Burden of hepatitis E virus infection in pregnancy and maternofoetal outcomes: a systematic review and meta-analysis

We carried-out a comprehensive search on major electronic databases including MEDLINE (through PubMed), EMBASE, Web of Knowledge, and Global Index Medicus to identify relevant studies on HEV infection among pregnant women. The search strategy was adapted to suit each database as illustrated by the search on PubMed (Supplementary Table 1). We considered studies published until January 26, 2020, without any language or country restriction. To supplement the electronic search, references of all relevant studies were also screened for potential consideration.

Two review authors (AF and SK) independently screened titles and abstracts of aggregated citations retrieved from the electronic search, and full texts of potentially eligible articles were further assessed for inclusion. Disagreements were resolved through discussion and unreached consensus was resolved by a third author (JJB).

Cross-sectional, case-control, and cohort studies were considered for inclusion. We excluded letters, reviews, commentaries, editorials, and studies without primary data. We also excluded studies that included participants who had been selected based on presence of other viral hepatitis or HIV and the description of method was incomplete. In these considered studies, HEV infection had to be diagnosed by serum detection of immunoglobulins (Ig) the major outcomes of interest, comparing HEV positive and negative pregnant women, included maternal mortality, foetal immaturity (low birth weight, preterm birth, small for gestational age), and foetal non-survival (intrauterine death, miscarriage, stillbirth). To estimate the prevalence of HEV vertical transmission, we calculated the proportion of HEV infected new-borns (HEV positive with polymerase chain reaction technique on neonatal cord-blood or peripheral blood from the new-borns) among HEV infected mothers.

Using a pretested data extraction form, two review authors (JJB and AFM) independently extracted relevant information, including first author, publication year and period of participants’ recruitment, country, site, area, setting, timing of data collection, study design, sampling method, sample size, sample tested for HEV, number of participants with IgM of HEV in blood or stool, number of participants with maternofoetal complications and the WHO region. Additionally, for each country of study recruitment, we retrieved data on human development index (HDI) [12]. We defined two groups of pregnant women based on clinical presentation at HEV screening: symptomatic and asymptomatic pregnant women. Symptomatic women were those with signs suggestive of acute hepatitis including jaundice and/or elevated transaminases. When relevant data from included studies were not available, corresponding authors were contacted at least twice for clarification. For the methodological quality and risk of bias assessment of included studies, the tool to be used was determined by the outcome of interest which guided study inclusion. Accordingly, for studies presenting the prevalence of HEV, we used an adapted version of the tool developed by Hoy and colleagues (Supplementary Table 2) [13]; for those presenting the association between HEV infection and maternofoetal outcomes, we used an adapted version of the ROBINS-I tool (Supplementary Table 3) [14]. Two review authors (AFM and JJB) independently ran the assessment; discrepancies were arbitrated by a third review author (JRN). Inter-rater agreements between investigators for study inclusion and methodological quality assessment were assessed using the Cohen’s κ coefficient [15].

We undertook data meta-analysis using the package ‘meta’ (version 4.9–2) of R (version 3.6.2, The R Foundation for Statistical Computing). For HEV seroprevalence estimates, we calculated unadjusted prevalence based on crude numerators and denominators provided by individual studies. Then, to minimise the effect of the size of study-specific estimates of prevalence on the overall estimate, we used the Freeman–Tukey double arc-sine transformation before pooling data with a random-effects meta-analysis [16]. We also performed sensitivity analyses to assess the robustness of our estimates when only studies with a low risk of bias were included. Symmetry of counter-enhanced funnel plots and the Egger test were used to assess reporting and publication bias [17]. Consideration of significant publication bias was at a threshold of a p-value < 0.10. To assess the association between HEV infection and maternofoetal outcomes, we used random-effects approach by the Der Simonian and Laird method, and reported pooled weighted results as odds ratios (OR) both with 95% confidence and 95% prediction intervals (CI and PI) [18]. A continuous correction of 0.5 was added to each cell frequency for studies with a zero cell count. Heterogeneity across studies was assessed by χ² test, and reported as I² statistics [19]. In the case of substantial heterogeneity (I² > 50%) [20], we carried-out subgroup analysis to investigate sources of residual heterogeneity.

