Systematic review and meta-analysis of the current evidence on the duration of protection by bacillus Calmette-Guérin vaccination against tuberculosis

Primary objective

To assess and quantify changes in protection by BCG vaccination over time, against all tuberculosis, tuberculosis mortality, pulmonary tuberculosis and extrapulmonary tuberculosis (meningeal, military and other extrapulmonary sites separately and together), based on controlled trials and observational studies.

Secondary objectives

To estimate, where the data are available:

1. overall efficacy for the above tuberculosis disease categories

2. variations in efficacy according to:

   1. latitude/geographic region

   2. whether prior tuberculin sensitivity was an exclusion criterion for BCG vaccination and, if so, the stringency of the test criterion

   3. time since vaccination and age at vaccination (neonatal, school age, adult/occupational), if possible separating the effect of time since vaccination from that of age

   4. risk of bias in the different study designs

   5. vaccine strain

   6. gender

   7. human immunodeficiency virus (HIV) status.

Randomised controlled trials

Studies that attempted to allocate participants (even those in which the mode of allocation was not clearly described) were classified as clinical trials. This decision was based on the fact that all reviewed trials were conducted several decades ago, prior to the development of modern standards for the reporting of trials. A few studies allocated a proportion of participants and not another. When data could be separated, the allocated section was considered as a clinical trial and the non-allocated section extracted as a cohort. If data could not be separated, the whole study was recorded as a cohort.

Case–control studies

In assessing case–control study eligibility for extraction, we assessed whether the control subjects were likely to have come from the same population as cases (scope for selection bias was reduced). For hospital-based studies, selection bias was judged to be minimal (1) if there was no detail on the disease status of control subjects but there was no mention of the hospital being a specialist reference centre for a particular disease or (2) if < 30% of control subjects were admitted for diseases that might indicate a different health-seeking behaviour than cases and other control subjects.

If either of the above criteria was not fulfilled for hospital control subjects, the study was not classed as having control subjects from the ‘same population’, and publications were excluded. To avoid exclusion of infected subjects from vaccinated but not from unvaccinated groups, we excluded case–control studies conducted in populations vaccinated outside the neonatal period, if tuberculin testing was undertaken to exclude positive reactors before vaccination. This was unless all unvaccinated subjects (cases or control subjects) were also tuberculin tested at the age of recommended vaccination. This is because studies that were not able to restrict the study population to those not already infected at the age when they would have received vaccination may be biased because the number of tuberculosis cases in the unvaccinated group would be higher than normal, because those infected with tuberculosis would be over-represented among those unvaccinated, and thus would lead to an overestimate of the effect of the BCG vaccination. 38 Case–control studies with a matched design but no matched analysis were included to avoid losing useful information; this was addressed in the analysis.

Cohort studies

Cohorts were extracted only of children, vaccinated and unvaccinated, who were tuberculin tested before vaccination and were tuberculin negative.

Case population studies

A case population study is undertaken on the total population of a geographical area. Data on all cases emerging during the study period in the population of interest are ascertained through surveillance and all other persons in the population are regarded as ‘control subjects’. The characteristics of cases occurring in the population (notably, their possible previous exposure to a given risk factor) are compared with those of the entire set of subjects in the population (supposing that demographic, health or drug use statistics for the whole population are available). The validity of this approach assumes the existence of a surveillance system capable of identifying all of the cases of an event within the population. The null hypothesis is that, in the absence of an association between risk factors and event, the odds of exposure are identical among cases and the rest of the population, barring sampling variations. 39

Cross-sectional and outbreak studies

Cross-sectional studies are those in which data on BCG vaccination and tuberculosis outcome are collected at the same point in time, whereas outbreak studies refer to investigation of BCG vaccination effectiveness that were undertaken using an observation study design following the occurrence of an outbreak of tuberculosis.

To be included case population studies, cross-sectional contact and outbreak studies had to account for tuberculin positivity in the design or analysis, or BCG vaccine must have been administered neonatally.

The detailed data extraction forms are included in Appendix 2.

Case–control studies

Following discussion within the BCG systematic review group, rules were agreed and case–control studies were assessed for risk of bias using the following criteria:

• Were BCG vaccination definitions the same for cases and control subjects?

• Was disease status blinded to BCG assessors?

• Were cases diagnosed assessed independently of vaccination status?

For case–control studies with a matched design, a criterion was developed based on whether or not matching was performed in the analysis, as an unmatched analysis may have introduced some bias.

Cohort studies

Cohort studies were assessed with regard to risk of loss to follow-up bias, treatment allocation concealment, case ascertainment bias and incomplete case ascertainment bias.

Case population studies

Case population studies were evaluated based on whether or not cases and the populations were consistent in terms of geography, time and age, as well as according to whether case ascertainment was blind to vaccination status, disease status was blind to BCG assessors and methods of case ascertainment were the same for vaccinated and unvaccinated.

Cross-sectional studies

Cross-sectional studies were assessed on whether BCG vaccination definitions were the same for cases and control subjects, whether disease status was blinded to BCG assessors and whether cases and control subjects were assessed independently of vaccination status.

Bacillus Calmette–Guérin strain

Randomised controlled trials were analysed according to strain of BCG used. Studies that provided information on the strain of BCG used were classified using information from Brewer et al. 17 to place within a phylogenetic group defined by Brosch et al. 41 When the BCG strain was not provided, efforts were made to identify the strain lineage from Brewer et al. 42 and other literature sources. 43 As the greatest number of trials reporting which BCG strain was used reported on efficacy for all types of tuberculosis not separately by site, this analysis was performed only for all tuberculosis morbidity outcomes, and not for pulmonary tuberculosis alone.

Overall efficacy by categories of tuberculosis outcomes

Results from each study, together with both fixed- and random-effects summary effect estimates, were displayed in forest plots. Meta-analyses were carried out using (log-) rate ratios, and results were displayed both as rate ratios and as VE (= 1 − rate ratios) if this was appropriate. Differences between fixed- and random-effects estimates suggest that there are differences between RRs estimated from smaller and larger studies.

In some studies, vaccination was restricted to individuals who were screened for M. tuberculosis infection and confirmed uninfected, whereas the control group may have included individuals who were not eligible for vaccination. Studies in which an attempt was made to adjust estimated efficacy for the tuberculin-positive population were included in meta-analyses. Studies that did not account for the tuberculin-positive population were reported separately, and excluded from meta-analyses.

Variation in efficacy according to characteristics of individuals and studies

Whenever possible, differences in efficacy according to characteristics of individuals (e.g. gender or age) were estimated by comparing efficacy [using ratios of RRs (RRRs)] between subgroups of individuals within studies. Investigation of the reason behind variation in efficacy in different studies used meta-regression, and quantified how much of the heterogeneity in estimates of efficacy was explained by selected factors.

Differences in efficacy between subgroups of studies (e.g. those classified as at low or high risk of bias, or those conducted at different latitudes) were quantified using random-effects meta-regression to estimate RRRs. Heterogeneity (differences between the true vaccine effects in the different studies) was quantified by estimating the between-study variance tau-squared (τ²). This is based on the assumption that the distribution of the true vaccine effects is normal: τ² corresponds with the variance of this distribution. To illustrate the meaning of this quantity, Table 2 shows the ratio of the effect (e.g. RR or rate ratio) at the 90th centile of the distribution to the effect in a study at the 10th centile.

In forest plots and meta-analyses, τ² was estimated using the method-of-moments estimator proposed by DerSimonian and Laird. Within meta-regression analyses, τ² was estimated by restricted maximum likelihood, using the metareg command in Stata. We conducted both univariable and multivariable meta-regression analyses: results from multivariable analyses were interpreted with caution because the number of studies was typically small compared with the number of study characteristics of interest.

We examined variation in efficacy according to the following characteristics.

Duration of protection

Analyses of the duration of protection by BCG vaccination included only studies in which efficacy can be estimated separately according to time since vaccination, age at vaccination, or both. Changes in efficacy with time since vaccination (or age for infant vaccination) were estimated within studies and these estimates were then combined in meta-analyses. Random-effects Poisson regression was used to estimate rates of change in the duration of protection with time, allowing for between-study heterogeneity. If data permitted, we examined evidence for non-linearity in the rate of waning protection, for example by including a quadratic term for time. These models were also used to examine associations of study characteristics with the extent and duration of protection. We also derived indirect estimates of changes in efficacy, for instance by comparing estimates from case–control studies in adults and children who were vaccinated at birth. However, such estimates should be interpreted with caution because of the potential for confounding by other study characteristics (ecological fallacy). Decisions were made not to plot studies in the duration plots for which there was no specified age at vaccination and/or age at outcome assessment. Time since vaccination was broken down into four categories: 0–5 years after vaccination, 5–10 years, 10–15 years, ≥ 15 years after vaccination. However, some studies present data only for 0–5 years and are therefore missing from these plots.

Design-specific issues: controlled trials

For each outcome in each controlled trial, we classified results as at low, unclear or high risk of bias, based on domain-specific assessments of risk of bias conducted using the Cochrane Collaboration's Risk of Bias tool. We compared estimates of extent and duration of protection according with risk of bias. We reported sensitivity analyses restricted to studies assessed as at low, and low or unclear, risk of bias if this was feasible.

Tuberculin sensitivity testing was assessed only in trials; cohort studies were included only if all participants were tested before vaccination.

Only trials reports contained sufficient data on BCG strain to allow analyses on the variation of efficacy by strain lineage.

Design-specific issues: observational studies

Meta-analyses of crude and adjusted rate ratios were derived separately for each observational study design. An a priori minimum set of confounders to designate a rate ratio as adjusted was agreed based on consideration (blind to study results) of a tabulation of all the confounders used in each study, stratified by study design. At minimum, results were not considered adjusted unless they were controlled at least for age and sex.

We reported sensitivity analyses restricted to studies that adjusted for socioeconomic status (SES) in their analyses.

Issues of interpretation

We investigated whether it is possible to classify regions or settings according to categories of VE (e.g. low, medium, high), based on whether it is possible to identify subgroups of settings and studies within which the effect of BCG vaccination appears relatively consistent.

All meta-analyses were carried out using the metan and metareg commands for Stata version 11.0 (StataCorp L P, College Station, TX, USA).

Pulmonary tuberculosis

This section of the report first presents overall results, followed by results stratified according to study characteristics (latitude, age at vaccination/stringency of tuberculin testing and risk of diagnostic detection bias). Unstratified analyses are ordered by year trial started.

Figure 3 shows the results of 18 trials evaluating the efficacy of BCG vaccination against pulmonary tuberculosis, listed by date of study start. Nine studies were conducted in the USA,5,44,45,48–50 four in India,28,51–53 and the others in Canada,13 Puerto Rico,15 the UK,14 Haiti55 and South Africa. 54 There was substantial variation in the protective efficacy of BCG vaccination against pulmonary tuberculosis, ranging from substantial protection, as in the UK MRC14 trial [rate ratio 0.22 (95% CI 0.16 to 0.31)], to absence of clinically important benefit, as in the large Chingleput28 trial [rate ratio 1.05 (95% CI 0.88 to 1.25)].

FIGURE 3.

Rate ratios (with 95% CI) comparing the incidence of pulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 3) in RCTs, by year of study start. a, Date of study publication was used if study start date was not available. CCH, Cook County Hospital; TB HH, tuberculosis households; TBPT, Tuberculosis Prevention Trial.

Stratified analysis by latitude (10°), ordered by year study started

Figure 4 shows estimated effects of BCG vaccination against pulmonary tuberculosis, stratified by latitude of the study location (10° bands either side of the equator). In general, the protective effect of BCG vaccination was either absent or low in studies conducted close to the equator, whereas there was reasonably consistent evidence of good protection observed in studies conducted at latitudes exceeding 40°. Relatively high protection was observed in studies (Saskatchewan Infants13 and MRC14) conducted above 50° latitude: rate ratio 0.22 (95% CI 0.16 to 0.30), corresponding to a VE of 78% (95% CI 70% to 84%). Latitude explained a substantial amount of the between-study variation in the protective effect of BCG vaccination.

