Observational study to estimate the changes in the effectiveness of bacillus Calmette Guerin (BCG) vaccination with time since vaccination for preventing tuberculosis in the UK

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 08/17/01. The contractual start date was in May 2011. The draft report began editorial review in November 2015 and was accepted for publication in November 2016. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

Jonathan Sterne was a member of the National Institute for Health Research (NIHR) Health Technology Assessment Clinical Evaluation and Trials Board while the study was being conducted.

Copyright statement

© Queen’s Printer and Controller of HMSO 2017. This work was produced by Mangtani et al. under the terms of a commissioning contract issued by the Secretary of State for Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

Tuberculosis epidemiology

Tuberculosis (TB) remains a significant and preventable cause of morbidity and mortality globally. Approximately 10% of infections with Mycobacterium tuberculosis progress to clinical disease. 1 The World Health Organization2 estimates that > 2 billion of the world’s population is infected. In 2014, 9.6 million people developed symptoms of TB disease and 1.5 million died from TB. 2 In the UK, after many decades during which both the risk of infection with M. tuberculosis and the incidence of TB decreased, the last decade of the 20th century and the first decade of the 21st century saw a steady rise in TB cases. 3 From 2005 to 2011, there were around 8000 cases of TB per year in England. This declined to 6520 cases in 2014, but England still has the highest rate of TB in Western Europe. 4 There has been no decline in TB rates among the UK-born population overall; however, the incidence of childhood TB in UK-born children, including miliary disease and meningitis, has started to fall. TB continues to be concentrated in urban areas, with much higher rates in the most deprived areas and in non-UK-born populations. 5 Drug-resistant TB has increased among culture-confirmed cases in the UK (the percentage resistant to any first-line drug increased from 5.6% in 1998 to 7.5% in 2005), mainly because of a rise in isoniazid resistance,6 and has remained stable. 5 However, the percentage of patients with multidrug resistance has started to fall and is < 1.4%, although complex long-term treatment requirements and poor completion rates make such an outcome an ongoing concern.

Bacillus Calmette–Guérin vaccine effectiveness and UK policies on bacillus Calmette–Guérin vaccination in relation to the changing epidemiology in the UK

Bacillus Calmette–Guérin (BCG) vaccination is widely used globally. In the UK, the vaccine has mainly been given either to infants or to adolescents at school. The protection against pulmonary TB in the UK is high when BCG is given to tuberculin-negative schoolchildren at around age 13 years. This was shown by a trial initiated by the Medical Research Council (MRC) in 19517 and in subsequent analyses of the effectiveness of the vaccine given in the routine school immunisation programme. 8 However, there have been variable findings with respect to the effectiveness of the vaccine against pulmonary disease in different countries or between different studies in the same country. 7,9,10 The effectiveness of the BCG vaccine given in infancy (to prevent pulmonary TB, miliary TB and tuberculous meningitis) has been found to be consistently high in all countries where it has been measured. 11,12 Although the World Health Organization13 recommends not to re-vaccinate, mostly because of lack of evidence of the efficacy of revaccination, many countries implement re-vaccination programmes. Trials in Malawi14 and more recently in Brazil15 found no increase in effectiveness or a modest increase in effectiveness associated with repeat BCG vaccination.

Although the MRC trial of adolescent vaccination with BCG demonstrated high levels of protection in the UK,7 there have been several subsequent policy changes in the UK with respect to BCG vaccination, prompted by changes in the epidemiology of TB. In brief, from 1953, BCG vaccine was given to tuberculin-negative [‘purified protein derivative (PPD)-negative’] schoolchildren at age 10–13 years, as part of the national vaccination programme. In 1972, as the proportion of cases of TB in ethnic minority groups increased, BCG vaccination in infancy was recommended for newborns of recent immigrants from countries with a high incidence of TB (e.g. Indian subcontinent and Africa) as well as all refugees and asylum seekers. It was also given to all newborns in some areas [health districts/primary care trusts (PCTs)] with a high TB incidence.

In 1991, a survey was conducted in the UK of how well the policies for BCG vaccination in the first year of life were implemented. 16 At that time, five districts offered the BCG vaccine to all newborn children, 31 districts offered it to none and 148 districts offered it to infants born to those in ethnic groups from the Indian subcontinent, Africa, the West Indies, China, the Middle East and South-East Asia. Of the 184 districts, 120 reported that they offered the vaccine to the newborn children of recent migrants from other countries with a high incidence of TB.

There was discussion over whether or not BCG vaccination of the general population should be discontinued when the risk of TB decreased, based primarily on the high number of vaccinations needed to prevent one case of TB in the UK and worldwide. 17 The International Union Against Tuberculosis and Lung Disease (IUATLD) developed a set of criteria for the discontinuation of mass BCG programmes in low-prevalence populations. 18 IUATLD recommends that BCG vaccination be discontinued if an efficient TB notification system is in place and:

• the average annual notification rate of smear-positive pulmonary TB is < 5 per 100,000 or

• the average annual notification rate of TB meningitis in children aged < 5 years has been < 1 per 10 million population over the previous 5 years or

• the average annual risk of infection is < 0.1%.

