. None is available.
  Rules out: Nothing, because it cannot be implemented. Named so that the limitation is on the record rather than silently absent.
  Vocabulary note: Considered and rejected. No inert injection reproduces a BCG scar, so the blind cannot be preserved by matching the local reaction, and the design has to defend the endpoint instead of the blind. This is the clearest vocabulary gap in the set: the kind exists and fits the problem, but nothing fills it.

Sensitivity

The numbers come from the M72/AS01E-4 phase 3 simulations of Dagnew and colleagues 2026: 110 events rule out a two-sided 95% confidence lower bound at or below 10% for vaccine efficacy against disease, and 185 events rule out 25%, both assuming at least 90% power and a true efficacy of 55%; 18,000 assay-positive participants give a 90% probability of accruing 110 events in 3.5 to 4 years at an assumed incidence of 0.4% per year. Carrying those numbers to BCG revaccination assumes a comparable true efficacy, which is the weak point, because the Malawi 30-year trial gives an odds ratio of 0.92 with a confidence interval of 0.80 to 1.05 and so already excludes a large effect. If the true efficacy for revaccination is 20% rather than 55%, the event requirement rises steeply and cannot be read off the published simulations. What is needed is a simulation rerun at efficacy values of 20% to 35% with the incidence of the chosen sites.

Would count as

Supported → positive · opposed → negative-no-effect
Inconclusive if: The active comparator arm also shows no effect, which would mean the trial could not detect protection in this population, or incidence falls so far that the event count is not reached within the funded follow-up, as happened to the infection endpoint here when COVID-19 school closures lowered conversion rates.

Known before

• A second BCG gave no protection against tuberculosis overall in nearly 47,000 people in northern Malawi, with an odds ratio of 0.92 and a confidence interval of 0.80 to 1.05 across 824 cases and up to 30 years of follow-up, although there was weak evidence of modest protection in HIV-negative disease and in those vaccinated as children. (Glynn and colleagues 2021, Lancet Global Health, doi:10.1016/S2214-109X(21)00309-0)
  So: This is the single most constraining piece of prior art: a disease-endpoint revaccination trial has effectively already been run and was null, which bounds the plausible efficacy and so the event count. Any new trial must say why its population or era differs.
• The BCG-REVAC cluster-randomised trial in Brazil reported evidence of an effect of revaccination on tuberculosis incidence in school-aged children. (Barreto and colleagues 2011, Vaccine, doi:10.1016/j.vaccine.2011.05.023)
  So: The main published counterweight to the Malawi result, and the reason age at revaccination belongs in the stratification. I take the claim from the title as cited in the manuscript's reference list and did not retrieve the abstract, so the effect size is not in hand.
• Revaccination may have a positive health impact and be cost-effective in India and South Africa under an assumed 45% prevention-of-infection efficacy, across a range of additional prevention-of-disease efficacy from plus 50% to minus 50%. (Clark, Sumner, Weerasuriya, Bakker, Scriba and White 2024, Journal of Infectious Diseases, doi:10.1093/infdis/jiae089)
  So: Shows that the policy case was built on the infection efficacy this trial has now removed, and that the modelling explicitly admitted the disease efficacy would remain unknown. The modelled range, including the possibility of harm, is the reason a trial rather than an extrapolation is needed.
• Modelling for India projected that M72/AS01E would avert about 40% more cases and deaths than revaccination by 2050, with uncertainty dominated by whether revaccination prevents disease at all. (Clark and colleagues 2023, BMC Medicine, doi:10.1186/s12916-023-02992-7)
  So: Identifies the disease question as the dominant uncertainty in the policy case, which is the argument for funding the trial, and makes the comparator arm choice a natural one.
• In adult Brazilian health-care workers, BCG-Denmark produced no protection against infection, using an intradermal 0.9% saline placebo and a web-based stratified randomisation. (dos Santos and colleagues 2024, Lancet Infectious Diseases, doi:10.1016/S1473-3099(23)00818-6)
  So: Supplies the placebo formulation and the practical precedent for blinding a revaccination trial, and is a second null on the infection endpoint that the trial's own discussion treats as weakening the case.

