Background

The hepatitis B virus (HBV) accounts for significant morbidity and mortality worldwide [1]. An estimated one-third of the worldwide population has been exposed to HBV, and 400 million people are chronic carriers [1]. HBV can be transmitted vertically, sexually, and through other routes, and pregnant women play important roles in each of those transmission routes. Chronic hepatitis B (CHB) infection during pregnancy often results in hepatic flare [2] and possibly other adverse outcomes [3–17], including gestational diabetes mellitus (GDM)[3–8, 10–13, 15–17], although conflicting evidence exists on this last point.

GDM is defined as glucose intolerance with an onset or first recognition during pregnancy. Asian women, including Chinese women, have been found to have the highest incidence of GDM [18]. Women with GDM are at an increased risk for adverse perinatal and maternal outcomes including macrosomia, caesarean section, birth trauma, and diabetes later in life, although treatment significantly decreases these risks [19]. Until now, liver disease of various aetiologies has been implicated as a cause of diabetes [20], and chronic hepatitis C virus (CHC) infection has been established to increase the risk of diabetes mellitus (DM) [21].

Similarly to CHC infection, CHB infection is considered to be associated with insulin resistance (IR) [22], which implies that it may increase the risk of developing DM and/or GDM. However, the association between CHB infection and the risk of DM and/or GDM remains less convincing. An increased risk was reported for DM only by Li-Ng [23], but this result was not confirmed by a recently published ten-year cohort study [24]. Several studied have reported an increased risk for GDM [4, 6, 7, 10, 11], but this result was contradicted by many others [3, 5, 8, 9, 12–17]. Due to the limited resources and time available for HBV vaccination, the prevalence of CHB infection among pregnant women remains high, especially in HBV epidemic areas such as China [25].

To address these issues, we performed a meta-analysis of cohort studies to investigate whether CHB infection during pregnancy is associated with an increased risk of GDM.

Methods

Data sources and search strategy

This meta-analysis was conducted in compliance with the Meta-Analysis of Observational Studies in Epidemiology guidelines (http://edmgr.ovid.com/ong/accounts/moose.pdf). Two independent investigators (Kong and Liu) searched PubMed from 1966 to September 5th, 2013 using the combinations of terms “hepatitis B” or “hepatitis B surface antigen (HBsAg)” and “pregnancy outcome” or “prenatal outcome” or “perinatal outcome” or “gestational diabetes”. We sifted through potentially relevant articles, firstly by titles and abstracts, and then we retrieved the full texts of articles for detailed review. Further, we scanned the reference lists of the articles that met the inclusion criteria in our analysis, and searched for those articles or citations in the Web of Knowledge, Google Scholar and Google to obtain additional studies.

Inclusion and exclusion criteria

Articles were included if they used a cohort design and reported a quantitative association between CHB infection during pregnancy and the risk of GDM among pregnant women versus a non-CHB control group. For studies that enrolled overlapping pregnant women, only more recently published studies and/or those with larger sample sizes were included.

Studies included in this analysis defined CHB infection status during pregnancy by the presence or absence of hepatitis B surface antigen (HBsAg) in blood during the first prenatal care visit, on admission to the labour ward, or before delivery. GDM diagnosis was performed according to the recommended clinical practice.

Data extraction

A form designed a priori was used to extract the information from the included studies. Two independent investigators (Han and Wei) performed the data extraction. A third investigator (Wang) examined the results, and a consensus was considered as agreement between at least two out of the three investigators. GDM was the primary outcome measure. We extracted the following data (Table 1): authors’ name, journal and year of publication, country of origin, enrolment period, incidence of GDM in pregnant women who had CHB infection and those who did not, general characteristics of pregnant women, unadjusted and/or adjusted effect size and the adjusted variables if available.

Table 1 General characteristics and quality scores of the 14 studies included in this meta-analysis
Full size table

Assessment of methodological quality

Two independent investigators (Kong and Hu) evaluated the quality of each study using the Newcastle-Ottawa Quality Assessment Scale (NOS) [26]. The third investigator (Lu) examined the results, and a consensus was considered as agreement between at least two out of the three investigators (Table 1). We did not consider the 2nd item of the NOS scale (Was follow-up long enough for outcomes to occur) or the 3rd item (Adequacy of follow up of cohort) because the final outcomes of all pregnant women were followed up to their delivery dates. We defined studies with NOS ≥ 7 as high quality and those with NOS < 7 as low quality.

Statistical analysis

Effect measurement: odds ratio and its adjustment

The cross-study effect of CHB infection vs. non-CHB control on the risk of GDM was measured using the odds ratios (OR) with 95% confidence intervals (CIs), which were reported as the effect measurements in all of the included studies except for one [4], which reported relative risk.

