Spatial-Temporal Analysis for Identification ofVulnerability to Dengue in Seremban District, Malaysia

Naim, M. R.,' Sahani, M..,^ * Hod, R.,' Hidayatulfathi, O.,^ idrus, S.,* Norzawati, Y.,' Hazrin, H.,' Tahir,A.,' Wen, T. H.,^ King, C. C., Zainudin, M. A.J'GIS Project Team, Institute of Public Health, Ministry of Health, Malaysia^Faculty of Health Sciences, Universiti Kebangsaan Malaysia, E-mail: ma2Tura@fsk.ukm.my^ Faculty of Medicine, Universiti Kebangsaan, Malaysia* LESTARI, Universiti Kebangsaan, Malaysia'Department of Geography, National Taiwan University, Taiwan^Graduate School of Public Health, National Taiwan University, Taiwan^Negeri Sembilan Health Department, Ministry of Health, MalaysiaCorresponding Author: Mazrura Sahani

.a

AbstractDengue is a major public health threat in Malaysia, which is known for the hyperendemicity with all the jourserotypes of the dengue virus circulating concurrently. Annual dengue cases reported were 43,000 cases for2013, and this imposed a heavy toll on the resources for dengue prevention and control program. Theobjective of mapping in our study is to determine the spatial clustering of the dengue cases and to identify theareas that are vulnerable to dengue outbreaks. A Geographical Information System (GIS) was used to assessthe vulnerability of Seremban district. Dengue data were obtained from the Ministry of Health. We determinedthe spatial distribution, the average distance of dengue cases and identified hotspots areas using the Moran-sI, Average Nearest Neighbourhood (ANN), Kernel density estimation. Vulnerability to dengue was assessedwith the spatial temporal analyses and Local Indicator for Spatial Autocorrelation (LISA). From 2003-2009Seremban recorded 6076 dengue cases. Moran-s I showed the cases occurred in clusters with a Z-score of16.384 (p

theoretical and practical values that will furtherincrease significant global awareness of theinfectious disease (Yang et al., 2007). The usage ofGIS to control and prevent dengue outbreak shouldbe explored extensively and put under seriousconsideration (Houghton et al , 1996, Yang et al.,2007, Aziz, 2011 and Er et al , 2010). Map showingthe mosquito vector data (presence or abundancepatterns, epidemiological data (location of dengueor dengue cases pattern scene), or vector controlcoverage performed a useful tool in vector anddengue control programs. Map can be used to guideand assess the progress of operations and activitiesto disseminate information to outside parties. Forexample, the map can help to alert people to the areain a city with a particularly high risk exposure todengue virus (Eisen and Lozano-Fuentes, 2009).Dissemination of spatial and temporalcharacteristics of the spread of the disease can bemonitored and used by the party responsible forcarrying out control measures. Identifying keynodes in the spread of infection, defined by timespent at unique locations, can help us understandpast infectious episodes, and predict futuredevelopments (Meliker and Sloan, 2011). With thespread of disease models available, this diffusionprocess is dynamically simulated and visualized intwo or three dimensions spatial scale (Yang et al.,2007). High-risk populations can be identified andthe location of the spatial distribution pattern isdescribed and transmission behaviors of the diseasemay be found. Prevention through more effectivedecisions can be made by government and publichealth institutions through the provision of bettermedical resources using GIS network analysismodel (Yang et al., 2007). The objective of thisstudy was to identify the vuhierable areas to dengueusing spatial temporal analysis. Vulnerable areas inthis study area will be explained using spatial-temporal indexes (the frequency index, the durationindex and intensity index) of dengue cases andLocal Indicator of Spatial Autocorrelation (LISA).

2. Materials and Methods2.1 Study Area and DataThe Seremban district is one of the 7 districts inNegeri Sembilan, which is located on the west coastof Peninsular Malaysia. The administration ofSeremban district is divided between the SerembanMunicipal Council (inner Seremban) and the NilaiMunicipal Council (outer Seremban). Seremban isthe capital of Negeri Sembilan and consists ofmostly urbanized area.

