A probable case was defined as someone presenting with fever and bleeding from their gums, nose, eyes, rectum, lungs, or gastrointestinal tract, between October 1997 (the onset of flooding) and February 8, 1998. If specimens were available, diagnostic laboratory tests for RVFV were performed.

In Garissa, active case-finding for hemorrhagic fever was performed by members of local law enforcement, village chiefs, and other government officials. Periodic reports were provided to the district health officers and the task force. We disseminated a case-report form to all district medical officers and international relief agencies involved in medical care in the area. When possible, reports of possible cases were investigated to obtain demographic and clinical information, as well as blood, stool, or tissue specimens. The WHO Hemorrhagic Fever Task Force was conducting similar surveillance throughout Kenya and in southern Somalia and northern Tanzania.

Initially, specimens were transported to the Africa Medical Research Foundation in Nairobi, where serum specimens were centrifuged, divided into aliquots, and forwarded to the Viral Research Center/Kenyan Medical Research Institute. Subsequently, a laboratory operated by Médècins du Monde was established in Garissa to process clinical samples. Within 12 hours of collection, serum specimens were stored at 4°C until packed and shipped to National Institute for Virology and CDC for RVFV-specific IgG and IgM antibody enzyme-linked immunosorbant assay (ELISA), virus isolation, RT-PCR and sequencing of the DNA product, or immunohistochemistry using previously described methods (16–19).

For the cross-sectional survey, a laboratory-confirmed case definition was used. A case of recent infection was defined as presence of RVFV-specific IgM antibodies by ELISA. Persons with RVFV-specific IgG antibodies by ELISA and no IgM antibody were considered to have been infected before this outbreak. In our initial calculations, we considered persons with only anti-RVFV IgG antibodies to be protected against infection. The background prevalence of past RVFV infection in Kenya is not known, but IgG positivity as high as 40% has been shown in populations at high risk after epizootics (20). In contrast, a 1983 study for serologic prevalence of hemorrhagic fever viruses in Kenya demonstrated that <1% of persons sampled had anti-RVFV IgG, but a less sensitive immunofluorescent antibody test was used (21). This study did not include the North Eastern Province. Additionally, the duration of IgM antibodies is not well documented. Therefore, our analysis included persons with either anti-RVFV IgG or IgM antibodies.

We selected participants for the cross-sectional survey from the non-refugee population of Garissa District using a modified multistage cluster design (22). In Kenya, each province is divided into divisions that are made up of locations and then sublocations. We used sublocations as the units for choosing clusters. Thirty clusters (sublocations) were selected from 12 divisions. The number of clusters per division was weighted to represent population density. Clusters were then chosen randomly. Seven households were chosen in each cluster by systematic random sampling for a total sample size of 210 persons. The investigating team identified one person in each household for recruitment into the study. To reflect the age distribution of the RVFV infection and hemorrhagic fever cases already identified in the outbreak, we included one child between 2 and 9 years old, five persons between 10 and 49 years old, and one person ³50 years in each cluster. Between February 8 and 14, 1998, three field teams interviewed and obtained samples from 202 persons from 29 clusters. Three of the selected cluster sites had been destroyed by the flood and could not be located. Replacement sublocations were sampled for two of the three destroyed clusters. In one cluster, only six persons were sampled. In four clusters, children <10 years of age were not sampled and were replaced by adults.

After informed consent was obtained, a blood specimen was collected and each participant was interviewed. We used a standardized questionnaire that included demographic characteristics (age, gender, family size), exposure information (e.g., slaughtering practices, butchering, consuming raw meat and milk), environmental factors (displacement by flood, type of settlement, loss of livestock), and history of illness between the start of the floods and the date of the interview. Exposure to mosquitoes was evaluated through questions about attempts to reduce bites (i.e., mosquito nets, fires, other methods). Otherwise, all persons were assumed to share a similar risk for insect bites. Local health workers fluent in English, Kiswahili, and Somali were trained to administer the questionnaire and were supervised by an epidemiologist. Interviewers recorded information in English.

