Demographic and ecological risk factors for human influenza A virus infections in rural Indonesia

Background Indonesia has the world's highest reported mortality for human infections with highly pathogenic avian influenza (HPAI) A(H5N1) virus. Indonesia is an agriculturally driven country where human‐animal mixing is common and provides a unique environment for zoonotic influenza A virus transmission. Objectives To identify potential demographic and ecological risk factors for human infection with seasonal influenza A viruses in rural Indonesia, a population‐based study was conducted in Cileunyi and Soreang subdistricts near Bandung in western Java from 2008 to 2011. Methods Passive influenza surveillance with RT‐PCR confirmation of influenza A viral RNA in respiratory specimens was utilized for case ascertainment. A population census and mapping were utilized for population data collection. The presence of influenza A(H3N2) and A(H1N1)pdm09 virus infections in a household was modeled using Generalized Estimating Equations. Results Each additional child aged <5 years in a household increased the odds of H3N2 approximately 5 times (OR=4.59, 95%CI: 3.30‐6.24) and H1N1pdm09 by 3.5 times (OR=3.53, 95%CI: 2.51‐4.96). In addition, the presence of 16‐30 birds in the house was associated with an increased odds of H3N2 (OR=5.08, 95%CI: 2.00‐12.92) and H1N1pdm09 (OR=12.51 95%CI: 6.23‐25.13). Conclusion Our findings suggest an increase in influenza A virus infections in rural Indonesian households with young children and poultry.


| INTRODUCTION
Influenza pandemics and seasonal influenza epidemics have caused high mortality and morbidity with devastating global economic losses. 1 4 Indonesia, a largely agricultural country where human-animal mixing is common, provides a unique environment for zoonotic influenza A virus transmission and an ideal ecological setting for the generation of novel influenza A viruses. 5 A significant amount of research has focused on understanding the risk factors associated with HPAI H5N1 viral infection. Studies from Indonesia and elsewhere show that the majority of human infections with HPAI H5N1 virus have been associated directly or indirectly with poultry exposure including close contact with sick or dead poultry, visiting a live poultry market and commercial poultry density. [6][7][8][9][10] A recent review of the literature concluded that direct exposure to birds was one of the most likely sources of human infection with A(H5N1). 11 Other studies investigating area-level ecological correlates of HPAI suggest that rice paddy fields, population density, and exposure to potentially contaminated water sources 6,7,12-14 all increase risk of infection.
While human infection with HPAI A(H5N1) virus in Indonesia has occurred sporadically, seasonal influenza A and B viruses regularly circulate within the population. Several influenza surveillance studies have been conducted in Indonesia which suggest that the burden of seasonal influenza is high and that influenza viruses appear to circulate year-round with increased activity between November and March. [15][16][17][18] These studies have focused on determining the burden of disease, characterizing circulating virus strains, and understanding seasonal trends. We are not aware of any studies of seasonal influenza in Indonesia that have examined risk factors beyond age and sex. In fact, much of what we know regarding the risk factors of communityacquired seasonal influenza is derived from research conducted in middle-and high-income countries. These studies reinforce that factors such as the number of school age children, household and population age structure, [19][20][21] contact patterns, 22,23 and interaction with birds or bird environments [24][25][26] are associated with influenza distribution in a community.
The intent of this study was to explore the role of the animalhuman interface in community-acquired seasonal influenza. This is a population-based exploration of ecological risk factors driving symptomatic influenza A virus infections defined by care seeking at firstlevel health facilities. We combined passive influenza surveillance with a complete population census to determine the influence of household demographic characteristics, birds in the household, and community-level population structure on the risk for symptomatic influenza A virus infections.

| Study area
This study was conducted in Cileunyi and Soreang subdistricts in rural West Java Province, approximately 18 km outside the provincial capital of Bandung city. A map of the study area is shown in Figure 1.