Searched: Europe PMC by digital object identifier for the Malawi follow-up, the Brazilian nested trial, both modelling papers and the M72 feasibility simulations; Europe PMC abstract search for BCG revaccination with prevention of disease, 2024 to 2026; and Europe PMC title searches for a BCG revaccination trial in India. I did not find a registry entry or protocol for the India trial, and I did not search ClinicalTrials.gov or the Clinical Trials Registry India directly.

Feasibility

A multi-country tuberculosis vaccine trial network with a sponsor at the scale of the Bill and Melinda Gates Medical Research Institute, the Indian Council of Medical Research, or the European and Developing Countries Clinical Trials Partnership · 6 to 9 years from protocol to primary analysis · high, because the event requirement implies tens of thousands of participants followed for at least four years with active disease surveillance
Blockers: A trial under way in India may make a second one unfundable, yet its protocol is not in the published record; The Malawi result bounds the plausible efficacy and so inflates the event requirement, perhaps past affordability; No inert comparator can reproduce the BCG scar, so participant blinding cannot be maintained; An active comparator arm requires access to another sponsor's licensed or investigational product; Incidence is falling in some sites and is sensitive to events like school closures

Strongest objection

A tuberculosis vaccine trialist would say this trial has in effect already been done and was negative. The Malawi follow-up randomised nearly 47,000 people, accrued 824 cases and returned an odds ratio of 0.92 with an upper bound of 1.05, which rules out the kind of effect a new trial would be powered to find. Add two null infection trials and the prior probability of a worthwhile disease effect is low enough that the money belongs with a different candidate. The counter-argument is that Malawi was a different era, population and HIV context, and that the subgroup vaccinated as children showed an odds ratio of 0.77, but that is a subgroup the authors themselves warn against overinterpreting.

Not derivable from what is published

• The protocol, population, endpoint and status of the India revaccination trial
• A plausible true efficacy for revaccination against disease, which the event count depends on
• Site-level tuberculosis incidence for candidate sites today
• Whether a comparator vaccine with demonstrated disease efficacy could be obtained for a third arm
• The effect size from the BCG-REVAC Brazil trial, whose abstract I did not retrieve

3. Send it to a lab

Send to a lab

This design is a study in people. It cannot be sent to a laboratory service; it needs a trial network, ethics approval and a sponsor.
No bench job exists: this is a multi-year randomised trial with a clinical disease endpoint, and its laboratory work is routine diagnostic microbiology at the trial sites rather than a discrete assay a laboratory-automation service could quote on.

This is the assumption the whole endpoint rests on. The paper states the interpretation openly: reversion from positive to negative could indicate successful immune control or clearance, which would make sustained conversion the thing worth preventing. It then notes that because no more than 5 to 10% of infected people progress, efficacy against sustained infection may not predict efficacy against disease. Nobody has shown that participants whose conversion persists go on to tuberculosis more often than participants who revert. Until that is shown, two trials have been run against an endpoint of unknown meaning.

1. Has anyone done this since?

Status

partly addressed

Works since

• Characterising progression and regression patterns across the spectrum of tuberculosis: a multistate modelling approach (Larsson and colleagues 2026, Lancet Global Health, doi:10.1016/S2214-109X(25)00454-1)
  The closest thing to an answer. In 1789 household contacts followed six-monthly for 12 to 24 months, transition probabilities by month 12 were 13% from non-infection to infection and 3% from infection to non-infectious disease, and 85% of those in the non-infectious disease state regressed. It quantifies movement between states using the International Consensus for Early TB framework, but its 34 infectious disease events are too few to compare progression between people who sustain and people who lose immunoreactivity.
• Antigen-specific Th1 cytokine markers and protection against tuberculosis: a systematic review and meta-analysis stratified by progression to active disease and sustained IGRA conversion (Lin, Liu and Pan 2026, Frontiers in Cellular and Infection Microbiology, doi:10.3389/fcimb.2026.1780600)
  Treats progression to disease and sustained assay conversion as two separate endpoints and finds immune markers predict neither, with pooled odds ratios of 0.97 to 1.11 for disease across six studies. By analysing them separately it documents the gap between the two endpoints rather than bridging it.
• The role of interferon-gamma release assays in tuberculosis vaccine trials: strengths, limitations, and implications beyond trial design (Dagnew, Schmidt, Cinar, Garcia-Basteiro, Noble, Pelzer, Churchyard, Hatherill, White and Cobelens 2026, Vaccine, doi:10.1016/j.vaccine.2026.128822)
  Written by this trial's first and last authors with others and aimed squarely at the limitations of the assay as a trial endpoint, so it is the field's considered response to this issue. Europe PMC holds no abstract or full text, so I can confirm the title, authors, journal and year but not its conclusions.
• Adult immunization initiative: which surrogate markers should be used for vaccine efficacy? (2026, Indian Journal of Medical Research, doi:10.25259/ijmr_780_2026)
  Named in the citing-works file and addressed to surrogate marker choice, which is this issue stated generally. Title only; I did not retrieve the abstract.

