Effectiveness of BCG Vaccination Against Mycobacterium tuberculosis Infection in Adults: A Cross-sectional Analysis of a UK-Based Cohort

Abstract

Background

BCG appears to reduce acquisition of Mycobacterium tuberculosis infection in children, measured using interferon-gamma release assays (IGRAs). We explored whether BCG vaccination continues to be associated with decreased prevalence of M. tuberculosis infection in adults.

Methods

We conducted a cross-sectional analysis of data from adult contacts of tuberculosis cases participating in a UK cohort study. Vaccine effectiveness (VE) of BCG, ascertained based on presence of a scar or vaccination history, against latent tuberculosis infection (LTBI), measured via IGRA, was assessed using multivariable logistic regression. The effects of age at BCG and time since vaccination were also explored.

Results

Of 3453 recent tuberculosis contacts, 27.5% had LTBI. There was strong evidence of an association between BCG and LTBI (adjusted odds ratio = 0.70; 95% confidence interval, .56–.87; P = .0017) yielding a VE of 30%. VE declined with time since vaccination but there was evidence that LTBI prevalence was lower amongst vaccinated individuals even >20 years after vaccination, compared with nonvaccinated participants.

Conclusions

BCG is associated with lower prevalence of LTBI in adult contacts of tuberculosis. These results contribute to growing evidence that suggests BCG may protect against M. tuberculosis infection as well as disease. This has implications for immunization programs, vaccine development, and tuberculosis control efforts worldwide.

Clinical trials registration

NCT01162265.

BCG, Bacillus Calmette-Guérin vaccine, tuberculosis, vaccine effectiveness

Tuberculosis is the leading cause of death from infectious disease worldwide [1]. The widely used Bacillus Calmette-Guérin (BCG) vaccine is the only licensed vaccine against Mycobacterium tuberculosis [2]. Infant BCG vaccination has shown consistently high efficacy of 70%–80% against childhood tuberculosis, namely meningitis and miliary tuberculosis [3]. The protective effect of BCG against adult pulmonary disease varies geographically [4–6], which may in part be associated with varying exposure to M. tuberculosis or environmental mycobacteria [7, 8] that may mask or block protection induced by BCG [4, 9–11].

Until recently it was not possible to determine if BCG vaccination prevents acquisition of M. tuberculosis infection or only limits progression from latent tuberculosis infection (LTBI) to active disease. This was due to limitations of the tuberculin skin test (TST), which can be positive following BCG vaccination and, to a lesser extent, exposure to environmental mycobacteria [12, 13]. More recently developed interferon-gamma release assays (IGRAs) are more specific for M. tuberculosis infection [14], as they measure interferon-gamma release from T lymphocytes stimulated with antigens not present in BCG or most environmental mycobacteria [15]. Since the protective effect of BCG against M. tuberculosis infection was first demonstrated [16], several observational studies have investigated this phenomenon. A metaanalysis of 14 observational studies in children aged <16 years with recent exposure to active tuberculosis found a vaccine effectiveness (VE) of 19% against M. tuberculosis infection. In the 6 studies that followed up children who were IGRA-positive, BCG was associated with a 58% reduction in the risk of active tuberculosis [17].

Data on whether this protection against M. tuberculosis infection continues into adulthood, as it appears to against disease [18, 19], are limited. Several observational studies of adults have found weak evidence of protection by BCG against infection measured via IGRA [20–25]. There has been no investigation of the influence of age at vaccination on VE of BCG against LTBI in adults, or of changes in protection over time. These issues are important as, in order to assess the role of BCG (or similar vaccines in development) in tuberculosis control programs, the effect of BCG on the full spectrum of tuberculosis should be understood [26]. In this cross-sectional study, we used baseline data from a large United Kingdom (UK) cohort study of adults at risk of M. tuberculosis infection to determine the presence and durability of BCG protection against M. tuberculosis infection, assessed by IGRA, and the potential factors that influence VE.