FIGURE 4.

Rate ratios (with 95% Cl) comparing the incidence of pulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 3) in RCTs, stratified by latitude of study location (10° bands), ordered by year of study start, a, Date of study publication was used if study start date was not available. CCH, Cook County Hospital; D + L, DerSimonian and Laird method; M-H, Mantel-Haenszel method; TB HH, tuberculosis households; TBPT, Tuberculosis Prevention Trial.

Stratified analysis by age at vaccination and tuberculin testing stringency, ordered by year study started

Figure 5 shows estimated effects of BCG vaccination against pulmonary tuberculosis, stratified by the age at which vaccination was administered and the stringency of tuberculin testing performed, grouped into neonatal vaccination, school-aged vaccination, both stringent and less stringent, and an ‘other’ age group which includes studies in which older persons were vaccinated as well as those in which BCG vaccination, was given at any age, also divided into stringent and non-stringent tuberculin testing. Within these strata, there was little evidence of between-study heterogeneity, other than for studies with non-stringent tuberculin testing (school-age non-stringent testing τ² = 0.095 and other age non-stringent testing τ² = 0.091). In studies of school-age vaccination with stringent tuberculin testing prior to vaccination, the overall rate ratio was 0.26 (95% CI 0.18 to 0.37), corresponding to a good protective effect, with VE of 74% (95% CI 63% to 82%). There was also evidence of moderate levels of protection with neonatal vaccination [summary rate ratio 0.41 (95% CI 0.29 to 0.58), equivalent to a VE of 59% (95% CI 42% to 71%)]. Other strata showed less clear-cut evidence of protection, ranging from a moderate protection with rate ratio 0.59 (95% CI 0.65 to 1.01), equivalent to VE of 41% (95% CI −1% to 65%) in school-age vaccination studies with non-stringent tuberculin testing, to 0.88 (95% CI 0.59 to 1.31) equivalent to a VE of 12% (95% CI −0.31% to 41%) in studies of vaccination in other age groups with stringent tuberculin testing protocols.

FIGURE 5.

Rate ratios (with 95% CI) comparing the incidence of pulmonary tuberculosis among BCG vaccinated individuals compared with unvaccinated individuals for the longest duration of follow-up (see Table 3) in RCTs, stratified by age at vaccination/tuberculin testing stringency, ordered by year of study start. a, Date of study publication was used if study start date was not available. CCH, Cook County Hospital; D + L, DerSimonian and Laird method; M-H, Mantel-Haenszel method; TB HH, tuberculosis households; TBPT, Tuberculosis Prevention Trial.

Stratified analysis by risk of diagnostic detection bias, ordered by year study started

Figure 6 presents results from trials classified according to the level of risk of diagnostic detection bias. Among studies in which the outcome assessors were adequately blinded to the vaccination status of participants, or that performed active surveillance (which would act to lower the risk of diagnostic detection bias), there was substantial between-study variation, but the average protective efficacy of BCG vaccination against pulmonary tuberculosis was good [overall rate ratio 0.40 (95% CI 0.25 to 0.64), corresponding VE of 60% (95% CI 36% to 75%)] and was greater than for studies with higher risk of diagnostic detection bias, which showed a low protective effect [rate ratio 0.78 (95% CI 0.64 to 0.95), corresponding to VE of 22% (95% CI 5% to 37%)].

FIGURE 6.

Rate ratios (with 95% CI) comparing the incidence of pulmonary tuberculosis among BCG vaccinated individuals compared with unvaccinated individuals for the longest duration of follow-up (see Table 3) in RCTs, stratified by risk of diagnostic detection bias,^b ordered by year of study start. a, Date of study publication was used if study start date was not available; b, Diagnostic detection bias is considered to occur if the assessor of BCG vaccination outcome is not blinded to vaccination status. CCH, Cook County Hospital; D + L, DerSimonian and Laird method; M-H, Mantel-Haenszel method; TB HH, tuberculosis households; TBPT, Tuberculosis Prevention Trial.

All tuberculosis disease morbidity outcomes

Figure 7 shows the overall results of 21 trials evaluating the efficacy of BCG vaccination against all tuberculosis morbidity outcomes, ordered by the year of study start.

FIGURE 7.

Rate ratios (with 95% CI) comparing the incidence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 3) in RCTs, ordered by year of study start. a, Date of study publication was used if study start date was not available. CCH, Cook County Hospital; TB HH, tuberculosis households; TBPT, Tuberculosis Prevention Trial.

Unstratified analysis ordered by year trial started

An additional three trials did not disaggregate results by type of disease, compared with results on pulmonary tuberculosis. Observation of data from all 21 RCTs shows that the estimates of BCG vaccination efficacy against all disease from tuberculosis (except death) vary considerably (see Figure 7). Similarly to the pulmonary tuberculosis outcome results, the observed effect varies from a good protective effect, as in the Native American,5 Chicago Infants CCH48 and MRC trials,14 to an absence of protection against all tuberculosis morbidity outcomes, as in the Georgia School,49 Chingleput28 and in a residential school for people with learning difficulties (Illinois mentally handicapped50) trials. Most of the efficacy point estimates suggest that BCG vaccination reduces risk of tuberculosis, although the CIs are wide and include 1 in 14 of 21 studies.

Figures 8–11 present the results of RCTs stratified by latitude, age at vaccination/stringency of tuberculin testing and diagnostic detection bias, stringency of tuberculin testing and BCG strain lineage; the last was only possible to examine for all trials with all tuberculosis morbidity as the outcome.

FIGURE 8.

Rate ratios (with 95% CI) comparing the incidence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 3), in RCTs, stratified by latitude of study location (10° bands), ordered by year of study start. a, Date of study publication was used if study start date was not available. CCH, Cook County Hospital; D + L, DerSimonian and Laird method; M-H, Mantel-Haenszel method; TB HH, tuberculosis households; TBPT, Tuberculosis Prevention Trial.

FIGURE 9.

Rate ratios (with 95% CI) comparing the incidence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 3), in RCTs, stratified by age at vaccination, ordered by year of study start. a, Date of study publication was used if the study start date was not available. CCH, Cook County Hospital; D + L, DerSimonian and Laird method; M-H, Mantel-Haenszel method; TB HH, tuberculosis households; TBPT, Tuberculosis Prevention Trial.

FIGURE 10.

Rate ratios (with 95% CI) comparing the incidence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 3), in RCTs, stratified by risk of diagnostic detection bias,^b ordered by year of study start. a, Date of study publication was used if study start date was not available; b, Diagnostic detection bias is said to occur if the assessor of BCG vaccination outcome is not blinded to vaccination status. CCH, Cook County Hospital; D + L, DerSimonian and Laird method; M-H, Mantel-Haenszel method; TB HH, tuberculosis Households; TBPT, Tuberculosis Prevention Trial.

FIGURE 11.

Rate ratios (with 95% CI) comparing the incidence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 3), in RCTs, stratified by BCG strain lineage, ordered by year of study start. a, As per Brosch et al. 41b, Date of study publication was used if study start date was not available. The efficacy data for two trials (Native American5 and Chingleput28), were provided for two different strains of BCG, accounting for two extra sets of results in this graph). CCH, Cook County Hospital; D + L, DerSimonian and Laird method; DU2-III^a, third form of tandem duplication 2; DU2-IV^a, fourth form of tandem duplication 2; M-H, Mantel-Haenszel method; TB HH, tuberculosis households; TBPT, Tuberculosis Prevention Trial.

Scatterplot of vaccine efficacy by bacillus Calmette–Guérin strain type over time since trial started

Figure 12 shows VE for each trial plotted by vaccine strain and year the trial was started. We hypothesised that as the strains evolved over time, at least over the first few decades, the distribution of effect sizes would change over time. Not only was there no noticeable pattern over time, but also results showing trials according to strains used are distributed over a wide range of effect sizes.

FIGURE 12.

Scatterplot of study estimate of BCG vaccine efficacy over time of study start, by BCG strain category. DU2-III, third form of tandem duplication 2; DU2-IV, fourth form of tandem duplication 2; DU1-DU2-IV, tandem duplication 1 and fourth form of tandem duplication 2, according to Brosch et al. 41

Combined meningeal and/or miliary tuberculosis

The following section reports results of RCTs assessing the efficacy of BCG vaccination against meningeal and/or miliary tuberculosis outcomes.

Unstratified analysis ordered by year trial started

The rate ratios and 95% CIs from the six trials assessing the efficacy of BCG vaccination on meningeal and/or miliary tuberculosis are shown in Figure 13. Most studies had very few events in the BCG group. Although there is evidence of variation, all point estimates apart from those of the Puerto Rico and Georgia/Alabama trials15 were consistent with a high protective effect against these severe forms of tuberculosis disease. Two trials (Native American5 and MRC14) provided stronger evidence of a protective effect of BCG vaccination against meningeal and/or miliary tuberculosis.

FIGURE 13.

Rate ratios (with 95% CI) comparing the incidence of meningeal and/or miliary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 3) in RCTs, ordered by year of study start. a, The outcome is miliary tuberculosis only. TB HH, tuberculosis households.

Stratified analysis by latitude (10°) group, ordered by year study started

Figure 14 shows the results of trials evaluating the effect of BCG vaccination on meningeal and/or miliary tuberculosis stratified according to 10° latitude groups. With the exception of two trials conducted between 10° and 40° in Puerto Rico15 and Georgia/Alabama,15 all other studies (conducted at higher latitudes) showed evidence of a high and consistent protective effect against meningeal and/or miliary tuberculosis. Studies conducted between 40° and 50° and > 50° latitude had a combined high VE of 90% (95% CI 71% to 97%) and 94% (95% CI 55% to 99%), respectively. Latitude appeared here to explain a substantial amount of the between-study variation in the effect of BCG vaccination, although the results should be interpreted with caution in view of the small number of studies contributing to this analysis.

FIGURE 14.

Rate ratios (with 95% CI) comparing the incidence of meningeal and/or miliary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 3) in RCTs, stratified by latitude of study location (10° bands), ordered by year of study start. a, The outcome is miliary tuberculosis only. D + L, DerSimonian and Laird method; M-H, Mantel-Haenszel method; TB HH, tuberculosis households.

Meta-regression analysis

Stratification on latitude appeared to explain the between-study heterogeneity (τ² values before and after stratification were 0.767 and 0.000, respectively); however, there were limited data to support the hypothesis that BCG vaccination efficacy varies according to latitude (p-value = 0.141); similarly, studies conducted between 20° and 40° latitude had a RR 60.58 (95% CI 0.34 to 10715.67) times than studies located at latitudes above 40° latitude, but the CIs were wide (Table 9).

Stratified analysis by age at vaccination and tuberculin testing stringency, ordered by year study started

Both school-age vaccination with stringent tuberculin testing and neonatal vaccination showed high BCG vaccination efficacy against meningeal and miliary tuberculosis, with the two neonatal vaccination studies showing a pooled rate ratio of 0.10 (95% CI 0.01 to 0.77) corresponding to a high VE of 90% (95% CI 67% to 99%). The two pooled studies of school-age vaccination with stringent tuberculin testing gave a rate ratio of 0.08 (95% CI 0.03 to 0.25) also corresponding to a high VE of 92% (95% CI 75% to 97%). Only one study assessed BCG vaccination15 at other ages with stringent and non-stringent tuberculin testing, respectively, showing little evidence of protection in the case of vaccination in other/all age groups with non-stringent tuberculin testing (Figure 15). Stratification on age at vaccination and tuberculin testing stringency appeared to explain a substantial amount of the between-study variation.

FIGURE 15.

Rate ratios (with 95% CI) comparing the incidence of meningeal and/or miliary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 3) in RCTs, stratified by age at vaccination and tuberculin testing stringency, ordered by year of study start. a, The outcome is miliary tuberculosis only. CCH, Cook County Hospital; D + L, DerSimonian and Laird method; M-H, Mantel-Haenszel method; TB HH, tuberculosis households.