Seremban was selected as our study area because ithas the highest recorded of dengue cases for NegeriSembilan. We focused on surveillance data, basedon reported dengue cases from the year 2003 to2009. The data was obtained from Seremban districthealth office. Dengue cases were based on clinicallyconfirmed cases. The MOH Clinical PracticeGuidelines (Ministry of Health Malaysia, 2010b)were used to diagnose and classify DF and DHF.The software used for mapping was ArcGIS 10. Thedengue data was used to map the location of eachdengue case using the Global PositioningSystem (GPS). The data obtained from the fieldobservations were then utilized to construct adatabase as incidence points. The statistical analysisused in our research was the spatial-temporalanalyses such as the Moran's Index, ANN, Kerneldensity and the spatial-temporal indexes. We alsoused the Local Indicator of Spatial Autocorrelation(LISA). LISA is a spatial risk index to identifyspatial patterns including clustering and outliers(Anselin, 1995, dland, 1998 and Wen et al , 2010).This research focused on the clustering of thedengue cases which can be identified using thefeatures of frequency-time-incident cases and thenumber of cases occurring in a certain period oftime or in an epidemic. These factors give a true andmore realistic picture of the epidemic events.

2.2 Moran 's IndexSpatial autocorrelation of Moran's index was usedto analyse whether the dengue cases withinthe study district is spatially correlated. Itis determined by calculating a mean for observationand then comparing the value of each incident withthe value at all other locations (Wen et al., 2010).Moran's I values range from -1 which is negativespatial autocorrelation (approaching scatteringincident) to +1 of positive spatial autocorrelation(approaching grouping incidents). Valueapproaching 0 refers to the random distributionpattern.

2.3 Average Nearest Neighbourhood (ANN)Average nearest neighborhood (ANN) was used toassess the pattern of clustering of the incidence ofdengue. ANN calculates the distance between eachfeature centroid to its nearest centroid. It thencalculates the average nearest neighbordistance. The average ratio was calculated as theaverage of the observed distance divided by theexpected average distance (Wen et al., 2010).

3 2 Spatial-Temporal Analysis for Identification of Vulnerability to Dengue in Seremban Distriet, Malaysia

2.4 Kernel Density EstimationHot spots analysis using Kernel density estimationinterpolation techniques was carried out Kerneldensity estimation is a technique used to identifyhigh-risk areas based on the pattern of diseaseincidence or the hotspot. It calculates the density ofpoint features around each output raster cell (Bithell,1990). This hotspot identification is an importanttool to focus on the control activities to preventirther transmissions of the mosquitoes-bornedisease.

2.5. Spatial-Temporal Indexes and LISAVuberability of the population towards the denguecases in this study was assessed by conducting aspatial autocorrelation in which all three of thespatial-temporal indexes (frequency index, durationindex and intensity index) were calculated andLISA. The frequency of a disease during anepidemic can be described by a probability that oneor more laboratory confirmed cases occurred incertain week(s) out of the total epidemic period(Wen et al., 2006).

Occurrence index (a) can be defined as:

a = number of weeks of the epidemic / number ofweeks in the year

An epidemic period usually involves severalepidemic waves. Duration index () of an epidemiccan be described as the average number of weeks inwhich occurrence of cases persists during the wholeepidemic period (Wen et al., 2006).

Duration index () ean be defined as:

= number of epidemic weeks / total epidemicwave

The duration index () is very important forpractitioners and administrators of public health andenvironment agencies because it refiects theeffectiveness of the prevention or control strategiesused during the epidemic. Intensity refers to thelikely magnitude within an epidemic wave whenmore than one case occurs (Wen et al.,2006). Incidence rate is taken as an index tomeasure the magnitude of new cases appearingduring a specified period.Intensity index (y) is measured as:Y = incidence rate/ total epidemic wave

The vulnerability assessment is presented with amap of the local indicators of spatial autocorrelation(LISA). LISA considers the possible combinationsof the three indices, namely O, D and I, and todistinguish these risk characteristics. Then, the high-risk areas were identified and mapped out andcompared (Wen et al, 2010). LISA provides eightpossible types of indexes summarized as O foroccurrence index, D for duration index and I forintensity index.

