Long-term impact of infant immunization on hepatitis B prevalence: a systematic review and meta-analysis

Abstract Objective To conduct a systematic review and meta-analysis of the long-term impact of infant vaccination on the prevalence of hepatitis B virus (HBV) infection at the population level. Methods We searched online databases for articles reporting comparisons between population cohorts aged ≥ 15 years who were exposed or unexposed to infant HBV immunization programmes. We categorized programmes as universal or targeted to infants whose mothers were positive for hepatitis B surface antigen (HBsAg). We included studies reporting prevalence of hepatitis B core antibody (HBcAb), HBsAg, or both. We evaluated the quality of the study methods and estimated the relative reduction in the prevalence of infection. Findings Of 26 studies that met the inclusion criteria, most were from China (20 studies). The prevalence of HBV infection in unvaccinated and universally vaccinated cohorts ranged from 0.6% (116 of 20 305 people) to 16.3% (60/367) and from 0.3% (1/300) to 8.5% (73/857), respectively. Comparing cohorts with universal vaccination to those without vaccination, relative prevalences were 0.24 (95% confidence interval, CI: 0.16–0.35) for HBsAg and 0.23 (95% CI: 0.17–0.32) for HBcAb. For populations with targeted vaccination, relative prevalences were 0.32 (95% CI: 0.24–0.43) and 0.33 (95% CI: 0.23–0.45), respectively. Conclusion The residual burden of infection in cohorts offered vaccination suggests that longer-term evaluations of vaccination coverage, timeliness and other aspects of programme quality are needed. As HBV-vaccinated infant cohorts reach adulthood, ongoing analysis of prevalence in adolescents and young adults will ensure that elimination efforts are on track.


Introduction
Infection with the hepatitis B virus (HBV) is a major global cause of ill health that particularly affects low-and middleincome countries. An estimated 257 million people have chronic HBV infection, and 686 000 deaths occur annually due to long-term complications including liver cirrhosis and hepatocellular carcinoma, a number that is projected to increase. 1,2 Most chronic infection is acquired in infancy or early childhood, primarily through mother-to-child transmission. 3,4 Exposure later in life, through sexual intercourse or blood contact, can also lead to chronic infection but more frequently results in viral clearance and immunity.
Vaccines against HBV infection are highly effective. A plasma-derived vaccine, first used in a national infant immunization programme in 1984, 5 was gradually replaced by recombinant vaccines, which can be manufactured at greater scale. 6 The global recommended schedule is a dose at birth, ideally within 24 hours, followed by two to three doses at monthly intervals. 7 A few countries have reported declines in the prevalence of HBV infection following the implementation of infant vaccination programmes, 8,9 and declining incidence of liver cancer in children and young adults. [10][11][12][13][14] The World Health Organization's (WHO's) Member States aspire to the goal of global elimination of HBV infection and its consequences, with the central strategy being infant vaccination programmes to achieve direct and herd protection. 15 WHO has set an elimi-nation target of a 90% reduction in prevalence by 2030, 16 and specified two primary target indicators: cumulated incidence of HBV infection in children aged 5 years; and deaths from hepatocellular carcinoma, cirrhosis and chronic liver diseases attributable to HBV infection. 17 Although these two indicators effectively represent programme goals, neither are straightforward to monitor in a consistent way over time. Incidence in 5-year-olds is estimated through surveys that require the collection of specimens from a representative sample, which can be difficult to obtain in this age group. 9 Cause-specific mortality is difficult to measure reliably, particularly in countries with constrained resources. It could take decades for mortality data to fully reflect vaccinerelated improvements, due to the long latency of chronic HBV infection in the causation of liver disease.
Prevalence of infection in the wider population is more straightforward to measure and is an earlier and more specific indicator of impact of immunization programmes than measuring prevalence in 5-year-olds alone. Population prevalence is also the core indicator recommended as the first priority by WHO. 17 With infant vaccination now in place for several decades in several countries, it is timely to consider how well it is achieving reductions in HBV infection. This is particularly important as vaccinated generations enter adulthood, a period of higher risk of exposure through sexual activity, 18 and become the potential source of transmission to the next generation. We therefore conducted a systematic review and meta-analysis of studies assessing changes in the prevalence Objective To conduct a systematic review and meta-analysis of the long-term impact of infant vaccination on the prevalence of hepatitis B virus (HBV) infection at the population level. Methods We searched online databases for articles reporting comparisons between population cohorts aged ≥ 15 years who were exposed or unexposed to infant HBV immunization programmes. We categorized programmes as universal or targeted to infants whose mothers were positive for hepatitis B surface antigen (HBsAg). We included studies reporting prevalence of hepatitis B core antibody (HBcAb), HBsAg, or both. We evaluated the quality of the study methods and estimated the relative reduction in the prevalence of infection. Findings Of 26 studies that met the inclusion criteria, most were from China (20 studies). The prevalence of HBV infection in unvaccinated and universally vaccinated cohorts ranged from 0.6% (116 of 20 305 people) to 16.3% (60/367) and from 0.3% (1/300) to 8.5% (73/857), respectively. Comparing cohorts with universal vaccination to those without vaccination, relative prevalences were 0.24 (95% confidence interval, CI: 0.16-0.35) for HBsAg and 0.23 (95% CI: 0.17-0.32) for HBcAb. For populations with targeted vaccination, relative prevalences were 0.32 (95% CI: 0.24-0.43) and 0.33 (95% CI: 0.23-0.45), respectively. Conclusion The residual burden of infection in cohorts offered vaccination suggests that longer-term evaluations of vaccination coverage, timeliness and other aspects of programme quality are needed. As HBV-vaccinated infant cohorts reach adulthood, ongoing analysis of prevalence in adolescents and young adults will ensure that elimination efforts are on track.