The UK met all of these criteria and the BCG vaccination policy for the UK was changed by the Department of Health in 2005 to the current policy. 19 The school vaccination programme was stopped and BCG vaccination was recommended for infants using a risk-based approach, in line with the IUATLD guidelines. In the UK, infants are eligible for vaccination if they have a parent/grandparents originating from a high TB incidence country or if they are born in a part of the UK with a high incidence of TB (> 40 per 100,000). It is also recommended that some occupational groups, and uninfected contacts of TB cases, receive BCG vaccination. 20

Evidence for the duration of bacillus Calmette–Guérin protection

In the UK MRC trial,7 the efficacy of the BCG vaccine by time since vaccination of adolescents at school was estimated as 84% during the first 5 years after vaccination, 68% at between 5 and 10 years since vaccination and 63% at between 10 and 15 years since vaccination. Although all of these estimates were statistically significantly different from zero, the number of cases at 10–15 years post vaccination was small and the efficacy estimate had a wide 95% confidence interval (CI) (17% to 84%). There were too few cases between 15 and 20 years after vaccination to assess efficacy. A summary of protection by time since vaccination, with 95% CIs (calculated by us based on the trial data presented in the paper), is provided in Table 1. The level of protective effect in the first 10 years after vaccination was confirmed in a subsequent cohort analysis of data from the school-aged BCG vaccination programme in England. 8 There are no data on long-term protection post-infant BCG vaccination in high-risk groups.

The National Institute for Health Research stated, and we agree, that it is not known how long protection from the BCG vaccine lasts, particularly in different age and population groups, and this hinders the development of evidence-based policies. Until recently there was little evidence of protection lasting beyond 10 years after vaccination at any age. In a review of published studies conducted by two of the current authors, the pooled estimate of protection after 10 years was 14% (95% CI –9% to 32%). 21 Considerable heterogeneity was observed between studies in the annual change in BCG vaccine efficacy (VE) with time since vaccination. There was no relation between average annual change in efficacy and overall efficacy. As with most vaccines, immunological memory may wane with time, leading to a lower level of protection. Other explanations proposed include decreasing susceptibility among the unvaccinated because of continued exposure to environmental mycobacteria and an increase in the proportion of disease caused by reactivation or reinfection, against which BCG may not protect. 22

An update of this systematic review of the duration of protection conferred by the BCG vaccine against TB was conducted by our group. 23 The review included the recent additional follow-up of a BCG vaccine trial in Native Americans (who were, on average, aged 7 years when vaccinated in the 1930s), which has reported protection lasting for several decades,24 and a cohort study in the control arm of the Brazilian BCG re-vaccination trial, suggesting that protection lasted for 15–20 years. 25

However, there is evidence from some countries of poor protection of the BCG vaccine in adult life and much of the existing research is of uncertain relevance to the UK. The aims of this research project were to estimate the duration of protection of the BCG vaccine given to high-risk infants in the UK and to school-aged children in the general population. If the study provided evidence of a long duration of protection, beyond 10 years, this would have several implications, including changing the estimates of the cost-effectiveness of the BCG vaccine, the number of vaccinations needed to prevent a case, the possible characteristics of new BCG-like vaccines and the timing of vaccination for any new TB vaccine developed, that is, it would provide evidence for vaccination policies as well as inform the research and development of new TB vaccines.

Primary objectives

• To estimate the effectiveness of BCG vaccination in preventing TB when given in the first year of life to high-risk groups, at 5-year intervals since vaccination.

• To estimate the effectiveness of BCG vaccination in preventing TB when given in adolescence to the general population, at 5-year intervals since vaccination, starting at 10 years since vaccination.

• To explore whether or not protection wanes with time since vaccination in high-risk groups and in the general population.

Health technology assessed

The health technology assessed was BCG vaccination in the UK given to:

• infants at higher risk of TB (referred to throughout as ‘infant BCG’)

• schoolchildren in the general population (referred to throughout as ‘school-aged BCG’).

Overview

Two main observational analytical studies aiming to estimate the effectiveness of the BCG vaccine against TB by time since vaccination were conducted, as well as three supporting studies. The two main studies were case–control studies aimed at estimating the effectiveness of (1) the BCG vaccine given to infants in high-risk groups (results generalisable to high-risk groups in the UK) and (2) the BCG vaccine given at school age to the general population (results generalisable to the UK population). The three supporting studies were (1) a survey of BCG vaccination policy in England, (2) a pilot study for the main observational studies and (3) a validation study of BCG scar reading.

In the two case–control studies, TB cases included in the study were sampled among those notified in the years 2003–12 to the UK national Enhanced Tuberculosis Surveillance (ETS) system. Control subjects (controls) were recruited from the community in the areas where sampled cases had arisen. BCG vaccination status was established based on BCG records when available, scar reading (inspection of both arms) and BCG history (recall of vaccination). Information on potential confounding variables (including demographic and social variables) was collected from cases and controls in a face-to-face computer-assisted interview conducted by trained staff from the National Centre for Social Research (NatCen), a leading centre for independent social research with > 40 years’ experience in nationwide surveys. Clinical and microbiological information, including type of disease for cases, was retrieved from the ETS system.