Blood specimens were kept at ambient temperature for <6 hours. Specimens were processed at the Médècins du Monde laboratory as noted above. RVFV-specific IgG ELISA, as well as IgM ELISA, was performed on all blood specimens. At the time of the cross-sectional survey, there were no reports of severe manifestations of illness and we believed that virus transmission was not ongoing. Therefore, virus isolation and PCR were not performed on these specimens.

Data from completed questionnaires were double-entered into databases by using Epi Info version 6.0 (CDC, Atlanta, GA). Univariate and multivariate analyses were performed by using SAS (version 6.12, Cary, NC). Poisson regression was used with a generalized estimating equations algorithm to control for the clustered nature of the data (23).

Of the 202 persons enrolled in the cross-sectional study, 31 (15%) were positive only for anti-RVFV IgG (i.e., previously infected). Although persons having only IgG antibody were widely dispersed geographically, the highest prevalence was found in the Hulugho Division (32%) and the Masalani Division (29%). The highest percentage of previous infection by age group was for persons ³65 years of age (Table 2). Characteristics of the sampled population are provided in Table 2.

Of the 171 susceptible persons in the sample, 31 (18%) were positive for anti-RVFV IgM. The percentage of those with detectable IgM antibody varied with age; this antibody was detected in 5% of children <15 years of age, 23% of persons 15 to 65 years of age, and 13% of persons >65 (Table 2). The age-adjusted prevalence of IgM antibody positivity standardized to the population of Garissa District was 14% (95% CI 11-16). Sixteen clusters, representing all 12 of the administrative divisions in Garissa District, had persons with anti-RVFV IgM antibody, indicating recent viral transmission. Of the 31 persons with serologic evidence of recent infection, 30 (97%) reported having a recent illness, compared with 116 (83%) of the antibody-negative persons (p=0.06); no symptoms distinguished recent RVFV-associated illness from other illness.

Assuming an age-adjusted IgM antibody prevalence of 14% in the susceptible population, we estimate that approximately 27,500 persons were infected with RVFV during this outbreak in Garissa District alone. If the presence of IgG antibody were included in the case definition for recent infection, then the standardized prevalence would increase to 23% (95% CI 20-26) and would represent approximately 53,000 infected residents.

Certain demographic characteristics (Table 2) were associated with higher rates of infection as determined by IgM ELISA, including rural habitation (p=0.02) and household size of less than four persons (p=0.001). Age <15 years was associated with a lower rate of recent infection (p=0.05). A large number of animal contact activities, including herding, milking, slaughtering, and sheltering animals in the home, were statistically associated with recent RVFV infection (Table 3). The association was greatest with sheep-related activities, especially for those resulting in contact with sheep blood or body fluids.

Many of the animal contact variables were highly associated with infection in univariate analysis, but were also highly colinear. Multivariate analysis of composite variables for species-specific activities that resulted in similar exposures and potential confounders demonstrated a significant association between recent RVFV infection and persons who had contact with sheep blood, amniotic fluid, or milk (not including milk consumption; RR 3.0, 95% CI 1.3-6.7) (Table 4). Sheltering any domestic livestock (mostly sheep and goats) in one=s home during the flood also remained independently and significantly associated with infection in the multivariate analysis (RR 3.5, 95% CI, 1.3-9.1). Although not significant in univariate analysis, being male also was significantly associated with infection (RR 1.6, 95% CI 1.0-2.8). Age <15 years was associated with a reduced risk for infection when controlling for the other risk factors (RR 0.3, 95% CI 0.06-1.0). Drinking raw sheep milk was independently associated with infection, but did not reach statistical significance (RR 1.6, 95% CI 0.9-2.9).

When the analysis was repeated using detection of either anti-RVFV IgM or IgG as the outcome variable, which effectively doubled the number of cases of RVFV, only contact with sheep blood or body fluids and sheltering animals in the home remained significantly associated with illness in the multivariate model (data not shown).