Abstract

Author summary

The increased geographic spread and intensity of dengue is due to numerous factors including, increased urbanization, human migrations and air travel, flooding and global warming. In the Caribbean, outbreaks continue to occur with hyperendemic occurrence of the disease. This is mainly due to the use of reactive programs and limited resources available to control the disease. Using the island of Trinidad as a case study, we show that higher rates of infection occur in areas with a history of dengue incidence. Also, a general pattern in the movement of dengue cases is found leading up to and transitioning away from an epidemic occurrence, and associated with the locations of transportation hubs. These findings can be used to contain the disease in a more efficient and effective manner. Also, few studies have examined the space and time relationship of dengue incidents at local scales in the Caribbean islands. Other islands can adopt the approach used to better allocate resources and understand the disease. This information can then be used to gain regional perspective and understanding about the spatio-temporal persistence of dengue in the Caribbean.

Background

The concept that human movement enhances the spatial distribution of dengue fever is not new and has been demonstrated along busy routes of human traffic in Asia [1-3] and more recently in the Americas [4,5]. In fact wherever people congregate in dengue infested countries whether at home, in church with friends and at relatives and festivals, the likelihood of man-vector-virus contact increases thereby enhancing disease transmission [6]. Studies conducted by Chao et al. [7] modeled the role human movement played in the spatial spread of dengue and suggested that dengue cases which clustered in space and time were probably due to the short flight range of the Ae. aegypti mosquito [8]. This clustering of cases and the concept of “casa segura” demonstrate the role human movement plays in dengue spatial transmission [1-3,9,10]. Similar studies have been recently conducted in India [11], Peru [12], Cambodia [10] and Puerto Rico [4].

Dengue is an acute tropical viral disease caused by one of four serotypes, Dengue-1, Dengue-2, Dengue- 3, and Dengue-4 (DEN-1, DEN-2, DEN-3, and DEN-4) transmitted by Aedes mosquitoes; Aedes aegypti (L.) and Aedes albopictus (Skuse) [13]. Of the two species of Aedes mosquitoes, Ae. aegypti has been identified as the principal vector for much of the morbidity and mortality associated with dengue [14]. Adult female Ae. aegypti acquire the virus by biting an infected person during the viremic phase, which usually lasts for four to five days, but may last up to 12 days [15]. The virus is then transmitted to non-infected persons via bites from infected mosquitoes. Clinical symptoms of infected persons vary from a mild, self-limited fever, which can be undiagnosed or misdiagnosed, to severe illness with acute hemorrhagic manifestations such as Dengue Hemorrhagic Fever (DHF) or Dengue Shock Syndrome (DSS) [16,17]. These severe forms of the disease are associated with several factors, including the immunity status of hosts, disease strain [18], age and genetic predisposition [19].

Over the last 25 years the overall incidence of dengue in developing regions of the world, including the Caribbean region [20,21] has increased with circa 40% of the world’s population at risk of infection with dengue [7]. It is now considered the most frequent arboviral disease in the world, with an estimated annual 390 million cases of dengue fever (DF), 250,000 cases of DHF, 25,000 deaths per year and 93 million asymptomatic cases [7,22]. Within Latin America and the Caribbean region, over 908,926 cases of DF and DHF have been reported up to 2008 [22]. In Trinidad, dengue outbreaks continue to occur despite conventional vector control efforts with the situation made worst by the absence of suitable vaccines for the disease.

