*Gretchen Stevens and Steven T. Wiersma are staff members of WHO. The authors alone are responsible for the views expressed in this publication and they do not necessarily represent the decisions, policy, or views of WHO.
*Gretchen Stevens and Steven T. Wiersma are staff members of WHO. The authors alone are responsible for the views expressed in this publication and they do not necessarily represent the decisions, policy, or views of WHO.
Potential conflict of interest: Nothing to report.
Supported by the Expanded Programme on Immunization (EPI), Department of Immunization, Vaccines and Biologicals (IVB), Family and Community Health (FCH), World Health Organization, through Agreement for Performance of Work (APW) Ref. no. I8-APW-189.
We estimated the global burden of hepatitis E virus (HEV) genotypes 1 and 2 in 2005. HEV is an emergent waterborne infection that causes source-originated epidemics of acute disease with a case fatality rate thought to vary by age and pregnancy status. To create our estimates, we modeled the annual disease burden of HEV genotypes 1 and 2 for 9 of 21 regions defined for the Global Burden of Diseases, Injuries, and Risk Factors Study (the GBD 2010 Study), which represent 71% of the world's population. We estimated the seroprevalence of anti-HEV antibody and annual incidence of infection for each region using data from 37 published national studies and the DISMOD 3, a generic disease model designed for the GBD Study. We converted incident infections into three mutually exclusive results of infection: (1) asymptomatic episodes, (2) symptomatic disease, and (3) death from HEV. We also estimated incremental cases of stillbirths among infected pregnant women. For 2005, we estimated 20.1 (95% credible interval [Cr.I.]: 2.8-37.0) million incident HEV infections across the nine GBD Regions, resulting in 3.4 (95% Cr.I.: 0.5-6.5) million symptomatic cases, 70,000 (95% Cr.I.: 12,400-132,732) deaths, and 3,000 (95% Cr.I.: 1,892-4,424) stillbirths. We estimated a probability of symptomatic illness given infection of 0.198 (95% Cr.I.: 0.167-0.229) and a probability of death given symptomatic illness of 0.019 (95% Cr.I.: 0.017-0.021) for nonpregnant cases and 0.198 (95% Cr.I.: 0.169-0.227) for pregnant cases. Conclusion: The model was most sensitive to estimates of age-specific incidence of HEV disease. (HEPATOLOGY 2012)
The hepatitis E virus (HEV) is an enterically transmitted RNA virus that can cause outbreaks or sporadic disease.1 HEV was first postulated as a unique infectious agent following a large outbreak of hepatitis in Kashmir in 1978, and was first isolated in the stool of Soviet military recruits stationed in Afghanistan in 1983.2, 3 HEV outbreaks are thought to result primarily from contamination of water supplies, although some evidence exists for person-to-person transmission.4
The prevalence of HEV infection varies genotypically by global region. HEV has one serotype and four reported genotypes. Genotypes 1 and 2 exclusively infect humans and are often associated with large outbreaks and epidemics in developing countries with poor sanitation conditions. Genotypes 3 and 4 infect humans, pigs, and other animal species and have been responsible for sporadic cases of disease in developed and developing countries.5 Although genotype 3 has been reported to cause chronic hepatitis in persons with chronic liver disease, those infected with human immunodeficiency virus (HIV), or organ transplant recipients, the extent to which genotype 3 and 4 infections result in disease in otherwise healthy patients is unknown and warrants further investigation.6-11
Like hepatitis A virus infection, only a portion of those infected with HEV develop symptoms and the risk of symptomatic illness may depend on age of infection. Symptomatic acute HEV infection are largely indistinguishable from those experienced during any acute phase of hepatic illness. Symptoms typically last 1 to 2 weeks and include anorexia, dark urine, nausea and vomiting, abdominal pain, and eventually icterus.2 In rare instances, acute hepatitis E can result in fulminant liver failure and death among nonpregnant individuals. Among pregnant women, HEV substantially increases the risk of mortality among the infected pregnant women and also is associated with intrauterine fetal death, preterm delivery, and stillbirths.12
In this article we attempt to estimate the annual global burden of HEV associated with genotype 1 and 2 infections that occur in Africa and Asia (insufficient scientific study of the probability of illness following infection precluded the inclusion of genotypes 3 and 4 in this analysis).13-15 The results from this study can be used to inform disease prevention policy and to understand key areas of uncertainty in creating the burden estimates and to create estimates of disability adjusted life years (DALYs) lost due to HEV genotype 1 and 2 infections.
