This study combined spatial and spatio-temporal analyses, which have been proven useful [
42] for assessing the distribution of the incidence of dengue in intra-urban settings. Although the observed clusters in Girardot shifted throughout time and space (Fig.
7), these patterns might be the result of different territorial dynamics related to urban functions, urban changes (i.e., urban expansion), demographic composition, or touristic activities. Further understanding of dengue persistence (space clusters) and outbreaks (space time clusters) in the context of hyper-endemic urban spaces could eventually help to guide prevention measures from a territorial perspective (i.e., where, when and how to intervene).
Space clustering of the incidence of dengue
The geography of dengue in Girardot is characterized by annual concentrations of cases in clusters of different sizes across disparate places throughout the city. Similar results were obtained in a study in Delhi [
15], where a similar number of cases were assessed over 2 years. The results showed concentrations of cases in different locations. The author emphasized that a system in apparent equilibrium (i.e., the same number of cases in two consecutive years) might hide an unstable configuration of the incidence of dengue on a finer scale. Although the number of cases in Girardot changed over years, similar numbers were reported in 2012 and 2015 (380 and 358, respectively). Nevertheless, the results for the general Getis-Ord index showed that the cases in 2015, unlike 2012, did not correspond to an agglomerated pattern; thus, assessments of local clustering for this particular year are not recommended. Even more drastic that the results from Delhi, the similar count of dengue cases in Girardot might not be significantly concentrated.
Nevertheless, this study described the persistence of significant concentrations of dengue cases, defined as the overlapping of clusters for 2 or 3 years of the study period as well as in both seasons (Figs.
4 and
5). A 3-year study in another hyper-endemic urban context [
16] found a low percentage of the spatial units registered dengue cases in 2 or 3 of the study years. Therefore, an absence of permanent clusters was described in the context of a highly populated city (Delhi) undergoing the reintroduction of the DENV-1 serotype 5 years after its last circulation. The overlap of significant clusters described for Girardot accounted for a small percentage of the total cases for each year or season (Table
2); nevertheless, this persistence was observed despite the local effect of monotypic herd immunity, which can weaken the persistence of the incidence as described by Telle et al. [
16]. Moreover, these clusters showed a distribution that was coherent with the local epidemiology of dengue and residential structures.
Non-significant median age differences were found among patients in clustered and non-clustered patterns in Girardot (Table
3). However, the higher percentage of cases in patients younger than 15 years old and the lower mean age of clustered cases supports the hypothesis that the presence of vulnerable populations shapes the distribution of the identified clusters, given that the most affected age group in Colombia shifted from 15 to 44 years old towards 0 to 15 years old [
41]. This result is a response to the actual hyper-endemic transmission in cities such as Girardot and characterized by the simultaneous circulation of all four dengue virus serotypes observed since 2004. A multi-country study in Asia and Latin America (5 countries in each region) assessed symptomatic dengue and seropositivity in children [
43]; Colombia contributed the largest subcohort of the region (
n = 3245) distributed among 9 municipalities, including Girardot. Although non-specific data were available for Girardot, Colombia reported the highest seropositivity at baseline (92.3%) in Latin America, and it was the only country of the region to report the circulation of all 4 dengue serotypes. Additional studies are needed to assess the impact of demographic composition at a fine scale on the incidence of dengue in the context of hyper-endemic transmission.
Regarding the distribution in the residential structures of the described persistent clusters, three of the five areas that showed this persistent behaviour corresponded to neighbourhoods of local residents (non-tourists) with low-to-medium socioeconomic conditions; one corresponded also to neighbourhoods of local residents with heterogenic conditions but with a predominance of medium socioeconomic conditions; and finally, one was related to a high stratum of secondary residence complexes (Brisas de Guadalquivir). Although the available information for this study did not enable an assessment of the specific socioeconomic conditions of each of the five mentioned areas, a relationship between the structural socioeconomic conditions and the persistence of dengue is likely. A previous study of Girardot [
44] that evaluated dengue cases (1998–2002) from an ecological approach obtained results consistent with those discussed here. Using a standardized incidence ratio (i.e., the incidence of dengue for each neighbourhood divided by the incidence of dengue for the whole city), a substantial number of neighbourhoods were identified at high risk [
44], including the same specific areas of persistent clusters discussed earlier (excluding the high-class complex Brisas de Guadalquivir, which did not exist at the time). Furthermore, a general relationship was found between high-risk neighbourhoods with low socioeconomic status and the presence of the vector in households. However, these general findings were not suitable for assessing specific intra-urban dynamics.