| Study population and data
This study used a prospective cohort design to identify ILI cases in Soreang and Cileunyi. Patients with symptomatic influenza virus infections were identified by passive surveillance in three government community health centers (puskesmas) in the study area between October 2008 and September 2011. Dedicated, trained study physicians screened and enrolled all patients visiting the clinics with signs and symptoms of influenza-like illness (ILI). To be included in the study, participants must have: (i) lived in the study area at the time of illness, (ii) presented with ILI defined as a fever (body temperature >37.5°C) with cough or sore throat, and (iii) had signed informed consent. Basic demographic data, address, and clinical information F I G U R E 1 Location of the study area in Bandung District, West Java Province, Indonesia.Map shows the location of Soreang and Cileunyi relative to other major Indonesian cities. The two subdistricts are located in West Java province just outside the city of Bandung. Bandung is Indonesia's third largest city located approximately 140 km southeast of Jakarta of all eligible subjects were recorded, and nasal and oropharyngeal swabs were collected for influenza virus testing at Hasan Sadikin General Hospital Laboratory. Approximately 2 weeks after enrollment, trained nurses conducted home visits and administered a survey to collect information on clinical outcomes and household and potential environmental risk factors. The geographic location of each participant's household of residence was collected using a handheld GPS receiver and linked to study data using a Geographic Information System (GIS). If a participant presented to the puskesmas with an ILI more than once during the study period, but at least 14 days apart, each visit was treated as a separate event. The patient was screened again, and a second nasal and oropharyngeal swab was obtained. If the patient was identified as having influenza from laboratory results, this was recorded as a separate case from the first and recorded as a second case in the household (ie, a household could have two separate records if both H3N2 and H1N1pdm09 viruses were identified during these two visits).
To collect data on the total population at risk, 448 trained local community health workers (CHWs) conducted a census of all households in the two study subdistricts. The CHWs used standardized forms, which collected data on: address of residence, the age, sex and education of all permanent household residents, and number of birds and poultry kept by the household. Twelve dedicated trained field surveyors conducted quality control to identify missing values and errors of transcription on the forms. They subsequently conferred with the CHWs to correct the data entry forms. Following the creation of the population list and address validation, the 12 field surveyors conducted door-to-door visits to geocode the household using handheld GPS receivers. This mapping activity was used as a secondary quality control check to validate the population data initially collected by the CHWs.

| Laboratory testing
Nasopharyngeal (children and adults) or oropharyngeal (adults) swab samples were obtained from each enrolled patient and transported at 4-8°C to the Research Laboratory in a Universal Virus transport medium (Becton-Dickinson, Franklin Lakes, NJ, USA).
Influenza A and B viral RNA was detected in a one-step multiplex real-time RT-PCR using primers and specific LNA-mediated TaqMan probes in two separate assays, using standard protocols. 27 Briefly, the first assay consisted of primers and probes specific to the matrix (M1) gene of influenza A virus, influenza B virus, and host glyceraldehyde-3phosphate dehydrogenase (GAPDH) gene. The second multiplex assay detected influenza A virus subtypes using primers and probes (1st Base, Singapore) specific for regions of the H1, H3, and H5 hemagglutinin (HA) genes. The probes were labeled with three different fluorescent reporter dyes (FAM, HEX, and Cy5 with emission wavelengths at 518, 556, and 667 nm, respectively). Specimens, which tested positive for influenza type A, were subtyped in a second real-time RT-PCR assay incorporating primers and probe specific for H1N1pdm09 using the standard CDC protocol. 28 Due to resource limitations, further characterization of the non-subtypeable specimens was not performed for this study.

| Household covariates
Using the population census in the GIS, we developed a number of demographic and ecological variables at the neighborhood and household levels that were examined for a relationship with influenza A cases. We focused on influenza A cases as our primary goal was to identify ecological risk factors at the human-animal interface. For each household, we calculated the average household size, whether the head of household had less than a high school education, the age structure of the household, the distance from each household to the nearest puskesmas and several indicators of the number of birds kept by the household (presence/absence and total number). Households were also grouped into "neighborhoods"-one of 224 subcommunities within the subdistricts. Using a Geographic Information System (GIS), neighborhood-level variables were created by drawing a 200-m radius buffer around each household and aggregating data within the buffer. For each household neighborhood, we calculated the population density, percent of households where the head of household had less that a high school education, percent of the population in one of five age categories (0-5, 6-15, 16-50, 51-65 and >65 years) and the total number of birds within a 200-m radius of each household. To control for potential bias due to underreporting, we created a measure of the neighborhood healthcare utilization and calculated as the percentage of households that used community healthcare services for ILI during the study period.

| Statistical approach
The occurrence of a subject with symptomatic influenza A virus infection seeking care at the puskesmas in each household was modeled using generalized estimating equations (GEE) with the logit link function. The GEE models used correlation matrices that were independent and exchangeable within neighborhoods. The use of exchangeable neighborhood matrices corrects for potential spatial correlation in the data due to local-level transmission dynamics of influenza viruses. GEE was implemented by R version 2.15.2 with the geepack library.
Each of the two different influenza A subtype virus infections (H3N2 and H1N1pdm09) were modeled separately. In the models, the occurrence (yes/no) of an influenza A case in each household was the primary outcome variable. Covariates with a known association with influenza virus infection were included to control for potential confounding. The full set of covariates we considered can be found in Table 1; the set of covariates and their specification used in the final model are shown in Table 2