An important issue in monitoring communicable diseases and planning intervention is to understand the spatial and temporal patterns of disease. This information can provide clues to understanding the dynamics of disease spread, especially the detection of spatial and space-time clustering of cases and identifying high risk areas and times, where disease surveillance and control should be targeted. In this study several methods are used for spatial and temporal analysis of DHF cases in Trinidad for the period 1998–2004. The Nearest Neighbor Index (NNI) and the Nearest Neighbor Hierarchical Clustering (NNHC) methods have been widely used in exploring first and second order spatial point processes [23]. The NNI is used to detect the presence of spatial clustering while NNHC is used to map the extent of the spatial clustering if present. Recent studies have demonstrated the utility of the NNI for detecting the presence of spatial clustering in dengue cases [24,25] while the NNHC was used to identify high mortality concentrations of yellow fever for each month in 1878 representing an epidemic year of the disease in New Orleans [26]. The NNHC clustering is particularly applicable if the nearest neighbor distance is believed to be of relevance to the spatial and temporal distribution of point features [27]. Work conducted in the County of St Patrick, Trinidad showed dengue cases closely clustered together and suggested that this clustering was possibly due to the short flight range of Ae. aegypti[28], thus making NNI and NNHC suitable methodologies for the current study. The present study was conducted to investigate the impact of different road networks on the spatio-temporal distribution of dengue fever cases associated with the locations of transportation hubs in Trinidad, West Indies.

Methods

Results

Spatial distribution of dengue

Figure 1 shows the spatial distribution of DHF cases by year over the 7 year study period. These results show the transition between the epidemic years 1998 and 2002, with a significant (G=57.4; d.f. 6; P<0.03) decline in the number of cases immediately after 1998. This was followed by the cases becoming more dispersed or scattered, after which a significant (G=87.2; d.f.6; P<0.01) increase in the number of cases occurred towards 2002 with a high density of cases in space (see Figure 1). Although the results showed where dengue cases were clustered, it was impossible to demonstrate the full pattern because all DHF cases, or at least a substantial quantity, were not geocoded for 2002. However, similar to activity soon after the epidemic in1998, immediately after 2002, a significant decline (G= 54.2; d.f. 6; P<0.02) in the number of cases is detected.

Open in a separate window

Figure 1

Spatial distribution of dengue hemorrhagic fever cases in Trinidad, 1998–2004.

Spatial cluster analysis

Nearest neighbor analysis

Table 4 shows the result of the Nearest Neighbor analysis with spatial clustering occurring for individual years 1998–2003 but absent during 2004. During 2004 only 17 (of 29) DHF cases were geocoded, which represented a year with a significant decline (P<0.02) in DHF cases with very dispersed spatial distribution (a mean value of 6638.1 m according to Table 4), (see Figure 1 and Table 3). Table 4 shows that DHF cases were very densely clustered across years in space (mean of 598 m), suggesting a higher probability of infection in an area with a history of infection.Figure 2 shows the number of geocoded cases and the Minimum Nearest Neighbor Distance (MNND) plotted for individual years. Values for MNND in Figure 2 were scaled by dividing the original value by 10 to aid in visual comparative analysis. Figure 2 shows a reversed trend when the number of DHF cases and MNND were compared, that is, as the number of DHF cases increases, the mean distance between cases decreases. In addition, there was significant (P<0.03) dispersion between cases soon after an epidemic has occurred, as indicated by the slope of the MNND line just after 1998 and 2002. It is expected that with more geocoded cases for 2002 the spatial dispersion (MNND) would have been significantly greater in the following years. The high spatial dispersal pattern observed represents the mean distance found between cases after outbreaks.

Table 4

Nearest neighbor analysis for DHF cases in Trinidad

Year	P-Value	Mean nearest neighbor distance (m)
1998	0.0001	1576.9
1999	0.0001	3815.8
2000	0.0001	1902.5
2001	0.0001	1396.7
2002	0.0001	1362.3
2003	0.0001	1877.5
2004	Not significant	6638.1
All years	0.0001	598.0

Open in a separate window

Open in a separate window

Figure 2

Geocoded DHF cases and the MNND for individual years 1998–2004 in Trinidad.