The adjusted OR (ORa) was automatically included in our analysis; the unadjusted or crude OR (ORc) was corrected using external estimates of confounding by the method of coefficient adjustment introduced by Greenland [27]. Briefly, the confounding effect U = ORc/ORa was calculated by an external study. The variance of U was estimated using the formula Vu = Va-Vc, where Va and Vc are the variances of the natural logarithm of the ORc and ORa from the same study, respectively. Then, we used U and its variance Vu for further correction. In our analysis, the ORc from the Lobstein study [12] and the Chinese or Thailand studies [3, 5, 8, 9, 15–17] were adjusted using the estimates of U and Vu from Reddick 2011 [13] and Lao 2007 [7], respectively.

Heterogeneity analysis and its sensitivity analysis

We used a random-effects model to pool the ORa across studies in Stata version 10.0 (Stata Corp). Heterogeneity was first explored by a χ2 test with a P value <0.10 considered statistically significant and then judged visually by a forest plot, funnel plot and Galbraith plot. The extent of heterogeneity was measured by Higgins’ I2. We also explored the influence of heterogeneity on the pooled ORa across studies using a sensitivity analysis introduced by Patsopoulos [28]. Briefly, for a meta-analysis of n studies, we perform n new meta-analyses where one study is excluded from the calculation each time. The study that leads to the largest decrease in I2 upon exclusion is dropped, and a new set of n-1 studies is created. This process is continued until the I2 is decreased to the lowest possible value.

For the heterogeneous studies that were excluded from further analysis, we explored the possible reasons for the heterogeneity using a causal diagram.

The combined effect estimate

For the homogenous studies, the Dersimonian and Laird random-effects model was used to pool the ORa across studies in Stata version 10.0 (Stata Corp). Sub-group analysis and meta-regression were carried out to examine the effect in relation to design, type of epidemic area, article quality and GDM diagnostic criteria. We used the interaction test to estimate the difference between two sub-groups [29]. Publication bias was assessed by a visual inspection of a funnel plot and by the Egger’s and Begger’s test [30], and a nonparametric trim and fill method was applied if possible [31]. The predictive intervals were calculated using a t-distribution with k-2 degrees of freedom, where k represents the number of studies [32].

Results

Literature search

The literature search retrieved 280 papers, 29 of which were identified as potentially relevant to the current analysis. Further backwards or citation searches produced six additional papers. Only fourteen papers [3–13, 15–17] representing fourteen independent studies were included in our analysis. The reasons for exclusion of the other studies are listed in Additional file 1: Figure S1.

General characteristics about the studies included in meta analysis

The fourteen studies included in the present analysis were published between 1999 and 2013 (Table 1). Two studies each were conducted in Thailand [8, 9] and Iran [10, 11], one in Germany [12], the United States [13], and mainland China [15], and the other seven in Hong Kong, China [3–7, 16, 17]. All fourteen were hospital-based retrospective cohort studies.

The CHB infection status was determined by the presence or absence of HBsAg during pregnancy, and the diagnosis of GDM was determined through routine clinical practices. This information was retrieved directly from medical records or indirectly from medical records based on a computerised database through the International Classification of Diseases codes.

Heterogeneity analysis and its sensitivity analysis

The high heterogeneity of the ORa across the fourteen studies was demonstrated by visual observation of the forest plot, funnel plot, and Galbraith plot and further verified by the Higgins’ I2 value of 58.9% (χ2 = 31.6, df = 13, p = 0.003).

I2 was reduced by removing one study at a time using a random-effects model based sequential algorithm [28]. As seen from Figure 1a-d, removing one study each time led to a gradual decrease in the Tau-squared value with decreasing I2, and the 95% CIs of the pooled ORa across studies became narrower, with the means, lower and upper 95% CIs ranging from 1.00 - 1.40, 0.82 - 1.13, and 1.21 - 1.78, respectively. Actually, 38% of the 95% lower CIs were below 1.0 during this process.

Figure 1
figure 1

Pooled odds ratios, their 95% confidence intervals, Tau-squared and Higgins’ I2values for the association between chronic hepatitis B infection during pregnancy and the occurrence of gestational diabetes mellitus using a random-effects model based sequential algorithm. OR: odds ratio; ORL: 95% lower confidence limit of OR; ORH: 95% higher confidence limit of OR; Notes: In each step of the analysis, one study was removed out of: (a) the original fourteen studies (Step 1); (b) the remaining thirteen studies after the exclusion of the Lao 2003 (Step 2); (c) the remaining twelve studies after the exclusion of Lao 2003 and Saleh-Gargari 2009 (Step 3); (d) the remaining eleven studies after the exclusion of Lao 2003, Saleh-Gargari 2009 and Tse 2003 (Step 4). The arrows point to the studies whose removal minimises the Higgins’ I2 value. The blue, green and orange hollow circles denote the pooled ORa, 95% ORL and ORH, respectively, and the red multiple signs denote the Tau-squared values.