METHODS

Study Design

A cross-sectional study was carried out amongst participants recruited to the UK PREDICT study who were recent contacts of patients with active tuberculosis (contacts). PREDICT was a prospective cohort study that aimed to assess the prognostic value of IGRAs in predicting the development of active tuberculosis among individuals with, or at risk of, LTBI, as previously described [27]. A total of 10 045 tuberculosis contacts, as well as recent migrants, aged ≥16 years were recruited between January 2011 and July 2015. Participants with evidence of active tuberculosis at baseline were excluded and migrants were excluded from this analysis as it was unclear when their primary infection occurred. Contacts were recruited at contact tracing appointments for screening in tuberculosis clinics (part of the routine public health management of tuberculosis in the UK).

After obtaining informed consent, study nurses completed a questionnaire with participants, including demographic, social, medical, and tuberculosis exposure history, took blood samples for IGRA testing, and administered a Mantoux TST (which was read 48–72 hours later).

Most recruitment sites were in London, with 1 clinic each in Birmingham and Leicester. To increase the probability that any detected LTBI was due to recent infection, participants who reported previous contact with active tuberculosis (prior to that resulting in recruitment) were excluded. Participants who reported a previous tuberculosis diagnosis were also excluded from this analysis, as a positive IGRA may reflect their prior tuberculosis disease rather than current LTBI.

The study was approved by the London School of Hygiene and Tropical Medicine MSc Research Ethics Committee (reference number 13431). Ethics approval for the PREDICT study was granted through the National Research Ethics Service (Brent Medical Ethics Committee, approval number 10/H0717/14) and is registered on clinicaltrials.gov (NCT01162265).

Primary Exposure, Outcome, and Covariates

The primary exposure of interest was previous BCG vaccination. Vaccination status was determined by inspection of both arms for a vaccination scar by a trained study or tuberculosis nurse, combined with a vaccination history and documentation from a personally held vaccination record (eg, the “red book”) if available. A positive response was recorded when a scar was observed or there was documented evidence of BCG administration, and a negative response recorded if no scar was seen. If participants thought they had been vaccinated but no scar was observed, the interviewer recorded “unsure” and this was treated as missing. Participants who had received vaccination were asked the year of vaccination.

The outcome of interest was LTBI, measured via IGRA. LTBI was defined as a positive result on either or both of QuantiFERON-TB Gold In-Tube (QFT-GIT) and T-SPOT.TB. Those who were negative on both assays, or negative on 1 and indeterminate on the other, were considered not to have LTBI. Most participants had both IGRAs performed; however, a small number were only tested with QFT-GIT and LTBI was determined based on the single assay. If both assays were indeterminate, the outcome was considered missing.

Self-reported questionnaire data included details of the tuberculosis exposure, country of birth, ethnicity (based on the enhanced tuberculosis surveillance system categories [28]), smoking status (never or ever), social risk factors (none, or any of homelessness, imprisonment, or use of controlled drugs), and medical details, including whether the participant had diabetes, human immunodeficiency virus (HIV), or other immunosuppression.

Age at vaccination was calculated as the difference between vaccination and birth year, and dichotomized based on the ages of vaccination recommended in typical BCG vaccination polices (≤2 years to reflect infant vaccination, >2 years for vaccination in childhood and at older ages). Time since vaccination was calculated as age at recruitment minus the stated age at vaccination and grouped into 3 categories to avoid data sparsity (≤10 years, 11–20 years, and >20 years) [5]. Tuberculosis incidence in country of birth was obtained from World health Organization (WHO) estimates for the year 2000 [29]. Absolute latitude of country of birth, found to affect VE in previous studies [9], was calculated using average country latitude data from Google Public Data Explorer [30], and collapsed into 20° groups [9, 17].

Statistical Analysis

To investigate the association between LTBI and BCG status (and all other covariates), cross-tabulation, unadjusted ORs, and likelihood ratio tests (LRTs) were calculated. For variables with no natural reference group, the largest group was used as the baseline.