Univariable meta-regression analysis was performed to identify and quantify (R²) sources of heterogeneity including clinical presentation, year of publication, HDI, WHO regions, country HEV endemic profile, and sampling method. We planned to integrate clinical profile in the final multivariable meta-regression model that was chosen based on the lowest corrected Akaike’s information criterion (AICc). A p value < 0.05 was considered statistically significant. Strength of association were reported with (adjusted) prevalence odds ratios (aPOR) and corresponding 95% CIs.

In total, we identified 597 records, of which 54 were finally included [8, 21‐73] (Supplementary Fig. 1). Agreement between review authors for study selection based on title and abstract (κ = 0.89) and data extraction (κ = 0.78) were moderate to high. Among the 54 included studies performed in 22 countries, 51 had been conducted to estimate HEV prevalence (Supplementary Table 4) and 5 to investigate the association between HEV and maternofoetal outcomes (Supplementary Table 5). Risk of bias was low, moderate and high respectively in 12 (23%), 37 (70%), and 4 studies (7%) addressing HEV prevalence estimation; all studies which assessed the association between HEV infection and maternofoetal outcomes had a moderate risk of bias. Overall, 53 (98%) studies were cross-sectional and one (2%) was case control. In the countries where studies were done, 29 (54%) were in South-East Asia, 10 in Eastern Mediterranean (18%), 6 in Africa (11%), 4 (7%) in Europe, 3 (6%) in Western Pacific, and 2 (4%) in The Americas. Clinically, pregnant women were symptomatic in 29 studies (54%) (Supplementary Table 6). The prevalence of viral hepatitis A, viral hepatitis B, viral hepatitis C, and viral hepatitis D varied from 0 to 14.6% (n = 23 studies), from 0 to 31.3% (n = 27 studies), from 0 to 13.5% (n = 23 studies), and from 0 to 1.5% (n = 3 studies), respectively.

To estimate the prevalence of HEV, a total of 13,153 pregnant women were included in the meta-analysis. All data on symptomatic women were only from high HEV endemic countries. In asymptomatic women, 14 studies were from high endemic, 3 from endemic, and 4 from not endemic countries. The HEV infection prevalence was 49.6% (95%CI: 42.6–56.7) in symptomatic (Fig. 1) and 3.5% (95%CI: 1.4–6.4) in asymptomatic pregnant women (Fig. 2) with substantial heterogeneity; p <  0.0001 (Table 1). Funnel plots suggested no asymmetry (Supplementary Figs. 2 and 3) confirmed by the Egger test (Table 1).

There was no difference in HEV prevalence considering HDI grouping for asymptomatic women (Table 1). There was no data on symptomatic women from high HDI countries (Table 1).

In the univariable meta-regression analysis, the HEV prevalence was associated with clinical presentation (R²: 76.1%), year of publication (R²; 9.8%), human development index (R²: 24.6%), WHO regions (65.3%), and HEV Endemic profile of countries (R²: 0.0%) (Table 2). In the final multivariable model, two variables were included: clinical profile and year of publication explaining 80.6% of the variance of HEV prevalence. The prevalence was higher in symptomatic women compared to asymptomatic (aPOR: 1.76; 95%CI: 1.61–1.91; p <  0.0001) and decreased with increasing year of publication (by 10-year) (aPOR: 0.90; 95%CI: 0.84–0.96; p = 0.003) (Table 2).

Three studies with a total of 155 women reported data on HEV vertical transmission. The pooled estimate of vertical transmission was 36.9% (95%CI 13.3–64.2) (Fig. 3). All three studies diagnosed HEV infection using neonatal cord-blood samples.

In total, 479 HEV positive and 490 HEV negative women from five studies were included to investigate the association between HEV infection and maternofoetal outcomes. HEV infection during pregnancy was associated with low birth weight (OR: 3.23; 95%CI: 1.71–6.10), small for gestational age (OR: 3.63; 95%CI: 1.25–10.49), preterm < 32 weeks (OR: 4.18; 95%CI: 1.23–14.20), and preterm < 37 weeks (OR: 3.45; 95%CI: 2.32–5.13) (Fig. 4). HEV infection during pregnancy was also associated with stillbirth (OR: 2.61; 95%CI: 1.64–4.14), intrauterine deaths (OR: 3.07; 95%CI: 2.13–4.43), but not with miscarriage (OR: 1.74; 95%CI: 0.77–3.90) (Fig. 5). HEV infection during pregnancy increased the likelihood of maternal deaths (pooled OR 7.17, 95%CI 3.32–15.47) (Fig. 5).