Chronic infection with the hepatitis B virus (HBV) is associated with an increased risk of liver cirrhosis and hepatocellular carcinoma (HCC) [1]. The prevalence of chronic hepatitis B (CHB) infection varies considerably according to geographical regions, as does the incidence of HCC, and highly endemic areas of HBV infection such as East Asia or sub-Saharan Africa have high HCC incidence [2]. However, the risk of HCC amongst people with CHB infection also varies markedly by geographical areas. Population-based cohort studies in hepatitis B surface antigen (HBsAg)-positive men have estimated age-standardized HCC incidence rates of 451–1,030/100,000 carrier-years in East Asia [3], [4], 68/100,000 carrier-years in sub-Saharan Africa [5] and 34/100,000 carrier-years in the Mediterranean basin [6]. This variation in incidence appears to be due to variation in HBV natural history as is discussed below.

Persistence of high HBV viral load [7] or e antigenemia [8] play an important role in increasing the risk of primary liver cancer, and variations in this may explain the observed geographical difference in HCC incidence. In Taiwan, about half of HBsAg-positive children remain hepatitis B e antigen (HBeAg)-positive into their twenties [9], while in sub-Saharan Africa where HBV endemicity is similar, the prevalence of HBeAg among HBsAg-positive people declines to 10% in the second decade of life [10]. Evans et al. attempted to identify factors responsible for the persistence of high viral replication in an ecological study comparing Chinese and African populations. They suggested that earlier age at HBV infection and the subsequent prolonged maintenance of viral replication in Asians with CHB infection is the cause of the higher HCC incidence in Asia [5]. Indeed, there is wide geographical variation in the mode of HBV transmission which may explain the differences in the average age at which persistent infection is established. In sub-Saharan Africa or the Mediterranean basin, horizontal transmission during childhood is the major mode [11]. In contrast, perinatal transmission from the mother is important in Asia where 40% of chronic infections were attributable to this mode in the pre-vaccination era [12].

Age at infection is known to influence the establishment of HBV infection. Infection persists in 80–90%, 20–30%, <10%, and <5% of people infected perinatally, in early childhood, adolescence, and adulthood, respectively [13]. However, it is unclear whether early HBV infection also increases the risk of liver cirrhosis and/or HCC through the persistence of high viral replication, in addition to increasing the risk of chronic infection [14]. We therefore undertook a systematic review of observational studies where the association between age at establishment of chronic HBV infection and risk of cirrhosis and/or HCC has been investigated among people chronically infected with HBV.

In addition to liver cirrhosis and HCC as endpoints, two indicators of ongoing viral replication (serum HBV DNA level, presence or persistence of viral replication (HBeAg)) and two of hepatic damage (elevated serum alanine transaminase (ALT) and degree of hepatic fibrosis) were included. These are important predictive factors for cirrhosis and HCC, and often used as indicators for antiviral treatment of CHB infection [15]. Our hypothesis is that earlier age at infection is associated with worse outcomes in addition to its effect of increasing the probability of chronic HBV infection.

The search of databases identified 7,077 potential articles (2,250 in Medline, 3,288 in Embase, 136 in CNKI and 1,403 in SinoMed) of which 1,593 were excluded due to duplication. Review of the titles and abstracts excluded 5,086. Forty-three papers were manually identified from reference lists. The full text of the 441 articles was examined in detail, and of these 422 was discarded leaving 19 papers eligible for the systematic review. Criteria of exclusion are described in Figure 1.

There was only one longitudinal study that examined the association between age at first HBsAg-positive result and persistence of HBeAg [21]. This study used an historical cohort of HBsAg-negative children in Alaska who were later identified to have a CHB infection through semi-annual serological follow-up. The cohort was followed up for greater than ten years to assess the timing of HBeAg loss (Table 1).

A total of 12 studies assessed the association of maternal HBV sero-status with various outcomes (Table 2 and 3); HCC (2 case-control studies) [22], [23], cirrhosis by liver histopathology (1 cross-sectional study) [24], persistence of HBeAg (4 cohort studies) [25]–[28], and presence of HBeAg at one time point (5 cross-sectional studies) [29]–[33]. The two cohort studies that evaluated HBeAg loss also examined ALT levels [25], [28]. Except for four studies [26], [30]–[32], the studies exclusively enrolled children. In all the cohort studies, treatment was not given during follow-up except one study of steroid withdrawal therapy [26]. All studies examined HBsAg as the only maternal HBV marker, and three also considered maternal HBeAg [28], [29], [33]. The proportion of participants with available maternal sero-status varied markedly between studies. In two studies having a mother alive [30] or maternal serology available [25] was one of the eligibility criteria while among the remaining studies the percentage of subjects with known maternal sero-status varied from 49% [32] to 95% [27], [28]. The timing of when maternal sero-status was defined in relation to the child’s age also varied. One study in England [29] assessed prenatal maternal HBeAg whereas 6 studies examined mothers’ HBV markers when their children were enrolled in the study (Table 2 and 3). In the remaining 5 studies, the timing was unclear.

For birth order, one cross-sectional study [34] and two case-control studies [16], [35] compared the frequency distribution of birth order by the presence of HBeAg-positivity and HCC, respectively (Table 4). Three Chinese case series of HBV-related HCC used the Greenwood-Yule method (Table 5) [36]–[38]. All birth order studies were conducted in adults.

Except for one study [28], all longitudinal studies compared proportions rather than incidence rates. Results of multivariable analyses were only reported in three studies [16], [28], [35] although some studies presented results stratified by age group [27], [30]. The author of a New Zealand study [32] provided additional data which enabled us to control for age. The risk of bias in the included studies is summarized in Appendix S5.