DALY, disability adjusted life years; HAV, hepatitis A virus; HEV, hepatitis E virus; HIV, human immunodeficiency virus.
Materials and Methods
We developed a simplified model of HEV infection and its outcomes based on expert opinion and clinical disease descriptions (Fig. 1).5 In the model, incident infections can occur in nonpregnant or pregnant individuals and infection can result in asymptomatic infection, symptomatic infection with icterus but without death, or symptomatic infection that results in death. All infections that occur in pregnancy also carry the additional risk of stillbirth.12 We did not consider pregnancies terminated due to mortality to be additional stillbirths.
We applied the model to the nine Global Burden of Diseases, Injuries, and Risk Factors Study (the GBD Study) defined regions in Africa and Asia for which genotype 1 (and to a lesser extent genotype 2) HEV is dominant: Asia Central, Asia East, Asia South, Asia Southeast, North Africa and the Middle East, Sub-Saharan Africa Central, Sub-Saharan Africa East, Sub-Saharan Africa Southern, and Sub-Saharan Africa West.16, 17 Collectively, these nine regions contained approximately 4.7 billion people or 72.8% of the global population in 2005.18 Figure 2 illustrates the steps taken to convert our raw data into burden estimates.
HEV Prevalence and Incidence Proportions.
We modeled seroprevalence and incidence of HEV infection using data gathered from a systematic review of the prevalence of HEV disease and the DISMOD III v. 3.0 generic disease model (DISMOD), which was designed for the GBD Study.5, 19 We abstracted primary data on the age-specific prevalence of HEV from 37 articles published between 1990 and 2010 that we identified in an existing literature review of the prevalence of HEV antibodies.5 We used DISMOD to estimate the age, sex, and region-specific prevalence and incidence of HEV based on these data.
We used DISMOD to generate estimates in two stages: First, using the data identified in the literature review we fit an empirical prior estimate of prevalence that had the following elements: a random effects model structure (so that estimates of the intercept in an individual region were derived from a joint function of data observed in the region of interest and a global estimate from all regions' data); a flexible age pattern; a linear global time trend; a fixed effect for data on males; a simple fixed effect to control for studies that used convenience sampling; and an assumption of a prevalence of zero at birth.
Second, for each time period (1980-1996 and 1997-present), sex, and region (of the nine included GBD regions), we used DISMOD to fit a Bayesian model based on the prevalence data and empirical prior estimates that generated posterior estimates of incidence, prevalence, and mortality that were internally consistent (see Barendregt et al. for additional documentation).20 In the second model stage, we used the following assumptions to generate accurate estimates of prevalence and incidence: First, a lifelong duration of antibodies following acquisition (which is equivalent to assuming no remission), and second, insignificant incremental mortality from the disease as compared with other lifetime causes of death. Although HEV causes mortality, the degree to which HEV-specific mortality affects the relationship between incidence and seroprevalence estimates is likely negligible. These assumptions allowed DISMOD to calculate incidence directly from the modeled increases in prevalence over age. We used a separate but similar model to estimate HEV seroprevalence in Egypt, a country that exhibited highly divergent seroprevalence patterns from the rest of the world.2 The DISMOD model produced estimates of age-specific HEV infection seroprevalence and incidence in 2005 for each region and Egypt, including a 95% credible interval (Cr.I.) for each age by region estimate.
Number of Infections.
We estimated the number of unique HEV infections by age and region by multiplying incidence rates generated by the DISMOD 3 software by the age-specific population of each region subcategorized into pregnant or nonpregnant groups.18 To calculate the pregnant population we calculated the number of pregnant individuals by age in each country by multiplying the estimated variant of the United Nations published crude birth rate per 1,000 population by the total population divided by 1,000 to estimate total births and then divided the total by 29—the number of years between ages 15 and 44—which we assumed accounted for the majority of pregnancies.18, 20 We multiplied this number by 0.77 to account for the fact that, for each birth, the mother is only pregnant for an average of 40 weeks. We then attributed this number of years of pregnancy risk to each age category between 15 and 44 and subtracted these pregnancies from the population in each of those age categories so as not to double-count individuals.
Asymptomatic, Symptomatic, and Fatal Outcomes of Disease.