Full size image

After four studies were removed (Lao 2003 [4], Saleh-Gargari 2009 [10], Tse 2003 [6] and Aghamohammadi 2011 [11]), the remaining ten studies were highly homogeneous, with a Higgins’ I2 value of 0.0% (χ2 = 8.06, df = 9, p = 0.53, Figure 2) and a Tau-squared value of 0.0.

Figure 2
figure 2

Effect of chronic hepatitis B infection during pregnancy on the occurrence of gestational diabetes mellitus using a random-effects model. OR: odds ratio; CI: confidence interval; CHB: chronic hepatitis B infection; GDM: gestational diabetes mellitus. Notes: The black boxes denote the ten studies used for this analysis; the hollow boxes with the same size denote the four studies that were excluded from the analysis. R1, R2, R3 and R4 denote the studies in the order of their removal during the process for the reduction of Higgins’ I2.

Full size image

The combined effect estimate

The ten highly homogeneous studies involved 99.4% (436,991/439,514) of the subjects among the fourteen studies, and together reported a GDM prevalence of 6.30% (884/14,028) among pregnant women with CHB infections and 3.63% (15,352/422,963) among uninfected pregnant women. The association between CHB infection during pregnancy and the risk of developing GDM was 1.11 [the adjusted odds ratio (ORa), 95% CIs: 0.96 - 1.28, 95% predictive intervals 0.94 - 1.31] (Figure 2). These findings suggest that CHB infection during pregnancy does not increase the risk of GDM, in agreement with all of the individual studies except for one [7], which suggests that CHB infection during pregnancy is associated with a higher risk of GDM.

Subgroup and meta-regression analysis showed that the pooled ORa across studies could not be modified by the following characteristics: quality of papers (high vs. low), designs (all non-CHB pregnant women as controls vs. an unmatched or matched sample of non-CHB pregnant women as controls), epidemic levels (high, middle vs. low), and GDM diagnostic criteria.

Visual inspection of the funnel plots of the ORa from the ten studies demonstrated a possible publication bias. After trim and fill methods were performed, the pooled ORa was 1.10 (95% CIs: 0.96 - 1.26; 95% predictive intervals: 0.94 - 1.30) (Figure 3).

Figure 3
figure 3

The Egger’s funnel plot with pseudo 95% confidence limits for the ten homogenous studies analysing the effect of chronic hepatitis B infection during pregnancy on the occurrence of gestational diabetes mellitus by using trim and filled methods. OR: odds ratio; Ln: natural logarithm. The hollow circles with or without enclosed boxes represented the original and filled studies, respectively.

Full size image

Possible reasons for the heterogeneity of the remaining four studies

The four heterogeneous studies that were excluded from further analysis all reported a higher risk of GDM for pregnant women with CHB infection (Figure 2). The effect of their removal on the heterogeneity was depicted within a causal diagram (Figure 4). The first and third excluded studies, Lao 2003 [4] and Tse 2003 [6], respectively, had both recruited a special subset of pregnant women from the population in Hong Kong, China. For the former study, CHB infected pregnant women were subjected to more rigorous oral glucose tolerance testing (OGTT), and exhibited a higher serum ferritin concentration which was considered to be both an outcome of the CHB infection and a risk factor of GDM [33]. The pregnant women in the latter study had multiple risk factors, with some suffering from significant medical diseases requiring active treatment, including pre-existing diabetes mellitus. Both CHB infection during pregnancy and GDM may lead to a high-risk status among pregnant women, which would affect their recruitment into the study. Thus, both studies may have enrolled more CHB infection-related GDM [34], a selection bias resulting in the overestimation of the association between CHB infection and GDM. The second and fourth excluded studies, Saleh-Gargari 2009 [10] and Aghamohammadi 2009 [11], respectively, were both conducted in Iran with similar designs. As we know, the prevalence of both CHB infection and GDM differs among ethnic groups [18], making it possible that the CHB infection-GDM association may exist in pregnant Iranian women but not the others. These issues should be addressed further in future studies.