Ethics

The protocol for all studies, information leaflets and data collection tools were reviewed and approved by the NHS National Research Ethics Service Committee – London and South East (reference number 11/H1102/11) and the London School of Hygiene & Tropical Medicine Observational/Interventional Research Ethics Committee (reference number 5996). NHS research and development permission was obtained with Public Health England (formerly the Health Protection Agency) as the ‘NHS participating organisation’.

We report first on the policy survey (one of the supporting studies) and then on the two main observational studies.

Background

There appeared to be widespread variation at the local level in the implementation of the recommendations for infant BCG vaccinations as well as in vaccine delivery pathways.

A survey of both past and current vaccination policies was conducted to support the main studies by assessing what infant BCG vaccination provision was in place prior to 2005 in local areas and to identify and engage with services or individuals on the delivery pathway who managed or had access to vaccination records.

Methods

We designed a standardised, mostly close-ended structured questionnaire covering both the historical and the current BCG vaccination policy in and outside infancy, eligibility criteria and their documentation, delivery pathways and constraints to service delivery. The questionnaire was tested by asking immunisation co-ordinators from four London PCTs to complete it and questions were then adjusted accordingly.

We surveyed all 152 PCTs, the local administrative areas for health-care services in England between November 2010 and March 2011. We also checked PCT websites and related NHS sources for publicly available documents to assess agreement with the questionnaire data received. Details of the survey are published as a peer-reviewed paper. 26

We also obtained the source data (original questionnaires/tables) from previous BCG policy surveys16,27 from the investigators to complement information on historical infant vaccination policies.

Results

Questionnaires were returned from 85% (129/152) of the PCTs in England. There were no differences in TB notification rates between responding and non-responding PCTs. We found publicly available current BCG policy documents for 114 of the 129 (88%) PCTs that responded. Two (2%) PCTs were excluded from the subsequent analysis because their BCG policies could not be determined clearly from their responses.

Discussion

The data for this survey were collected during and after a major reorganisation of the NHS. This complicated access to key informants, as staff responsibilities at this time were not clear. The potential implications of the recent NHS reorganisation for BCG vaccination policies at the local level are also not now reflected in this report.

Objectives

The primary objectives of the observational studies were to:

• estimate the effectiveness of BCG vaccination in preventing TB when given in the first year of life (‘infant BCG’) to high-risk groups, in 5-year intervals since vaccination

• estimate the effectiveness of BCG vaccination in preventing TB when given at school age (‘school-age BCG’) to the general population, in 5-year intervals since vaccination, starting 10 years after vaccination.

For both exposures it was also explored whether or not protection wanes with time since vaccination.

Methods

Study sampling

Selection of cases

For both studies, eligible TB cases were identified from the ETS database based on the date of diagnosis and reported date of birth, residential address at the time of diagnosis and self-reported ethnic background. Cases with a reported previous TB episode or with a previous notification of TB in the database were not included. For the infant BCG study, cases were included if they resided in the study area (defined to include small areas where ≥ 30% of the population were from BAME groups). For the school-age BCG study, the study area included all of England.

Selection of controls

The procedure for selection of controls was amended after the pilot study. We required community-based controls who represented the population in which the cases occurred. We had two potential strategies for this: the first, which was lower cost but high risk, was nominated controls. We piloted this strategy to recruit individually matched controls among unrelated acquaintances nominated by cases. The pilot study indicated that recruiting nominated controls was not feasible: people either were reluctant to nominate friends or reported that they did not have friends of an eligible age or ethnicity. The control recruitment strategy was thus changed to our second-choice strategy (more resource intensive but lower risk): self-weighted multistage stratified sampling of the target populations across the area from which cases were recruited. Multistage sampling was preferred to straightforward simple random sampling (SRS) to ensure wider geographical coverage of the respective study areas while maintaining reasonable clustering of field data collection for optimal logistical efficiency.

For the infant BCG study we used a two-stage self-weighted sampling design. Based on previous work by the Health Survey for England28,29 and the 2001 census,30 we estimated that, on average, screening 12 residential addresses would result in one eligible (based on BAME group and birth cohort) control successfully recruited from the community. A total of 7750 addresses were selected probability proportional to size (PPS) of the eligible BAME population in the study area, which consisted of geographical areas with ≥ 30% BAME residents based on the 2001 census. Intercensus estimates were not thought to be as robust and were not available by small areas.

The first stage consisted of sampling the lower-level super output areas (LSOAs) with a ≥ 30% resident BAME population by PPS. LSOAs are designed to have a fairly socially homogeneous population of, on average, 1500 residents each. 31 A total of 2659 out of 32,482 (8.2%) LSOAs had ≥ 30% BAME residents, accounting for 50% (2,027,398/4,024,287) of the total BAME population in England. The study area also included 60% of all eligible TB cases in these groups notified to the ETS system.

The second stage involved SRS of seven residential addresses within each LSOA selected in the first stage. This was carried out using the small-user Postcode Address File. To ensure equal geographical spread of selected addresses within the LSOA, we randomly sampled seven distinct postcode units in each LSOA and then one address per postcode unit. Postcode units each include, on average, 15 residential addresses and LSOAs include, on average, 30 postcode units each.