Meta-regression analysis

Based on meta-regression analyses, age at vaccination and stringency of tuberculin testing accounted for a substantial amount of the heterogeneity: τ² values before and after stratification = 0.767 and 0.000, respectively. However, there was insufficient evidence to test the hypothesis that BCG vaccination efficacy varied by age at vaccination/tuberculin testing stringency (p-value 0.282) (see Table 9).

Stratified analysis by risk of diagnostic detection bias, ordered by year study started

Results from trials were stratified according to risk of diagnostic detection bias (Figure 16). The most consistent protective effect was seen in studies with a lower risk of diagnostic detection bias. The overall rate ratio was 0.09 (95% CI 0.03 to 0.23), consistent with high VE of 91% (95% CI 77% to 97%), with little evidence of within-strata heterogeneity (p-value assuming a fixed-effects model = 0.959), whereas the trials with higher risk of bias showed an overall rate ratio of 1.26 (95% CI 0.14 to 11.27), equivalent to no clinical benefit. Diagnostic detection bias accounted for a substantial amount of the between-study variability.

FIGURE 16.

Rate ratios (with 95% CI) comparing the incidence of meningeal and/or miliary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 3) in RCTs, stratified by risk of diagnostic detection bias,^b ordered by year of study start. a, Outcome is miliary tuberculosis only; b, Diagnostic detection bias occurs if the assessor of BCG vaccination outcome is not blinded to vaccination status. CCH, Cook County Hospital; D + L, DerSimonian and Laird method; M-H, Mantel-Haenszel method; TB HH, tuberculosis households.

Meta-regression analysis

Stratification on risk of diagnostic detection bias accounted for a very substantial amount of variability (τ² values before and after stratification = 0.767 and 0.000, respectively) (see Table 9). However, the evidence that BCG vaccination efficacy varied with risk of diagnostic detection bias was weak: studies with a higher risk of diagnostic detection bias had a rate ratio 12.13 (95% CI 0.81 to 181.79) times that of studies with a lower risk of bias, with the VE for studies at higher risk of diagnostic detection bias being correspondingly lower.

Results from univariable meta-regressions suggested that latitude, age at vaccination/tuberculin testing stringency and risk of diagnostic detection bias were closely correlated and could each explain the between-study variation in overall BCG vaccination efficacy against meningeal and/or miliary tuberculosis with a τ² value of 0.000.

Extrapulmonary tuberculosis

Studies providing data on outcomes of extrapulmonary tuberculosis which were not specified to be tuberculosis meningitis and/or miliary tuberculosis were analysed in the extrapulmonary tuberculosis section below.

Unstratified analysis ordered by year trial started

Figure 17 shows a forest plot for the six trials providing data on the efficacy of the BCG vaccination on any extrapulmonary tuberculosis disease (excluding those referring to miliary tuberculosis or tuberculosis meningitis). There was substantial variation in the protective efficacy of the BCG vaccination against extrapulmonary tuberculosis, ranging from a high protective effect [rate ratio 0.14 (95% CI 0.03 to 0.63)] in Saskatchewan Infants,13 to absence of a clinically important benefit in the Georgia/Alabama trial15 [rate ratio 1.06 (95% CI 0.26 to 4.22)].

FIGURE 17.

Rate ratios (with 95% CI) comparing the incidence of extrapulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 3) in RCTs, ordered by year of study start.

Stratified analysis by latitude (10°), ordered by year study started

Figure 18 shows the estimated effect of BCG vaccination from trials with extrapulmonary tuberculosis outcome data excluding tuberculosis meningitis and miliary tuberculosis, stratified by location using 10° bands of latitude of study. Overall, the protective effect of the BCG vaccination was low or absent in studies conducted nearer the equator, whereas there was strong evidence of higher protection observed in studies carried out at latitudes exceeding 40°. The highest protective effect was observed in studies conducted above 50° latitude: rate ratio 0.17 (95% CI 0.11 to 0.29), corresponding to a high VE of 83% (95% CI 71% to 89%). Stratification by latitude appeared to explain a substantial amount of the between-study variation in effect of BCG vaccination.

FIGURE 18.

Rate ratios (with 95% confidence intervals) comparing the incidence of extrapulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 3) in RCTs, stratified by latitude of study location (10° bands), ordered by year of study start. D + L, DerSimonian and Laird method; M-H, Mantel-Haenszel method.

Meta-regression analysis

The between-study variation in estimates was explained by stratification on latitude (τ² values before and after stratification on 20° latitude were 0.275 and 0.000, respectively) (see Table 9). There was some evidence of protection varying with latitude (p-value = 0.100): the overall rate ratio for the 0–20° latitude strata was 2.89 times (95% CI 0.74 to 11.37) the rate ratio for studies above 40° latitude (thus VE correspondingly lower). Similarly, the overall protective effect for the strata 20–40° was 4.64 (0.70 to 30.87) times lower that of studies above 40° latitude.

Stratified analysis by age at vaccination and tuberculin testing stringency, ordered by year study started

The results of trials with outcome data on extrapulmonary tuberculosis disease (excluding outcomes specified as miliary disease or tuberculosis meningitis) were stratified according to age at vaccination and tuberculin testing stringency (Figure 19). A consistent, high protective effect was seen in studies undertaking stringent tuberculin testing for school-age vaccination [rate ratio 0.20 (95% CI 0.14 to 0.30)] equivalent to a VE of 80% (95% CI 70% to 86%). The one study of BCG vaccination in neonates also showed substantial high protection with rate ratio 0.14 (95% CI 0.03 to 0.63) [VE equivalent to 86% (95% CI 37% to 97%)]. The Georgia/Alabama study15 of vaccination in other age groups provided little evidence of an effect of BCG vaccination. Tis stratification accounted for a substantial amount of heterogeneity.

FIGURE 19.

Rate ratios (with 95% CI) comparing the incidence of extrapulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 3) in RCTs, stratified by age at vaccination/tuberculin testing stringency, ordered by year of study start. D + L, DerSimonian and Laird method; M-H, Mantel-Haenszel method.

Meta-regression analysis

Based on meta-regression analyses, the between-study variation appeared to be explained by age at vaccination/tuberculin testing (null model τ² = 0.275, after stratification τ² = 0.000) (Table 10). There was little evidence, however, that the effect of BCG vaccination on extrapulmonary tuberculosis varied with age at vaccination and tuberculin testing (p-value = 0.245); the overall effect of BCG vaccination in those vaccinated at school-age with stringent tuberculin testing was 1.44 times (95% CI 0.05 to 42.74) that of neonatal vaccination studies.

Stratified analysis by risk of diagnostic detection bias, ordered by year study started

Figure 20 shows estimated effect of BCG vaccination on extrapulmonary tuberculosis (excluding tuberculosis meningitis and miliary tuberculosis) stratified according to risk of diagnostic detection bias. There was reasonably consistent evidence of higher protection against extrapulmonary tuberculosis observed in pooled results of the three studies with a lower risk of diagnostic detection bias [rate ratio 0.19 (95% CI 0.12 to 0.28)] corresponding to a VE of 81% (95% CI 72% to 88%). By contrast, all three trials that did not adequately mask vaccination status and without active surveillance showed, overall, less effective results. Risk of diagnostic detection bias appears to explain a substantial amount of the between-study variation in the protective effect of BCG vaccination.

FIGURE 20.

Rate ratios (with 95% CI) comparing the incidence of extrapulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 3) in RCTs, stratified by risk of diagnostic detection bias,^a ordered by year of study start. a, Diagnostic detection bias is said to occur if the assessor of BCG vaccination outcome is not blinded to vaccination status. D + L, DerSimonian and Laird method; M-H, Mantel-Haenszel method.

Meta-regression analysis

Stratification on risk of diagnostic detection bias accounted for the between-study variation (τ² = 0.275 and 0.000, before and after stratification respectively) (see Table 10). There was evidence (p-value = 0.032) that efficacy varied with risk of diagnostic detection bias: overall rate ratios for studies with a higher risk of diagnostic detection bias were 3.35 (95% CI 1.19 to 9.48) times that of studies with a lower risk and hence a lower VE.

Results from univariable meta-regressions indicate that latitude, age at vaccination/tuberculin testing stringency and risk of diagnostic detection bias each explained a large amount of between-study variation in overall BCG vaccination efficacy with a τ² value of 0.000 for each of these variables, compared with the baseline τ² value of 0.275. These results should be interpreted with caution in view of the small number of studies contributing to this analysis.

Tuberculosis mortality

Unstratified analysis ordered by year trial started

Only 8 of the 21 trials presented data on the effect of BCG vaccination on mortality from tuberculosis (Figure 21). The Native American trial5 was the only study with a large number of events observed, all other studies having very few events and wide CIs. Two trials (Illinois mentally handicapped50 and Puerto Rico Children15) found a substantially increased risk of death in the vaccinated compared with unvaccinated group.

FIGURE 21.

Rate ratios (with 95% CI) comparing the incidence of tuberculosis mortality among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 3) in RCTs, ordered by year of study start. CCH, Cook County Hospital; TB HH, tuberculosis households.

Stratified analysis by latitude (10°), ordered by year study started

Figure 22 shows estimated effects of BCG vaccination against tuberculosis mortality, stratified by latitude of study location. In general, the protection afforded by BCG vaccination is low or absent in studies close to the equator, while there was reasonable consistent evidence of higher levels of protection against mortality in studies conducted at latitude exceeding 40°. The overall effect for studies conducted above 50° was rate ratio 0.17 (95% CI 0.11 to 0.29), corresponding to a high protection [VE of 83% (95% CI 71% to 89%)]. Again this should be interpreted with caution in view of the small number of studies contributing to this analysis. A substantial amount of the between-study variation in protective effect of BCG vaccination appeared to be explained in part by latitude, although residual heterogeneity remained between studies conducted in the same latitude bands.

FIGURE 22.

Rate ratios (with 95% CI) comparing the incidence of tuberculosis mortality among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 3), in RCTs stratified by latitude of study location (10° bands), ordered by year of study start. CCH, Cook County Hospital; D + L, DerSimonian and Laird method; M-H, Mantel-Haenszel method; TB HH, tuberculosis households.

Meta-regression analysis

Stratifying these studies by latitude (20° bands) accounted for 45% of the between-study variation (null model τ² = 1.067, after stratification τ² = 0.582) (Table 11). There was, however, insufficient evidence on which to suggest that efficacy varied with latitude (p-value = 0.138). The estimate of the rate ratio for studies at latitudes 0–20° was 5.63 (95% CI 0.48 to 66.58) times that of studies of > 20° latitude, pointing towards a corresponding lower VE.

Stratified analysis by age at vaccination and tuberculin testing stringency, ordered by year study started

The results in Figure 23 show the effect of BCG vaccination on tuberculosis mortality, stratified by age at which vaccination was administered and stringency of the tuberculin testing method. The most consistent protective effect of BCG vaccination was seen in neonatal vaccination and school-age vaccination with stringent tuberculin testing [overall good protection with rate ratio from five studies of 0.34 (95% CI 0.12 to 0.92) corresponding to a VE of 66% (95% CI 8% to 88%) and rate ratio from one study of 0.26 (95% CI 0.17 to 0.40) corresponding to a VE of 74% (95% CI 60% to 83%), respectively]. There was no evidence of a protective effect of BCG vaccination in the one study each of ‘other’ age groups of vaccination, with stringent or non-stringent tuberculin testing. Stratification by age at vaccination appeared to explain some of the between-study variation observed.

FIGURE 23.

Rate ratios (with 95% CI) comparing the incidence of tuberculosis mortality among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 3) in RCTs, stratified by age at vaccination/tuberculin testing stringency, ordered by year of study start. CCH, Cook County Hospital; D + L, DerSimonian and Laird method; M-H, Mantel-Haenszel method; TB HH, tuberculosis households.

