key: cord-0261379-p8g8dk5t
authors: Azman, A.; Paul, K. K.; Bhuiyan, T. R.; Koyuncu, A.; Salje, H.; Qadri, F.; Gurley, E. S.
title: Hepatitis E in Bangladesh: Insights from a National Serosurvey
date: 2021-06-18
journal: nan
DOI: 10.1101/2021.06.14.21258872
sha: 84fc6e539c0dff433e842336e12e40376fc06619
doc_id: 261379
cord_uid: p8g8dk5t

Background Hepatitis E virus, typically genotypes 1 and 2, is a major cause of avoidable morbidity and mortality in South Asia. Although case fatality risk among pregnant women can reach as high as 25%, a lack of population-level disease burden data has been cited as a primary factor in key global policy recommendations against the routine use of licensed hepatitis E vaccines, one of the only effective tools available for preventing disease and death. Methods We tested serum from a nationally-representative serosurvey in Bangladesh for anti-HEV IgG. We estimated the proportion of the population with evidence of historical HEV infection and used Bayesian geostatistical models to generate high resolution national maps of seropositivity. We examined variability in seropositivity by individual-level, household-level, and community-level risk factors using spatial logistic regression. Results We tested serum samples from 2924 individuals from 70 communities representing all divisions of Bangladesh and estimated a national seroprevalence of hepatitis E of 20% (95% CI 17-24%). Seropositivity increased with age and male sex (OR: 2.2, 95% CI: 1.8-2.8). Community-level seroprevalence ranged from 0-78% with the seroprevalence in urban areas being higher, including Dhaka, the capital, with 3-fold (95%CrI 2.3-3.7) higher seroprevalence than the rest of the country. Conclusion Hepatitis E infections are common throughout Bangladesh, though 90% of women reach reproductive age without any evidence of previous exposure to the virus, thus likely susceptible to infection and disease. Strengthening clinical surveillance for hepatitis E, especially in urban areas may help generate additional evidence needed to appropriately target interventions like vaccines to the populations most likely to benefit.

Methods 24 We tested serum from a nationally-representative serosurvey in Bangladesh for anti-HEV IgG. We 25 estimated the proportion of the population with evidence of historical HEV infection and used 26 Bayesian geostatistical models to generate high-resolution national maps of seropositivity. We 27 examined variability in seropositivity by individual-level, household-level, and community-level risk 28 factors using spatial logistic regression.