Searched

The citing-works file by title for anything on surrogate endpoints, reversion or progression; Europe PMC by digital object identifier for the multistate modelling paper, the Th1 marker meta-analysis and the assay review. I did not find any study that reports tuberculosis incidence separately for sustained converters and for reverters.

Still open

The direct comparison. No published analysis gives the rate of progression to tuberculosis among people whose QFT conversion persisted against those who reverted, which is the single number that would make prevention of sustained infection a defensible surrogate or retire it.

Registry query

0 of 49 records in the open node mention any of: sustained igra conversion, igra reversion, surrogate endpoint, progression to tuberculosis, quantiferon threshold, prevention of infection. The open node holds biology and chemistry so far, so this is the expected answer; a private node of a clinical group would be asked the same question.

2. A proposed experiment

Question

Among people who convert on an interferon gamma release assay, do those whose conversion is sustained progress to tuberculosis disease more often than those who revert?

Hypothesis

Sustained QFT conversion carries a higher rate of progression to laboratory-confirmed tuberculosis than conversion followed by reversion, by a margin large enough to justify using it as a trial endpoint.

Attempt type

secondary-analysis

Design

A pooled individual participant analysis of existing cohorts and trial populations that have serial interferon gamma release assay results and at least two years of disease follow-up in the same people. Candidate contributors include the two revaccination trials, the adolescent cohort studies in the Western Cape, and the ERASE-TB household contact cohort in Zimbabwe, Tanzania and Mozambique. Classify every converter as sustained or reverted using this trial's own definition, two further positives at 3 and 6 months, and compare incidence of laboratory-confirmed tuberculosis between the two groups with a Cox model stratified by cohort, entering time from conversion as the clock and the quantitative value at conversion as a covariate. Pre-specify a persistently assay-negative stratum as the lower reference. Report the absolute risks, not only the hazard ratio, because the surrogate's usefulness depends on the difference being large.

Controls

• known-positive-task Recovery of the already-published association between a higher quantitative interferon gamma value at conversion and a higher risk of progression, which this trial cites as established.
  Rules out: That a null comparison means the pooled dataset is too noisy or too small to detect any progression signal at all, rather than that reversion carries no information.
• permutation-null Sustained and reverted labels shuffled within cohort and within quantitative value band, a thousand times.
  Rules out: A hazard ratio appearing meaningful when it is driven by a handful of cases in one cohort.
• instrument-noise-floor Assay repeatability near the 0.35 IU/mL threshold, used to flag conversions and reversions that sit inside the measurement band.
  Rules out: That the reverter group is mostly measurement noise around the cut-off, which would make the comparison a test of assay precision rather than of biology.
• within-subject The full quantitative trajectory for each participant, used to confirm that reverters show a sustained decline across visits rather than one low reading.
  Rules out: Misclassification of reversion from a single aberrant result.

Sensitivity

This cannot be derived from the published record, and the binding constraint is disease events rather than participants. The anchors available are this trial's 6 laboratory-confirmed cases in 1835 participants over a median 30 months, ERASE-TB's 34 infectious disease events in 1789 contacts, and the statement in this paper's introduction that no more than 5 to 10% of infected people progress. Those together imply that detecting even a twofold difference in progression between sustained converters and reverters needs disease events from several large cohorts pooled. What is needed before this design can carry a sensitivity statement is the per-cohort count of laboratory-confirmed cases, broken down by conversion pattern, from each candidate contributor. That number is not in any of the papers read here.