Meta-regression analysis

Based on meta-regression analyses, age at vaccination and tuberculin testing stringency accounted for 31% of the between-study variation observed (null model τ² = 1.067, after stratification τ² = 0.731) (see Table 11). There was, however, insufficient evidence (p-value = 0.369) that BCG vaccination efficacy against tuberculosis mortality varied according to age at vaccination/tuberculin testing stringency. The point estimate of effect (rate ratio) for studies of school-age vaccination with stringent tuberculin testing was 0.53 (95% CI 0.02 to 13.26) times that of neonatal vaccination studies, whereas rate ratio for BCG vaccination at other ages with stringent tuberculin testing was 3.81 (95% CI 0.07 to 202.48) times higher and VE correspondingly lower.

Stratified analysis by risk of diagnostic detection bias, ordered by year study started

Figure 24 presents the estimated effects of BCG vaccination on tuberculosis mortality stratified by risk of diagnostic detection bias. The most consistent protective effect was seen in studies assessed as having a lower risk of diagnostic detection bias, with rate ratio 0.32 (95% CI 0.17 to 0.59), corresponding to a good protective effect [VE of 68% (95% CI 41% to 83%)], whereas there was an absence of protective effect of BCG vaccination in two trials with a higher risk of diagnostic detection bias. Stratification on risk of diagnostic detection bias appeared to explain a substantial amount of the between-study variation, although residual heterogeneity remained within strata.

FIGURE 24.

Rate ratios (with 95% CI) comparing the incidence of tuberculosis mortality among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 3) in RCTs, stratified by risk of diagnostic detection bias,^a ordered by year of study start. a, Diagnostic detection bias occurs if the assessor of BCG vaccination outcome is not blinded to vaccination status. CCH, Cook County Hospital; D + L, DerSimonian and Laird method; M-H, Mantel-Haenszel method; TB HH, tuberculosis households.

Meta-regression analysis

Stratification on risk of diagnostic detection bias accounted for 76% of the between-study variation (null model τ² = 1.067, after stratification τ² = 0.260) (see Table 11). There was more evidence (p-value = 0.034) that efficacy varied with latitude: rate ratio for studies with higher risk of bias was 6.65 (95% CI 1.22 to 36.09) times greater than that of studies with a lower risk of bias.

Results from univariable meta-regressions indicate that latitude, age at vaccination/tuberculin testing stringency and risk of diagnostic detection bias each appeared to explain some of the between-study heterogeneity in overall BCG vaccination efficacy with τ² values 0.582, 0.731 and 0.260, respectively, compared with the baseline τ² value (1.067), but only with risk of diagnosis detection bias was there sufficient strong evidence of an effect.

Pulmonary tuberculosis

A variety of observational studies have assessed the effectiveness of BCG vaccination in protecting against pulmonary tuberculosis: eight eligible case–control studies (in which eligibility was that control subjects were selected from the same population from which cases arose), 12 eligible cohort studies (in which the comparison group was tuberculin negative individuals at the start of follow-up), four case population studies (where the rate in BCG vaccinated was compared with rates in the estimated tuberculin negative populations), six cross-sectional studies and three outbreak studies. These studies were conducted in a range of countries between 1936 and 2004. Figures 25–28 present the unstratified results of these observational studies, ordered chronologically by date of study start. While there is some degree of variation in the estimates of effectiveness of BCG vaccination, the majority of studies for each study design showed evidence consistent with a protective effect of BCG vaccination against pulmonary tuberculosis.

FIGURE 25.

Odds ratios (with 95% CI) comparing the BCG vaccination status of pulmonary tuberculosis cases and control subjects in case–control studies, ordered by year of study start. a, Date of study publication was used if study start date was not available.

FIGURE 26.

Rate ratios (with 95% CI) comparing the incidence of pulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 4) in cohort studies, ordered by year of study start. a, Date of study publication was used if study start date was not available.

FIGURE 27.

Rate ratios (with 95% CI) comparing the incidence of pulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 5) in case population studies, ordered by year of study start.

FIGURE 28.

Risk ratios (with 95% CI) comparing the prevalence of pulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals in cross-sectional studies, ordered by year of study start. a, Date of study publication was used if study start date was not available.

Estimates of protection by BCG vaccination against pulmonary tuberculosis varied across the eight case–control studies, ranging from a strong protective effect in Tailand65 [OR 0.30 (95% CI 0.24 to 0.38)], to clinical benefit in Bangalore children56 [OR 0.88 (95% 0.53 to 1.17)]. In 11 of the 12 cohorts in Figure 26 there was strong evidence of a protective effect, ranging from rate ratio 0.56 (95% CI 0.39 to 0.81) in the control arm of the Brazil revaccination study6 [equivalent to VE of 44% (95% CI 19% to 61%)] to rate ratio 0.01 (95% CI 0.00 to 0.14) in UK medical students100 [VE 99% (95% CI 86% to 100%)], with one study from Karonga, Malawi24 showing no evidence of clinical benefit. Two of the case population studies, those conducted in Poland160 and Canada (Quebec Pulmonary158), showed evidence of a high level of protection against pulmonary tuberculosis, whereas the other two studies showed some evidence of a protective effect. Four of the six cross-sectional studies found strong evidence of BCG vaccination reducing the risk of pulmonary tuberculosis (see Figure 28), with the Kenyan187 and Lebanese children193 studies suggesting no clinical benefit from BCG vaccination. Two of the three outbreak studies suggested limited clinical benefit from BCG vaccination, whereas the Cork toddler outbreak198 showed evidence of a protective effect [rate ratio 0.09 (95% CI 0.01 to 1.52)].

Unstratified analysis ordered by year study started

Outbreak studies

See Figure 29.

FIGURE 29.

Risk ratios (with 95% CI) comparing the incidence of pulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals in outbreak studies, ordered by year of study start.

Stratified analysis by latitude (10°), ordered by year study started

Forest plots in Figures 30–33 present results from observational studies on the protective effect of BCG vaccination against pulmonary tuberculosis stratified by 10° latitude bands of study location. No case–control studies were identified above 40° of latitude, but there was reasonable evidence of the strongest protective effect in all studies, apart from the cross-sectional ones, at latitudes further from the equator. All outbreak studies were located above 50° latitude.

FIGURE 30.

Odds ratios (with 95% CI) comparing the BCG vaccination status of pulmonary tuberculosis cases and control subjects in case–control studies, stratified by latitude of study location (10° bands), ordered by year of study start. a, Date of study publication was used if study start date was not available. D + L, DerSimonian and Laird method; I – V, inverse variance method.

FIGURE 31.

Rate ratios (with 95% CI) comparing the incidence of pulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 4) in cohort studies, stratified by latitude of study location (10° bands), ordered by year of study start. a, Date of study publication was used if study start date was not available. D + L, DerSimonian and Laird method; I – V, inverse variance method.

FIGURE 32.

Rate ratios (with 95% CI) comparing the incidence of pulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 5) in case population studies, stratified by latitude of study location (10° bands), ordered by year of study start. D + L, DerSimonian and Laird method; I – V, inverse variance method.

FIGURE 33.

Risk ratios (with 95% CI) comparing the prevalence of pulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated in cross-sectional studies, stratified by latitude of study location (10° bands), ordered by year of study start. a, Date of study publication was used if study start date was not available. D + L, DerSimonian and Laird method; I – V, inverse variance method.

Meta-regression analysis

Using such an analysis, however, indicated that stratification on latitude explained only a small amount (9%) of the between-study heterogeneity within case–control studies (null model τ² = 0.141, after stratification τ² = 0.128). There was little evidence to suggest that BCG vaccination effectiveness varies with latitude (p-value = 0.199) in this study type (Table 12). Latitude explained a more substantial amount of heterogeneity (42%) in cohort studies (τ² values before and after stratification = 0.872 and 0.509, respectively, Table 13), and some evidence (p-value = 0.060) that BCG vaccination effectiveness varies with latitude and with study design (retrospective cohort studies were more likely to show a protective effect than prospective cohort studies). None of the between-study variation seen within cross-sectional studies was found to be associated with latitude (τ² values before and after stratification = 0.249 and 0.599, respectively (Table 14). There were an insufficient number of case population studies to undertake meta-regression analyses.

Stratified analysis age at vaccination, ordered by year study started

Estimated effects of BCG vaccination on pulmonary tuberculosis from observational studies were stratified by age at vaccination (Figures 34–37). Only one case–control study of pulmonary tuberculosis did not have neonatal BCG vaccination (Figure 34). This study did not show a protective effect of vaccination, whereas the pooled estimate of neonatal vaccination studies was consistent with a moderate protective effect: OR 0.53 (95% CI 0.39 to 0.72), equivalent to a VE of 47% (95% CI 28% to 61%).

FIGURE 34.

Odds ratios (with 95% CI) comparing the BCG vaccination status of pulmonary tuberculosis cases and control subjects in case–control studies, stratified by age at vaccination, ordered by year of study start. a, Date of study publication was used if study start date was not available. D + L, DerSimonian and Laird method; I – V, inverse variance method.

FIGURE 35.

Rate ratios (with 95 CI) comparing the incidence of pulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 4) in cohort studies, stratified by age at vaccination, ordered by year of study start. a, Date of study publication was used if study start date was not available. D + L, DerSimonian and Laird method; I – V, inverse variance method.

FIGURE 36.

Rate ratios (with 95% CI) comparing the incidence of pulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 5) in case population studies, stratified by age at vaccination, ordered by year of study start. D + L, DerSimonian and Laird method; I – V, inverse variance method.

FIGURE 37.

Risk ratios (with 95% CI) comparing the prevalence of pulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals in cross-sectional studies, stratified by age at vaccination, ordered by year of study start. a, Date of study publication was used if study start date was not available. D + L, DerSimonian and Laird method; I – V, inverse variance method.

Results from the cohort studies showed some variation with neonatal BCG vaccination appearing to provide a good protective effect [rate ratio 0.21 (95% CI 0.08 to 0.56)], compared with school-age vaccination [rate ratio 0.35 (95% CI 0.21 to 0.58)] (see Figure 35). Results from studies of vaccination in other age groups showed highly protective effect, with the rate ratio ranging from 0.01 (95% CI 0.00 to 0.14) in the Dublin nurses study115 to 0.12 (95% CI 0.04 to 0.37) in the Bornholm study. 122

One of the two case population studies evaluating neonatal BCG vaccination (Bydgoszcz children),160 suggested a very high level of effectiveness, whereas the other (Israel neonatal) study,162 showed very little reduction in the risk of pulmonary tuberculosis, similar to the other two case population studies of school-age and other age vaccination, respectively. Only one of the cross-sectional studies assessed the effect of neonatal vaccination on pulmonary tuberculosis and showed a moderate protective effect, as did the two investigating school-age vaccination. Little protection was noted in those examining vaccinations in other age groups (see Figure 37). Similarly, little protection was noted in the outbreak study in other age groups (Figure 39), whereas overall moderate efcacy was seen in neonatal vaccination outbreak studies.

Meta-regression analysis

Using this method, stratification on age at vaccination accounted for only some of the heterogeneity in cohort studies (τ² values before and after stratification were 0.872 and 0.546, respectively, see Table 13) and some evidence (p-value = 0.049) that age at vaccination is associated with the size of BCG vaccination protective effect.

Outbreak studies

See Figure 38.

FIGURE 38.

Risk ratios (with 95% CI) comparing the prevalence of pulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals in outbreak studies, stratified by age at vaccination, ordered by year of study start. D + L, DerSimonian and Laird method; I – V, inverse variance method.

Stratified analysis by study design, ordered by year study started

Cohort studies

Figure 39 presents estimated effects of BCG vaccination against pulmonary tuberculosis in cohort studies, stratified by study design. Te strongest protective effect of BCG vaccination was seen in retrospective cohort studies with rate ratio 0.15 (95% CI 0.13 to 0.18), corresponding to a high level of protection with a VE of 85% (95% CI 82% to 87%). Stratification by study design accounted for a substantial amount of heterogeneity, but residual between-study variation remained, particularly in the retrospective cohort study group.