Results 31 We tested serum samples from 2924 individuals from 70 communities representing all divisions of 32 Bangladesh and estimated a national seroprevalence of hepatitis E of 20% (95% CI 17-24%). 33 Seropositivity increased with age and male sex (OR: 2.2, 95% CI: 1.8-2.8). Community-level 34 seroprevalence ranged from 0-78% with the seroprevalence in urban areas being higher, including 35 Dhaka, the capital, with 3-fold (95%CrI 2.3-3.7) higher seroprevalence than the rest of the country. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted June 18, 2021. ; https://doi.org/10.1101/2021.06.14.21258872 doi: medRxiv preprint 3 Background 48 Hepatitis E is estimated to cause over 3 million cases of acute jaundice each year with more than 49 70,000 these leading to death and another 3000 still births [1] . Human infections from hepatitis E 50 viruses, part of the orthohepevirus genus, are caused by four main genotypes (genotypes 1-4), with 51 only genotypes 1 and 2 known to cause epidemics. HEV genotypes 1 and 2 are associated with self-52 limiting acute jaundice in the majority of infections although special populations, like pregnant women, 53 have particularly poor outcomes with case fatality risk as high as 65% [2,3]. 54 While HEV was only identified in 1981 CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Wantai Biological, Beijing, China) following manufacturer's instructions. As suggested in the packet 02 insert, samples with a standardized optical density > 1.1 were considered positive, those < 0.9 were 03 considered negative, and those in the range 0.9-1.1 were considered indeterminate. 04 Statistical Analyses 05 We estimated the national seroprevalence by including survey design weights and post-stratifying by 06 age and sex to the 2011 Bangladesh census, with confidence intervals estimated using the Rao-Scott 07 method implemented in the survey package for R [16, 17] . We used the same approach to estimating 08 seroprevalence by urban/rural locations, sex and age (only post-stratifying by sex for age-group-09 specific estimates). We excluded indeterminate results from all primary analyses. 10 We explored the relationship between individual, household and community-level factors and 11 seropositivity using hierarchical logistic regression models including a spatial random field assuming CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted June 18, 2021. ; https://doi.org/10.1101/2021.06.14.21258872 doi: medRxiv preprint 6 available data sources. In the main analyses we included household and community random effects 17 in addition to the spatial random field, but in sensitivity analyses estimated models with different 18 combinations of random effects to understand variability in our estimates (Supplement). We explored 19 univariate models, a fully saturated model, a model with only variables significant in the univariate 20 analyses and two simplified models selected a priori and compared their fit with Wanatabe-Akaike 21 information criterion (WAIC) [22] . 22 Using the same INLA modeling framework we estimated seroprevalence on a 5 km by 5 km grid 23 across Bangladesh. To do this we assigned each community to a grid-cell by its centroid, estimated 24 the mean seroprevalence in each cell containing observations and fit spatial regression models to 25 these data. We then used these fitted models to predict seroprevalence in the unobserved grid-cells. 26 We fit both a fully saturated model, including population size, distance from a major water body, a 27 poverty index, travel time to the nearest city, and altitude as linear predictors in addition to spatial 28 random effects using a Matern covariance structure and a null model with only the spatial random 29 effects. To quantify the out of sample performance of this approach we used leave-one-community 30 out cross validation and compared predictions to a 'naive' model that predicted the mean grid-cell 31 seroprevalence for all but the held-out cells and calculated the mean absolute error. 32 We predicted seropositivity among girls reaching child-bearing age (15 years) by fitting generalized 33 additive models with penalized cubic splines to age-seroprevalence curves in each first-level 34 administrative unit (division). We estimated simultaneous 95% confidence intervals by re-sampling 35 from estimates of the variance-covariance matrix of the fitted model using a simulation-based 36 approach with 1,000 draws [23]. Seroprevalence varied greatly across communities from 0% to 77.5% across the country (Figure 1 ). 60 Urban areas (29.7%; 95%CI 21.4-39.6) had 1.8 times higher seroprevalence than rural areas (16.8%; 61 95%CI 13.5-20.6). Household-level seroprevalence ranged from 0 to 100% and we found no 62 evidence of increasing household seroprevalence with household size. 63 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted June 18, 2021. ; https://doi.org/10.1101/2021.06.14.21258872 doi: medRxiv preprint 8 While age, sex and living in an urban area were associated with the risk of being seropositive, other 64 community and household-level risk factors may also be important. To explore the relationship 65 between these potential risk factors we used a series of univariable and multivariable logistic 66 regression models with spatially correlated errors. In univariate models (Table 1) , we found significant 67 positive associations with age, being male, travel, urbanicity, population density, a community poverty 68 index, and protective effects of various indicators of socioeconomic status (e.g., having completed 69 primary school compared to having no formal education, having cattle or other animals in the 70 household, being a household owner). However, in our primary multivariable model with spatial 71 random effects, none of these factors were independently associated with seropositivity except for 72 age and sex, though the effect sizes were largely consistent across various models considered ( 

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted June 18, 2021. ; https://doi.org/10.1101/2021.06.14.21258872 doi: medRxiv preprint 10 We fitted Bayesian geostatistical models to make a national map of seroprevalence. Our primary In this nationally representative serosurvey we found that one in five people in Bangladesh had 94 evidence of prior HEV infection with men having more than 1.5 times higher risk than women. water and sanitation (e.g. household income, education level) were not significantly predictive of 04 seropositivity in our study. Our data are from a study not originally designed to study hepatitis E, 05 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted June 18, 2021. ; https://doi.org/10.1101/2021.06.14.21258872 doi: medRxiv preprint 11 therefore, we did not have data on household water sources and sanitation. Despite this, our findings were estimated to occur in Dhaka, inhabitants had lower than average risk overall. In contrast, 21 inhabitants of Dhaka had 3 times higher risk of HEV seropositivity than others in the country. Some of 22 the differences in spatial risk profiles may be due to the fact that cholera estimates capture only a 23 snapshot of transmission (one year) compared to the lifetime exposures captured by HEV antibodies. 24 Additionally, men had significantly higher HEV seroperelance though we found no significant 25 difference by sex for cholera. These differences might be in part explained by HEV transmission CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted June 18, 2021. ; https://doi.org/10.1101/2021.06.14.21258872 doi: medRxiv preprint 12 This study comes with a number of limitations. We assumed that these serologic assays had perfect 30 sensitivity and specificity for detecting historical HEV infections. Previous studies have estimated 31 sensitivity and specificity of these assays to be high [33,34] however, without a gold standard assay 32 to compare against these estimates are unlikely to be perfect nor generalisable to all settings. 33 Furthermore, seropositivity has been shown to decay over time so those infected many years before 34 the serosurvey may be differentially misclassified as seronegative [35, 36] . That seroprevalence does 35 not significantly increase past the age of ~30 years old is likely due to a combination of seroreversion 36 and changes in the force of infection over the decades. Future work synthesizing assay validation 37 data may be valuable for correcting seroprevalence estimates appropriately. As there is only one 38 HEV serotype, our estimates likely include immunologically meaningful exposures not only to 39 genotypes 1 and 2, the most concerning for outbreaks, but other genotypes, which have not been including those that go on to become pregnant, are underway [13] We thank all the participants of the study throughout Bangladesh for their willingness to participate in this 92

