We conducted an immediate investigation in response to notification of the confirmed human H5N1 case in Banteay Meas District. From March 25 through 27, 2005, all households located within a radius of 1 km from the H5N1 case-patient's household were mapped and positioned with a hand-held global positioning system (Garmin, Olathe, KS, USA). We collected information on illness suggestive of H5N1 among animals in each household by interviewing the head of the family with a standardized questionnaire. Households where the head of the family was not at home or could not be found were omitted.

A household chicken flock was considered likely to have been infected by H5N1 virus during the previous 6 months if all of the following characteristics were reported: flock death >60%, 100% case-fatality ratio, and sudden death of young and mature birds within 1 or 2 days of becoming sick. We collected sick poultry and carcasses for H5N1 virus testing. Cloacal swabs of 10 to 14 randomly selected, live, healthy poultry were also collected from each household where birds remained.

We conducted a seroepidemiologic investigation June 3–7, 2005, ≈2 months after the village reported high poultry deaths. It consisted of interviews of household members with a standardized 39-question questionnaire on demographic information and data on specific exposures to animals and the environment during the last 12 months; a 5-mL venous blood specimen was also collected from participants. Four investigation teams of 3 members each visited all households in 4 different directions, starting from the household of the confirmed human case-patient, until 300 participants were enrolled in the study. Each household was visited once, and no further attempts were made to interview absent adult household members. The sample size was estimated to have a 95% chance of detecting >1 seropositive person, if one assumes a 2% prevalence of H5N1 antibodies in the village. Written informed consent was obtained from adults or from a parent or guardian for children <18 years of age. The study was approved by the Cambodian Ethics Committee.

All animal samples were placed into viral transport medium in sterile tubes in the field, kept cold, and transported daily to the National Animal Health Laboratory of the Ministry of Agriculture, Forestry and Fisheries in Phnom Penh. Cloacal specimens and organ samples from sick and dying poultry were tested for influenza A with an indirect fluorescent antibody assay. Positive results were forwarded to the virology unit of the Institut Pasteur in Cambodia for confirmation with real-time RT-PCR to detect H5 viral RNA. Human blood specimens were centrifuged, and sera were aliquoted and frozen at –80°C. Sera were shipped on dry ice to the Hong Kong Government Virus Unit Laboratory for detection of H5N1 neutralizing antibodies by microneutralization assay and confirmatory Western blot assay. Serologic evidence of H5N1 virus infection was defined as an H5N1 neutralizing antibody titer >80 with a confirmatory Western blot assay (9).

The position codes of all surveyed households were entered into ArcGIS version 9.0 (ESRI Systems, Redlands, CA, USA). We used the space-time scan statistic to determine the cluster of households most likely to have been affected by H5N1 virus in the previous 6 months. Analysis was performed with SaTScan version 5.1.3 (10); cases were assumed to follow a Poisson distribution. The space-time statistic is defined by a cylindrical window with a circular geographic base and height pertaining to time. The window is moved in space and time for each geographic location and size. We obtained several overlapping cylinders of different sizes and shapes covering the study area; each cylindrical window reflected a possible cluster. The most likely cluster is the one least likely to have occurred by chance, according to the maximum likelihood ratio test statistic. Individual and household data were entered into EpiData version 3.02 (EpiData Association, Odense, Denmark) and validated with a duplicate data file. STATA version 8.0 (StataCorp LP, College Station, TX, USA) was used for all statistical analyses. Odds ratios were estimated with bivariate logistic regression. Independent associations between demographic and behavioral data and households that were likely to be affected by H5N1 in poultry were also analyzed by logistic regression models. We accounted for the cluster effect of households with STATA's "cluster" option for logistic regression, which specifies that observations are independent across households but not within households. For multivariate analysis, variables with a p value <0.1 were retained in the models. Selected variables whose correlation coefficient was >0.4, which indicates colinearity between these variables, were not included in the logistic regression model.