We divided unique HEV infections into mutually exclusive categories of asymptomatic (i.e., anicteric), symptomatic (i.e., icteric), and fatal cases (i.e., icterus with liver failure and death). For cases that occurred in pregnant women, we also estimated stillbirths associated with asymptomatic and symptomatic cases. We conducted a systematic review of published studies to estimate the probability of icterus given infection, the probability of death given icteric illness for nonpregnant individuals, and the probability of death given icterus for pregnant individuals. We searched PubMed for articles published between 1950 and 2010, abstracts from the American Association for the Study of Liver Disease Meetings from 2009 and 2010, and abstracts presented at the International Symposium on Hepatitis E in 2010. We did not search the European Association for the Study of the Liver Abstracts due to a lack of access. We identified 644 possible articles for inclusion using the search criteria “hepatitis E outbreak,” “hepatitis E epidemic,” “hepatitis E jaundice,” or “hepatitis E mortality,” of which 602 either contained no relevant data, presented data found in another publication, or dealt with genotype 3 disease. Another five articles were published in a language other than English, which we lacked resources to translate, and 41 contained possibly relevant information. Our search terms yielded no relevant American Association for the Study of Liver Disease abstracts and four abstracts from the International Symposium on Hepatitis E, for a total of 45 possibly relevant articles or abstracts. Of these when reviewed, 37 contained information useful for identifying the risk of symptomatic infection or death: 25 from Asia (two from Central, one from East, 21 from South, and one from Southeast) and 12 from Africa/Middle East (six from East, four from North Africa and the Middle East, and two from Sub-Saharan South).4, 6-9, 12, 22-52 We extracted data from these 37 studies on the number of people who were serologically documented to be infected with HEV during the study period, the number of people who developed icteric illness, and the number of people who died. We divided these data into categories of nonpregnant, pregnant, and unknown/mixed population status; we also recorded the continent (Africa or Asia) in which the data were collected and the age ranges of the affected patients.
Using the data from these studies, we estimated separate models of the probability of symptomatic infection among all persons infected with HEV, the probability of death among those who developed symptoms and were pregnant, and the probability of death among those who developed symptoms and were not pregnant. Data were insufficient to stratify estimates of the probability of symptomatic disease by pregnancy status. Following published methodology, our first model used Monte Carlo Markov Chain simulation methods programmed in Proc MCMC of the SAS 9.2 Software (SAS Institute, Cary, NC) to estimate the probability of symptomatic illness from the observed number of seropositive and symptomatic patients across studies seen in the literature review after assuming that probability was a binomially distributed variable with a uniformly distributed prior mean.53 Our second model was similar to the first except that it estimated the probability of death among infected and symptomatic pregnant women. The third model estimated the probability of death for infected and symptomatic nonpregnant persons similar to the first two models but used random effects to control for studies of mixed population of pregnant and nonpregnant women where pregnancy status could not be identified. We also fit additional models to identify differences in mortality rates between estimates observed in Africa versus Asia, but the data did not support systematic differences in these estimates.
Symptomatic Illness by Age.
Most HEV clinical experts believe that, similar to hepatitis A virus (HAV), the risk of symptomatic illness given infection with HEV increases with age at infection. Unfortunately, we found no data with which to estimate this trend. Instead, to be conservative, we assumed that a similar function governed the age-specific risk of infection for HEV as for HAV. Our model modified a previously estimated function of the age-specific risk of symptomatic illness given infection with HAV to estimate the age-specific risk for HEV after substituting our estimate of the adult risk of symptomatic illness from HEV into the equation.54 This yielded the following baseline age-specific risk of symptomatic illness:
where MAXRATE is the maximum adult rate of symptomatic infection and Age is the infected individual's age in years.
We estimated the increased risk of HEV-related stillbirth as the incremental difference between the probability of stillbirth observed in one study of women infected with HEV and the United Nations stillbirth rate (estimated per GBD Region as a weighted average of component country results) among all pregnancies (Table 1).12, 21 To estimate the number of stillbirths, we multiplied this rate by the number of anicteric and icteric infections that occurred among pregnant women in the region. We assumed that pregnancies that result in death of the mother did not also result in a stillbirth.
Table 1. Key Model Parameters and Their Standard Errors Used in Probabilistic Sensitivity Analysis
All parameters beta distributed. Age-specific incidence estimates (Fig. 3) were used to compute baseline estimates and sensitivity analysis.