research. We also wish to thanks the field and laboratory staff for their dedicated work. 93 94

The authors have no conflicts of interest to declare. 96 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted June 18, 2021. ; https://doi.org/10.1101/2021.06.14.21258872 doi: medRxiv preprint CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted June 18, 2021. ; https://doi.org/10.1101/2021.06.14.21258872 doi: medRxiv preprint

with Hepatitis E Virus outbreak in Napak District

Evidence for a virus in non-A, non-B 15 hepatitis transmitted via the fecal-oral route

Study of an epidemic of non-A, non-B hepatitis. Possibility of another human 17 hepatitis virus distinct from post-transfusion non-A, non-B type

Enterically transmitted 19 non-A, non-B hepatitis associated with an outbreak in Dhaka: epidemiology and public health 20 implications

Hepatitis E as a cause of adult hospitalization in 22

Bangladesh: Results from an acute jaundice surveillance study in six tertiary hospitals

Outbreak of hepatitis E in urban Bangladesh resulting in 25 maternal and perinatal mortality

Viral load and genotype of recent hepatitis E virus 27 outbreak in Chittagong

Estimating the burden of maternal and neonatal 30 deaths associated with jaundice in Bangladesh: possible role of hepatitis E infection

A large outbreak of hepatitis E among a displaced 33 population in Darfur, Sudan, 2004: the role of water treatment methods

Recommendations to assure the quality, safety and efficacy of 36 recombinant hepatitis E vaccines

HEV study protocol : design of a cluster

trial to assess the safety, immunogenicity and effectiveness of the hepatitis 41 E vaccine HEV 239 (Hecolin) in women of childbearing age in rural Bangladesh

Hepatitis E vaccine: WHO position paper

Nationally-representative serostudy of dengue in Bangladesh 46 allows generalizable disease burden estimates

Available from: the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted

Complex Surveys: A Guide to Analysis Using R

An explicit link between Gaussian fields and Gaussian Markov 53 random fields: the stochastic partial differential equation approach

A global map of travel time to cities to assess inequalities 57 in accessibility in 2015

Mapping poverty using mobile phone and satellite 59 data

An alternative view of variational Bayes and asymptotic approximations of free 62 energy

Semiparametric Regression

Vibrio cholerae O1 transmission in Bangladesh: 66 insights from a nationally-representative serosurvey. medRxiv [Internet]. Cold Spring Harbor 67 Laboratory Press

Hepatitis E Virus Reinfection in Experimentally Infected Rhesus Macaques

The global prevalence of hepatitis A virus infection and susceptibility: 73 a systematic review

Population seroprevalence of hepatitis E virus 77 antibodies in rural Bangladesh

Epidemiology and risk factors of incident hepatitis E 79 virus infections in rural Bangladesh

Risk Factors Associated with Blood Exposure for 81

Setting a course for preventing 83 hepatitis E in low and lower-middle-income countries: A systematic review of burden and risk 84 factors. Open Forum Infect Dis

Vibrio cholerae O1 transmission in Bangladesh: insights 87 the author/funder, who has granted medRxiv a license to display the preprint in perpetuity

The copyright holder for this preprint this version posted June 18, 2021. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)The copyright holder for this preprint this version posted June 18, 2021. ; https://doi.org/10.1101/2021.06.14.21258872 doi: medRxiv preprint