Environmental contamination and risk factors for transmission of highly pathogenic avian influenza A(H5N1) to humans, Cambodia, 2006-2010

Background Highly pathogenic avian influenza A (H5N1) virus has been of public health concern since 2003. Probable risk factors for A(H5N1) transmission to human have been demonstrated in several studies or epidemiological reports. However, transmission patterns may differ according to demographic characteristics of the population and local practices. This article aggregates these data from three studies with data collected in the previous surveys in 2006 and 2007 to further examine the risks factors associated with presence of anti-A(H5) antibodies among villagers residing within outbreak areas. Methods We aggregated 5-year data (2006–2010) from serology survey and matched case-control studies in Cambodia to further examine the risks factors associated with A(H5N1) infection among villagers in the outbreak areas. Results Serotesting among villagers detected 35 (1.5 % [0–2.6]) positive cases suggesting recent exposure to A(H5N1) virus. Practices associated with A(H5N1) infection among all ages were: having poultry cage or nesting area under or adjacent to the house (OR: 6.7 [1.6–28.3]; p = 0.010) and transporting poultry to market (OR: 17.6 [1.6–193.7]; p = 0.019). Practices found as risk factors for the infection among age under 20 years were swimming/bathing in ponds also accessed by domestic poultry (OR: 4.6 [1.1–19.1]; p = 0.038). Association with consuming wild birds reached borderline significance (p = 0.066). Conclusion Our results suggest that swimming/bathing in contaminated pond water and close contact with poultry may present a risk of A(H5N1) transmission to human.


Background
Highly pathogenic avian influenza A (H5N1) virus has been of public health concern since 2003, particularly in south-east Asia and Egypt. The virus causes mass die-off in domestic poultry, occasionally infecting humans. As of 04 April 2016, 850 confirmed cases of A(H5N1) human infections worldwide with 449 deaths have been reported to the World Health Organization [1]. Probable risk factors for A(H5N1) transmission to human have been demonstrated in several studies or epidemiological reports [2]. However, transmission patterns may differ according to demographic characteristics of the population and practices leading to contact with poultry [3] and may be impacted by domestic waterfowl density and local geo-ecological settings [4].
In 2005, the first human case of A(H5N1) infection was reported in Cambodia, a country of approximately 15 million population. As of 04 April 2016, 56 confirmed cases have been identified, 37 of these fatal [1]. An investigation followed the identification of the first four human cases which found no evidence of mild or asymptomatic infection in the population [5]. Further investigations around the 5th, 6th and 7th human cases between 2006 and 2007, however, found evidence of mild or asymptomatic infections among villagers. These early studies explored the extent of contact with poultry and potential risk factors for A(H5N1) infections in humans, underscoring the possibility of an environmental source [6,7]. As Cambodia continued to identify more A(H5N1) human cases during 2008-2010, we conducted three community investigations. This article aggregates these data from the three studies with data collected in the previous surveys in 2006 and 2007 to further examine the risks factors associated with presence of anti-A(H5) antibodies among villagers residing within the outbreak areas.

Documenting A(H5N1) circulation in poultry
Within one day after identification of a A(H5N1) human case, investigation teams from the National Veterinary Research Institute (NaVRI) and from Institut Pasteur in Cambodia (IPC) went to the village of the patient and conducted household visits within a one-kilometer radius of each case's house. Household heads were interviewed using a standardized questionnaire, collecting information on poultry morbidity and mortality during the past 3 months including number of sick and dead poultry, dates of onset and symptoms. Venous blood, tracheal swab and cloacal swab specimens were taken from ≤10 randomly selected ducks in each household with duck flocks (chickens being extremely sensitive to A(H5N1) infection, no samples were taken from apparently healthy chickens). Blood samples were stored in 5 mL tubes without additive. Swab samples were put in tubes containing viral transport media (VTM) and shipped at temperature 4-8°C to NaVRI laboratory in Phnom Penh within 48 h.

A(H5N1) serological surveys in humans
During 2006-2010, six confirmed A(H5N1)-infected human cases were identified in five villages of five Cambodian provinces. Teams from IPC, the Cambodian Communicable Disease Control Department (Ministry of Health) and from the World Health Organization (WHO) conducted serological investigations in humans within a one-kilometer radius area approximately four to ten weeks after confirmation of the human A(H5N1) index cases. The objective was to interview and sample all villagers using a standardized questionnaire to collect demographic, poultry exposure and environmental exposure data for the three months preceding symptom onset in index cases. For children aged <12 years old, parents or guardians were interviewed. We used major social events such as Khmer or Lunar New Year, or Water Festival as temporal markers to help respondents recall past activities [3]. Informed consent was obtained and blood samples (5 mL) collected in tubes without additive were taken from all members of households and sent to IPC at temperature 4-8°C within 48 h for serology testing.
As two human cases were detected in the same village successively in April 2007 and December 2009, blood specimens were collected twice from each subject with an interval of one month during the second investigation to draw laboratory conclusion on recent infection and to serologically rule out previous infection unrelated to the current event.