Methods
This review was registered and conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. Registration is available at: http://www.crd. york.ac.uk/PROSPERO/display_record. asp?ID=CRD42017060309 (the checklist is available from the corresponding author).

Search strategy
We searched the online databases of Embase®, MEDLINE®, Web of Science and LILACS to January 2017. The search terms used were: "hepatitis B" AND ("vaccination" OR "mass vaccination" OR "immunization" OR "immunization programmes") AND ("adolescent" OR "adult"). The search was supplemented with manual searches of reference lists of articles for additional studies.

Inclusion criteria
We included any study addressing the long-term impact of an infant hepatitis B vaccine programme at a population level if it satisfied the following criteria: (i) HBV infection status, as defined by hepatitis B core antibody (HBcAb) or hepatitis B surface antigen (HBsAg) status (or both), was assessed and reported in the study population; (ii) the study population, or a specified subgroup for which data were reported separately, was aged ≥ 15 years at the time when HBV infection status was assessed; (iii) the study population, or a specified subgroup meeting criterion ii, included a cohort not offered vaccination through an infant programme (referred to here as the unvaccinated cohort); and (iv) the study population, or a specified subgroup meeting criterion ii, included a cohort offered vaccination through an infant programme (referred to as the vaccinated cohort).
We categorized infant vaccination programmes as either universal, if hepatitis B vaccine was reported as being available to all newborns; or targeted, if vaccine was available to infants born to women screening positive for chronic HBV infection or to women at high risk for some other reason. In some countries, catch-up vaccination programmes were made available to children born a few years before implementation of universal infant vaccination. In these settings, cohorts classified as targeted may have been involved in both a targeted vaccination programme and a catch-up programme.
The primary outcomes of interest were the serological prevalences of (i) HBsAg, which defines chronic HBV infection, and (ii) HBcAb, which defines past, cleared HBV infection in people who are negative for HBsAg.
Two authors independently reviewed the abstracts of the studies identified by the search strategy for studies that met the inclusion criteria. Chinese language articles were translated by a third author to determine if they met the inclusion criteria. The full texts of articles that appeared to be relevant were reviewed by the two authors independently for inclusion. If the same cohort of individuals appeared to have been included in more than one publication, we contacted the authors of the articles to define overlap and avoid duplication.