Nearest neighbor hierarchical clustering

The hierarchical ellipses show the identified hotspots of dengue cases for individual years 1998 to 2003 (Figure 3), and represent the spatial distribution of DHF cases detected using only the first order clusters. The years 1999 and 2004 are not shown in this figure since there were no clusters identified based on the criteria set for determining the presence of clustering (a minimum of 10 points and using a mean random distance threshold based on the dataset as explained in the previous section). Figure 3 shows three main regions of clustering indicated in Figure 4 by regions A, B and C (encircled by red ellipses) followed by 4 smaller regions D, E, F and G. Region A and B are the main city capitals in Trinidad, Port of Spain and San Fernando respectively. Region C represents an equally important major town of Arima, which is smaller in size than to the identified cities in regions A and B.Figure 4 shows hierarchical clustered regions superimposed onto a road network of Trinidad. This figure clearly shows the relationship between dengue clusters and areas within and around transportation hubs on the island. Regions A, B and C represent the island’s major transportation hubs between major cities while region E (town of Sangre Grande) represents another transportation hub which links most of the North Eastern part of the island to highways and by extension, the major cities in Trinidad. Region G (Penal) represents a smaller town near the suburban part of the island while region F represents an even smaller town in the rural part of Trinidad. Figure 4 also reveals a distinct movement of dengue clusters between types of transportation hubs and epidemic years with clustering occurring in and around major cities during an epidemic (through major transportation hubs). Following this, the results show migration towards more rural settings, transitioning away from an epidemic and towards areas with smaller transportation hubs (and nearer to smaller towns) on the island. The reverse relationship is seen transitioning towards an epidemic year.

Open in a separate window

Figure 3

Nearest neighbor hierarchical clustering.

Open in a separate window

Figure 4

Transportation hubs and dengue clusters in Trinidad (A-Port of Spain, B-San Fernando, C-Arima, D-Chaguanas, E-Sangre Grande, F-Biche, and G-Penal).

Space-time analysis

Using the Knox test the results showed a significant number of dengue cases were clustered in space (G= 76.7; d.f.6; P<0.05) and time (G=67.9; d.f. 6; P<0.04) within and across years (∑²=345.97 d.f. 10; <0.05) (see Table 3). Table 5 shows that the clustering time between years is approximately 727 days (about 2 years) and represents the transition between epidemic years 1998 and 2002, where in 2000 (2 years after) cases began appearing once more and transitioned towards the epidemic year 2002. Figure 5 shows the 'close in distance' and 'close in time' values for individual years. Values for 'close in distance' in Figure 5 were scaled by dividing the original value by 1000 to aid in visual comparative analysis.Figure 5 shows a reversed relationship between 'close in distance' and 'close in time' variables with the exception of the transition between years 2001 to 2002. This is possibly due to the fact that only 44% of the DHF cases were geocoded in 2002, but overall the results show when the number of days decreases between clustering in time the distance between events gets larger and vice versa. This relationship suggests that when an outbreak of disease occurs within short time periods there is a faster rate of distribution of the disease across space (see Figure 5).

Table 5

Space-time analysis of DHF cases in Trinidad (1998–2004)

Year	P-Value	Close in time (days)	Close in distance (m)
1998	0.0001	95.1	33799.0
1999	0.0001	125.6	31692.9
2000	0.0001	90.4	35235.9
2001	0.0001	138.3	33049.1
2002	0.0001	124.3	26386.1
2003	0.0001	71.5	30467.4
2004	0.0010	90.1	32158.2
All years	0.0001	727.0	506774.5

Open in a separate window

Open in a separate window

Figure 5

Close in distance and close in time values for individual years dengue occurred in Trinidad.

Discussion

In the 1970s and 1980s dengue outbreaks occurred in the Caribbean region at 8–10 year intervals when new serotypes were introduced [6]. However, this epidemiological pattern changed with the outbreak of dengue in Cuba in 1981 when a DEN-2 outbreak was followed by a DEN-1 episode [34]. However, in 1994, the fourth serotype DEN-3 was introduced from Southeast Asia to the Caribbean region after an absence of 17 years. This introduction has increased the frequency of epidemics from 8 to 10 years to 3 to 5 year intervals, especially in Trinidad [35]. The present results show the frequency of dengue outbreaks at approximately 727 days or circa every 2 years which suggest that the transition between outbreaks have further shortened and the frequency of dengue outbreaks reduced from every 3 to 5 years to every 2 to 3 years. This close frequency of outbreaks is cause for concern for DF/DHF in the Caribbean region and especially in Trinidad can emerge as a major public health problem with an increase in DHF/DSS burden of disease and the accompanying impacts on morbidity and mortality rates, DALYs and economic cost ([35,36]).