A systematic review argued that there is no consistent evidence to affirm that poverty is a predictor of dengue [
13]. This lack of evidence might also be related to the scale of analysis usually adopted to assess this type of relationship. Considering the persistence of dengue at an intra-urban scale might help define a more appropriate scale to measure the relationship between dengue and socioeconomic conditions, which might also help explore how other systems interact to explain why not all deprived areas have concentrated cases or why dengue cases are also concentrated in high-stratum areas such as Brisas de Guadalquivir in Girardot. Nevertheless, we agree with other authors [
7] that this type of finding should strengthen rather than discourage public policy to attain sustainable urban development, including dengue prevention and control programmes, especially for those with greater need.
Olivier Telle [
15] found that intra-urban deprived and densely populated spaces in Delhi were associated with dengue concentration and the permanent presence of the dengue vector throughout the year. This relationship might play an important role in maintaining viral circulation during inter-epidemic periods. The degree of vulnerability of deprived urban areas and their relationships with vector presence have yet to be adequately measured in Girardot. Nevertheless, the persistence of the incidence of dengue was found near the Bogotá and Magdalena Rivers and in the oldest, more consolidated and less fragmented regions of Girardot. This finding contrasted with the northern part of the city, which presented with a more fragmented residential morphology, an important commercial and tourism-related infrastructure (e.g., shopping malls), and the two urban expansion areas of the city (i.e., the northern and western fronts).
Related to the urban morphology of the city, dengue clusters identified during the rainy season (Fig.
5) were located near small natural drainages (caños) or open, undeveloped terrain. Although the vector productivity in Girardot has been primarily related to indoor recipients [
45], this result suggests that specific places exist where the rainy season increases the productivity of breeding sites in public spaces and therefore augments the risk of dengue and the number of cases. This suggestion is a plausible hypothesis, given that undeveloped terrains and “caños” are usually unattended and filled with rubbish, creating a potential accumulation of rainwater and consequent breeding sites for
Aedes aegypti. Major breeding sites in public spaces (specifically sewers), have been reported in other Colombian cities such as Cali [
45].
One overall point about the persistent behaviour of dengue concentrations is that it seems to be related to low and medium socioeconomic conditions. In addition, this persistence was correlated with more consolidated and stable urban systems, unlike the north and west expansion areas of Girardot.
Space-time clustering of the incidence of dengue
Regarding the results of the spatio-temporal analysis, the general spatial distribution differed from the pattern discussed regarding persistent agglomerations of the incidence of dengue. Outbreaks have been described along a diagonal from the northeast towards the southwest of the city, following the axis of urban expansion described earlier, whereas a persistence of cases was present in the southern part of the city. However, common places remained (5 of the 10 spatio-temporal reported clusters) where these two distributions overlapped, specifically throughout the onset and initial evolution of the 2013 epidemic (November 2012–February 2013).
However, the observed cases within the space-time clusters represent a small percentage of the total reports (4.25%). This result diverges from that obtained by a similar analysis that assessed a specific outbreak with the same measurement, finding 65.3% of total dengue cases [
46]. Another study (with the same model parameters: 300 m and 20-day windows) observed 67% and 62% of the total cases in spatio-temporal clusters over two consecutive years with similar numbers of cases [
15]. In the context of hyper-endemic dengue transmission in Girardot, cases were distributed throughout the city year-round. Therefore, space-time clusters might represent particular transmission dynamics more than focalized starting points for generalized epidemics. Nevertheless, 68% of the cases grouped in space-time clusters were located in areas with tourist infrastructures and secondary residences; furthermore, they occurred during the dry season, which is the longest and most tourism-related period of the year in Girardot (December and January).
This temporal distribution contrasts with the described relationship during the rainy season and the increase in dengue cases in Colombia [
4,
41]. In Girardot, a clear pattern of dengue cases is notable during the dry season. The results of an entomologic study support this finding, stating that Girardot reported (in 20 randomly selected clusters composed of approximately 100 households each) more recipients with water, a slight increase in low tanks (albercas) with high water levels and an overall increased productivity of
Aedes aegypti pupae during the dry season [
47]. Importantly, 19.6% of 3228 reported cases in the initial processing of the SIVIGILA database (2012–2015) corresponded to rural residents and residents from other municipalities. This finding led to the perception of important city migration fluxes due to tourism, low-wage construction workers and population from surrounding municipalities in search of services. These migration dynamics are commonly reported for Girardot but have not been well documented or measured.