Overall, 1107 LSOAs were sampled with PPS out of 2659 LSOAs with ≥ 30% BAME residents and, in each selected LSOA, seven residential addresses were selected using SRS.

For the school-age BCG study we used a three-stage self-weighted sampling design to reflect the much wider study area (the whole of England). We estimated that a total of about 9500 residential addresses would have to be screened to meet our target, based on an average of one eligible control successfully recruited from every seven to eight addresses screened, based on the Health Survey for England28,29 and 2001 census30 data broken down by age and ethnicity.

The first stage was the selection of 449 mid-level super output areas (MSOAs) out of a total of 6781 in England, with PPS of their 2010 mid-year estimates of the population aged 25–49 years. 32 MSOAs consist of contiguous LSOAs, constrained by the 2003 local authority boundaries; each has a minimum population of 5000 and an average of 7200 inhabitants. 31

The second stage was the random selection of three LSOAs in each MSOA, by PPS.

The third stage was the selection of seven residential addresses from each LSOA by SRS using the same procedure as for the infant BCG study.

In total, we selected 449 out of 6781 MSOAs and a total of 1347 LSOAs across England (three per MSOA). Seven randomly sampled addresses were selected in each LSOA. A few addresses (n = 6) no longer existed, leaving a total of 9423 addresses screened for eligible controls.

Summaries of the sampling strategies for the infant BCG study and the school-age BCG study are provided in Figures 1 and 2, respectively.

FIGURE 1.

Summary of the sampling strategy for the infant BCG study. a, LSOA sampled at random using probability proportional to size of the number of residents from ethnic minority background in the LSOA.

FIGURE 2.

Summary of the sampling strategy for the school-age BCG study.

Results: pilot study

Results: concordance between different measures of bacillus Calmette–Guérin vaccination

These results are based on eligible participants successfully recruited to the respective studies, as detailed in the subsequent sections. For the infant BCG study, it was found that participants had difficulties distinguishing BCG vaccination in infancy from other childhood vaccines. As the self-reported vaccination information was clearly of poor quality, the three other BCG indicators were examined: scar inspection, personally held records (red book or vaccination card) and NHS records. For the school-age BCG study, the quality of recall was better, probably because vaccination was offered at an older age, that is, to schoolchildren aged 12–13 years on average.

Results: infant bacillus Calmette–Guérin study

Overview of recruitment

Recruitment of cases

Of the 1390 potentially eligible cases from the ETS system who were invited to take part in the study, it was possible to contact 1138. Of these, 6% were excluded because they were not born in the UK, they were away, as reported by a household member, or, during interview, they were noted to have been vaccinated because of contact with a case of TB. Of the 1076 subjects contacted and eligible, 797 (74%) were enrolled (Figure 3).

FIGURE 3.

Overview of recruitment of infant BCG cases. a, No contact because changed address (n = 142), no contact with anybody at the address provided (n = 87), moved from England to another part of the UK (n = 10), moved abroad (n = 8) or unspecified (n = 5). b, Ineligible because not born in the UK (n = 20), not from a BAME group (n = 2), reported contact with a TB case in the red book and consistent with the date of TB diagnosis on the ETS system (n = 2) or unspecified (n = 5). c, Other difficulties include away or in hospital during the survey (n = 22), too frail to take part (n = 5) and various other difficulties (n = 6). d, In total, 797 cases were successfully recruited but 53 were not included in the analysis because they developed TB before they had a chance to receive the BCG vaccination (i.e. before their first birthday).

Recruitment of controls

We sampled 7755 residential addresses, of which 1089 (14%) could not be screened to establish eligibility for the study because they could not be located or were inaccessible, or because no contact could be established following the standard visit pattern of visiting more than once on different days and more than once at different times. Among the 6666 addresses that were screened, 1073 (16%) had at least one eligible resident, with 694 (65%) eligible subjects successfully recruited to the control group (Figure 4).

FIGURE 4.

Overview of recruitment of infant BCG controls. a, Not screened because no contact with anybody at the address provided after several visits following a preset visit pattern. b, Includes 4861 addresses screened with no eligible subjects and 720 addresses with non-residential, vacant, derelict or demolished buildings.

Comparison of recruitment between cases and controls

The proportion of cases who could be contacted was fairly comparable across quintiles of area-level IMD (Table 13). Among those contacted and eligible, the refusal rate did not vary much by quintile of IMD. Similarly, the proportion of addresses successfully screened for eligibility was similar across quintiles of IMD. Among subjects identified as being eligible to be controls, there was a similar trend for refusal as for cases, but the refusal rate was consistently higher for controls than for cases.