These investigations were conducted as a collaborative effort between the Cambodian Ministry of Health and Ministry of Agriculture, Forestry and Fisheries, Institut Pasteur in Cambodia, the World Health Organization, Hong Kong Public Health Department, the Centers for Disease Control and Prevention, and the Food and Agricultural Organization of the United Nations.

Of 194 households located within 1 km of the H5N1 patient's house, 163 (84%) had occupants who were home at the time of the survey. No household refused to participate.

Among interviewed households, 155 (95%) raised chickens (median 20, range 1–80 per household) and 52 (32%) raised ducks (median 4, range 1–50). Fifty households (31%) reared both ducks and chickens. Sixty-three households owned pigs (range 1–4 animals). From January 1 through March 26, 2005, 102 (66%) of 155 households reported deaths of chickens. Of these households, 73 (72%) recorded deaths during the last 4 weeks of this 3-month period (Figure 1 and Figure 2). The median poultry flock death ratio was 90% (range 4%–100%). According to our definition, 42 households were likely to have had an outbreak of H5N1, for an overall attack rate of 27% among households with chickens. Flock death ratio was >80% in 31 of these households. The initial mean flock size in households likely to have had H5N1 in chickens was significantly larger than in households without chicken deaths (31 vs. 20, p = 0.003).

Eleven households with a high likelihood of H5N1 (35%) in chickens also owned ducks, although only 2 of these described simultaneous deaths of ducks: 1 reported a duck flock death ratio of 80%, the other a ratio of 100%. Seven other households reported high levels of death among duck flocks (>60%), but the number of deaths among chickens did not suggest H5N1. Overall, raising ducks with chickens was not associated with deaths in chickens (p = 0.57).

Cloacal swabs were collected from 28 chickens and 14 ducks. Specimens from 2 sick chickens were positive for H5N1 by RT-PCR. These chickens belonged to a household located ≈50 m from the household of the confirmed human H5N1 case-patient. The owner of the sick chickens reported that the farmer with H5N1 virus infection spent daylight hours in his compound.

The space-time scan statistic detected a significant cluster of 25 (60%) households with an overall relative risk of 7.9 (log likelihood ratio 34.1, p = 0.001) (Figure 1). The cluster was confined to the period from February 25 through March 26, with a radius of 444 m, which contained both the household of the confirmed H5N1 case-patient and the house with the 2 H5N1-infected chickens.

Among 93 households that were surveyed, 351 persons participated and 3 refused. An average of 4 people resided in each household, the median age was 23 years (range 1 month–81 years), and 150 (42.7%) of the sample were male; 207 (59%) were farmers of both crops and livestock. The rest of the participants were students (29.3%), had no stated occupation (18.8%), or were construction or factory workers (0.9%). Reflecting the rural setting and the common means of livelihood, ownership of animals, including poultry, was high (Table 1). The number of households with chickens decreased by 17.5% (p<0.001) after the outbreak, while the number of households that possessed ducks and pigs remained similar (–15.7% and 5.7%, p = 0.43 and 0.74, respectively).

Because many households owned poultry or pigs, a substantial proportion of the surveyed population had regular, high-intensity contact with these animals in the 12 months before the survey; this contact included collecting, processing, and eating sick birds or birds that had recently died when H5N1 viruses were thought to be circulating among flocks in the village. Despite this finding, none of the villagers interviewed reported having a febrile or respiratory illness during the same period, and none of the 351 participants had neutralizing antibodies suggestive of H5N1 virus infection on microneutralization assay.

We compared exposures of residents from households with a high probability of having had an outbreak of H5N1 in their chicken flock (n = 96) with occupants from households where no chickens died (n = 166) (Table 1). Bivariate analysis showed that households that purchased live poultry in the preceding year were almost 4× more likely to have had H5N1 in their flock than households that did not buy live chickens. In contrast, certain behavior by household members appeared to reduce the risk for H5N1 virus infection in their household flock by half (and the difference remained significant after controlling for poultry purchasing): cleaning cages or stalls, cleaning up poultry feathers, and handling live poultry (Table 1). Slaughtering chickens was not a significant risk factor after controlling for exposures that were significant on multivariate analysis (Table 2).