Multivariable analysis was performed using logistic regression and LRTs. Age group and sex were considered a priori confounders. Age group was treated as a categorical, rather than continuous, variable as this produced a better fit in bivariable analysis. Further confounding variables were included in the multivariable model based on the change-in-estimate method. All variables that were associated with BCG status or LTBI on bivariable analysis with P ≤ .2 (with none deemed to lie on the causal pathway) were added separately to the model. Any variable that resulted in a ≥10% relative change in the odds ratio (OR) was kept in the model (with the variable that caused the largest change in the OR selected first if more than 1 variable caused a ≥10% change). All other variables were then individually re-added to the model until the addition of no further variables resulted in a 10% change in OR. All analyses used a complete-case approach. VE for LTBI was calculated as VE = 1 − OR. This OR is the prevalence OR [31], which approximates the incidence rate ratio in cross-sectional studies [32, 33].

The roles of age at vaccination and time since vaccination on the association between BCG and LTBI were then explored. Two further multivariable logistic regression models were fitted separately (for age at BCG and time since BCG), using the same modelling strategy as above. The analysis of age at BCG was stratified by place of birth (UK or non-UK). Due to collinearity between age at vaccination and time since vaccination, analysis of time since BCG was stratified by age at BCG (≤2 years or >2 years).

Six sensitivity analyses of the primary multivariable model were performed: (1) with additional adjustment for smoking status and any social risk factor; (2) including adjustment for tuberculosis incidence of country of birth; (3) including only participants with concordant IGRAs; (4) using binomial regression with a log-link to directly estimate the prevalence ratio (PR), which may not be well approximated by the OR as the outcome was common [34]; (5) restricting the analysis to household contacts of active tuberculosis cases, who likely had the most defined single tuberculosis exposure [35]; and (6) including participants who reported earlier contact with people with active tuberculosis. We also performed additional analyses using TST results to measure LTBI. Two cutoff criteria were used to define positivity: firstly, TST ≥5 mm [36] and, secondly, TST ≥5 mm in BCG-naive or ≥15 mm in BCG-vaccinated participants (which better predicts disease progression [27]).

To examine bias from missing data, observations with missing values were investigated for association with the outcome and exposure using cross-tabulation, together with multivariable analysis of factors associated with missing BCG status.

Analysis was conducted using Stata v15.0.

RESULTS

Participant Inclusion

Of 10 045 participants enrolled in the PREDICT study, 9515 had no evidence of active tuberculosis and did not report a prior history of tuberculosis. Of the total enrolled, 4310 participants were recent tuberculosis contacts, of whom 857 had missing data on LTBI and/or BCG status, leaving 3453 participants in this cross-sectional study (Figure 1).

Figure 1.

Recruitment of participants to the PREDICT study and inclusion in the cross-sectional study. *Contacts were defined as people with a cumulative duration of exposure to the index case (pulmonary or extrapulmonary tuberculosis) of >8 hours in a confined space during the period of infectiousness. Abbreviations: BCG, Bacillus Calmette-Guérin; LTBI, latent tuberculosis infection; TB, tuberculosis.

Baseline Characteristics and Bivariable Analysis

Of 3453 participants, 86.9% (3000/3453) had received BCG vaccination. The median age was 32 years (interquartile range 25–43), 50.0% were male, and 2420/3444 (70.3%) were born outside the UK (Table 1).

Table 1.