Variables extracted
We extracted the following study variables for each eligible study, if available: study design, location (country and region); study period; participants' age and sex distribution; vaccine programme coverage as reported in the article, including coverage of timely birth dose; number of participants in unvaccinated and vaccinated cohorts; and the number positive for HBsAg and/or HBcAb by serology testing in each cohort. We contacted the authors of the articles if data were unavailable in the original publication.

Quality assessment
In addition to data extraction, two authors separately reviewed the quality of each study using an adapted Cochrane method. 19 We considered whether studies had addressed potential confounding variables, used a repeatable sampling frame and calculated a response rate. We also judged whether there was potential for bias through assessment of the serological markers, selective reporting of data or the representativeness of populations surveyed. 19

Statistical analyses
We classified studies according to outcome measures (prevalence of HBsAg, HBcAb or both) and whether they reported on universal or targeted infant vaccination programmes. We calculated the relative prevalence (RP) and corresponding 95% confidence intervals (CI) as the ratio of the prevalence in vaccinated and unvaccinated cohorts within each study. Meta-analyses were performed by pooling across studies using the Mantel-Haenszel method. We assessed statistical heterogeneity using I 2 , with a value of > 30% as the cut-off. 20 A fixed-effects model was used when there was no significant heterogeneity and a random-effects model otherwise. Publication bias was evaluated visually with a funnel plot. We also conducted separate meta-analyses restricted to studies in which the comparisons involved subjects of the same age, and to determine if effects differed by HBsAg prevalence (< 10% versus ≥ 10%) in the unvaccinated cohort or by geographical location. All analyses were conducted using RevMan version 5.3 (Review Manager, The Cochrane Collaboration, Copenhagen, Denmark).

Results
The electronic search yielded 5781 unique articles which we screened for inclusion ( Fig. 1). Of these, we assessed 114 full-text articles for eligibility and excluded 88 that were ineligible for various reasons. As three publications 21-23 reported on the same subject group in overlapping study periods, only data from the most recent article was included. 21

Study quality
Few studies reported on participation rates for serological testing in the target populations. Those that did were conducted in institutional settings and participation rates were very high. Two studies restricted recruitment to those whose immunization history was recorded. 30,43 Potentially confounding variables were reported in a minority of studies, and were generally confined to age, sex and region. The sampling method appeared to be repeatable in studies based in institutions or on recruitment of blood donors, but was not clear for the other studies. There was no evidence of bias in the strategies used for serological testing or in the data selected for presentation in reports. Generalizability to wider regional or national populations was not addressed in any of the studies.

Prevalence of HBV markers
We analysed 21 studies reporting on HB-sAg prevalence in populations exposed to universal vaccination compared with those given no vaccination (Fig. 2). The prevalence in the universally vaccinated cohorts ranged from 0.3% 27 to 8.5% 33 (median: 2.0%). The prevalence of HB-sAg in the corresponding no vaccination cohorts were substantially higher, ranging from 0.6% 48 to 16.3% 34 (median: 9.8%). All except one study 45 reported a decrease in HBsAg prevalence; the combined RP in vaccinated populations was 0.24 (95% CI: 0.16-0.35; Fig. 2). Among 15 studies that reported on HBsAg in a targeted vaccination cohort (Fig. 3), prevalences ranged from 0.6% 41 to 11.4% 31 (median: 3.2%), compared with a range of 3.5% 38 to 16.3% 34 (median: 10.9%) in the corresponding no vaccination cohorts, leading to a combined RP of 0.32 (95% CI: 0.24-0.43; Fig. 3). Highly significant statistical heterogeneity between studies was found in the meta-analyses of HBsAg prevalence in studies comparing both universal and targeted vaccination with unvaccinated cohorts (I 2 = 98%, P < 0.001 and I 2 = 89%, P < 0.001, respectively). We therefore used random-effects models to estimate combined RPs and CIs.
Funnel plots for each of the four analyses were symmetrical around the combined RPs (Fig. 6).