In Trinidad DF/DHF outbreaks occur during the wet season and are usually associated with thousands of people seeking medical treatment and hospitalization. Although not reported in the results section of this paper, this study found most reported cases for dengue recorded between June and November, coinciding with the temporal periodicity of dengue epidemics during the wet season on the island. This pattern is very similar to that reported previously for Trinidad [28] and Puerto Rico [37] but is different from that observed in Paraguay and Argentina [38]. In the latter two locations, dengue transmission occurs during the first half of the year in the southern hemisphere and during the second half of the year in the northern hemisphere, which is equivalent to the wet season for both hemispheres when temperature, humidity and precipitation are elevated.The spatial and temporal patterns of reported cases of DHF in Trinidad were analyzed using the minimum nearest neighbor distance method showed that DHF cases were clustered in both space and time within and across years (Figure 3). These results show that during an epidemic year, the number of DHF cases significantly (P<0.02) increases and then significantly (P<0.02) declines with a transition away from the epidemic. These results suggest that as herd immunity increases, that is as the population at risk becomes infected and then immune to the circulating dengue serotype, the susceptible population size is reduced and transmission declines. Therefore, during this time, the dengue virus and by extension DHF cases tend to occur away from intense spatial clustering towards greater spatial distribution (see Figure 3). Interestingly, the reverse effect is seen during an epidemic year, however when the number of days decrease between clustering in time, the distance between events gets larger and vice versa. This relationship suggests that when an outbreak occurs within a smaller time period there is a faster rate of distribution of the disease across space. A good example of this feature may involve what happens when a new serotype is introduced into a new location. This observed dengue epidemiologic pattern is quite unique and has been suggested as a mechanism of dispersal in many countries but the present study provides for the first time evidence for this dengue dispersal pattern and association with transportation hubs in space and time for Trinidad.

During this study DHF cases were observed clustered in space and time (Knox test) and this pattern is similar to that observed in Florida, Puerto Rico [4]. The space-time clustering of the cases suggested the identification of ‘hot spots’ which is similar to that identified by Levine [23] who used this approach to detect crime waves and crime hot spots in the USA. The adoption of this ‘hotspot’ approach has clearly graphically identified areas with the highest dengue activity and once introduced can be an invaluable tool for the analysis of disease dispersal and diffusion from one community to another, for infectious diseases, as well as dengue.

It is also evident that during an epidemic there is a greater probability of infection in areas near and within cities where previous infections or outbreaks had occurred (Table 4), e.g. clusters are frequently observed along the Eastern Main Road (highway running from East to West in the northern part of the island) as shown in Figure 4. This observed pattern explains the emergence of DHF outbreaks in Trinidad which usually occur when there are co-circulating dengue antibodies within a population, generally the leading risk factor for the development of DHF/DSS within clusters in these townships [6]. This circulation pattern explains the increase in DHF cases especially in Trinidad where all 4 dengue serotypes are now endemic [28] and can possibly lead to a massive epidemic in the future with high morbidity and mortality rates.

The present results show that DHF distribution is correlated with transportation hubs as indicated by significant clustering in and around these travel centers. These findings extend the work conducted by Mahabir et al. [5] who identified areas around less voluminous road networks as conducive locations for outbreaks of DHF in Trinidad, especially those associated with or connected to major transportation hubs. Although dengue fever is considered a residential disease its dispersal has been found to be limited to movement of infected people within and outside their home environment [28,39], house to house human movement [12], dispersal of infected Aedes aegypti mosquitoes based on the type of housing patterns and type of road network available in the location [5] and the association with travel hubs (present study). Conversely, areas of high traffic volumes were found to limit the movement of mosquitoes [40] and therefore limit the transmission of dengue fever to certain parts of Trinidad [5].