Adult probability of symptomatic infection
Intercept term in Eq. 1 (determines age at which adult probability of symptomatic infection is reached)
Armstrong and Bell (53), Assumption
Probability of death given infection and not pregnant
Probability of death given infection and pregnant
Incremental probability of stillbirth by region
Patra et al. (12); United Nations (20)
1. Asia, central
2. Asia, east
3. Asia, south
4. Asia, southeast
5. North African, middle east
6. Sub-Saharan, central
7. Sub-Saharan, east
8. Sub-Saharan, southern
9. Sub-Saharan, west
Probabilistic Simulation and Sensitivity Analysis.
We estimated a probabilistic simulation of global burden using 10,000 replications. In each simulation our model selected incidence and key model parameters from their plausible distributions using their standard errors and assuming each parameter was beta-distributed.55 For each simulation replication we used a single multiplier for the model's incidence parameters (one parameter used across the nine GBD Regions plus Egypt) and a second for the model's nine stillbirth rates. Each multiplier was equal to a beta-distributed decimal between 0 and 1 with a standard error of 0.25. Once selected, the incidence multiplier was multiplied by the range between the minimum and maximum incidence parameter generated by the DISMOD 3 model for each age in each region and that value was added to the minimum incidence parameter. The same method was used for stillbirth rates. Our results present the mean estimate across 10,000 simulations, as well as a credible interval based on the 250th and 9,750th observations after our observations were sorted numerically.
Prevalence and Incidence Rates.
Globally, we estimated a consistent prevalence pattern across the nine GBD Regions. In each region we observed the largest increases in prevalence occurring between 5 and 20 years of age. We observed the highest seroprevalence rates in South Asia South and East Asia, each with peak seroprevalence rates in excess of 25% of the age-specific population (Fig. 3). North Africa and the Middle East (Egypt excluded) exhibited the lowest seroprevalence of the nine regions, and the remaining six GBD Regions exhibited similar seroprevalence rates between 15% and 25% for ages greater 25. For Egypt (not shown), our model predicted seroprevalence rates in excess of 50% for all persons age 5 years or older.
Our model predicted annual incidence rates roughly between 0.5% and 1.0% for ages 0 to 15 years, with rates increasing to between 1.0% and 1.4% for ages 15 to 20 years, then falling rapidly to a lower rate of 0.2% and below at ages older than 30 years (Fig. 4). Our incidence estimates exhibited a great deal of uncertainty, with a 95% Cr.I. range between approximately 40% and 150% of the estimated incidence parameter. For each region the model exhibited the greatest uncertainty in incidence estimates between the ages of 10 and 20 years, with uncertainty diminishing at younger and older ages. Across all regions we estimated an average age of infection of 17.1 years with a low of 8.1 in North Africa and a high of 21.1 in Asia East. North Africa's average age of infection was a relative outlier. The next youngest average age of infection was 15.5 years observed in Sub-Saharan Africa.
Risk of Symptoms.
We estimated a probability of symptomatic illness in adults given infection (MAXRATE) of 0.198 (95% Cr.I.: 0.167, 0.229). We had insufficient data to test whether this probability differed by continent of infection, age, gender, or pregnancy status.
The probability of death given symptomatic illness differed substantially between nonpregnant and pregnant persons. For nonpregnant persons, we estimated a probability of death given symptomatic illness of 0.019 (95% Cr.I.: 0.017-0.021). For pregnant persons we estimated a probability of death given symptomatic illness of 0.198 (95% Cr.I.: 0.169-0.227). The probability of death given symptomatic illness did not differ meaningfully by continent of illness. We had insufficient data to test whether this probability differed by age or between nonpregnant women and men.
In 2005 we estimated a total of 20.1 (95% Cr.I.: 2.8-37.0) million incident HEV infections in the nine GBD regions we evaluated. These 20.1 million infections resulted in 3.4 (95% Cr.I.: 0.5-6.5) million (17.0%) cases of symptomatic illness, 70,000 (95% Cr.I.: 12,400-132,732) deaths, and 3,000 (95% Cr.I.: 1,892-4,424) stillbirths (Table 2). Our estimates contained a great deal of uncertainty. Globally, 60.6% of cases occurred in East and South Asia, and these two regions accounted for 64.7% of deaths from HEV infection. Although the North Africa region accounted for 14.3% of all global infections, it only contributed 8.3% of global symptomatic cases and 8.1% of global deaths due to the younger average age of infection in that region.