Laboratory testing of poultry specimens
Poultry specimens were processed and tested at NaVRI laboratory as previously described [6]. Tracheal and cloacal swabs were inoculated into embryonated hen's eggs. Allantoic fluid was tested to detect the presence of influenza virus by hemagglutination, followed by identification of H5 virus by hemagglutination inhibition assay (HIA). Positive (titer ≥1:16) specimens were sent to IPC for confirmation and virus subtyping using real-time reverse-transcriptase polymerase chain reaction (qRT-PCR) targeting the HA, M and NA genes [8]. Duck serum samples were tested using HIA method using reference H5 antigen provide by Office International des Epizooties (OIE). A poultry flock was considered infected with A(H5N1) virus if one swab sample or one serum sample tested positive. Sera from 2006 were tested by microneutralization (MN) assay and when the titer was ≥1:80, the result was confirmed by a Western blot assay, as recommended by WHO [6,9].

Serological testing in humans
Sera from 2007-2009 were screened using H5 hemagglutinin pseudotyped lentiviral particles (H5pp) expressing the hemagglutinin of a 2005 Cambodian H5N1 virus and those with a titer ≥1:20 were further tested by MN and hemagglutination inhibition (HI) assay using a 2007 Cambodian H5N1 strain [7,10]. Antibody titers ≥1:80 in duplicate MN assays and ≥1:160 in duplicate HI test using horse red blood cells (HRBCs) were considered positive as per WHO recommendations [9].
Because the antigenic characteristics of the H5pp were no longer adapted to the strains detected after 2009, we modified the testing strategy. Beginning 2010, sera were initially screened by HI test using HRBCs and all samples with titers ≥1:10 were further tested by MN. In both assays, the viral strain isolated during the corresponding outbreak was used. Based on our findings suggesting that compared to severe cases, patients asymptomatically-infected developed lower antibody titers [11], in 2010 we slightly revised the criteria for positivity as follow: MN titer ≥1:80 associated with HI titer using HRBCs ≥1:80.

Matched case-control studies
When seroprevalence surveys laboratory results became available, we conducted matched case-control studies by randomly selecting up to four A(H5N1)-seronegative subjects as controls for each case tested positive for anti-H5 antibodies. Controls were identified during the same investigation as cases and were matched to cases on age (+/-3 years), gender, village of residence and households with suspected/confirmed A(H5N1) infection in poultry (chicken flock with >20 % mortality or ducks with confirmed A(H5N1). A standardized questionnaire was used to comprehensively document participants' usual and recent activities in each affected village with regards to direct exposure to domestic poultry or wild birds (preparing food, children playing with poultry, husbandry practices), and indirect exposure with poultry in their living environment (house type, poultry cages/area, water source, ponds, rice field). Interviewers were not informed of respondents' casecontrol status. Parents or guardians were interviewed if participants were <12 years old.
As we were unable to conduct in-depth interviews in 2009, exposure information was drawn from seroprevalence data documented for seropositive cases and randomlyselected, matched control subjects.

Statistical analysis
Statistical analyses were conducted using Stata software (StataCorp, College Station, TX, USA). All tests were twotailed and statistical significance level was set at p < 0.05.
For statistical comparisons, we used Chi 2 or Fisher Exact tests for proportions, Student test for means and Kruskal-Wallis test for medians. Conditional logistic regression models and Wald Chi 2 test were used in bivariate and multivariate analyses to calculate maximum likelihood estimates for the matched odds ratios (OR) and 95 % confidence intervals (CI). All variables showing association at p < 0.20 with serology in bivariate analysis were included in the multivariate model using a backward process.
Three multivariate models were constructed with regards to the age of seropositive cases: (1) one model for all ages; (2) one for subjects aged less than 20 years; and (3) one model for subjects aged ≥20 years. Age and sex were taken into account in all multivariate models as possible confounding factors. All possible two-way interaction terms were tested separately on the final models and the likelihood ratio test was examined to find whether interaction terms significantly improved the model. Population attributable fractions (PAF) were computed for adjusted ORs associated with increased risk of infection.