The spatio-temporal clusters also describe the displacement from the northeast (November 2012) to the southwest (February 2013). In another study, the possible effect of wind was discussed, which was associated with the dispersal of dengue epidemics [
46]. In this case, however, it seems that social structures and the migration of susceptible populations remain as possible explanations. The onset date (1 January) of three simultaneous space-time clusters might be correlated with the long tourism season of the city because two clusters overlap with high-stratum secondary residence complexes. One possible hypothesis is that the arrival of susceptible populations to secondary residence complexes during a season when the vector densities are higher [
47] results in conditions conducive to the spread of the virus, thereby generating symptomatic reported cases as well as asymptomatic infections that generally affect locals and foreigners. Although the influence of tourism has been explored in other studies [
4,
19,
27], more evidence is needed for Colombian cities such as Girardot. Findings concerning touristic economies and risk of outbreaks are sensible and have an important effect in local economies, as reported in the Indian Ocean during the chikungunya outbreak (2004–2006) [
48]. In the case of Girardot, recent outbreaks of chikungunya and Zika have concerned local authorities regarding health and economic impacts (mainly in tourism).
To attain a better understanding of the possible transmission dynamics relative to the interactions between ecological and social systems, much remains to be learned about the intra-urban circulation of dengue serotypes and their relationships with the movement of people (at different scales), especially with regard to Girardot and Colombia. Much has been said about the effects of daily movements, including the methodological challenges regarding accurate measurements and the estimated effect of daily movement on dengue transmission [
7,
30,
31,
49‐
51]. Intra-urban daily movements have been suspected in hyper-endemic cities such as Girardot, given the low percentage of cases represented in spatio-temporal clusters. Assessing cases by their residential address can affect a spatio-temporal analysis, given that people might be infected away from home. Additional evidence is needed to understand how daily movements shape dengue transmission in Girardot.
Dengue in Girardot: Among places of persistence and places of outbreaks
The incidence of dengue in Girardot was characterized with space and space-time manifestations throughout an urban space. Each cluster had different spatial patterns; however, a general pattern was evident, in which persistent clusters (local Getis-Ord indices) and space-time concentrations (Kulldorff’s scan statistics) occurred along the periphery of the city, identifying a weak presence of the disease in the historical and geographical centre of Girardot. Although the incidence analysis controlled for the effect of population density, spatial units with low population denominators (which are usually present in city centres) can overestimate the incidence, even when they present a reduced number of cases [
15]. This finding was not the case for Girardot, even that population density is lower in the city centre because of other non-residential land uses; nevertheless, no apparent dengue concentrations were detected.
In other contexts, the city centre has been described as having a high incidence of dengue [
7,
52]. However, peripheral vulnerable areas have also been emphasized to play a role in understanding dengue transmission, and interventions are recommended in these places even if incidence of dengue is low [
7,
15]. The periphery of Girardot was not specifically correlated with vulnerable conditions because of the mixed residential pattern already described. Nevertheless, the peripheral spaces in Girardot are dynamic, and they demonstrate the recent urban changes that the city has undergone in terms of housing construction and commercial/tourism activity emplacement. In contrast to dengue persistence, spatio-temporal clustering seems to be related to areas where seasonal fluxes are present, specifically with regard to susceptible populations. In the same sense, dynamic and changing urban spaces (expansion and changes in urban functions) were related to the spatio-temporal concentrations of the incidence of dengue.
Complementary use of space and space-time analysis can reveal different underlying territorial dynamics. Structural or long-term urban conditions (e.g., deprived areas) might be related to persistent concentrations of dengue to a certain extent. By contrast, more unstable and temporal urban dynamics (urban expansion and the movement of people on different scales) might enhance spatio-temporal concentrations. This possible pattern would certainly need to be evaluated in the context of and particularities of specific urban spaces before arriving at generalized conclusions.
Study limitations
Working with notified dengue cases (SIVIGILA) posed a series of challenges mentioned throughout this paper. The absence of serotype information related to reported dengue cases limits the interpretation of the space and space-time clusters described in this study. One of the most important limitations concerning this type of study is the accuracy of address information in Colombia, which required an important and comprehensive process of database editing. In addition, differential access to the healthcare system (in which cases are officially reported) might have underestimated the incidence of dengue among specific social groups (e.g., inhabitants from more deprived areas). Finally, no evidence exists as to whether the distribution of reported symptomatic cases is comparable to the asymptomatic non-registered cases; if important differences exist, then the results might be biased.
Context-related limitations were also present, such as the availability of updated populations and socioeconomic information, particularly with regard to the detailed spatial units that enabled additional exploration of the distribution of cases in urban areas. Population denominators are crucial for this type of analysis, and although approximate results can be achieved using the available information, having more accurate data that can lead to better results and more useful recommendations is always desirable.