FIGURE 39.

Rate ratios (with 95% CI) comparing the incidence of pulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 4) in cohort studies, stratified by cohort study design, ordered by year of study start. a, Date of study publication was used if study start date was not available. D + L, DerSimonian and Laird method; I – V, inverse variance method.

Meta-regression analysis

Stratification on study design for cohorts explained 61% of the between-study variation observed within these studies (τ² null model = 0.872, τ² values after stratification on study design = 0.342) (see Table 13). There was evidence that BCG vaccination effectiveness varies depending on cohort study design (p-value = 0.018). Te overall rate ratio for retrospective cohort studies was 0.27 (95% CI 0.10 to 0.76) times the overall rate ratio for prospective studies, corresponding to a protective effect approximately 73% higher in retrospective compared with prospective studies.

Meta-regression analysis

Unstratified analysis ordered by year study started

More observational studies assessed the effectiveness of BCG vaccination in preventing all forms of tuberculosis: 16 case–control studies, 32 cohort studies, 12 case population studies, 13 cross-sectional studies and seven outbreak studies (Figures 40–44). All case–control studies showed evidence of protection against all forms of tuberculosis disease (see Figure 40). Similarly, the majority of cohort studies found a protective effect of BCG vaccination on all tuberculosis morbidity outcomes (see Figure 41), with only four showing no evidence of a clinical benefit from BCG vaccination. Te observed effect varied from a high protective effect (rate ratio 0.01; 95% CI 0.00 to 0.05), in the German Siblings cohort,107 to no clinical benefit (rate ratio 6.08; 95% CI 0.00 to 5.97e7) in the Edinburgh study. 96 All case population studies showed evidence of a strong protective effect of BCG vaccination against all tuberculosis morbidity outcomes (see Figure 42) and the majority of cross-sectional studies, except for the Surui Indians180 study, found evidence of substantial protection against tuberculosis (see Figure 43). Te majority of outbreak studies provided evidence of a protective effect against all forms of tuberculosis morbidity (see Figure 43).

FIGURE 40.

Odds ratios (with 95% CI) comparing the BCG vaccination status of all tuberculosis outcome cases and control subjects in case–control studies, ordered by year of study start. a, Date of study publication was used if study start date was not available.

FIGURE 41.

Rate ratios (with 95% CI) comparing the incidence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 4) in cohort studies, ordered by year of study start. a, Date of study publication was used if study start date was not available.

FIGURE 42.

Rate ratios (with 95% CI) comparing the incidence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 5) in case population studies, ordered by year of study start.

FIGURE 43.

Risk ratios (with 95% CI) comparing the prevalence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals, in cross-sectional studies ordered by year of study start. a, Date of study publication was used if study start date was not available.

FIGURE 44.

Risk ratios (with 95% CI) comparing the incidence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals, in cross-sectional studies ordered by year of study start.

Stratified analysis by latitude (10°), ordered by year study started

Figures 45 and 46 present the estimated effects of case–control and cohort studies against all tuberculosis morbidity outcomes stratified by latitude. Studies on all types of tuberculosis morbidity, unlike pulmonary tuberculosis studies, showed evidence of high effectiveness with no defined pattern of latitude effect on BCG vaccination effectiveness.

FIGURE 45.

Odds ratios (with 95% CI) comparing the BCG vaccination status of all tuberculosis outcome cases and control subjects in case–control studies, stratified by latitude of study location (10° bands), ordered by year of study start. a, Date of study publication was used if study start date was not available. D + L, DerSimonian and Laird method; I – V, inverse variance method.

FIGURE 46.

Rate ratios (with 95% CI) comparing the incidence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 4) in cohort studies, stratified by latitude of study location (10° bands), ordered by year of study start. a, Date of study publication was used if study start date was not available. D + L, DerSimonian and Laird method; I – V, inverse variance method.

Figures 47 and 48 show forest plots of the results of case population and cross-sectional studies, respectively, stratified by 10° latitude. The majority of case population studies were conducted in sites located above 50° latitude. The small number of case population studies in each stratum other than the above 50° latitude band limits any conclusion, although there was some evidence that the protective effect was strongest in studies conducted at latitudes furthest away from the equator. The majority of cross-sectional studies show a protective effect of BCG vaccination against tuberculosis; however, there was no clear evidence of variation by latitude. Stratification by latitude therefore appeared to explain very little of the heterogeneity in both case population and cross-sectional studies. All outbreak studies were conducted above 50° latitude.

FIGURE 47.

Rate ratios (with 95% CI) comparing the prevalence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 5) in case population studies, stratified by latitude of study location (10° bands), ordered by year of study start. D + L, DerSimonian and Laird method; I – V, inverse variance method.

FIGURE 48.

Risk ratios (with 95% CI) comparing the incidence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals, in cross-sectional studies stratified by latitude of study location (10° bands), ordered by year of study start. a, Date of study publication was used if study start date was not available. D + L, DerSimonian and Laird method; I – V, inverse variance method.

Meta-regression analysis

Meta-regression with stratification on latitude partially explained the heterogeneity between case–control studies (null model τ² = 0.208; after stratification by 20° latitude group τ² = 0.180)

(Table 15). Latitude also accounted for a proportion of the heterogeneity between cohort studies (null model τ² = 0.253; after stratification by 20° latitude group τ² = 0.185) (Table 16) but there was little evidence (p-value = 0.174) that effectiveness varied by latitude. There was some evidence (p-value = 0.069) that BCG vaccination protection against all tuberculosis varies according to latitude in cohort studies: the overall rate ratios for the 0–20° and 20–40° latitude groups were, respectively, 2.30 (95% CI 1.10 to 4.78) and 0.94 (95% CI 0.53 to 1.68) times the overall rate ratio in the 40° and above group. Based on meta-regression analysis (Table 17), stratification on latitude did not account for any of the between-study variation seen in case population studies (τ² values before and after stratification by 20° group were 0.619 and 0.683, respectively) or cross-sectional studies (τ² values before and after stratification by 20° group 0.716 and 0.754, respectively) (Table 18).

Stratified analysis by age at vaccination, ordered by year study started

Forest plots showing the results of observational studies stratified by the age at which BCG vaccination was given to participants are presented in Figures 49 and 52. All but one case–control study evaluated the effect of neonatal BCG vaccination against all forms of tuberculosis (see Figure 49). This study did not show a protective effect, whereas the estimates of neonatal vaccination studies ranged from substantial protection, as in Nepal75 [OR 0.11 (95% CI 0.07 to 0.17)], to reduced protection [OR 0.69 (95% CI 0.47 to 1.01)], in Madagascar children. 79 Estimates of effect from cohort studies indicated that, in general, the level of protection was similar among studies investigating neonatal BCG vaccination [rate ratio 0.22 (95% CI 0.12 to 0.40)], and school-age vaccination [rate ratio 0.24 (95% CI 0.17 to 0.34)] (the results being based only on tuberculin-negative participants at study start), corresponding to VE of 76% (95% CI 66% to 83%). The Karonga study124 did not show evidence of a protective effect from school-age vaccination, and the majority of studies failed to detect a substantial protective effect from BCG vaccination in other age groups. The majority of case population studies showed strong protective effects of BCG vaccination against all forms of tuberculosis, ranging from rate ratio 0.05 (95% CI 0.04 to 0.08) in the Bydgoszcz children160 neonatal vaccination study to rate ratio of 0.80 (95% CI 0.72 to 0.89) in the Ireland Survey study of school-age BCG vaccination. 167 All cross-sectional studies evaluated either neonatal or school-age vaccination. Both groups were associated with substantial protection against tuberculosis; the highest level was observed in studies evaluating the vaccination of school-age children [RR 0.16 (95% CI 0.04 to 0.65)], equivalent to VE of 84% (95% CI 35% to 96%) (see Figure 52). Age at vaccination appears to explain very little of the heterogeneity seen between case–control studies, whereas evidence of substantial heterogeneity remained after stratification for age at vaccination for all study designs. In outbreak studies, there was little evidence of a clinical benefit from neonatal vaccination, whereas outbreaks in which vaccination was undertaken at school age showed evidence of a high protective effect RR of 0.02 (95% CI 0.01 to 0.07), equivalent to VE of 98% (95% CI 93% to 99%).

FIGURE 49.

Odds ratios (with 95% CI) comparing the BCG vaccination status of all tuberculosis morbidity outcomes cases and control subjects in case–control studies, stratified by age at vaccination, ordered by year of study start. a, Date of study publication was used if study start date was not available. D + L, DerSimonian and Laird method; I – V, inverse variance method.

Meta-regression analysis

Stratification of case–control studies on age at vaccination did not explain the between-study variation (τ² = 0.215 after stratification, compared with τ² = 0.208 in the null model, as shown in Table 15). Age at vaccination did not explain any of the variation between cohort studies [τ² values before and after stratification = 0.253 and 0.268, respectively (see Table 16)], case population studies [τ² values before and after stratification = 0.619 and 0.705, respectively (see Table 17)] or cross-sectional studies [τ² values before and after stratification = 0.736 and 0.716, respectively (see Table 18)]. There was no evidence that effectiveness of BCG vaccination against all forms of tuberculosis varied by age at vaccination for case–control studies, cohort studies or cross-sectional studies (p-values = 0.537, 0.487 and 0.398, respectively).

Cohort studies

See Figure 50.

FIGURE 50.

Rate ratios (with 95% CI) comparing the incidence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 4) in cohort studies, stratified by age at vaccination, ordered by year of study start. D + L, DerSimonian and Laird method; I – V, inverse variance method.

Case population studies

See Figure 51.

FIGURE 51.

Rate ratios (with 95% CI) comparing the incidence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 5) in case population studies, stratified by age at vaccination, ordered by year of study start. D + L, DerSimonian and Laird method; I – V, inverse variance method.

Cross-sectional studies

See Figure 52.

FIGURE 52.

Risk ratios (with 95% CI) comparing the prevalence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals in cross-sectional studies, stratified by age at vaccination, ordered by year of study start. a, Date of study publication was used if study start date was not available. D + L, DerSimonian and Laird method; I – V, inverse variance method.

Outbreak studies

See Figure 53.

FIGURE 53.

Risk ratios (with 95% CI) comparing the incidence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals in outbreak studies, stratified by age at vaccination, ordered by year of study start. D + L, DerSimonian and Laird method; I – V, inverse variance method.

Stratified analysis by study design, ordered by year study started

Cohort studies

Figure 54 shows the result of cohort studies grouped by retrospective and prospective design and contact studies. As with pulmonary tuberculosis the most consistent protective effect of BCG vaccination against all types of tuberculosis disease was seen in retrospective cohort studies. Stratifying on cohort study design explained a small amount of the heterogeneity seen.

FIGURE 54.

Rate ratios (with 95% CI) comparing the incidence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 4) in cohort studies, stratified by cohort study design, ordered by year of study start. D + L, DerSimonian and Laird method; I – V, inverse variance method.

Meta-regression analysis

Based on meta-regression analyses, however, only 7% of the between-study variation was explained by study design [τ² values before and after stratification were 0.253 and 0.236, respectively (see Table 16)].

Meta-regression analysis

Unstratified analysis ordered by year study started

A total of 31 observational studies provided data on the effect of BCG vaccination against miliary tuberculosis and tuberculosis meningitis. The majority of studies assessed this effect on meningeal tuberculosis (14 case–control studies, six cohorts, six case population studies and six cross-sectional studies), seven studies provided data on the combined outcome of both miliary tuberculosis and meningitis, whereas one cohort and one cross-sectional study provided data on miliary tuberculosis only. Figures 55–58 present the unstratified results of these studies. Despite variation in the protective effectiveness of BCG vaccination against meningeal and/or miliary tuberculosis between individual studies, the majority (21) of these showed an overall protective effect.