3. ResultsA total of 6076 dengue cases were recorded inSeremban distriet from 2003 to 2009. The highestincidence of dengue, during that period was in 2003,followed by 2004 and 2008 with respectiveincidence rates of 37.4, 23.5 and 20.5 per 10,000populations. The spatial distribution of dengue casesin the study area was identified using spatialautocorrelation i.e Moran's index. This indexmeasures the autocorrelation based on the locationand distribution of the dengue cases. The Moran'sIndex of dengue incidence was 0.16 at p

district A

district B

Figure 1 : Hotspot areas consist of twosub districts

Figure 2: Vulnerability areas in 2003. Eight types ofvulnerable area were defined fi-om all the possiblecombinations of high values of the three temporalindices in year 2003. Red area is the most vulnerable

Sub district A

Figure 3: Vulnerability areas in 2004. Eight typesof vulnerable area were defined firom all thepossible combinations of high values of the threetemporal indices in year 2004

Figure 4: Vulnerability areas in 2005. Eight typesof vulnerable area were defined from all thepossible combinations of high values ofthe threetemporal indices in year 2005. Most of red areasare in 3 subdistricts in Seremban

3 4 Spatial-Temporal Analysis for Identification of Vulnerability to Dengue in Seremban District, Malaysia

Figure 5: Vulnerability areas in 2006. Eight typesof vulnerable area were defined from all thepossible combinations of high values of the threetemporal indices in year 2006

Figure 6: Vulnerability areas in 2007. Eight typesof vulnerable area were defined from all thepossible comhinations of high values ofthe threetemporal indices in year 2007. Highest vulnerableareas is only in a subdistrict in Seremban

.g

SD

Sub district A

Figure 7: Vulnerability areas in 2008. Eight typesof vulnerahle area were defined from all thepossible combinations of high values of the threetemporal indices in year 2008

Figure 8: Vulnerahility areas in 2009. Eight typesof vulnerahle area were defined from all thepossible combinations of high values of the threetemporal indices in year 2009

hucniational Journal of Geoinformatics. Vol. 10, No. 1, March, 2014 3 5

In 2006, there were also three sub districts with highODI. There were 12 units with high ODI (Figure 5).In 2007, LISA analysis showed there was only onehigh ODI area consisting of two units (Figure 6). In2008, LISA analysis showed clearly the 3 high ODIareas were located at 3 different sub districts (Figure7). This area consisted of 8 units of the HighODI. In 2009, there were 10 units with high ODIlocated in two of sub-districts (Figure 8).

4. DiscussionThis study evaluated the spatial-temporaldistribution of dengue cases in Seremban, an urbandistrict in Malaysia. Data manipulation and GISpresentation and spatial statistics were used to mapthe distribution of the dengue incidence in the studyarea. The study found a significant clustering patternof the dengue eases in the study area with Moran'sIndex of 0.16. The Average Nearest Neighborhood(ANN) found was a distance of 55 meters betweendengue eases and other cases (with z score of -109.73 and p value less than 0.05). This differs withEr et al., (2010) that found a Moran's Index of 0.75with an average nearest neighborhood (ANN)analysis of 380 m. The difference could beattributed to the density of population and thepattern of housing in both areas. The district in thisstudy is more developed compared to the areas ofstudy in Er et al, (2010).The significance of the 55meters ANN in this study means that the control andprevention activities such as destruction of breedingsites, fogging and health promotion activities couldbe focused on a smaller area, hence the control oftransmission should be easier to achieve. Moran'sIndex reported in by Nakhapakom andJirakajohnkool (2006), and in Rio de Janeiro, Brazilby Almeida et al, (2009) showed similar findings ofsignificant clustering patterns of dengue cases. Thisstudy showed the potential of an area at risk even inthe absence of data on the mosquito density or otherenvironmental factors. However, Aziz (2011)reported a spatial relationship between theincidences of dengue with the environmental factorsthat were associated with the breeding ofmosquitoes. For dengue risk, areas marked withdark color signified the locations with the maximumnumber of dengue incidence. Dengue density mapusing LISA analysis showed that most cases wereconcentrated in two sub-districts as discussed below.In 2003, LISA analysis showed that the high-ODImostly occurred in the two sub districts. The year2004 recorded the absence of high-ODI. There wasa decline in dengue cases as compared with thoserecorded in 2003. However, in 2005, the high ODI