Cross-sectional Study Baseline Characteristics and Unadjusted ORs of Their Association with LTBI (n = 3453)

Variable	Group	No. Positive for LTBI	Percentage Positive for LTBI	Crude OR	95% CI	P Valuea
BCG vaccination	No	142/453	31.4	1		.054
	Yes	809/3000	27.0	0.81	.65–1.00
Sex (m = 16)	Male	538/1726	31.2	1
	Female	410/1711	24.0	0.70	.60–.81	<.001
Age group, y (m = 1)	16–25	213/965	22.1	1		<.001
	26–35	328/1175	27.9	1.37	1.12–1.67
	36–45	195/601	32.5	1.70	1.35–2.13
	>45	215/711	30.2	1.53	.24–.33
Country of birth (m = 9)	UK	155/1024	15.1	1		<.001
	Non-UK	790/2420	32.6	2.72	2.25–3.29
Ethnicity (m = 76)	Indian	292/967	30.2	1		<.001
	White	137/709	19.3	0.55	.44–.70
	Black African	190/536	35.5	1.27	1.01–1.59
	Mixed	145/474	30.6	1.02	.80–1.29
	Pakistani	71/245	29.0	0.94	.69–1.28
	Bangladeshi	22/142	15.5	0.43	.26–.68
	Black Caribbean	31/162	19.1	0.55	.36–.83
	Black other/ Chinese/other	35/142	27.3	0.76	.50–1.13
Any social risk factor	No	897/3227	27.8	1		.20
	Yes	54/226	23.9	0.82	.59–1.12
HIV (m = 111)	No	913/3317	27.5	1		.96
	Yes	7/25	28.0	1.02	.43–2.50
Other immunosuppression (m = 6)	No	928/3338	27.8	1		.02
	Yes	20/110	18.2	0.58	.35–.94
Diabetes (m = 6)	No	869/3236	26.9	1		.0007
	Yes	80/211	37.9	1.66	1.25–2.22
Smoking status (m = 16)	Ever	207/822	25.2	1		.078
	Never	741/2615	28.4	1.17	.98–1.41
Latitude of country of birth (m = 37)	<20°	265/779	34.0	1		<.001
	20–40°	455/1378	33.0	0.96	.79–1.15
	>40°	221/1259	17.6	0.41	.34–.51
TB incidence per 100 000 of country of birth (m = 46)	≤10	26/115	22.6	1		<.001
	11–40	200/1252	16.0	0.65	.41–1.03
	41–100	36/129	28.0	1.33	.74–2.37
	101–150	15/69	22.1	0.97	.47–1.99
	151–300	584/1621	36.0	1.93	1.23–3.02
	≥300	77/222	34.7	1.82	1.08–3.05
Setting of TB exposure (m = 274)	Household	671/2149	31.2	1		<.001
	Non-household	204/1030	19.8	0.54	.46–.65

Abbreviations: BCG, Bacillus Calmette-Guérin; CI, confidence interval; HIV, human immunodeficiency virus; LTBI, latent tuberculosis infection; n, number; m, missing; OR, odds ratio; TB, tuberculosis.

^aDerived from likelihood ratio test.

The overall prevalence of LTBI was 27.5% (951/3453), with 27.0% (809/3000) in those vaccinated and 31.4% (142/453) in the unvaccinated, yielding an unadjusted OR of 0.81 (95% confidence interval [CI], .65–1.00; P = .054; Table 1). Characteristics associated with increased LTBI prevalence on univariate analysis were male sex, older age, being born outside the UK, some ethnicities, lack of immunosuppression, having diabetes, lower latitudes of country of birth, tuberculosis incidence in country of birth, and household tuberculosis exposure.

Multivariable Analysis

Of the participants, 3399 (98.4%) were included in the complete-case analysis. Apart from sex and age (included a priori), latitude of country of birth was the key confounding variable, shifting the OR away from the null, and the only covariate retained in the final model. After adjusting for sex, age group, and latitude of country of birth, there was strong evidence of an association between BCG and LTBI (OR, 0.70; 95% CI, .56–.87; P = .0017; Table 2), thus the VE of BCG for LTBI in adult contacts of tuberculosis cases was 30% (95% CI, 13%–44%).

Table 2.