FIGURE 55.

Odds ratios (with 95% CI) comparing the BCG vaccination status of meningeal and/or miliary tuberculosis cases and control subjects in case–control studies, ordered by year of study start. D + L, DerSimonian and Laird method; I – V, inverse variance method. a, Combined tuberculosis meningitis and miliary tuberculosis outcomes; b, Tuberculosis meningitis outcome only; c, Date of study publication was used if study start date was not available.

FIGURE 56.

Rate ratios (with 95% CI) comparing the incidence of meningeal and/or miliary tuberculosis among cases with that in control subjects for the longest duration of follow-up (see Table 4) in cohort studies, ordered by year of study start. a, Combined tuberculosis meningitis and miliary tuberculosis outcomes; b, Tuberculosis meningitis outcome only; c, Miliary tubeculosis outcome only. D + L, DerSimonian and Laird method; I – V, inverse variance method.

FIGURE 57.

Rate ratios (with 95% CI) comparing the incidence of meningeal and/or miliary tuberculosis among vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 5) in case population studies, ordered by year of study start. a, Tuberculosis meningitis outcome only.

FIGURE 58.

Risk ratios (with 95% CI) comparing the prevalence of meningeal and/or miliary tuberculosis among vaccinated individuals with that in unvaccinated individuals in cross-sectional studies, ordered by year of study start. a, Combined tuberculosis meningitis and miliary tuberculosis outcomes; b, Tuberculosis meningitis outcome only; c, Miliary tuberculosis outcome only. D + L, DerSimonian and Laird metod; I – V, inverse variance method.

All case–control studies showed evidence of a protective effect of BCG vaccination against the combined outcome of meningeal and miliary tuberculosis, and meningeal tuberculosis only (see Figure 55). Similarly, the effect of BCG vaccination observed in cohort studies was protective, although in three of the six studies the sample sizes were small and CI crossed one. All case population studies showed evidence of a strong protection against meningeal tuberculosis. Only one out of the six cross-sectional studies did not find evidence of a clinical benefit (see Figure 58).

Stratified analysis by latitude (10°), ordered by year study started

Figures 59–62 show the estimated effects of BCG vaccination against meningeal and miliary tuberculosis, stratified by latitude of study location. All case–control studies were conducted at latitudes of < 40°. The most consistent, protective effect was high, and seen in studies conducted between 30° and 40° with an overall OR of 0.19 (95% CI 0.12 to 0.31), corresponding to a VE of 81% (95% CI 69% to 88%). There was evidence of heterogeneity for studies from locations between 20° and 30°, with some studies showing a protective effect similar to that seen at higher latitudes, whereas in others the protective effect was low. Studies at lower latitudes (0° to 20°) showed evidence of a good protective effect. There was consistent evidence of high protection in cohort studies conducted above 50° latitude [rate ratio 0.23 (95% CI 0.12 to 0.44)] equivalent to a good VE of 77% (95% CI 56% to 88%), whereas only one cohort study each was conducted at 30–40° and 40–50° latitude. Results from observational case population and cross-sectional studies showed no evidence that protection by BCG vaccination against meningeal and miliary tuberculosis varied substantially by latitude.

FIGURE 59.

Odds ratios (with 95% CI) comparing the BCG vaccination status of meningeal and/or miliary tuberculosis cases and control subjects in case–control studies, stratified by latitude (10°), ordered by year of study start. a, Meningitis only; b, Combined meningitis and miliary; c, Date of study publication was used if study start date was not available. D + L, DerSimonian and Laird method; I – V, inverse variance method.

FIGURE 60.

Rate ratios (with 95% CI) comparing the incidence of meningeal and/or miliary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 4) in cohort studies, stratified by latitude (10°), ordered by year of study start. a, Date of study publication was used if study start date was not available; b, Combined tuberculosis meningitis and miliary tuberculosis outcomes; c, Tuberculosis meningitis outcome only; d, Miliary tuberculosis outcome only. D + L, DerSimonian and Laird method; I – V, inverse variance method.

FIGURE 61.

Rate ratios (with 95% CI) comparing the incidence of meningeal and/or miliary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 5) in case population studies, stratified by latitude (10°), ordered by year of study start. a, Tuberculosis meningitis outcome only. D + L, DerSimonian and Laird method; I – V, inverse variance method.

FIGURE 62.

Risk ratios (with 95% CI) comparing the prevalence of meningeal and/or miliary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals in cross-sectional studies, stratified by latitude (10°), ordered by year of study start. a, Date of study publication was used if study start date was not available; b, Combined tuberculosis meningitis and miliary tuberculosis outcomes; c, Tuberculosis meningitis outcome only; d, Miliary tuberculosis outcome only. D + L, DerSimonian and Laird method; I – V, inverse variance method.

Meta-regression analysis

Stratification by bands of 20° latitude did not explain any of the between-study variance (null model τ² = 0.275; after stratification band τ² = 0.291) in case–control studies (Table 19). There was no clear evidence that BCG vaccination effectiveness in case–control studies varied with latitude (p-value = 0.570). The stratification on 20° latitude group for cohort (Table 20) accounted for the between-study heterogeneity (null model τ² = 0.035; after stratification by 20° band τ² = 0.000), but did not explain the heterogeneity for case population studies: null model τ² = 0.292; after stratification by 20° band τ² = 0.434, respectively) and no evidence (p-value = 0.839 and 0.723, respectively) indicated that BCG vaccination effectiveness differed according to latitude in cohort and case population studies. Similarly, latitude did not account for any of the between-study heterogeneity seen within cross-sectional studies (Table 21), with τ² value before and after stratification on latitude = 1.433 and 2.140. There was no evidence (p-value = 0.734) that BCG vaccination effectiveness varied with latitude (Table 22).

Stratified analysis by age at vaccination, ordered by year study started

The estimated effects of BCG vaccination against meningeal and miliary tuberculosis from observational studies were stratified by age at vaccination (Figures 63–66). Overall, the protective effect of BCG vaccination from case–control studies of school-age vaccination [OR 0.23 (95% CI 0.16 to 0.33), VE of 77% (95% CI 67% to 84%)] was similarly good, and similar to studies of neonatal vaccination [OR 0.31, 95% CI 0.20 to 0.47, equivalent to VE of 69% (95% CI 53% to 80%)]. The small number of cohort studies in each age at vaccination category makes results difficult to interpret (see Figure 64).

FIGURE 63.

Odds ratios (with 95% CI) comparing the BCG vaccination status of meningeal and/or miliary tuberculosis cases and control subjects, stratified by age at vaccination, ordered by year of study start. a, Date of study publication was used if study start date was not available; b, Combined tuberculosis meningitis and miliary tuberculosis outcomes; c, Tuberculosis meningitis outcome only; d, Miliary tuberculosis outcome only. D + L, DerSimonian and Laird method; I – V, inverse variance method.

FIGURE 64.

Rate ratios (with 95% CI) comparing the incidence of meningeal and/or miliary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 4) in cohort studies, stratified by age at vaccination, ordered by study start. a, Tuberculosis meningitis outcome only; b, Miliary tuberculosis outcome only; c, Combined tuberculosis meningitis and miliary tuberculosis outcomes. D + L, DerSimonian and Laird method; I – V, inverse variance method.

FIGURE 65.

Rate ratios (with 95% CI) comparing the incidence of meningeal and/or miliary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 5) in case population studies, stratified by age at vaccination, ordered by year of study start. a, Tuberculosis meningitis outcome only. D + L, DerSimonian and Laird method; I – V, inverse variance method.

FIGURE 66.

Risk ratios (with 95% CI) comparing the prevalence of meningeal and/or miliary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals in cross-sectional studies, stratified by age at vaccination, ordered by year of study start. a, Combined tuberculosis meningitis and miliary tuberculosis outcomes; b, Tuberculosis meningitis outcome only; c, Miliary tuberculosis outcome only. D + L, DerSimonian and Laird method; I – V, inverse variance method.

All but one case population study evaluated neonatal vaccination, with high overall levels of effectiveness (see Figure 65). This study of school-age vaccination showed a substantial high protective effect [rate ratio 0.13 (95% CI 0.02 to 0.74)], similar to the overall effect of neonatal vaccination studies [rate ratio 0.12 (95% CI 0.06 to 0.24) equivalent to VE of 88% (95% CI 76% to 94%)]. Results from cross-sectional studies in Figure 66 also showed strong evidence of high protection from vaccination at birth [RR 0.19 (95% CI 0.12 to 0.29, VE of 81% (95% CI 71% to 88%)]. The Tohoku outpatients study,143 which performed vaccination in other age groups, also shows a high estimate of vaccine effectiveness.

Meta-regression analysis

Based on meta-regression analyses, age at vaccination explained very little of the between-study variability (τ² before and after stratification = 0.275 and 0.266, respectively): see Table 19. Stratification on age at vaccination explained some of between-study heterogeneity within cohorts, in Table 20 (τ² before and after stratification = 0.035 and 0.000, respectively) although heterogeneity was already low prior to stratification. Table 21 showed that age at vaccination did not account for any of the between-study heterogeneity in case population studies (τ² before and after stratification = 0.292 and 0.340, respectively). Age at vaccination was not found to account for any of the heterogeneity seen in cross-sectional studies (τ² before and after stratification = 1.433 and 2.140, respectively), as shown in Table 22. However, there was insufficient evidence that BCG vaccination protection varied according to age at vaccination in case–control, cohort, case population and cross-sectional studies (p = 0.386, 0.475, 0.969, 0.734, respectively).

Stratified analysis by study design, ordered by year study started

Cohort studies

Figure 67 shows the forest plot for cohort studies stratified by study design. Contrary to results on protection against pulmonary and all types of tuberculosis disease, prospective studies of BCG vaccination against miliary and or meningeal tuberculosis showed evidence of overall higher protection against tuberculosis meningitis with rate ratio 0.10 (95% CI 0.02 to 0.50), corresponding to VE of 90% (95% CI 50% to 98%), compared with retrospective studies (which include a study with data on miliary tuberculosis only), with an overall good protective effect rate ratio 0.27 (95% CI 0.14 to 0.52) which corresponds to VE of 73% (95% CI 48% to 86%) VE. Cohort study design thus appears to explain a substantial amount of between-study variability.

FIGURE 67.

Rate ratios (with 95% CI) comparing the incidence of meningeal and/or miliary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 4) in cohort studies, stratified by cohort study design, ordered by year of study start. a, Tuberculosis meningitis outcome only; b, Combined tuberculosis meningitis and miliary tuberculosis outcomes; c, Miliary tuberculosis outcome only. D + L, DerSimonian and Laird method; I – V, inverse variance method.

Meta-regression analysis

Stratification of cohort studies by study design accounted for the between-study heterogeneity (τ² before and after stratification = 0.035 and 0.000) (see Table 20). There was, however, insufficient evidence that BCG vaccination effectiveness varies according to study design in cohort: rate ratio for retrospective cohort studies was 2.79 (95% CI 0.22 to 34.99, p-value = 0.323) times the rate ratio of prospective cohort studies.

Meta-regression analyses

Unstratified analysis ordered by year study started

The results presented in forest plots in Figures 68–71 show estimated effects of BCG vaccination against non-meningeal and non-miliary forms of extrapulmonary tuberculosis from observational studies. Case–control studies showed relatively high protection of BCG vaccination against extrapulmonary tuberculosis, ranging from substantial protection [OR 0.12 (95% CI 0.06 to 0.26)] in Argentina72 to a lower effect [OR 0.63 (95% CI 0.30 to 1.32)] in Bangalore Children. 56 There was more substantial variation in estimates of BCG vaccination effectiveness in cohort studies (see Figure 68), ranging from high levels of protection [rate ratio 0.14 (95% CI 0.10 to 0.19)] in Dusseldorf children,149 to no evidence of clinical benefit in the Lyon students study108 [rate ratio 1.71 (95% CI 0.31 to 9.36)]. High levels of protection against extrapulmonary tuberculosis were seen in all case population (see Figure 70) and cross-sectional studies (see Figure 71).

