Iracoubo is a small rural municipality located on the coastal plain of French Guiana with a population of 1,428 inhabitants (26), most of whom live in the main town or in the Bellevue village, located 5 km from the main town (Figure 1). Housing areas are surrounded by rain forest, mangrove forest, and coastal wetlands.

All patients who visited the healthcare center of Iracoubo with a temperature of >38.5°C, arthralgia, headache, or myalgia, were suspected of having DF. Blood samples were taken for evaluation of probable and confirmed DF cases. The terms suspected, probable, and confirmed cases of DF were used according to the definitions adopted by the Council of State and Territorial Epidemiologists and the Centers for Disease Control and Prevention (CDC), Atlanta, Georgia (27). A suspected case is defined as an illness in a patient whose serum was sent to National Reference Centre for Arboviruses (Institut Pasteur de la Guyane, Cayenne, French Guiana) for the diagnosis of DF. A probable case was an illness in a person that is clinically compatible with dengue, combined with supportive serologic test results (a single convalescent-phase serum specimen containing dengue virus immunoglobulin [Ig] M antibody, or a dengue virus IgG antibody titer of >1,280 by hemagglutination inhibition assay [HI]). A confirmed case was defined as having any of the following criteria: isolation of dengue virus from serum, demonstration of a dengue virus cDNA fragment by amplification (reverse transcription–polymerase chain reaction [RT-PCR]) from a serum sample, IgM antibody seroconversion, or a fourfold or greater increase in reciprocal titers of IgG antibody to one or more dengue virus antigens in paired serum samples.

During a dengue epidemic in a disease-endemic area such as French Guiana, the predictive positive value for a probable dengue case to be a confirmed case is very high (24). For this reason, we decided to include the probable dengue cases in the group of confirmed cases. Thus, we use the term confirmed case for both probable and confirmed dengue cases, and the term suspected case for all reported cases during the epidemic.

All tests were performed at the Institut Pasteur de la Guyane, National Reference Centre for Arboviruses (Cayenne, French Guiana).

Two techniques were used to detect antibodies to dengue viruses. The first was detection of IgM dengue virus antibodies by using an IgM capture enzyme-linked immunosorbent assay (MAC-ELISA) with a tetravalent dengue virus antigen. The procedure was modified from a previously described method (28).

HI was also used. HI titers were determined by using the method of Clarke and Casals (29) that was adapted to a microtechnique. Antibody responses to dengue virus were interpreted according to the World Health Organization criteria (30).

Acute-phase serum samples from febrile patients (<4 days after the onset of fever) were diluted 10-fold in Leibowitz medium containing 3% fetal calf serum, and dilutions were injected into subconfluent AP 61 cell cultures as previously described (31). After 7 days of culture, cells were harvested, and dengue viruses were identified according to serotype by an indirect immunofluorescence assay (IFA) with monoclonal antibodies specific to DEN-1, -2, -3, and -4 viruses (provided by CDC, Fort Collins, CO).

Viral RNA was extracted from a 50-μL aliquot of acute-phase serum with TRIzol (Invitrogen Life Technologies, Paisley, Refrewshire, UK), according to the manufacturer’s recommendations and precipitated with isopropanol and 1 μL of glycogen (5 μg/μL) (Roche Diagnostics, Mannheim, Germany). Air-dried RNA pellets were suspended in 20 μL of water. Then, 5 μL of RNA were mixed with 200 ng of random hexamer primers, and first-strand cDNA synthesis was performed with the SuperScript First-Strand Synthesis System for RT-PCR (Invitrogen Life Technologies), according to the manufacturer’s recommendations. The first run of RT-PCR analysis and subsequent seminested PCR analysis were performed following a previously described procedure (32).

For all suspected dengue patients, patient’s home was recorded with a cadastral map (paper copy, scale 1/1,000). Georeferenced aerial photographs were used to improve the identification and the location. The geographic coordinates were integrated into a GIS (Geoconcept software) (33), with the following information about the patient: identification number, date of onset of symptoms, age, sex, diagnosis.