Adjusted ORs for the Association of LTBI and BCG Vaccination From a Multivariable Logistic Regression Model (n = 3399)

Variable	Group	Adjusted ORa	95% CI	P Valueb
BCG vaccination	No	1		.0017
	Yes	0.70	.56–.87
Sex	Male	1		<.001
	Female	0.70	.60–.82
Age group, y	16–25	1		.0016
	26–35	1.31	1.07–1.61
	36–45	1.55	1.23–1.97
	>45	1.38	1.10–1.74
Latitude of country of birth	<20°	1		<.001
	20–40°	0.92	.76–1.12
	>40°	0.42	.34–.52

Abbreviations: BCG, Bacillus Calmette-Guérin; CI, confidence interval; LTBI, latent tuberculosis infection; OR, odds ratio.

^aAdjusted for all other variables in the table.

^bLikelihood ratio test P value.

Sensitivity analyses produced largely similar results (Supplementary Table 1). Additional analyses using TST results found that BCG was positively associated with LTBI using a 5 mm TST threshold, and negatively associated using a stratified threshold based on BCG status (OR, 0.50; 95% CI, .40–.63; P < .001; Supplementary Table 1).

There was no evidence of an association between missing LTBI data and BCG vaccination (OR, 1.05; 95% CI, .74–1.49; P = .79). However, those without data on BCG status were more likely to have LTBI (OR, 1.33; 95% CI, 1.09–1.63; Supplementary Table 2).

Vaccine Effectiveness by Age at Vaccination and Time Since Vaccination

Data on reported age at (and time since) vaccination were available for 2195 (73%) of 3000 BCG-vaccinated participants. Of the 641 people born in the UK, 497 (78%) were vaccinated aged >2 years. Of the 1554 people born outside the UK, 1268 (82%) were vaccinated aged ≤2 years (Supplementary Table 3). Protection against infection was observed following both infant and older age vaccination, among those born within and outside the UK, after adjusting for age group, sex, and latitude of country of birth (Table 3). There was no clear pattern by age at vaccination, as the CIs for the adjusted ORs overlapped.

Table 3.

Crude and Adjusted ORs for the Association of LTBI and Age at BCG Vaccination, by Place of Birth

Birthplace	Age at BCG	Proportion With LTBI (%)	Crude OR (95% CI)	P Valuea	Adjusted ORb (95% CI)	P Valuea
UK (n = 792)	No BCG	32/151 (21.2)	1		1
	≤2 y	14/144 (9.7)	0.40 (.20–.79)		0.39 (.19–.77)
	>2 y	72/497 (14.5)	0.63 (.40–1.00)	.021	0.62 (.38–1.00)	.018
Non-UK (n = 1854)	No BCG	110/300 (36.7)	1		1
	≤2 y	434/1268 (34.2)	0.90 (.69–1.17)		0.83 (.64–1.09)
	>2 y	83/286 (29.0)	0.71 (.50–1.00)	.123	0.63 (.44–.90)	.040

Abbreviations: BCG, Bacillus Calmette-Guérin; CI, confidence interval; LTBI, latent tuberculosis infection; OR, odds ratio; y, years.

^aLikelihood ratio test P value.

^bAdjusted for age group, sex, and latitude of country of birth.

The association between BCG and LTBI, adjusted for age group, sex, ethnicity, and country of birth, appeared to vary with time since vaccination in both age at vaccination strata (Table 4). In those vaccinated at age 2 or younger, the protective association was greater in those vaccinated 11–20 years ago (OR, 0.55; 95% CI, .33–.90) than those vaccinated >20 years ago (OR, 0.78; 0.60–1.00). As this was an adult cohort, there were no participants vaccinated ≤10 years ago in the group vaccinated in infancy. In participants vaccinated at age >2 years, the protective association was greatest in those vaccinated ≤10 years ago (OR, 0.31; 95% CI, .15–.63), versus those vaccinated 11–20 and >20 years ago (OR, 0.73; 95% CI, .48–1.10, and OR, 0.67; 95% CI, .46–.98, respectively). However, there appeared to be evidence of protection more than 20 years later in both those vaccinated in infancy and those vaccinated in childhood and older ages, compared to those without BCG vaccination.