FIGURE 68.

Odds ratios (with 95% CI) comparing the BCG vaccination status of extrapulmonary tuberculosis cases and control subjects in case–control studies, ordered by year of study start. a, Date of study publication was used if study start date was not available.

FIGURE 69.

Rate ratios (with 95% CI) comparing the incidence of extrapulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 4) in cohort studies, ordered by year of study start.

FIGURE 70.

Rate ratios (with 95% CI) comparing the incidence of extrapulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 5) in case population studies, ordered by year of study start.

FIGURE 71.

Risk ratios (with 95% CI) comparing the prevalence of extrapulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals in cross-sectional studies, ordered by year of study start.

Stratified analysis by latitude (10°), ordered by year study started

Forest plots showing stratifications of observational studies assessing the protective effect of BCG vaccination on extrapulmonary tuberculosis according to latitude are provided in Figures 72 and 73. Only a few studies fell into each latitude category and the findings should therefore be treated with caution. The overall protective effect of BCG vaccination in case–control studies was lower in the four studies conducted close to the equator, whereas there was reasonably consistent evidence of higher protection observed in three studies conducted at latitudes above 20°. There is also evidence from cohort studies of relatively higher protection observed in studies conducted above 50° with substantial variation between cohort studies conducted at high latitudes. Estimated effects from the one case population study (above 50° latitude) showed a higher protection effect compared with one located between 30° and 40° while there was little variation in the effect of BCG vaccination effectiveness according to latitude in the two cross-sectional studies, one at 30–40° latitude and the other above 50° latitude.

FIGURE 72.

Odds ratios (with 95% CI) comparing the BCG vaccination status of extrapulmonary tuberculosis cases and control subjects in case–control studies, stratified by latitude of study location (10° bands), ordered by year of study start. a, Date of study publication was used if study start date was not available. D + L, DerSimonian and Laird method; I – V, inverse variance method.

FIGURE 73.

Rate ratios (with 95% CI) comparing the incidence of extrapulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 4) in cohort studies, stratified by latitude of study location (10° bands), ordered by year of study start. D + L, DerSimonian and Laird method; I – V, inverse variance method.

Meta-regression analysis

Only case–control and cohort studies had a sufficient number of studies of the protection by BCG vaccination against extrapulmonary tuberculosis to perform meta-regression analyses. Based on these results, latitude was found to explain respectively 26% and 42% of the between-study variation (τ² value before and after stratification = 0.262 and 0.193, respectively for case–control studies and τ² values before and after stratification = 0.895 and 0.520, respectively, for cohort studies, Tables 23 and 24). However, there was insufficient evidence, from either study design, that BCG vaccination effectiveness was higher at higher latitudes (estimates of protective effect from case–control studies conducted between 0° and 20° latitude were 2.04 (95% CI 0.66 to 6.31; p-value = 0.168) times that of studies conducted at higher latitudes (20–40°). Similarly, rate ratios from cohort studies of 0–20° latitudes were 3.65 (95% CI 0.48 to 28.03; p-value = 0.2) times the rate ratio for studies of latitudes > 40°.

Stratified analysis by age at vaccination, ordered by year study started

There were insufficient data from observational studies to investigate the difference in effectiveness of BCG vaccination protection against extrapulmonary tuberculosis according to age at vaccination. All case–control subjects and case population studies assessed neonatal BCG vaccination, and only one cross-sectional study provided data on school-age vaccination (Liege children182) and on vaccination in an ‘other’ age group (Tohoku Outpatients study181), respectively. Figure 74 shows the estimated effects of BCG vaccination on extrapulmonary tuberculosis from cohorts stratified by age at vaccination. There was substantial variation in estimates of effect for neonatal and school-age vaccination. Stratification did not appear to explain the between-study variation.

FIGURE 74.

Rate ratios (with 95% CI) comparing the incidence of extrapulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 4) in cohort studies, stratified by age at vaccination, ordered by year of study start. D + L, DerSimonian and Laird method; I – V, inverse variance method.

Meta-regression analysis

Based on meta-regression analyses, stratification by age at vaccination does not account for any of the heterogeneity in cohort studies (τ² before and after stratification, respectively = 0.895 and 1.455). There was therefore no evidence that the effect of BCG vaccination varied according to age at vaccination (p-value = 0.973) (see Table 24).

Stratified analysis by study design, ordered by year study started

Cohort studies

Figure 75 presents estimated effect of BCG vaccination against extrapulmonary tuberculosis excluding meningeal and miliary tuberculosis, stratified by study design. The highest protective effect of BCG vaccination was seen in the pooled estimate for retrospective study design. Study design appears to explain a small amount of the between-study variation, with a substantial amount of variation remaining after stratification.

FIGURE 75.

Rate ratios (with 95% CI) comparing the incidence of extrapulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 4) in cohort studies, stratified by cohort study design, ordered by year of study start. D + L, DerSimonian and Laird method; I – V, inverse variance method.

Meta-regression analysis

Stratification on study design accounted for 9% of the heterogeneity: τ² values = 0.895 and 0.817 before and after stratification, respectively. There was no evidence (p-value = 0.320) that the protective effect of BCG vaccination varied according to cohort study design, as shown in Table 24.

Meta-regression analysis

Unstratified analysis by year study started

Figures 76 and 77 show the results of cohort and cross-sectional studies, respectively, which presented data on the effect of BCG vaccination on mortality. No case–control or case population studies provided data on tuberculosis mortality.

FIGURE 76.

Rate ratios (with 95% CI) comparing the incidence of tuberculosis mortality among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 4) in cohort studies, ordered by year of study start.

FIGURE 77.

Risk ratios (with 95% CI) comparing the prevalence of tuberculosis mortality among BCG vaccinated individuals with that in unvaccinated individuals in cross-sectional studies, ordered by year of study start.

Despite the small number of events in each cohort study, all estimates were consistent with a good to high protective effect of BCG vaccination against tuberculosis mortality. Both cross-sectional studies provided strong evidence of high BCG vaccination effectiveness.

Stratified analysis by latitude (10°), ordered by year study started

Figure 78 shows the estimates of effect of BCG vaccination against tuberculosis mortality from all nine cohort studies stratified by 10° bands of latitude. None were conducted close to the equator. The highest pooled estimate of protection was seen in cohort studies conducted between 30° and 40° latitude [rate ratio 0.11 (95% CI 0.05 to 0.24; VE 88%)], while there was reasonably consistent evidence of lower, albeit still good protection from cohort studies at latitudes further from the equator: rate ratio 0.29 (95% CI 0.11 to 0.77; VE 71%) for studies conducted between 40° and 50° latitude, and rate ratio 0.28 (95% CI 0.17 to 0.46; VE 72%) for studies conducted above 50° latitude.

FIGURE 78.

Rate ratios (with 95% CI) comparing the incidence of tuberculosis mortality among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 6) in cohort studies, stratified by latitude of study location (10° bands), ordered by year of study start. D + L, DerSimonian and Laird method; I – V, inverse variance method.

Meta-regression analysis

Stratifying cohort studies by latitude (20° bands) accounted for a very substantial amount of the heterogeneity between-studies (τ² values before and after stratification 0.128 and 0.000, respectively) (Table 25). There was some evidence (p-value = 0.075) that effectiveness of BCG vaccination against tuberculosis mortality varies with latitude.

Stratified analysis by age at vaccination, ordered by year study started

Figure 79 presents the estimated effects of BCG vaccination, stratified by age at vaccination, for cohort studies. Studies of the effect of BCG vaccination given at school and other ages showed slightly lower but still good overall protective effect of BCG vaccination against mortality, compared with neonatal vaccination studies which showed high levels of protection. Age at vaccination appears to explain a substantial amount of between-study variation. Both cross-sectional studies conducted vaccination in neonates.

FIGURE 79.

Rate ratios (with 95% CI) comparing the incidence of tuberculosis mortality among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 4) in cohort studies, stratified by age at vaccination, ordered by year of study start. D + L, DerSimonian and Laird method; I – V, inverse variance method.

Meta-regression analysis

Among cohort studies, stratification on age at vaccination explained 98% the heterogeneity seen within studies (null model τ² = 0.128, τ² value after stratification = 0.003) (see Table 25). However, there was no evidence that BCG vaccination effectiveness varied with age at vaccination (p-value = 0.248).

Stratified analysis by study design, ordered by year study started

Figure 80 presents the estimated effects of BCG vaccination against tuberculosis mortality, stratified by study design. In general, there was no difference in the effectiveness of BCG vaccination between prospective and retrospective cohort studies. Study design did not explain the between-study heterogeneity of cohorts' estimates of BCG vaccination effectiveness against tuberculosis mortality.

FIGURE 80.

Rate ratios (with 95% CI) comparing the incidence of tuberculosis mortality among BCG vaccinated individuals with that in unvaccinated individuals for the longest duration of follow-up (see Table 4) in cohort studies, stratified by cohort study design, ordered by year of study start. D + L, DerSimonian and Laird method; I – V, inverse variance method.

Meta-regression analysis

Stratification by cohort study design did not explain any of the between-study heterogeneity (null model τ² = 0.128, τ² value after stratification = 0.135) (see Table 25).

Meta-regression analysis

Pulmonary tuberculosis

Ten trials provided sufficient data to investigate the duration of protection afforded by BCG vaccination against pulmonary tuberculosis (Figure 81). In the majority of studies there was evidence of protection during the first 5 years of follow-up, while one study (Chingleput28) found an adverse effect of vaccination during this period. In the majority of studies, the protection afforded by BCG vaccination appeared to decline after the first 5 years. The study with the longest follow-up period (Native American trial5) found evidence of protection up to 20 years. By contrast with the trend in the majority of trials, the negative effect of BCG vaccination in the early years of follow-up in the Chingleput study28 was attenuated, with rate ratios close to 1, in subsequent years. Possible explanations for this result are explored further in the discussion (see Chapter 5).

FIGURE 81.

Vaccine efficacy and rate ratios (with 95% CI) comparing the incidence of pulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals, in RCTs, over time. CCH, Cook County Hospital; TB HH, tuberculosis households.

We examined changes in efficacy over time in each study by estimating the ratio of rate ratios per 5 years of follow-up. Such estimates are based on the assumption that, within each study, there is a linear trend in the log-rate ratio over time, and correspond to the assumption of a straight-line relationship for each study in Figure 81. The forest plots in Figures 82 and 83 display these estimated ratios of rate ratios, before and after excluding the first 5 years of follow-up (during which tuberculosis rates may be influenced by prior infection in studies that did not exclude tuberculin skin test positive individuals at enrolment). Figure 82 shows that, on average, the protective effect of BCG vaccination declined over time (summary ratio of rate ratios = 1.14 per 5 years' follow-up; 95% CI 1.03 to 1.27), with little evidence of between-study heterogeneity (τ² = 0.00). Figure 83 shows that excluding data from the first 5 years of follow-up for each study resulted in a more pronounced decline in the protective effect of BCG vaccination (summary ratio of rate ratios 1.47; 95% CI 1.03 to 1.24), and this effect was also more consistent across studies. Again there was little evidence of between-study heterogeneity (τ² = 0).

FIGURE 82.

Change per 5 years (with 95% CI) in the rate ratio comparing the incidence of pulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals in RCTs for the longest follow-up period (see Table 3), including the first 5 years of data for each study. CCH, Cook County Hospital; D + L, DerSimonian and Laird method; I – V, inverse variance method; TB HH, tuberculosis households; TBPT, Tuberculosis Prevention Trial.

FIGURE 83.

Change per 5 years (with 95% CI) in the rate ratio comparing the incidence of pulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals in RCTs for the longest follow-up period (see Table 3), excluding the first 5 years of data for each study. CCH, Cook County Hospital; D + L, DerSimonian and Laird method; I – V, inverse variance method; TB HH, tuberculosis households; TBPT, Tuberculosis Prevention Trial.