Of 21 outcomes examined in 19 studies, univariable analysis of 14 outcomes supported our hypothesis with P≤0.05 and two with weak evidence (0.05<P<0.1). Although the other studies did not reach statistical significance, the direction of the associations was consistent with our hypothesis.

McMahon et al. directly assessed the age at HBV infection and its association with HBeAg persistence in Alaskan natives whose approximate date of infection was known because of consecutive serological tests. They showed that the median age at first HBsAg-positive result was lower in those who did not clear HBeAg during the follow-up period than those who HBeAg sero-converted, suggesting that the early age at infection is associated with delayed HBeAg loss. However, the length of follow up is different for the two groups, which is a crucial determinant of the chance of HBeAg loss.

Studies of maternal HBV serology showed that having a mother with a positive HBV marker is associated with worse outcomes (HCC, cirrhotic changes, persistence of HBeAg or possession of HBeAg at one particular time) although three of these were not supported by statistical evidence. Two longitudinal studies from East Asia which evaluated both HBeAg loss and peak ALT levels during follow-up demonstrated good evidence that children with HBsAg-positive mothers experienced elevated levels of hepatic enzymes less frequently and fewer episodes of HBeAg loss than those whose mothers are negative. It is well established that having a highly elevated serum ALT level during the immune-tolerant phase of CHB infection is a factor leading to early HBeAg seroconversion [39].

The only maternal serological study which reported the results of a multivariable analysis [28] showed that the association with maternal HBsAg observed in the univariable analysis was no longer significant after controlling for maternal HBeAg. This reflects the fact that maternal HBeAg is a stronger predictor of perinatal transmission than maternal HBsAg. In fact, the risk of perinatal transmission ranges from 10–20% in HBsAg-positive mothers without HBeAg to 90% in mothers with positive HBeAg [40]. Nevertheless, because HBeAg sero-clearance (loss of the maker) occurs faster than HBsAg sero-clearance in chronically infected persons (2–15% versus 1% per year) [15], misclassification of maternal HBeAg status between when the index child was born and when he/she entered the study is greater than misclassification of maternal HBsAg status. Apart from 5 studies not reporting when mothers were bled, all but one examined maternal sero-status at the child’s study entry, which might have led to non-differential misclassification of the exposure, resulting in underestimation of its association. The only study which investigated prenatal maternal HBeAg [29], the best proxy for perinatal transmission, was inconclusive because of a small sample size.

Recent evidence suggests that viral genotypes modify the natural history of CHB infection. Certain genotypes are associated with delayed HBeAg seroconversion [41] and also HCC risk [42]. Moreover, associations between certain viral genotypes and mother-to-infant transmission have been reported [43]. For example, Tseng et al. [28] found that genotype C was associated with delayed HBeAg seroconversion after accounting for maternal HBsAg and HBeAg status. The study also showed that maternal HBeAg was associated with persistence of HBeAg after adjusting for viral genotype.

Two Greek case-control studies of birth order demonstrated that HCC cases are concentrated in higher birth orders, suggesting that later-born children who might have been exposed to HBV at a young age through their older siblings have a higher risk of developing HCC than early-born children. However this finding was not confirmed in three subsequent Chinese birth order studies which applied the Greenwood-Yule methods. Furthermore, one Taiwanese study showed that HBeAg was more prevalent in earlier birth orders although the evidence was weak [34]. These heterogeneities can be understood in terms of geographic differences in the main mode of transmission. In East Asia, perinatal infection through vertical transmission is relatively important compared to other parts of the world [12]. In such places, first-born children have a higher chance of having been born to a HBeAg-positive mother than later-born children, because young mothers with positive-HBeAg will clear it with increasing age [15]. Consequently, early-born children have a higher chance of having been infected earlier from their mother than later-born children. In contrast, in Mediterranean countries where children are horizontally infected through other children with positive-HBeAg [11], birth order acts as an indicator for number of older siblings, and later-born children have an increased chance of being exposed to HBV at an early age through infectious older siblings than early-born children. Perinatal transmission also occurs in Greece but is much less common than in China. This could explain the results of the first Greek study where people born as the second child have lower odds of HCC than the first birth rank (OR 0.5), with the odds increased for higher birth orders.

The funnel plot and Egger’s test suggest the presence of publication bias in the studies assessing the association between maternal sero-marker and persistence/presence of HBeAg. In general, observational studies are more vulnerable to publication bias than randomized trials. Epidemiological studies are often conducted retrospectively by analysing existing databases, and unless published, this kind of study is hard to trace. In addition, often only statistically significant variables are reported [19]. Although we were unable to assess the publication bias in the birth order studies due to the small number of studies included, these studies are prone to selective reporting bias because birth order is easily obtained by interview without additional cost.

Although a robust conclusion cannot be drawn due to the potential role of publication bias and the heterogeneity of studies included in the review, this systematic review supports the hypothesis that earlier age at HBV infection is associated with an increased risk of HCC through persistence of viral replication. This review has highlighted some issues in interpreting this association. Studies using maternal HBsAg are only informative when prenatal maternal HBeAg status is available in the study population. The relationships between surrogate markers of disease endpoints, such as viral load and/or HBeAg, are now so clear that they can be used to shorten the time period over which the association is observed. Finally, confounding factors such as viral genotype should be appropriately controlled in the analyses.

The implications of the effect of early age at infection on the prevention of cirrhosis and primary liver cancer are twofold. First it means that the impact of hepatitis B vaccination on raising the average age at infection will not simply be in reducing the prevalence of chronic infection but also in reducing the adverse effects of that chronic infection in those acquiring it at an older age. Second it adds emphasis to the critical importance of interrupting perinatal transmission – as reflected in the WHO recommendation for a birth dose within 24 hours of birth in all countries [44].