| RESULTS
The population census recorded 163 014 individuals in 42 775 households, for an average of 3.8 people per household. Of the 3356 enrolled ILI subjects, 402 had influenza A and 105 had influenza B. Of the former, 193 were H3N2, 157 were H1N1pdm09, 9 were seasonal H1N1, and 43 were not subtypeable. On follow-up 2 weeks later, none of these subjects were hospitalized or died. The multivariable statistical analysis modeled household-level risk, so the 193 H3N2 cases corresponded to 171 unique households, the 157 H1N1pdm09 cases corresponded to 149 households, the 43 non-subtypeable were found in 38 households, and the 9 seasonal H1N1 were in 9 households. These sample sizes are reflected in Table 1. The analysis presented here focuses on households with influenza A as we had a sample size suitable for multivariable analysis and our primary goal was to study the animal-human interface. Furthermore, resource limitations did not allow for characterization of the non-subtypeable specimens, so we also excluded these in multivariable modeling.  33 Therefore, the association of family age structure with influenza A in the household is likely not solely due to the age-related immune response or susceptibility, but may also be associated with age-related contact pattern heterogeneity, such as interaction with other persons between school or workplace. Unfortunately, we did not collect contact or activity pattern information to explore these dynamics further.
The number of birds or poultry in the household was also associated with symptomatic influenza A virus cases. This association was relatively weak with the ownership of 1-5 birds, but became stronger with an increasing number of birds in the household. Figure 3 illustrates this dose-response relationship. This trend was consistently found across both influenza A subtype outcomes. While it is biologically implausible that there is a causal link of seasonal influenza A virus infection of humans to the presence of birds, there could be a potential environmental/immunological explanation for our observations. A larger number of poultry kept by rural households would be associated with more fecal and hence bacterial environmental contamination. 34,35 Bacterial lipopolysaccharide (LPS) has been shown to protect mice against HPAI virus infection through toll-like receptor stimulation 36 and might explain the relative mildness of the illness seen in our rural population, as none of them were subsequently hospitalized or died.
Bacterial LPS has also been shown to inhibit the induction of CD8+ Tcell immunity by influenza virus infection. 37 One possibility is that high levels of bacterial LPS in the environment, by suppressing T-cell responses, increase the ratio of asymptomatic to symptomatic influenza virus infections, leading to our observed higher rates. Our observation that the relationship is dose-dependent supports this hypothesis.
Another potential explanation for this finding is related to sanitation and associated health behaviors or household socioeconomic status. Handwashing and hand hygiene have been highly publicized as a core management strategy for influenza and other respiratory disease. Although handwashing is effective in reducing the incidence of common diseases such as acute respiratory infections, data on its ef- idation on bird ownership were blinded to the outcome result, and therefore, we believe any systematic exposure misclassification is minimal. We assume any bias occurred randomly and therefore would be more likely to bias the result toward the null.
Some migration and mobility are inevitable over the course of the 3 years of the study. We note that the location of the patient at the time of diagnosis is accurate, as a nurse visited their home 2 weeks after the community health center visit and collected household information. However, household data on the "control" population in the study-all households with no reported influenza case-were only collected once at the during the initial population census. Some changes to household structure could certainly happen during the study period, perhaps leading to a misrepresentation of the control population.
However, we note that there is little migration out of this study area, and people do not typically move residences as most households are engaged in agriculture and more or less tied to their land. Thus, we believe these biases are minimal.

| CONCLUSION
We found that age structure and the number of birds in the household were significantly associated with influenza A virus infections and these ecological determinants operate at the household level rather at the neighborhood level. In this exploratory study, the positive association between household birds and seasonal influenza A virus infections suggests an underlying common ecological factor, as yet unidentified. Regardless of the mechanism, in these populations where avian influenza A viruses may be prevalent among poultry, the possibility of reassortment of HPAI H5N1 virus or other avian influenza A viruses with seasonal influenza A viruses is a theoretical possibility. Our results suggests that further studies examining the relationship between human behavior, animal exposure, and influenza A virus evolution and novel influenza A virus emergence are indicated in rural Indonesia.
for maintaining the CIRAI web-based data entry and database system, and The FKUP-RSHS laboratory team (Nur Izzatun Nafsi, Yeni Rendiani) for performing real-time RT-PCR. This study was funded by the Centers for Disease Control and Prevention (CDC), Atlanta, Georgia, USA. We acknowledge the support of Diane K. Gross, who was the original program officer for the CDC funded project.

DISCLAIMER
The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the Centers for Disease Control and Prevention.