Assuming that DF spread within a community leads to the creation of transmission focus, the distance between neighboring housing would be an important factor in the spatial extension of these foci. We used the GIS and geocoded aerial photographs to locate all houses and to calculate the mean distance between neighboring houses (Geoconcept software) (33).

We used the Knox (34) test to identify possible space-time interactions, i.e., to determine whether cases which are close in distance will also be close in time. This method evaluates whether the number of pairs of cases found at a fixed temporal and spatial distances is substantially different from the number of pairs of cases expected at these distances by chance, when the times of occurrence of cases are randomly distributed across the case locations.

The ratio between real number of pairs of cases found at the space-distance s (in meters) and the time-distance t (in days) and the number of pairs of cases found at these distances by chance could be considered as the RR of occurrence of another dengue case, t days later and s meters away from the first case of dengue.

The Knox test was first computed for the population of patients with confirmed cases and for the population of those with suspected cases. Results were calculated for time distances varying from 1 to 200 days (duration of the epidemic) by 1-day step and spatial distances varying from 5 to 6,500 m (step: 5 m). An “RR map” was then obtained by interpolating the significant values (p = 0.05) (Surfer software) (35).

The final result is a representation of the RR, when space-distance and time-distance from a hypothetical dengue patient vary. The correlation between RR values derived from the confirmed cases and those derived from the suspected cases was evaluated.

In Iracoubo center and Bellevue, 161 patients with suspected dengue cases were reported between April and November 2001, which corresponds to 11.3% of the population. Blood samples from 57 patients were analyzed in the National Reference Center for Arboviruses, Arbovirology Laboratory of the Pasteur Institute of French Guiana. Among the 57 patients, 32 cases of DF were confirmed (56.1%). A total of 25 suspected cases were not confirmed; among them 4 cases were indeterminate and 21 were negative (Table). Virus isolation results show that the majority of confirmed cases were caused by DEN-3 (90%).

The first suspected dengue case was reported on April 10 but samples were not analyzed. The first confirmed dengue case occurred on April 22. Then, the epidemic spread rapidly through the community (Figure 2), with a temporal lag between cases occurring in Iracoubo center and those occurring in Bellevue (Figure 3). Indeed, 100% of the confirmed cases in Iracoubo center occurred between April and July, whereas 76.9% of confirmed cases in Bellevue occurred in October and November. The first glimpse of the spatial distribution of confirmed and suspected cases shows the existence of apparent spatial clusters (more than 2 confirmed cases or 3 suspected cases in the same neighborhood) (Figure 2).

We considered 406 buildings in the calculation. The mean distance between adjacent houses was 24.6 m for the whole municipality, including means of 22.6 m for Iracoubo center and 29.6 m for Bellevue.

A first analysis of the RR variations when space and time distances vary over all the epidemic’s extent highlights a main risk area, with RR > 1 (p < 0.001) (Figure 4). This area corresponds to a substantial increase in the theoretical risk of the occurrence of another dengue case. This area is active inside the boundaries of 400 m and 40 days.

A more detailed analysis of this risk area shows a strong heterogeneity: an area is very high risk (RR > 5) at short distances (15 m) and over brief periods (6 days). Beyond these space-time limits, the RR rapidly decreases (Figure 5). Moreover, particular patterns are observed, like a temporal periodicity, with peaks of risk every 3 days (Figure 6a). Spatial breaks seem to appear at the approximate distances 20–25 m, 45–50 m, and 80–85 m (Figure 5 and Figure 6a), showing three different risk levels (Figure 5).

A strong concordance exists between the results obtained by using the dengue laboratory-positive cases and those obtained by using all suspected cases: the space and time boundaries are roughly the same (Figure 6b). Although the RR values are different for the same space and time distances, they are correlated with a high correlation coefficient (r = 0.93; p < 0.05).