Coverage
A birth dose of vaccine, either within 24 hours or at birth, was included in the vaccination schedule for all studies except one from Italy, 27 where birth dose was only included in the schedule for targeted vaccination. The main parameter used for coverage was administration of three doses of HBV vaccine. There was limited information on coverage for vaccinated cohorts. Five studies used self-reporting, and found coverage ranging from 55.9% (7 700 of 13 765 people) to over 98% (872/878 people). 25,28,36,37,45 Two cohort studies were able to verify the vaccination status of the entire vaccinated group. 30,43 One study reported 76.1% (842/1107 people) coverage of the third vaccine dose overall, but 34.5% (35/99 people) for those older than 15 years in the universal vaccination cohort and 19.0.% (10/53 people) in the targeted vaccination cohort. 37 Only one study reported on timeliness of the birth dose, stating that 65 of 215 participants (30.2%) who had received the full schedule received their first dose within 24 hours. 44 While some studies reported on coverage at a national or regional level, coverage was not always specified for the age group included in the review and numerators and denominators were not stated. Overall coverage of universal vaccination in Taiwan, China, as cited by some studies, was generally high, ranging from 86.9 to 98.0%. 24,28,[31][32][33]39,47 In Italy coverage was cited in the paper to have been 63% in 1991 increasing to 99% in some regions by 1996. 42 In mainland China, studies cited coverage ranging from 30% in the earlier years of the vaccination programme to 99.9% in 2010. 40,48 In the Australian study, coverage among residents of remote indigenous communities was cited to be 90%. 38

Discussion
Our meta-analysis of the long-term impact on infection of infant hepatitis B vaccination programmes at the population level focused on long-term impact by restricting the time period to at least 15 years following vaccination. An earlier review of impact did not report a combined estimate and did not specifically focus on long-term impact. 49 We found that adolescents and adults in birth cohorts that were offered universal infant vaccination had a 76% lower prevalence of HBV infection and a similar reduction in risk (77%) of HBcAb prevalence compared with cohorts for whom infant vaccine programmes were unavailable. The effect was similar for both higher and lower levels of HBsAg prevalence in the unvaccinated population. The impact of targeted vaccination programmes was slightly lower (68% and 67%, respectively). As a result of overlap in the relevant cohorts, it was not possible to distinguish the effect of these programmes from the impact of catch-up programmes. It was also not possible to separate the direct effect of targeted programmes from herd effects of reduced horizontal transmission, both in vaccinated same-age peers and slightly younger universally vaccinated cohorts.   Table 1, as only age-eligible participants were included in our analysis.  50 have implemented such programmes, but few have so far reported on long-term impact. As cohorts of adolescents and young adults are entering a period of exposure risk through sexual activity, and will become parents of the next generation, it is appropriate to measure prevalence of infection in this age group. 18 The similarity between the impact on HBcAb and HBsAg suggests that the vaccine's mechanism of protection applies equally to preventing infection and    Table 1, as only age-eligible participants were included in our analysis.   Table 1, as only age-eligible participants were included in our analysis.  28,37,42,45 Studies have shown that perinatal HBV transmission is higher in infants who received the birth dose late or re-ceived fewer than the three scheduled doses. 51,52 Global coverage of the full HBV vaccination schedule is estimated at 84%, 50 while administration of the birth dose remains low at an estimated    Table 1, as only age-eligible participants were included in our analysis.  4 The few studies in our systematic review that reported on vaccine coverage recorded highly varying rates. Most of the vaccinated cohorts from the studies in Taiwan, China, were born in the early years of the infant vaccination programme when coverage was lower. The only study reporting on administration of the birth dose within 24 hours, from mainland China, found coverage of 30.2%. 44 No studies reported on the actual timing of the birth dose. Lack of timeliness of the birth dose and inadequate coverage of the full schedule are likely to have contributed to the residual infection that we found in vaccinated cohorts. Further monitoring of more recent cohorts, including reporting of timeliness of the birth dose, is needed to determine the impact of improved coverage on infection rates.