Would count as

Supported → positive · opposed → negative-no-association
Inconclusive if: The pooled disease events number in the tens rather than the hundreds, or contributing cohorts used different assay generations and conversion definitions that cannot be harmonised, or the known-positive association between quantitative value and progression does not reproduce in the pooled data.

Known before

• Reversion is common and most plausibly reflects reduced or absent exposure rather than immune control, and neglecting it has skewed infection risk estimation and transmission modelling. (Dale and colleagues 2024, European Respiratory Review, doi:10.1183/16000617.0007-2024)
  So: Supplies the competing hypothesis the design must distinguish: if reversion tracks exposure, then sustained conversion marks continued exposure rather than established infection, and preventing it is not the same as preventing anything inside the host.
• The paper's own framing, that reversion could indicate successful immune control or clearance, and that efficacy against sustained infection may not predict efficacy against disease because no more than 5 to 10% of infected individuals progress. (Schmidt and colleagues 2025, New England Journal of Medicine, Introduction, doi:10.1056/NEJMoa2412381)
  So: Fixes the question in the authors' own terms and gives the 5 to 10% progression figure that any event-count calculation starts from.
• The tuberculin skin test and QuantiFERON differ in predicting tuberculosis disease in a South African adolescent cohort. (Mahomed and colleagues 2011, PLoS One, doi:10.1371/journal.pone.0017984)
  So: Demonstrates both that the needed design has been run in this setting before, with serial testing and disease follow-up, and that the choice of test changes who is classified as infected, so harmonisation is the first analytic task.
• In the 2018 trial, the rate of sustained conversion, described by its authors as possibly reflecting sustained Mtb infection, was reduced by revaccination while initial conversion was not. (Nemes and colleagues 2018, New England Journal of Medicine, doi:10.1056/NEJMoa1714021)
  So: Shows the surrogate assumption was hedged from the beginning, in the paper that created the endpoint, which is why settling it changes the interpretation of both trials rather than only the later one.

Searched: Europe PMC by digital object identifier for the reversion review, the 2018 trial and the multistate modelling cohort, plus the manuscript's reference list for the adolescent cohort work and for the earlier reports linking higher interferon gamma values to progression risk. I searched the citing-works file for any study reporting progression stratified by conversion pattern and found none.

Feasibility

A consortium of cohort holders, for example the South African Tuberculosis Vaccine Initiative with the ERASE-TB investigators, or a modelling group with data-sharing agreements across adolescent and household contact cohorts · 9 to 18 months once data-sharing agreements exist · low to moderate, because the analysis is statistical and the main cost is harmonising datasets and negotiating access, not new measurement
Blockers: Per-cohort disease case counts stratified by conversion pattern are unpublished, so the pooled event count is unknown before access is granted; Cohorts used different assay generations, conversion definitions and testing intervals; Disease follow-up in vaccine trials is often shorter than progression needs; Data-sharing agreements across several sponsors and countries

Strongest objection

An epidemiologist would say the comparison is confounded by exposure in a way no covariate fixes. If reversion marks the end of exposure and sustained positivity marks continuing exposure, then sustained converters will progress more often whatever happens inside the host, and the analysis will return a hazard ratio that looks like a validated surrogate while measuring force of infection. That would be worse than a null, because it would license continued use of an endpoint on the wrong grounds. Any honest version of this design needs an exposure measure, and the cohorts that have one are the household contact cohorts rather than the adolescent trials.

Not derivable from what is published

• Per-cohort laboratory-confirmed case counts broken down by sustained conversion against reversion
• Whether the contributing cohorts' disease follow-up extends far enough past conversion
• Assay repeatability near the threshold for each assay generation used
• An individual-level exposure measure in the adolescent trial populations
• Whether consent in each cohort permits pooled reuse for this question

3. Send it to a lab

Send to a lab

This design is a reanalysis of existing data. There is nothing for a bench to run; it needs the data holder.
No bench job exists: this is a pooled reanalysis of existing serial assay results and disease outcomes. The only measurement it would commission is assay repeatability near the threshold, and that data most likely already exists in the contributing laboratories.