Figure 4
figure 4

Possible causal diagram related to the four excluded studies analysing the relationship between chronic hepatitis B infection during pregnancy and gestational diabetes mellitus. CHB: chronic hepatitis B infection; GDM: gestational diabetes mellitus; Pop: population. The arrow represents a causal relationship between the two variables at both edges of the arrow. The variable that the arrow points to denotes the outcome of the variable at the other end of the arrow. The red arrow denotes the relationship under study. The blue arrow lines denote the other variables that could affect this relationship. The author names and publication years are associated with the variables as appropriate.

Full size image

Discussion

Our results suggest that CHB infection during pregnancy does not represent a significant general risk for the development of GDM as compared to non-CHB infected controls.

The high initial between-study heterogeneity in this meta-analysis prevented us from drawing strong conclusions. To overcome this problem, we firstly used a sequential algorithm based sensitivity analysis of between-study heterogeneity [28], which differentiated those homogenous studies from the other heterogeneous ones. This strategy was helpful in identifying and sequentially excluding the heterogeneous studies, i.e., at least those with outliers, which made the analysis based on the random-effects model more reasonable [32]. Although the acceptable level of heterogeneity remains unknown [35], we were conservative and chose the lowest Higgins’ I2 as our standard for this differentiation. As a result, this strategy resulted in a parallel decrease in both Higgins’ I2 and Tau-squared, with values approaching 0.0% and 0.0, respectively. Finally, we differentiated ten homogenous studies from the remaining four heterogeneous ones.

Those ten homogeneous studies reported GDM prevalences of 6.30% and 3.63% among pregnant women with and without CHB infection, respectively. The risk of GDM for CHB-infected pregnant women was comparable to the risk for uninfected women (ORa = 1.11, 95% CI: 0.96-1.28). These findings suggest that CHB infection during pregnancy is not associated with an additional risk of GDM.

Two of the heterogeneous studies, Lao 2003 [4] and Tse 2003 [6], provided further support. The study of Lao 2003 [4] enrolled pregnant women with higher levels of ferritin, which independently predicts the severity or advanced fibrosis of non-alcoholic fatty liver disease [36]. The study of Tse 2003 [6] enrolled high-risk pregnant women. Thus, both of these studies may have enrolled women with more progressive stages of CHB infection, which would have resulted in the overestimation of the association between CHB infection and GDM.

It remains unclear why only Iranian pregnant women with CHB infection appear to have an increased risk of GDM. Both Iranian studies, Saleh-Gargari 2009 [10] and Aghamohammadi 2011 [11], were conducted on Iranian pregnant women with similar designs [10, 11] but reported a significantly higher association between CHB infection and GDM than the other studies analysed here. Since the occurrence of both CHB infection and GDM differs among ethnic groups [18], there may be an ethnic difference in the association between CHB infection and GDM.

GDM is characterised by pancreatic β-cell dysfunction that results in insulin resistance (IR). IR has been shown to be associated with both CHC and CHB infection [22], but CHC related IR was mainly due to steatosis development [37] or fibrosis progression [38], which seems to be insignificant for CHB infection as most pregnant women with CHB infections are asymptomatic or experiencing mild hepatic injury [39].

Routine screening for GDM is recommended near the beginning of the third trimester, and the identification of GDM will depend on the diagnostic criteria and guidelines. For economic, psychological and other reasons, GDM screening is conducted using a risk factor-based selective screening approach in most countries rather than universal screening. These differences in strategies did not seem to affect the non-significant association between CHB infection and the risk of GDM [12–14, 16, 17], which was further supported by our sub-group and meta-regression analysis. Only one out of the ten homogenous studies reported a positive association between CHB infection and the risk of GDM [7]. This difference can be explained by the fact that the replacement of the Australian standard for GDM diagnosis with the WHO standard would have resulted in the inclusion of individuals with impaired glucose tolerance into the GDM group. Routinely, pregnant women will receive the same screening and are subjected to the same diagnostic criteria in the same study regardless of CHB infection status, so issues of this type seem to play insignificant roles in most of the studies [3, 5, 8, 9, 13, 14, 16, 17]. Future studies should use a universal screening strategy together with the new diagnostic criteria established for GDM.

Conclusions

Our findings imply that GDM is not attributable to CHB infection during pregnancy. Thus, any extra GDM screening for this population is not necessary, which is a commonly accepted practice throughout the world except in Hong Kong, China. This extra screening will likely result in a greater economic burden and additional psychological concerns such as fear or guilt, especially in epidemic areas, like China. Probably, advanced clinical stages of liver diseases of varied aetiologies may be more important in the development of GDM, and the significantly high association between CHB infection and GDM in Iranian pregnant women is worth further study.