The dengue outbreak in Cat Ba revealed a clustered distribution of cases with the highest numbers of cases among floating village residents. Accounting for the differences in population sizes, the odds ratio to get infected by dengue in the floating villages as compared to the rest of the island was 4.9 (95 % CI: 3.6-6.7). This was particularly unexpected given that the 2012–2013 entomological surveillance in Cat Ba Island found the densities of both
Aedes aegypti and
Aedes albopictus to be lower in floating villages than on the island (no specifics of entomological surveillance methods are provided in this reference and results should be interpreted with caution) [
29]. Specifically, in winter (December, 2012), the number of
Aedes aegypti larvae per 100 houses (larvae index) in the floating villages was 0 while the index was higher in three other locations on the island (residential area: 0.8, tourism area: 0.9, Cat Ba national park: 0.5). In summer (July, 2013), the larvae index of these locations was 0.7, 8.3, 1.6, 3.0, respectively [
29]. However, as the population in the floating villages reportedly come from rural districts on the mainland where dengue transmission is relatively rare (oral communication with Cat Ba public health staff), those people were likely more susceptible to dengue infection as compared to the local people living on the island. This may explain why the chance of becoming infected in the floating villages was higher than on the island. In addition, the average distance between cases was less than 100 m and the duration of the occurrence lasted 31 days in the floating villages, indicating that vector control activities implemented in response to the outbreak in the area were likely not effective in reducing dengue transmission. This might be due to poor efficiency of mosquito coils used in the floating houses as compared to malathion 95 % ultra-low volume (MOH guidelines - 973QD/BYT) spray used. ULV was not used in floating houses because of a concern of harming aquatic organisms being farmed beneath the floating houses. In this study, the highest ER was found in a cluster with a short distance (15 m) and just 4 days. This finding is in accordance with previously published results on dengue using Knox statistics, despite different setting in choosing incremental steps [
16,
30‐
32]. For example, Tran et al. [
16] used the Knox test over a range of distances in space and time (6500 m and 200 days with steps: 5 m and 1 day) to examine a dengue outbreak of 161 suspected cases in Iracoubo, a small rural municipality in French Guiana. They found a space-time interaction with a relatively high risk within 15 m and 6 days of confirmed dengue cases. When the distances increased from this space-time limit, the risk rapidly decreased. Likewise, in a larger study area, Tartagal city in Argentina, Rotela et al. [
32] identified three clusters of 467 suspected dengue fever cases in a range of space and time (8800 m and 109 days with steps: 100 m and 1 day). The first two clusters were found within a short temporal distance (1–3 days) and long spatial distance (100 and 500 m to 2800 m). The third cluster was observed as a second epidemic wave, occurring since day 12 from the index cases, with the same space distance of the second cluster (500 - 2800 m). In principle, these studies found commonly that dengue transmission was very local and constrained by space and time. These patterns were attributed to the blood feeding behavior of the
Aedes aegypti mosquitos [
20,
33]
. As the
Aedes mosquitos often take a blood meal on multiple hosts, it increases the chance of infecting dengue viruses to multiple individuals from infectious mosquitos within short spatial and temporal distances (determined by the mosquito flight range). In the case of Cat Ba, this should be true for dengue patients in the floating houses with limited mobility. The particularly narrow clusters of dengue cases observed in Cat Ba Island was likely due to the introduction of a few infected mosquitos infecting multiple individuals, rather than by mobility of viremic humans.
There are two factors that prevent dengue to persist in Cat Ba island: the cold winters and the small size of the human population. The outbreaks usually start with an introduction of dengue viruses by travellers in the warm season, as previously observed [
10]. As the island receives a large number of travelers from the mainland each year, this is the most likely route of introduction. This hypothesis is further supported by the dengue surveillance on the island documenting that many domestic travelers from Hanoi and southern provinces (where dengue is endemic) arrived in Cat Ba from June to September in 2013 [
29]. While in previous years DENV1 and DENV2 dominantly co-circulated in northern Vietnam, there was a massive replacement of these two serotypes by DENV3 in 2013, as also observed in other places in Southeast Asia [
27,
28]. Our phylogenetic results also showed that the two DENV3 isolates in Cat Ba were likely originated from the mainland.
Our study had some limitations. First, household locations of dengue patients may not have been the places where the patients acquired infection; it is possible that the persons became infected elsewhere due to their mobility. In addition, as dengue cases are largely unapparent or represent mild symptoms [
2], using notified dengue cases only represents only a small fraction of all infections. In the case of Cat Ba outbreak, the population setting is relatively small and there is good access to healthcare, we thus consider that underreporting likely does not reflect a significant proportion of the total number of dengue cases in the floating village in comparison to Cat Ba town. Despite these limitations, it is worth noting that conspicuous clustering was still observed in the data. Second, with 3 % (6/192) of dengue patients confirmed as DENV3 by PCR, it is insufficient to conclude that DENV3 was the dominant cause of the outbreak. However, since DENV3 had reportedly replaced other previous serotypes on mainland [
27,
28], the introduction of DENV3 was likely responsible for the large dengue apparent cases in Cat Ba. Third, we only were able to sequence the E gene for two patients, limiting our phylogenetic analysis. However, in such a short timeframe in a single clear outbreak it is unlikely that another strain co-circulated.