Temporal changes in efficacy

Figures 84 and 85 are scatterplots of the ratio of rate ratios per 5 years of follow-up compared with the overall rate ratio for each study, including and excluding the first 5 years' follow-up. There was little evidence of an association between change in efficacy over time and overall efficacy across studies.

FIGURE 84.

Scatterplot of change per 5 years in rate ratio compared with overall rate ratio, comparing incidence of pulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals, in RCTs, including the first 5 years of data for each study. The area of the circles is proportional to the inverse of the variance of the log-rate ratio comparing vaccinated with unvaccinated individuals. CCH, Cook County Hospital; TB HH, tuberculosis households.

FIGURE 85.

Scatterplot of change per 5 years in rate ratio compared with overall rate ratio, comparing incidence of pulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals, in RCTs, excluding the first 5 years of data for each study. The area of the circles is proportional to the inverse of the variance of the log-rate ratio comparing vaccinated with unvaccinated individuals. CCH, Cook County Hospital; TB HH, tuberculosis households.

Study efficacy by follow-up period

Figures 86–89 are forest plots of the effect of BCG vaccination in each trial, within the successive follow-up periods 0–5 years, 5–10 years, 10–15 years and > 15 years after vaccination. During the first 5 years after vaccination (see Figure 86), there was substantial between-study heterogeneity in the effect of BCG vaccination (τ² = 0.49). Although there was evidence of protection in most studies (ranging from low VE of 31% in the Puerto Rico trial15 to high efficacy of 89% in the Haiti trial55), the Chingleput trial28 found evidence of increased rates of pulmonary tuberculosis in vaccinated individuals (rate ratio 2.07, 95% CI 1.10 to 3.88) (see Chapter 5, Discussion, for explanations).

FIGURE 86.

Rate ratios (with 95% CI) comparing incidence of pulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals, in RCTs, in the first 5 years of follow-up. a, Date of study publication was used if study start date was not available. CCH, Cook County Hospital; D + L, DerSimonian and Laird method; M-H, Mantel-Haenszel method; TB HH, tuberculosis households; TBPT, Tuberculosis Prevention Trial.

FIGURE 87.

Rate ratios (with 95% CI) comparing incidence of pulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals, in RCTs, from 5 to 10 years of follow-up. a, Date of study publication was used if study start date was not available. CCH, Cook County Hospital; D + L, DerSimonian and Laird method; M-H, Mantel-Haenszel method; TB HH, tuberculosis households; TBPT, Tuberculosis Prevention Trial.

FIGURE 88.

Rate ratios (with 95% CI) comparing incidence of pulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals, in RCTs, from 10 to 15 years of follow-up. a, Date of study publication was used if study start date was not available. CCH, Cook County Hospital; D + L, DerSimonian and Laird method; M-H, Mantel-Haenszel method; TBPT, Tuberculosis Prevention Trial.

FIGURE 89.

Rate ratios (with 95% CI) comparing incidence of pulmonary tuberculosis among BCG vaccinated individuals with that in unvaccinated individuals, in RCTs, beyond 15 years of follow-up. a, Date of study publication was used if study start date was not available. This graph contains two additional studies (Puerto Rico Children15 and Chingleput study28) compared with the graphs in Figure 81 with data beyond 15 years. This is because the effects in Figure 81 were averaged over some years. D + L, DerSimonian and Laird method; M-H, Mantel-Haenszel method; TBPT, Tuberculosis Prevention Trial.

Among the seven trials that reported relevant outcome data for the period 5–10 years after vaccination, there was substantial between-study heterogeneity (τ² = 0.57) (see Figure 87). Although there was little evidence of protection in the Chicago infants,48 Georgia/Alabama15 and Chingleput trials,28 estimated vaccine efficacy was 72% in the MRC trial14 and 86% in the Native American trial. 5

Figure 88 displays results from the seven trials that reported relevant data for the period 10–15 years after vaccination. The number of events in the intervention and control arms of most studies was small. There was clear between-study heterogeneity, but this was less marked than for earlier time periods (τ² = 0.19).

Relatively few events were reported for the period after 15 years of study follow-up (see Figure 89). Although there was some evidence of between-study heterogeneity, this was less than for earlier time periods (τ² = 0.13). Only the Native American trial5 found evidence of protection.

All tuberculosis disease outcomes

Ten trials provided data to investigate duration of protection by BCG vaccination against all forms of tuberculosis (Figure 90). In the majority of studies, there was evidence of protection during the first 5 years of follow-up except in the Chingleput trial,28 with protective efficacy appearing to decrease over time. Trials including the Saskatchewan infants13 and the two Chicago studies48 showed an initially high protective effect which decreased within 5 to 10 years. The Chingleput trial28 showed BCG vaccination to be potentially harmful in the first 5 years of study, which attenuated to rate ratios close to 1 in subsequent years [further discussion in the Discussion (see Chapter 5) provides an explanation for this observation]. The remaining four studies showed low initial protective effects with a rapid decline in efficacy over the MRC and Native American studies5,14 showed a highly protective effect of BCG vaccination beyond 10 and 15 years, respectively.

FIGURE 90.

Vaccine efficacy and rate ratios (with 95% CI) comparing the incidence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals, in RCTs, over time. CCH, Cook County Hospital; TB HH, tuberculosis households.

Based on an assumption of a linear trend in log of rate ratio over time, we investigated the change in efficacy over time, by estimating the ratio of rate ratios per 5 years for the whole of the study periods in Figure 91 and excluding the first 5 years of study in Figure 92. Apart from the Chingleput28 and Puerto Rico15 studies, all studies showed an increasing rate ratio (declining protection) for each 5 years of follow-up, with a summary ratio of rate ratios per 5 years of follow-up of 1.16 (95% CI 1.01 to 1.33) and little between-study heterogeneity (τ² = 0.013) (see Figure 91). Figure 92 shows that if data from the first 5 years of follow-up were excluded, the decline in protective effect was more pronounced [summary ratio of rate ratio 1.32 (95% CI 1.10 to 1.59)]. There was again little between-study heterogeneity (τ² = 0.02).

FIGURE 91.

Change per 5 years (with 95% CI) in the rate ratio comparing the incidence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals in RCTs for the longest follow-up period (see Table 3), including the first 5 years of data for each study. CCH, Cook County Hospital; D + L, DerSimonian and Laird method; I – V, inverse variance method; TB HH, tuberculosis households; TBPT, Tuberculosis Prevention Trial.

FIGURE 92.

Change per 5 years (with 95% CI) in the rate ratio comparing the incidence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals in RCTs for the longest follow-up period (see Table 3), excluding the first 5 years of data for each study. CCH, Cook County Hospital; D + L, DerSimonian and Laird method; I – V, inverse variance method; TB HH, tuberculosis households; TBPT, Tuberculosis Prevention Trial.

Temporal changes in efficacy

The scatter plots displayed in Figures 93 and 94 show the change in rate ratio per 5 years' follow-up compared with the overall rate ratio for each study, including and excluding the first 5 years of follow-up. There is little evidence of an association between change in efficacy over time and overall efficacy across studies, similarly to Figures 84 and 85.

FIGURE 93.

Scatterplot of change per 5 years in rate ratio compared with overall rate ratio, comparing incidence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals, in RCTs, including the first 5 years of data for each study. The area of the circles is proportional to the inverse of the variance of the log-rate ratio comparing vaccinated with unvaccinated individuals. CCH, Cook County Hospital; TB HH, tuberculosis households.

FIGURE 94.

Scatterplot of change per 5 years in rate ratio compared with overall rate ratio, comparing incidence of all tuberculosis outcomes among BCG vaccinated individuals with that in unvaccinated individuals, in RCTs, excluding the first 5 years of data for each study. The area of the circles is proportional to the inverse of the variance of the log-rate ratio comparing vaccinated with unvaccinated individuals. CCH, Cook County Hospital; TB HH, tuberculosis households.

Temporal changes in efficacy in relation to overall efficacyStudy efficacy by follow-up period

The forest plots in Figures 95–98 show the effectiveness of BCG vaccination against all forms of tuberculosis, within the successive follow-up periods 0 to 5 years, 5 to 10 years, 10 to 15 years and > 15 years following vaccination. During the first 5 years, there was substantial heterogeneity in the effect of BCG vaccination (τ² = 0.56). While there was evidence of a protective effect, ranging from low protection 31% in Puerto Rico children15 to high protection 89% (95% CI 0.09% to 99%) in Haiti,55 the Chingleput Study28 showed a potentially negative effect of BCG vaccination in first 5 years of study [rate ratio 2.26 (95% CI 1.36 to 3.76)] (see Chapter 5 , Discussion, for an explanation).

FIGURE 95.

Rate ratios (with 95% CI) comparing incidence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals, in RCTs in the first 5 years of follow-up. a, Date of study publication was used if study start date was not available. CCH, Cook County Hospital; D + L, DerSimonian and Laird method; M-H, Mantel-Haenszel method; TB HH, tuberculosis households; TBPT, Tuberculosis Prevention Trial.

FIGURE 96.

Rate ratios (with 95% CI) comparing incidence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals, in RCTs, from 5 to 10 years of follow-up. a, Date of study publication was used if study start date was not available. CCH, Cook County Hospital; D + L, DerSimonian and Laird method; M-H, Mantel-Haenszel method; TB HH, tuberculosis households; TBPT, Tuberculosis Prevention Trial.

FIGURE 97.

Rate ratios (with 95% CI) comparing incidence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals, in RCTs, from 10 to 15 years of follow-up. a, Date of study publication was used if study start date was not available. CCH, Cook County Hospital; D + L, DerSimonian and Laird method; M-H, Mantel-Haenszel method; TBPT, Tuberculosis Prevention Trial.

FIGURE 98.

Rate ratios (with 95% CI) comparing incidence of all tuberculosis morbidity outcomes among BCG vaccinated individuals with that in unvaccinated individuals, in RCTs, beyond 15 years of follow-up. a, Date of study publication was used if study start date was not available. This graph contains one additional study (Chingleput study28) compared with the graphs in Figure 90 with data beyond 15 years. This is because the effects in Figure 81 were averaged over some years. D + L, DerSimonian and Laird method; M-H, Mantel-Haenszel method; TBPT, Tuberculosis Prevention Trial.

Among the eight trials5,13–15,28,48 with data on protection between 5 and 10 years after vaccination, there was substantial between-study heterogeneity (τ² = 0.54) (see Figure 96). Although there was little evidence of protection in the Chicago infants (TB HH),48 Puerto Rico,15 Georgia/Alabama15 and Chingleput trials,28 estimated vaccine efficacy was good, at 73% (95% CI 49% to 85%) in the MRC study14 and high [91% (95% CI 31% to 99%)] in the Saskatchewan trial. 13

Figure 97 shows results from nine trials,5,13–15,28,48,49,53 with relevant data on the protective effect of BCG vaccination between 10 and 15 years after vaccination. The number of events in the vaccinated and unvaccinated arms was small, for most studies; however, data from the MRC14 and Native American trials5 show evidence of substantial protection, with VEs ranging from good [VE 74% (95% CI 49% to 87%)] to high [VE 88% (95% CI 44% to 97%)], respectively. There was evidence of between-study heterogeneity but this was lower than for earlier time periods of follow-up (τ² = 0.16).

Seven studies5,14,15,28,49,53 provided estimates of BCG vaccination efficacy beyond 15 years after vaccination (see Figure 98). There was evidence of between-study heterogeneity (τ² = 0.16), although this was less than for periods before 10 years' follow-up. Only the Native American study5 showed evidence of a protective effect beyond 15 years [rate ratio 0.48 (95% CI 0.32 to 0.72)], equivalent to a moderate VE of 52% (95% CI 28% to 68%), whereas the remaining studies showed very little evidence of protection by BCG vaccination beyond 15 years after vaccination.