TABLE 13 - Comparison of recruitment of infant BCG cases and controls

Area-level deprivation quintiles	Group
	Cases		Controls
	Cases invited to participate		Addresses screened for eligibility
	Total, n	Contacted, n (%)	Total, n	Screened, n (%)
Least deprived	159	134 (84)	1556	1377 (88)
2	219	174 (79)	1559	1379 (88)
3	273	217 (79)	1544	1305 (85)
4	330	283 (86)	1547	1307 (84)
Most deprived	407	328 (81)	1549	1298 (84)
	Eligible cases contacted		Eligible controls contacted
	Eligible, n	Refused, n (%)	Eligible, n	Refused, n (%)
Least deprived	125	38 (30)	217	82 (38)
2	160	42 (26)	223	87 (39)
3	205	55 (27)	186	65 (35)
4	271	67 (25)	197	71 (36)
Most deprived	312	76 (24)	250	74 (30)

The distribution of interviewer visits by time of the day and day of the week was also comparable between cases and controls (Figure 5).

FIGURE 5.

Distribution of interviewer visits by (a) time of the day and (b) day of the week for invited cases and sampled control addresses in the infant BCG study.

Fifty-three subjects who developed TB in the first year of life were censored (i.e. exited the study) before the start of ‘follow-up’, as some may have developed TB before having the opportunity to receive the infant BCG vaccination (which could have been offered at any point before the age of 1 year); hence, these subjects were not included in the analysis. Table 14 shows the number of cases by age range at the time of TB diagnosis and birth cohort.

TABLE 14 - Distribution of casesa and controlsb in the infant BCG study: number of cases by birth cohort and age range at the time of TB diagnosis and number of controls from the same birth cohort

The study included cases notified between 2003 and 2012 and aged 1–19 years at the time of diagnosis; therefore, some birth cohorts do not contribute cases at some follow-up times (left-truncation, e.g. birth cohort 1985–9 would be aged 13–19 years at the start of ‘follow-up’ in 2003 and so subjects from the birth cohort who developed TB at age 1–12 years would not be included).

Controls could contribute observation time to more than one time period; hence, columns by time since vaccination for controls do not sum to the column totals for controls since vaccination (see Statistical methods for details).

Birth cohort	Group	Total	Time interval since vaccination (years)
			1–4	5–9	10–14	15–19
1985–9	Cases	54	–	–	9	45
	Controls	67	0	0	33	67
1990–4	Cases	188	–	2	72	114
	Controls	130	0	55	130	130
1995–9	Cases	202	3	59	109	31
	Controls	151	61	151	151	90
2000–4	Cases	189	93	73	17	–
	Controls	168	168	168	56	0
2005–11	Cases	117	103	14	–	–
	Controls	178	178	67	0	0

Trends in the association between time since bacillus Calmette–Guérin vaccination and risk of tuberculosis

Table 20 shows the results from modelling of the association between BCG vaccination and the log-hazard of TB as a linear function of time since vaccination, centred at 5 years post vaccination. The results are displayed graphically in Figure 6. The model suggests a 12% increase (95% CI –1% to 26%) in the HR with each year post vaccination, although this was of only borderline statistical significance. We also explored the trend in time using restricted cubic splines with three knots, at 5, 10 and 15 years post vaccination. The results were similar to those found using the linear model and are not shown here. The results suggest a statistically significant protective effect of the vaccine up to about 10 years post vaccination, with a gradual reduction in the protective effect up to that time. After 10 years post vaccination, the confidence bounds are wide and there is insufficient information to draw conclusions about vaccine effectiveness.

TABLE 20 - Association between BCG vaccination status and risk of TB in the infant BCG study as a smooth function of time since vaccination, using the multivariable adjustment model

Time since BCG vaccination	HR (95% CI)	p-value
Unvaccinated	1 (reference)
HR at 5 years post vaccination	0.32 (0.16 to 0.64)	0.001
Multiplying factor for each year increase in time since vaccination	1.12 (0.99 to 1.26)	0.062

FIGURE 6.

Results from modelling of the time-varying effect of the vaccine as a linear function of time (on the log-scale) in the infant BCG study. The left-hand vertical axis shows the HR and the right-hand vertical axis shows the VE, both on the log-scale. The results are based on the multivariable adjusted model. The dashed lines show the 95% confidence bounds.

Results: school-age bacillus Calmette–Guérin study

Overview of recruitment

Recruitment of cases

A total of 1602 potentially eligible cases was identified from the ETS system and invited to take part in the study, of whom 1047 (65%) were successfully contacted. About 11% were not included either because they were ineligible or because of other difficulties (Figure 7). Of those contacted and eligible, 257 (28%) declined to participate and 677 (72%) were enrolled.

FIGURE 7.

Overview of recruitment of school-age BCG cases. a, No contact because changed addresses (n = 353), no contact with anybody at address provided (n = 165), moved from England to another part of the UK (n = 10), moved abroad (n = 12), inaccessible (n = 6) or unspecified (n = 9). b, Ineligible because not born in the UK (n = 13), not white (n = 23), reported vaccination at a later age and following contact with a TB case in the red book and consistent with the date of TB diagnosis on the ETS system (n = 4) or unspecified (n = 20). c, Other difficulties includes away or in hospital during the survey (n = 23), too frail to take part (n = 20), died (n = 6) and various other difficulties (n = 4).

Recruitment of controls

We sampled 9424 residential addresses, of which 1248 (13%) could not be screened because the address no longer existed or no-one was contactable at the household after several visits to establish if any of the residents were eligible for the study (Figure 8). Among the 8176 addresses that were screened, 1790 (22%) had at least one eligible resident, with 1170 (65%) eligible subjects successfully recruited to the control group.