Fig. 4. Relative prevalence of hepatitis B core antibody in the meta-analysis of the long-term impact of immunization programmes: comparison of universal vaccination and unvaccinated cohorts
Vaccine failure, primary (no initial immune response) or secondary (waning of immunity), may have been another cause of infections in the vaccinated cohorts. 53 Poor immunological response was demonstrated in reports of both active and cleared HBV infection detected among fully vaccinated people. [54][55][56][57] HBV prevalence in vaccinated populations could also be attributed to mutations in the HBV S-gene, which can allow the virus to avoid being neutralized by the vaccine-generated anitbodies. 53,58 However, while they need to be monitored, these mutants probably have little significance at a public health level and are unlikely to be a major cause of prevalence in vaccinated cohorts. 58 More than half of the studies included in this meta-analysis were from Taiwan, China, reflecting a strong, early commitment to HBV prevention and research in this formerly high-prevalence area. The reduction in HBsAg prevalence was markedly higher for these studies than those in other areas (83% versus 64%). The early attainment of high newborn coverage of hepatitis B vaccine in Taiwan, China, over 95% by 2002, 11 could explain this observation.
The studies we included had limitations. Most were based on a single timepoint, and compared people who had received vaccination with those who had not, on the basis of their birth cohort. As these two groups were inevitably of different ages at the time of surveying, it is not possible to remove any effect of age from the comparison. However, it is likely that most chronic infections were acquired in infancy and early childhood, and that prevalence would have been relatively stable in adolescence and adulthood, with at most a slight agerelated increase. On the other hand, the estimated reduction in prevalence was 88% for the few studies that compared cohorts of the same age, as opposed to 70% in the single time-point surveys. This finding was surprising, as we might expect that studies comparing younger vaccinees to older unvaccinated cohorts would generate a more favourable estimate if there had been any increase of prevalence with age. However, comparisons involving same-age cohorts may have been affected by other, unknown confounders.
Limitations of the meta-analysis included the considerable heterogeneity in estimated effect sizes, and the statistically dominant role played by a small number of studies that contributed large numbers. Heterogeneity may have been due to differences in study designs, populations and immunization coverage, and to the characteristics identified in our assessment of study quality, such as response rates. Nevertheless all studies except one reported substantially lower HBsAg prevalence in the vaccinated cohort compared with the unvaccinated group, and the one exception reported low coverage in the vaccinated cohort. 45 Another issue to consider in interpreting the findings is the relatively narrow geographical spread of studies. Although many countries have had historically high prevalence of hepatitis B, particularly in the WHO Western Pacific and African regions, 59 most of the published studies that met our inclusion requirements were from China. Few studies from Africa met the inclusion criteria, so we were unable to separately assess impact in this region, where horizontal transmission is believed to play a greater role of transmission compared with Asian countries where perinatal transmission dominates. 60 Determining the impact of vaccination in African countries would be important, particularly the impact of timely birth dose vaccination on reductions in prevalence in this setting.
With infant HBV vaccination programmes now widely in place, many countries could now evaluate the longer term impact on the prevalence of   infection. The ideal mechanism for this evaluation is a time-series comparison with observations at regular timepoints, from comparable populations. In many countries, an efficient approach is serological testing done routinely in pregnant women or other populations, such as new entrants to university, the military or other institutions. All these populations are likely to be accessible for routine, repeatable monitoring. As identified in our quality assessment, such designs should incorporate repeatable sampling frames, measure participation rates and adjust for potential confounding variables. While prevalence in 5-year-olds provides a more immediate indicator of impact, large representative samples in this age group cannot be easily accessed in many settings.
Our estimated reduction in prevalence is consistent with findings of modelling studies which suggested that routine infant vaccination from birth with global coverage of 90% could prevent 4.3 million new infections from 2015 to 2030, 61 and could prevent 84% of HBVrelated deaths in the hypothetical birth cohort from the year 2000. 61,62 Our findings strongly support the importance of the WHO target of 90% coverage of the hepatitis B vaccine in infancy, 15 as low coverage is a potential contributor to residual HBV prevalence. The evidence presented here shows that elimination is achievable, but still on the far horizon. Analyses of available serological data on HBV prevalence in adolescents and young adults can provide information on gaps in the pathway to elimination that can then be addressed through programmatic measures.