Documenting A(H5N1) circulation in poultry in villages where A(H5N1) human cases occurred
Between years 2006 − 2010, we surveyed 453 households within a 1-km radius from where the six index A(H5N1) human cases resided. The proportions of households with poultry morbidity or mortality ranged from 34 % − 82 % depending on the year. An above 20 % mortality in chickens was documented in 16 % − 75 % of chickenrearing households. Among households whose ducks were tested, 30 % − 73 % had ≥1 duck which tested positive for anti-A(H5) antibody (Table 1). Additionally, we investigated one military training center by testing poultry raised there and identified H5-positive ducks.  Among study participants at a military center in Kandal province in 2008 (n = 394), most were military trainees (80.7 %), male (87.3 %), aged between 18 and 35 years old (76.4 %), and trainees who stayed at the center for at least 3 months. These participants all tested negative.

Factors associated with A(H5N1) infections to human
In total 35 cases and 115 matched control subjects were enrolled in the case-control studies (pooled data for 2006, 2007 and 2009 excluding military recruits). Bivariate analysis using conditional logistic regression for matched case-control analysis identified variables significantly associated with seropositivity at a <20 % level (Table 3), which were included in the multivariate analyses.
Among subjects of all ages, the following exposures were more frequently reported by cases positive by A(H5N1) serology compared to matched controls: having a poultry cage or nesting area located under or right next to the house (77.8 % vs 52.9 %; p = 0.060); swimming in ponds also accessed by poultry (40 % vs 30.4 %; p = 0.099); swimming in ponds also accessed by wild birds (47.1 % vs 29.2 %; p = 0.157); handling sick/dead wild birds (20.0 % vs. 8.7 %; p = 0.131); and transporting poultry to market (12.0 % vs. 2.2 %; p = 0.050).
In the subgroup of 22 cases aged <20 and their 77 matched control subjects, the following practices were more frequently reported among seropositive cases: Results from three conditional logistic regression models are presented in Table 4. Among subjects of all ages (35 cases and 115 controls), practices that appeared to be risk factors for A(H5N1) infection in humans after adjusting for the other variables were: having a domestic poultry cage or nesting area located under or adjacent to the subject's house (matched OR: 6.7