FIGURE 8.

Overview of recruitment of school-age BCG controls. a, Not screened because no contact with anybody at the address provided after several visits following the preset visit pattern. b, Includes 5692 addresses screened with no eligible subjects and 694 addresses with non-residential, vacant, derelict or demolished buildings.

Comparison of recruitment between cases and controls

The proportion of cases who could be contacted was slightly lower in the more deprived quintiles of area-level IMD (Table 21). Among those contacted and eligible, the refusal rate was slightly lower in the lower quintiles of area-level IMD. The proportion of addresses successfully screened for eligible controls was similar across quintiles of IMD. Among subjects identified as eligible to be controls, the refusal rate was similar across quintiles. The refusal rate was slightly higher for controls than for cases.

TABLE 21 - Comparison of recruitment of school-age BCG cases and controls

Seventeen could not be allocated a deprivation index as the postcode no longer existed.

Area-level deprivation quintiles	Group
	Cases		Controls
	Cases invited to participatea		Addresses screened for eligibility
	Total, n	Contacted, n (%)	Total, n	Contacted, n (%)
Least deprived	165	119 (72)	1889	1634 (87)
2	218	157 (72)	1885	1599 (85)
3	233	148 (64)	1886	1631 (86)
4	394	254 (64)	1886	1659 (88)
Most deprived	575	357 (62)	1878	1653 (88)
	Eligible cases contacted		Eligible controls contacted
	Eligible, n	Refused, n (%)	Eligible, n	Refused, n (%)
Least deprived	107	35 (33)	429	155 (36)
2	146	36 (25)	363	130 (36)
3	137	35 (26)	358	127 (35)
4	217	63 (29)	359	113 (31)
Most deprived	316	86 (27)	281	95 (34)

The distribution of interviewer visits by time of the day and day of the week was comparable between cases and controls (Figure 9).

FIGURE 9.

Distribution of interviewer visits by (a) time of the day and (b) day of the week for invited cases and sampled control addresses in the school-age BCG study.

We found that retrospective ascertainment of the results of the TST was challenging in the school-aged BCG study, with very poor recall and no records to support or validate participants’ self-reports. Further investigation of the literature47 indicated that the initially high TB risk in those who were PPD positive dropped to be the same as that for those who were PPD negative and unvaccinated (see Figure 11). Thus, we included all those who were unvaccinated, irrespective of PPD status prior to the offer of school-aged BCG vaccination.

Trends in the association between time since bacillus Calmette–Guérin vaccination and risk of tuberculosis

Table 27 shows the results from modelling of the association between BCG vaccination and log-hazard of TB as a linear function of time since vaccination, centred at 10 years post vaccination. The results are displayed graphically in Figure 10. The model suggests a 7% increase (95% CI 0.2% to 12%) in the HR with each year post vaccination and this was statistically significant. We also explored the trend in time using restricted cubic splines with three knots, at 15, 20 and 25 years post vaccination. The results were similar to those found using the linear model and are not shown here.

TABLE 27 - Association between BCG vaccination status and risk of TB in the school-age study as a smooth function of time since vaccination

The partially adjusted model is stratified on birth cohort and adjusted for the confounding variables of sex, area-level deprivation and educational level.

The fully adjusted model is further adjusted for the confounding variables of area-level deprivation and educational level, lifestyle variables (tobacco smoking, alcohol consumption and misuse/abuse of controlled drugs), history of homelessness, history of prison stays and TB infection risk from regular travels abroad.

Time since BCG vaccination	Adjusted model
	Partiallya		Fullyb
	HR (95% CI)	p-value	HR (95% CI)	p-value
Unvaccinated	1 (reference)		1 (reference)
HR at 10 years post vaccination	0.310 (0.188 to 0.511)	0.000	0.374 (0.215 to 0.652)	0.001
Multiplying factor for each year increase in time since vaccination	1.07 (1.00 to 1.12)	0.015	1.057 (1.002 to 1.116)	0.042

FIGURE 10.

Results from modelling of the time-varying effect of the vaccine as a linear function of time (on the log-scale) in the school-age study. The left-hand vertical axis shows the HR and the right-hand vertical axis shows the VE, both on the log-scale. The results are based on the multivariable adjusted model. The dashed lines show the 95% confidence bounds.

The results showed a statistically significant protective effect of the vaccine up to about 23 years post vaccination, with a gradual reduction in the protective effect up to that time. The protective effect of the vaccine appears to reduce more steeply after about 20 years post vaccination. However, we had only a relatively small number of cases with > 25 years post vaccination and, thus, have insufficient evidence to assess protection beyond that time.