Table 4.

Crude and Adjusted ORs for the Association of LTBI and Years Since BCG Vaccination, by Age at Vaccination

Age at BCG	Years Since BCG	Proportion With LTBI (%)	Crude OR (95% CI)	P Valuea	Adjusted ORb (95% CI)	P Valuea
≤2 y (n = 1867)	No BCG	142/453 (31.4)	1		1
	≤10	…	…		…
	11–20	27/160 (16.9)	0.44 (.28–.70)		0.55 (.33–.90)
	>20	423/1254 (33.7)	1.11 (.89–1.40)	<.001	0.78 (.60–1.00)	.026
>2 y (n = 1236)	No BCG	142/453 (31.4)	1		1
	≤10	12/106 (11.3)	0.28 (.15–.53)		0.31 (.15–.63)
	11–20	51/243 (21.1)	0.58 (.40–.84)		0.73 (.48–1.10)
	>20	92/434 (21.2)	0.59 (.43–.80)	<.001	0.67 (.46–.98)	.0014

Abbreviations: BCG, Bacillus Calmette-Guérin; CI, confidence interval; LTBI, latent tuberculosis infection; OR, odds ratio; y, years.

^a Likelihood ratio test P value.

^bAdjusted for age group, sex, ethnicity, and country of birth (UK or non-UK).

Discussion

Principal Findings

In this cross-sectional study, we found strong evidence of an association between BCG and LTBI in recent adult contacts of tuberculosis, with VE estimated as 30% (95% CI, 13%–44%). Protection against infection was seen following both infant and older age vaccination, among participants born within and outside the UK. The association appeared to differ by time since vaccination, in participants vaccinated at both ≤2 and >2 years of age. In those vaccinated aged ≤2 years, protection was seen in those vaccinated 11–20 years ago and, though less so, vaccinated >20 years ago. Among participants vaccinated at ages >2 years, the protective association was greater for those vaccinated <10 years ago than those vaccinated 10–20 years ago, where evidence of a protective effect was weak. In both groups, there remained evidence of some protection in participants vaccinated >20 years ago.

Strengths and Limitations

To our knowledge, this is the largest adult study on the association between BCG and LTBI measured via IGRA in recent contacts of tuberculosis patients. The large sample size and rich dataset are key strengths. This enabled analysis by strata of age at BCG and time since vaccination and broad sensitivity analyses. Although the analysis of time since vaccination could not be adjusted for age at vaccination (and vice versa) due to collinearity, stratification helped address this. Using both commercially available IGRAs allowed a sensitivity analysis of those with concordant results, reducing the potential for outcome misclassification.

There are some limitations in a cross-sectional study. The observational design cannot prove causality, and there may be confounding by other factors such as health-seeking behavior. Furthermore, as many participants were born in high-burden countries, some may have been exposed to M. tuberculosis before vaccination. However, to increase the probability that vaccination preceded M. tuberculosis exposure, we only included recent tuberculosis contacts without prior reported tuberculosis exposure. Also, most participants born outside the UK were vaccinated in infancy, limiting exposure to M. tuberculosis before vaccination.

Measurement Error

Similar to prior studies, this study relied predominantly on BCG scar to measure vaccination. BCG does not always result in scar formation but may still confer protection in these cases [35, 37, 38]. Therefore some vaccinated participants may have been misclassified as nonvaccinated, if there was no documented evidence otherwise. This would be nondifferential with respect to IGRA, potentially biasing our VE estimates towards the null. Data on vaccination year was also likely subject to recall error but again this is likely nondifferential with regards to outcome.

Most data on covariates were self-reported. Prevalence of HIV and smoking were relatively low, possibly reflecting social desirability biases. Residual confounding is also possible. For example, latitude of country of birth was an important confounder and may be a proxy measure of potential exposure to nontuberculous mycobacteria, as well as M. tuberculosis, but does not take into account the time lived at that latitude. Similarly, tuberculosis incidence in country of birth is an incomplete measure of M. tuberculosis exposure.