Discussion
Four consecutive serological surveys in humans conducted around identified cases during four of five years (2006-2010) by the same Khmer-speaking Cambodian field epidemiology team experienced in administering the same standardized questionnaire and collecting samples documented a low prevalence of A(H5N1) antibodies. These samples were analyzed by a reference diagnosis laboratory virology for A(H5N1). Although reference A(H5N1) diagnostic procedures improved with time, seroprevalence remained comparable across the years and even receded. At 1.5 % (ranging from 1.0 % in 2006 − 2.6 % in 2007), this low prevalence is in line with findings among populations not involved in poultry culling or sales, mainly in rural settings [1,[12][13][14][15][16][17].
The nested matched case-control study was undertaken to identify risk factors for transmission among those with evidence of recent infection by A(H5N1).
Exposure to A(H5N1) translates into infection risks at any age including young adults. When considering all age groups, having a poultry cage directly below the floor of stilt houses or appended to non-stilted houses was a significant independent risk factor. This confirms findings from studies in China [18] and Thailand [19] showing that living in close proximity to dead or sick poultry shedding virus or poultry increases infection risk. Transporting poultry to markets or collection sites was also an independent factor when all age groups were considered.
Childhood activities such as swimming in village ponds was significantly associated with recent A(H5N1) infection (OR: 4.6; p = 0.038), confirming data from previous studies in Cambodia [6,7]. In villages located far from the river, the alternative water source and water storage can be a small pond for washing/bathing, drinking/cooking,  [20][21][22]. A(H5N1) infections in two children in Vietnam may be linked to swimming/bathing in canal water also accessed by ducks [23]. Testing of environmental specimens taken from the ponds in A(H5N1)-affected area detected viral RNA in several samples including mud, water and aquatic plants [24,25]. Experiments have shown that infectious particles of A(H5N1) virus may survive in water and fauna for days even at high temperature, demonstrating the potential threat of contaminated ponds [26]. Moreover, studies in Vietnam [27] and Thailand [28] showed that documented human cases of A(H5N1) infection were significantly associated with lack of indoor water sources. Another identified activity common among children in the Cambodian countryside is preparing and consuming wild birds such as sparrows and cranes. The association of consuming wild birds with markers of recent A(H5N1) infection bordered on significance (OR: 4.9; p = 0.066). These birds are traditionally shot down or trapped and grilled over a small fire in the fields. Often, the deepmost internal parts of the bird are not fully cooked. Human cases of infection with A(H5N1) virus following direct exposure to dead wild birds were reported in Azerbaijan [29] and consuming uncooked duck blood has been suggested as a source of infection in some cases in Vietnam [30]. It has also been experimentally shown that small sparrows could theoretically be infected after contact with infected poultry and serological evidence of influenza contamination of these birds has been found in Cambodia [31].
In Cambodia, gathering or placing poultry in cages is usually done by older persons (aged 20 or above). The association of this practice with biological signs of recent infection bordered on significance (OR: 4.3; p = 0.072). The link between this activity and A(H5N1) infection was also well documented in Egypt [32].
Some factors were statistically associated with a lower risk of having markers of recent A(H5N1) infection. In subjects of all age, having a (private) pond in the house yard or presence of sick or dead poultry in the house yard were protective factors (OR: 0.2, p = 0.027 and OR: 0.2, p = 0.008 respectively). This is not paradoxical when the Cambodian context is taken into account. Having a pond in the yard entails a lesser probability of interacting with sick poultry in these private, well-maintained and protected ponds than in the communal ponds described above. Additionally, it provides better access to water and perhaps better personal hygiene practices such as hand washing after handling sick poultry. Having sick or dead poultry in the house yard is not in itself a good proxy for A(H5N1) circulation among poultry in Cambodia, where many co-circulating pathogens cause poultry death [33]. Furthermore, households in which poultry deaths occurred may be more prone to observe prevention messages and implement preventive measures.
The latter was equally true of "providing care for an identified human case of A(H5N1)", also found to be protective. Despite one human-to-human transmission event in Thailand, available epidemiological data show that A(H5N1) transmission to those providing care is extremely rare [34,35]. Furthermore, our detailed interviews showed that most study subjects who declared they provided care actually did little more than visit the human cases, without close or prolonged contact.
Paradoxically, eating food prepared from sick or dead poultry tended to be a protective factor. Although consuming well-cooked meat of infected poultry is well documented as posing little or no risk [36], the fact that this bordered on significance and appeared in independent analyses for both age groups but not in all-age analysis suggest that this is an artefact.
Among subjects aged below 20, feeding poultry was associated with a lower risk of recent A(H5N1) infection. Feeding poultry was understood in these surveys as a habitual activity of feeding live poultry. As backyard flocks in our settings were overwhelmingly constituted of chickens and not ducks, this activity was likely interrupted by signs of disease in the poultry and did not incur exposure to A(H5N1).
Our findings are subject to bias and limitations. Firstly, factors investigated were suggested by the literature and profound knowledge of the Cambodian setting but some additional factors may have been missed, such as individual host genetic factors [36]. However, this casecontrol study with cases matched on age (±3 years) and gender and the presence of A(H5N1) detected in at least some household poultry in the same villages naturally adjusts for many unidentified but possibly linked factors, such as host genetics in these Cambodian villages with low ethnic intra-village diversity.
Secondly, technical limitations may have biased our findings. During the five-year period, progressively improved assays may alter the year-to-year comparability of seroprevalence findings and the characteristics of cases. This improvement of diagnostic methods is to be expected from a reference laboratory and is unavoidable, but using the most recent reference strains improved detection in cases and therefore increased the sensitivity in our detection of factors associated with higher risk. In addition, if the screening approach differed over the time, the criteria for positivity essentially only slightly changed in 2010 after one of our study suggested that asymptomatically-infected individuals were expected to have lower antibody titers than severe human H5N1 cases. Thirdly, A(H5N1) antibodies decay could theoretically be associated with false-negative findings in study subjects [11]. Our seroprevalence surveys, however, were conducted within four weeks of the detection of index cases and poultry die-off.
Finally, our study is based on a limited number of surveys and cases. Although low power may explain borderline significance of some findings, a total of 35 cases and 115 controls provides adequate power to detect factors with a strong association, which are the most relevant in terms of public health and prevention. Furthermore, the fact that these were documented across several years in different villages reduces the risk of documenting risk or protective factors in a single setting at a single moment in time. Investigators who interviewed and sampled participants were de facto blind to subsequent serological results, adding to the reliability of the study findings.

Conclusion
Our findings from surveys conducted over several years in several different villages identifying proximity to poultry cages and swimming in communal ponds accessed by humans and poultry can reliably be extrapolated to the general Cambodian setting. Additionally, these surveys were conducted in the general population, not in high-risk groups such as veterinarians or poultry workers.
Findings were communicated to public health authorities who have since reinforced prevention messages. The identification of possible increased risks of A(H5N1) infection linked with hunting and consuming wild birds in Cambodia warrants further and careful study. Additional work is being conducted on documenting possible genetic polymorphisms associated with increased vulnerability or resistance to A(H5N1) infection in humans [34].