Discussion

The results from our studies of the duration of protection of BCG vaccination indicate that protection persists for at least 10 years for infant vaccination in a population at high risk for TB and for at least 20 years for school-age vaccination in a low-risk population. A protective effect of 48% (95% CI 17% to 68%) was seen 10–15 years after school-age BCG vaccination and a protective effect of 55% (95% CI 30 to 71%) was seen 15–20 years after school-age vaccination, beyond which protection appeared to wane. For infant BCG vaccination, a protective effect was seen for up to 10 years post vaccination (< 5 years post vaccination: VE 66%, 95% CI 17% to 86%; 5–10 years post vaccination: VE 75%, 95% CI 43% to 89%), but there was less evidence of an effect 10–15 post vaccination (VE 36%, 95% CI negative to 77%; p = 0.396). In the infant BCG study, in the analysis adjusted for several confounders, including birth cohort and ethnicity, the results were based on subjects for whom vaccine records were available. By adjusting only for ethnicity and birth cohort, slightly more records with complete data could be included, giving weak evidence of 50% VE (95% CI negative to 78%; p = 0.096) 10–15 years post vaccination. The higher than expected infant BCG vaccine uptake rate in this high-risk ethnic minority study population and the sparsity of vaccine record data in the later time periods precluded further assessment.

The study findings are consistent with recent findings from Norway41 and Brazil,25 and provide additional evidence to that from the seminal MRC trial,7 in which protection of 63% was seen 10–15 years after vaccination, with wide CIs (95% CI 17% to 84%), with no evidence of protection 15–20 years after vaccination (VE 9%, 95% CI negative to 71%).

The BCG vaccine composition and changes in administration were not considered to be important sources of variation over the study period. Although the Danish liquid BCG vaccine was replaced in the UK in the 1960s by the Glaxo freeze-dried BCG vaccine (Glaxo 1077 vaccine strain developed by Glaxo from the Danish strain and produced in the UK by Evans-Medeva), a study comparing both showed non-inferiority. 42,43 From 2002, the UK has been able to source the BCG vaccine only from Denmark’s Statens Serum Institut (SSI vaccine, also based on the Danish strain 1331). 44 A recent systematic review of the BCG vaccine trials by our research group also suggested that protection did not vary by vaccine strain. 10 Multipuncture vaccination use was limited. In our analysis of policy we noted that the 1983 survey indicated that 76% of districts used intradermal BCG vaccination, which increased in the 1992 survey to 96% of districts. No data were available comparing protection using multipuncture with protection using intradermal vaccination, but sensitisation based on the TST was similar. 45,46

There was a good response rate in both studies and we were able to successfully recruit population-based controls to represent both the children born in the UK to populations from high TB burden settings and the general population. We were unable to link to vaccine records for most of the school-aged BCG subjects, despite their willingness to consent to such linkage. Self-reported history of BCG vaccination and scar inspection were used instead to measure BCG vaccine uptake, having been found to have good agreement with each other (this agreement was not observed, by contrast, in those offered the BCG vaccine in infancy). Although interviewers were not blind to case–control status or to knowledge of a history of BCG vaccination before scar reading, they were specifically trained to identify a BCG scar, which included examining volunteers with and without scars. In addition, 10% of interviews selected at random were checked by senior staff and standardised reporting was required. The interviewers were aware that the study was to assess the duration of protection from BCG. As there was little information on how long BCG protects, it is unlikely that the interviewers would have been biased in one direction with the presence or absence of a BCG scar. There was some confounding because of lower BCG vaccine uptake in poorer subjects, who had a higher risk of TB. We were, however, able to control for this.

We were able to evaluate the effectiveness of the BCG vaccine in BAME population groups despite initial concerns that this group would be difficult to assess. Consent was obtained to link to vaccine records, but records could be retrieved for only about half of the participants, from either NHS administrative areas or patient-held records. There was a considerable lack of agreement between the different sources of information on infant BCG vaccination. In the infant study, self-reported BCG vaccine uptake was also found to be a poor measure of vaccination and scar reading was poorly concordant with vaccine records, when available. Red book and NHS records also had poor concordance, indicating that BCG vaccination was being noted in one record but not necessarily the other. We also noted a higher refusal rate for reading a BCG scar in some BAME young adult female cases, who were not comfortable having their upper arms examined compared with controls. For reasons of poor concordance of the two sources of records and probably biased assessments of scars, we therefore used vaccine records when available, despite not having information on uptake for about 40% of subjects. Future work may include exploring different approaches to determine vaccination status, which could include scar reading, despite the relatively poor specificity of scar reading noted in this study, to reduce the level of missing information on vaccination status in the infant vaccination study.

Addressing the issue of prior infection as the reason for not receiving bacillus Calmette–Guérin vaccination

A theoretical limitation in the school-age BCG study was the inability to assess, and exclude, controls for a prior positive TST. Retrospective ascertainment of TST results based on recall was clearly not feasible and there were no records to validate participants’ memory recall. Under the routine school vaccination programme, prior infection by M. tuberculosis or sensitisation by environmental mycobacteria was investigated through the TST and such subjects were not offered the BCG vaccine. These subjects are usually not included in studies measuring the effect of the BCG vaccine in the short term, as there is evidence that the risk of TB is higher in TST-positive subjects during the first few years after infection. Their inclusion would, thus, act to overestimate the protective effect of BCG vaccination. It is also thought that the BCG vaccine does not confer protection in TST-positive subjects. However, follow-up data from the UK MRC BCG trial of adolescents7,47 showed that, at least in settings with a low transmission rate, the risk of TB in TST-positive participants declined over time and was similar to that of subjects who were TST negative at baseline by about 10 years after enrolment (Figure 11). Thus, the TST was thought unlikely to play a major role when measuring the association between BCG vaccination and TB beyond 10 years after vaccination, as in our study. Extrapolation of earlier modelling work by one of the authors of the paper (extrapolation not published)48 and this report also suggested that the prevalence of tuberculin positivity in the white population in general would have been around 4% at the time of screening for vaccination.

