A BCG replacement vaccine missed its target in 6,950 African newborns

The limits of BCG are well documented: it protects children against severe forms of tuberculosis, but its efficacy against pulmonary tuberculosis in adolescents and adults is highly variable, and it offers limited and inconsistent protection against infection and transmission [s1]. VPM1002, a recombinant BCG, was built to do better. A phase 3 trial across sub-Saharan Africa has now reported that it did not — and the more interesting part of the result is why the trial cannot fully say so.

The design

The trial was double-blind, randomised, active-controlled and designed as a non-inferiority comparison, run at ten main sites and four satellite sites across sub-Saharan Africa, in settings ranging from rural to urban [s1]. Healthy newborns aged 0–14 days weighing at least 2·3 kg were randomly assigned 1:1 to a single 0·05 mL intradermal dose of either VPM1002 or BCG [s1]. Randomisation was stratified by maternal HIV status at a 1:10 ratio [s1].

The primary endpoint was not tuberculosis disease. It was infection, defined as incident conversion on the QuantiFERON-TB Gold Plus interferon-γ release assay, assessed every six months from month six [s1]. Non-inferiority required the upper bound of the 95% confidence interval for the hazard ratio to fall below 1·25 [s1].

What happened

Between 9 November 2020 and 21 June 2022, 6,950 infants were enrolled [s1]. The trial was terminated early, in October 2024, because the rate of QuantiFERON conversion was lower than expected [s1]. After ten withdrawals, 6,940 infants were randomly assigned — 720 born to mothers living with HIV and 6,220 HIV-unexposed — and 6,897 were vaccinated, 3,452 with VPM1002 and 3,445 with BCG [s1]. Median follow-up was 35 months (IQR 30–38) [s1].

QuantiFERON conversion occurred in 184 of 3,471 infants (5·3%) in the VPM1002 group and 150 of 3,469 (4·3%) in the BCG group [s1]. The hazard ratio was 1·23 (95% CI 0·99–1·53) [s1]. The point estimate sits just under the 1·25 non-inferiority margin; the upper bound sits far above it. VPM1002 was therefore not shown to be non-inferior to BCG [s1].

Safety was the one clean result. Adverse event profiles, including serious adverse events and deaths, were similar between the two groups, and no vaccine-related serious adverse events were reported [s1].

Why the trial says "uncertain" rather than "worse"

The authors are unusually explicit that their own trial does not have the statistical certainty to definitively compare the two vaccines [s1]. Three specific problems are named.

First, the event count. The trial was powered on 632 accrued cases of M. tuberculosis infection. It accrued 334, despite the follow-up window being extended from 36 months up to 48 months precisely to catch more [s1]. A non-inferiority trial that reaches roughly half its planned events produces a confidence interval too wide to exclude either conclusion.

Second, the endpoint itself. The authors report discordance between the QuantiFERON surrogate and confirmed tuberculosis endpoints [s1] — meaning the marker used to define "infection" did not track neatly with the disease the vaccine is meant to prevent.

Third, and most awkwardly, exposure was not balanced. Household tuberculosis exposure was approximately 33·0% higher in the VPM1002 group [s1]. Randomisation is supposed to prevent exactly this, and with 6,940 infants it usually would; that it did not is a caution about how much weight the head-to-head comparison can bear. When the analysts tried to adjust for exposure as a covariate, the proportional hazards assumption was violated — the effect of tuberculosis exposure on QuantiFERON conversion was not constant over time — which precluded reliable hazard ratio estimation without penalised modelling [s1].

The authors' own framing is that these findings highlight the challenges of using IGRA-defined infection endpoints in infant vaccine trials [s1]. That is a methods conclusion as much as a product one.

What it means for the pipeline

A negative or indeterminate phase 3 result matters more in tuberculosis than it would in most fields, because there is so little else in late-stage development. A five-year review of the global tuberculosis vaccine roadmap published earlier this year noted that while the pipeline has diversified and several candidates are in phase 3, the number of candidates in active clinical trials remains small and product development challenges persist [s2]. That review also flagged that manufacturer engagement is hampered by unclear demand and insufficient committed procurement, leaving investment in tuberculosis vaccine research and development risky and inadequate [s2].

Against that backdrop, the practical readings of this trial are narrow. VPM1002 has not displaced BCG as the newborn standard, and on this evidence there is no basis for changing what newborns receive. The vaccine's safety profile in newborns looked comparable to BCG's [s1], which is not nothing for a recombinant product being given in the first two weeks of life. And the field now has a concrete, expensively acquired lesson about endpoint choice: a trial designed around interferon-γ release assay conversion in infants may run for four years and still not answer the question it asked.

The trial is registered as NCT04351685 and PACTR202007868402718 and is complete [s1]. It was funded through the EDCTP2 programme, supported by the European Union and co-funded by DLR and the European Investment Bank [s1].

Sources

• Comparison of VPM1002 with BCG in the prevention of tuberculosis in newborn infants: a multicentre, double-blind, randomised, phase 3, non-inferiority trial, The Lancet Infectious Diseases, 18 August 2026
• Accelerating research and development of new vaccines against tuberculosis: 5-year progress on the global roadmap, The Lancet Infectious Diseases, 18 March 2026