There is currently no gold standard for diagnosing LTBI, and a positive IGRA cannot distinguish between true ongoing latent M. tuberculosis infection with dormant yet viable bacteria and an immunological memory response following exposure to M. tuberculosis. Long-term follow-up studies (or development of a gold standard test for diagnosing LTBI) would help assess if IGRA does measure true infection preceding risk of disease. We found increased prevalence of TST reaction ≥5 mm amongst vaccinated individuals, consistent with TST being affected by BCG. Stratified thresholds can improve the specificity of TST, as demonstrated by the protective association seen in our tiered 5/15 mm analysis. However, IGRAs are a more sensitive and specific surrogate marker for M. tuberculosis infection in BCG-vaccinated individuals [14, 15].

Missing Data

Twenty percent of otherwise eligible participants were excluded due to missing data on LTBI and/or BCG. Missing data on LTBI were predominantly due to logistic factors (eg, failed blood collection), so likely missing completely at random. This is supported by the lack of association between missing outcome data and BCG vaccination, and is therefore unlikely to be an important source of bias.

BCG vaccination status was missing for 13% of participants and these people were more likely to have LTBI. As it is harder to be certain of lack of vaccination, more unvaccinated participants may have been recorded as unsure and therefore coded as missing (possibly leading to the association of missing BCG status with LTBI). However, in these cases, when the missingness mechanisms are related to the exposure and covariates, but not the outcome, a logistic regression complete-case approach can still provide an unbiased estimate of the OR [39], dependent on the model including the covariates with which missingness is related. While being non-UK born and of Pakistani ethnicity were associated with missing BCG status (and not included in the final model as they were not found to be important confounders), these associations and numbers were small, so the overall bias of VE from missing BCG status is likely minimal.

Comparison With Existing Literature

The estimated VE of 30% is broadly consistent with other studies in adults [20–25]. Three studies in tuberculosis contacts had higher VEs than this study (ORs, 0.11–0.5), though these were small studies restricted to close contacts and had wide CIs that crossed 1 [23–25]. The lower VE in our analysis may reflect the limitations discussed above, which could have biased the estimate towards the null.

Although the CIs overlapped, our results suggest a possible decline in protection with increased time since vaccination. This concurs with studies of the duration of protection against active tuberculosis [5, 18]. Our finding that protection against infection was seen following both infant and older age vaccination is less consistent with other studies reporting that protection against active tuberculosis is greater when vaccination occurs at younger ages, in immunologically naive subjects [5, 9, 40]. Our findings are likely influenced by several factors. We do not have data on why participants were vaccinated at particular ages. However, those born outside the UK and vaccinated aged >2 years are likely to have been vaccinated after moving to the UK, where they would have been offered school-aged BCG vaccination only after negative TST results under the UK’s school-based BCG program, discontinued in 2005 [41]. Usually, only infant vaccination programs are offered in high tuberculosis burden settings. In addition, infant vaccination is less likely to result in scar formation compared with vaccination at older ages [42]. Therefore nondifferential misclassification is more likely in those vaccinated earlier, leading to an underestimate of VE in the younger age group. Finally, measurement of these variables was susceptible to bias, due to missing data and recall errors for age at vaccination. These exploratory results should therefore be interpreted cautiously and confirmed in other studies, including with longitudinal design.

Implications and Future Research

This study suggests that BCG may protect against the acquisition of M. tuberculosis infection in adults and not only progression to disease [17, 26]. This has implications for tuberculosis control strategies. The most recent WHO position paper on BCG reflects the growing evidence that BCG confers “a modest protective effect… against M. tuberculosis infection, representing a significant additional benefit of the BCG vaccine” [43]. Furthermore, our data are consistent with evidence of durability of protection against disease [18, 19], which should be considered in cost-effectiveness and modelling studies by low-burden countries considering moving away from universal BCG vaccination to targeted vaccination of high-risk groups [43].