FIGURE 11.

Comparison of TB rates in vaccinated and unvaccinated subjects by TST status at the start of follow-up in the UK MRC BCG trial of adolescents. TU, tuberculin units. Figure based on data from the UK MRC trial published in the Fourth Report to the Medical Research Council by its Tuberculosis Vaccines Clinical Trials Committee. 47

Conclusions

In summary, our two observational studies suggest that BCG vaccination provides longer-lasting protection than previously described, particularly for BCG vaccination at school age, for which we were able to use scar identification and history as measures of BCG uptake. Protection was noted for more than two decades after vaccination, but waned 20–25 years after vaccination. The findings are consistent with the limited data emerging from other trials and observational studies,23 as well as from a more recent Norwegian historical cohort study of the duration of protection of the BCG vaccine. 41 BCG vaccination at school age may thus have helped in the control of TB, including reducing the potential for developing multidrug-resistant disease, as those vaccinated at about 13 years of age have moved into adulthood.

Our current analysis of the infant BCG study, based on just over 50% of the subjects with vaccine records and complete information on covariates, was able to confirm the known protective effect of infant BCG vaccination against TB for up to 10 years after vaccination, adjusting for confounders. However, numbers were too sparse after 10 years to provide robust evidence of VE for this time period. Vaccine coverage was higher than expected in those with records. Further exploratory analyses to deal with missing records could include multiple imputation of missing BCG vaccination status, among other sensitivity analyses. This assessment might inform a review of incidence levels when countries are recommended to suspend BCG vaccination, as well as estimates of the cost-effectiveness of the BCG vaccine in the prevention of TB, and might have implications for the testing and scheduling of new vaccines against TB, which will need to provide better protection than the BCG vaccine and be as durable. More systematic recording of BCG vaccine uptake at the population level by ethnic group using the CHIS and linkage to TB events could help to provide ongoing data for assessing protection and the duration of protection over the life course.

Contributions of authors

Punam Mangtani co-designed and cowrote the proposal with Laura Rodrigues, provided the managerial lead for the project with Patrick Nguipdop Djomo, provided academic leadership and cowrote a first draft of the report.

Patrick Nguipdop-Djomo provided the managerial lead for the project with Punam Mangtani, provided academic leadership, assisted with the analyses, carried out data cleaning and merging of data across sources, and cowrote a first draft of the report.

Ruth H Keogh devised and carried out the analyses, carried out data cleaning and merging of data across sources, provided valued academic expertise and advice at key points, and cowrote a first draft of the report.

Lucy Trinder assisted with the analyses, carried out data cleaning and merging of data across sources, and provided valued academic expertise and advice at key points.

Peter G Smith provided valued academic expertise and advice at key points, and provided additional expertise in BCG vaccine epidemiology and study design.

Paul EM Fine provided valued academic expertise and advice at key points, and provided additional expertise in BCG vaccine epidemiology and study design.

Jonathan Sterne provided valued academic expertise and advice at key points, and provided additional expertise in statistics.

Ibrahim Abubakar provided valued academic expertise and advice at key points, and provided additional expertise in TB epidemiology, the ETS system, BCG vaccine records, public health policy and clinical aspects.

Emilia Vynnycky provided valued academic expertise and advice at key points, and provided additional expertise in exploring PPD positivity levels in the general population.

John Watson provided valued academic expertise and advice at key points, and provided additional expertise in TB epidemiology, the ETS system, BCG vaccine records, public health policy and clinical aspects.

David Elliman provided valued academic expertise and advice at key points, and provided additional expertise in TB epidemiology, the ETS system, BCG vaccine records, public health policy and clinical aspects.

Marc Lipman provided valued academic expertise and advice at key points, and provided additional expertise in TB epidemiology, the ETS system, BCG vaccine records, public health policy and clinical aspects.

Laura C Rodrigues co-designed and cowrote the proposal with Punam Mangtani, and provided academic leadership.

All authors contributed to the final report.

Publications

Pilger D, Nguipdop-Djomo P, Abubakar I, Elliman D, Rodridgues LC, Watson JM, et al. BCG vaccination in England since 2005: a survey of policy and practice. BMJ Open 2012;2:e001303.

Nguipdop-Djomo P, Mangtani P, Pedrazzoli D, Rodrigues LC, Abubakar I. Uptake of neonatal BCG vaccination in England: performance of the current policy recommendations. Thorax 2014;69:87–9.

Data sharing statement

A copy of the final data will be kept on the London School of Hygiene & Tropical Medicine’s data archive [LSHTM Data Compass; see http://datacompass.lshtm.ac.uk/ (accessed 24 March 2017)] and will be available after the main results are published. At this time it can be obtained from the corresponding author through the London School of Hygiene & Tropical Medicine repository.

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health.