Additionally, there appears to be a distinct association between dengue clustering and types of transportation hubs between epidemic years. During an epidemic year DHF cases are clustered tightly around major city hubs. As the epidemic becomes less intense clusters are found at smaller transportation hubs where smaller cities and towns exist. The reverse effect is seen when transitioning towards an epidemic period. These epidemics occur especially in areas in and around transportation hubs and near cities, which support large commuting and residential populations. These hot spot areas provide suitable breeding habitats and blood meals for Ae. aegypti mosquitoes. These results further suggest that the Ae. aegypti peak feeding times coincides with high traffic commute times to and from these hubs, that is, during the late afternoon and early morning periods [41]. It is therefore expected that Ae. aegypti mosquitoes that feed in these environments will stay in and around these sites since access to blood meals are regularly available for completion of their gonotrophic cycle. Consequently, this results in the rapid growth of mosquito populations along with possible intensification of dengue transmission at these hubs. This may explain the observed dengue transmission patterns at home and at travel hubs and may be responsible for the apparent relationship found between the distance in both space and time between DHF cases: as the number of days decreases between appearance of clusters; the distance between cases increases and vice versa (see Figure 5). This relationship suggests that when an outbreak of disease occurs within a smaller time period than in previous outbreaks, there is a faster rate of geographic dispersal of the disease. In addition, this finding suggests that areas around transportation hubs are suitable environments for the spread of dengue fever. In these environments, people live in close proximity to one another and provide suitable breeding sites for the immature stages and resting sites (indoors and outdoors) for adult Ae. aegypti mosquitoes [42]. It is well established that large populations tend to flock around city areas and in close proximity to transportation terminals or routes due to available job opportunities and for quick access to other locations. It is expected that Ae. aegypti mosquitoes feeding in these environments will not move very far because of the regular supply of blood meals and artificial breeding sites (8). However, the movement of DF/DHF to smaller towns and cities during an epidemic can lead to outbreaks especially where less emphasis may be placed on vector control thus providing suitable conditions for the establishment or renewal of dengue transmission.

Conclusions

In conclusion, the results of this study can be used by public health officials to develop maps which can aid in the presentation, visualization and understanding of the geographic distribution of DF/DHF cases, as well as demonstrate trends in the spatial and temporal patterns of reported DF/DHF cases. The clusters observed near and within travel hubs can also be used for developing health education and community awareness programs. It is clear that the current dengue space-time patterns and ‘hotspots’ detection method can be used to help public health officers plan more effective control programs and provide ‘early warning’ alerts with better emergency responses thus allowing better allocation of resources. In 2008–2009 this approach was adopted in Trinidad and significantly reduced the incidence of dengue but was unfortunately discontinued due to logistic and management issues. This component is absolutely vital to program planners since there is already evidence of Ae. aegypti resistance to organophosphate and pyrethroid insecticides in the Caribbean [43]. Therefore, in the future, it would be helpful if governments within the Caribbean region could improve their data collection and recording systems for dengue and other diseases to allow time series analyses and thus enable early detection, analysis of epidemiologic patterns, and the management of outbreaks in a more efficient and timely manner.

Competing interests

All authors declare that they have no competing interests.

Authors’ contribution

KDS, RSM, JMS, DDC conceptualized the study, conducted field work, conducted data analysis and manuscript preparation, KDC assisted in the data analysis and preparation of manuscript and JBA provided logistic support and assistance for the field study. All authors read and approved the final manuscript.

Authors’ information

KDS, Ministry of Health, 63 Park Street, Port of Spain, Trinidad, West Indies, RDM and KDC, Department of Geography and Geoinformation Science, George Mason University, Fairfax, Virginia, USA, JMS,JBA and DDC, Department of Life Sciences, The University of the West Indies, St. Augustine, Trinidad and Tobago.

References