The effect that BCG may have on wider transmission dynamics is unclear, though preliminary modelling studies suggest a vaccine that protects against M. tuberculosis infection would have a high population-level impact on tuberculosis incidence, across a range of exposure intensities [44]. Future modelling studies of tuberculosis transmission and burden should therefore include sensitivity analyses that account for some degree of protection against infection [43].

Finally, these findings may contribute to understanding the mechanism of action of BCG. This is vital in the development and evaluation of new tuberculosis vaccines, because many vaccine candidates rely on a BCG-boosting strategy [45]. Novel vaccines must, and are starting to show they do, confer additional protection beyond the effect of BCG on both M. tuberculosis infection and disease progression [46]. Prevention of infection measured via IGRA is being examined as a shorter-term endpoint in efficacy trials of new tuberculosis vaccines, accelerating their assessment [47]. A recent trial of BCG revaccination versus vaccination with a novel subunit vaccine assessed vaccine efficacy using IGRA conversion as a measure of M. tuberculosis acquisition and sustained IGRA positivity to reflect sustained infection [48]. Our study suggests BCG-like vaccines that prevent infection may also provide long-term durability of protection.

Our study contributes to the growing evidence that suggests BCG can act partially by providing protection against M. tuberculosis infection as well as disease. This has important implications for immunization programs, vaccine development and assessment, and tuberculosis control efforts worldwide.

Supplementary Data

Supplementary materials are available at The Journal of Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.

Notes

Author contributions. A. L. K., R. K. G., and I. A. had full access to all data in the study and take responsibility for the integrity of the data and accuracy of the data analysis. P. M., C. J., A. L. K., and I. A. carried out the study conception and design, and acquisition, analysis, and interpretation of data. A. L. K., C. J., and R. K. G. carried out statistical analysis. A. L. K., C. J., and P. M. drafted the manuscript. I. A., A. L., F. D., and M. L. obtained funding. C. J., R. K. G., and J. S. carried out administrative, technical, or material support. C. J., P. M., and I. A. carried out supervision. All authors reviewed the manuscript.

Acknowledgments. We are grateful to all members of the PREDICT Study Group: Ibrahim Abubakar, David Adeboyeku, Nabeela Bari, Jack Barker, Helen Booth, Graham Bothamley, Felix Chua, Dean Creer, Mathina Darmalingam, Robert N. Davidson, Martin Dedicoat, Jonathan J Deeks, Francis Drobniewski, Anne Dunleavy, Jose Figueroa, Chris Griffiths, Pranab Haldar, Mimi Haseldean, Andrew Hayward, Norman Johnson, Onn Min Kon, Heinke Kunst, Ajit Lalvani, Marc Lipman, Stefan Losewicz, Joanne Lord, William Lynn, Bobby Mann, Heather Milburn, John Moore-Gillon, Geoff Packe, Anton Pozniak, Frances Sanderson, Jo Southern. We also acknowledge Antonio Gasparini for his advice on the statistical methods.

Disclaimer. The funders had no influence on the contents of this paper.

Financial support. This work was supported by the National Institute for Health Research (NIHR) (grant numbers NIHR HTA 08/68/01 and NIHR SRF-2011-04-001); and the NIHR Imperial Biomedical Research Center (grant to F. D.).

Potential conflicts of interest. C. J., J. S., and A. J. report grants from the UK National Institute for Health Research during the conduct of the study. A. J. reports grants from the Wellcome Trust during the conduct of the study and is a named inventor on patents pertaining to T-cell-based diagnosis, including interferon-gamma release assays technologies. Some of these patents were assigned by the University of Oxford to Oxford Immunotec PLC resulting in royalty entitlements for the University of Oxford and A. L. All other authors report no potential conflicts. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.

Author notes

P. M. and I. A. contributed equally.