Table 2. Year 2005 Incident Infections and Outcomes of Hepatitis E by Global Burden of Disease Project Region (95% Simulated Credible Intervals)
North Africa, Middle East
This article represents the first attempt to estimate the annual global impact of HEV infections caused by HEV genotypes 1 and 2 in Africa and Asia. We found that in 2005 HEV genotypes 1 and 2 accounted for approximately 20.1 million incident HEV infections, 3.4 million cases of symptomatic disease, 70,000 deaths, and 3,000 stillbirths. Incident infections increased through childhood to peak levels between the ages of 15 and 19 and fell thereafter to lower levels in adulthood and disease outcomes followed a similar pattern.
This article is also the first to use meta-analytic techniques to summarize published reports into estimates of the rate of symptomatic illness given infection and the rate of death given symptomatic illness. We found strong evidence that the death rate differed between nonpregnant and pregnant symptomatic individuals, but we did not find evidence that the rate differed by continent of infection (Africa versus Asia).
This study is limited in several respects. First, we did not attempt to estimate the burden of HEV genotypes 3 and 4. HEV genotype 3 is most prevalent in Europe and the United States, but its capacity to cause symptomatic illness and disease is not extensively documented.56, 57 If evidence becomes available, future estimates of the burden of HEV should incorporate additional genotypes to create complete global estimates.
Second, the data used to estimate the prevalence and incidence of HEV infection are sparse and uncertain. Disease incidence was by far the dominant source of uncertainty in our model, and this uncertainty led to wide credible intervals for our estimates of annual infections and outcomes. A large degree of uncertainty is inherent in the measurement of any emergent infection, and assuming interest in HEV increases, prevalence and incidence estimates of HEV infection will likely improve over time as the disease is increasingly recognized and measured across different countries.
Third, our estimate of incidence and symptomatic illness relied on assumptions about HEV that are yet to be verified. Specifically, we assumed that all infections lead to seroconversion that can be detected by way of anti-HEV tests and that the presence of anti-HEV antibodies is lifelong. These assumptions were necessary to convert seroprevalence evidence into annual incidence estimates, but they may not be accurate. Several studies that we reviewed identified individuals during HEV outbreaks who reported jaundice and/or other symptoms indicative of infection but who exhibited no detectable serologic signs of infection.4, 38, 39 Furthermore, anti-HEV protection may not be lifelong.5 We also assumed that the risk of symptomatic disease given HEV infection varies based on age in a manner that is consistent with HAV infections.54 Although we made this assumption to be consistent with expert opinion about HEV infection, additional epidemiological observations will be required to verify its accuracy.
The above limitations bias our results toward more conservative estimates of disease burden. Restricting our study to genotypes 1 and 2 excludes any burden caused by HEV in many parts of the world. Our assumption that all infections of HEV cause antibody evidence of disease and that all antibody protection persists for life eliminates the inclusion of any possible infections that do not result in detectable antibody or in reinfections of individuals who lost antibody protection. For these reasons, we believe our study presents a conservative estimate of the annual burden of HEV.
In contrast, our study used an older global stillbirth estimate instead of updated estimates published in 2010.20, 58 The newer estimates are approximately 14.5% lower than the rates we used. Assuming the relative risk of stillbirth given HEV remained constant, our use of a higher population stillbirth rate would have led to an overestimate of stillbirths attributable to HEV of between 500 and 600 globally.
Despite these limitations, this study offers the first attempt to estimate the global burden of HEV infections and disease outcomes. We predicted a substantial global burden of HEV concentrated especially in the regions of East and South Asia. Future studies work should explore the probability of disease from genotype 3 HEV infections; the probability of symptomatic illness given infection and how this probability varies by age, gender, and pregnancy status; the link between infection and the production of antibodies and the durability of that antibody protection over time; and also collect population-based data on HEV antibody status.
This work was undertaken as part of the Global Burden of Diseases, Injuries, and Risk Factors study. The results in this paper were prepared independently of the final estimates of the Global Burden of Diseases, Injuries, and Risk Factors study. We thank Abraham Flaxman for providing access to Dismod III and invaluable guidance on its use.