areas of dengue cases were seen to beincreasing. This showed that in some areas, thedengue control program was still not adequate. For2005, the areas that have high ODI was mainly inthe three sub-districts. In 2006, the High ODI is stilllocated at the same sub-district but with fewchanges. In 2007, there was only one area with highODI. However, 2008 showed that the same area hada high ODI and with addition of two areas located intwo different sub districts. In 2009, there were onlytwo sub-districts that have high ODI. The sameareas located at same sub-districts were found ashigh ODI throughout the study period. Similarfindings were found by Wen et al., (2010) insouthern Taiwan where in 2002 during a dengueoutbreak, there were 20 high ODI areas which wereconcentrated in two areas in the city of Kaohsiungand Fengshan. In the high ODI area, 13 out of 52epidemiological weeks have eonfirmed cases ofdengue. During this dengue outbreak, the averageduration is longer than a month with an averageintensity of 3.2 cases per 10 000 people perepidemic wave. Similar findings were reported in aprevious study of Wen et al., (2006). Galli andNetto (2008) also used LISA to identify the area atrisk in the city of Sao Jos do Rio Preto,Brazil. Their research reported that in the year2001/2002, there were 6 units of ODI area where themost vulnerable area was in the city. Their findingsshowed that the most vulnerable area for denguewas in the northern region of the eity. This studyshowed the advantage of using spatial temporalanalysis which enables us to identify the specificareas that are vulnerable to dengue outbreaks. Thisfinding could guide the planning andimplementation of dengue control and preventionprograms and prioritizing the resources in riskmanagement. It will enable the health authorities aswell as other stakeholders such as the municipalagencies, solid waste management, urbandevelopment agencies to plan and implement thedengue prevention and control programs morecomprehensively and effeetively (Wen et al., 2010).The limitation of our study is due to our dependenceon the surveillance data which is based on thedengue ineidence. We could not include data on themosquito density due to the unavailability of thisdata. In understanding the dynamics of dengueoutbreaks, the inclusion of the mosquito density datais important because the disease transmissiondepends on the contacts between human and themosquito which carries the dengue virus. Data onthe number of breeding sites and mosquito densitywould be extremely useful in measuring the risk.

3 6 Spatial-Ternporal Analysis for Identification of V'ulncrabilit) to Dengue in Seremban District, Malaysia

The application of this analysis could be used in thefuture by incorporating other variables that maycontribute to the dengue outbreaks, such asinformation on the human population density,human movement and habits, availability ofbreeding sites and the entomological data. Thisanalysis could also be applied to other localities tohelp identify the high risk area more effectively.

5. ConclusionDengue eases showed a clustering pattern in thestudy area. The Average Nearest neighborhoodanalysis (ANN) showed that the average distance ofthose dengue cases was within 55 m (p

Seng, S. B., Chong, A. K. and Moore, A., 2005,Geostatistikal Modelling, Analysis and Mappingof Epidemiology of Dengue Fever in JohorState, Malaysia. The 17th Annual Colloquium ofthe Spatial Information Research Centre,University of Otago, Dunedin, NewZea/ai/. November 24th-25th.

Wen, T. H., Neal, H. L., Chao, D. Y., Hwang, K. P,Kan, C. C, Lin, K. C. M., Wu, J. T. S., Huang,S. Y. J., Fan, I. C. and King, C. C, 2010,Spatial-Temporal Patterns of Dengue in Areas atRisk of Dengue Hemorrhagic Fever inKaohsiung, Taiwan, 2002. Intemational Journalof Infectious Diseases 14: 334-343.

Wen, T. H., Lin, H. N., Lin, C. H., King, C. C. andSu, M. D., 2006, Spatial Mapping of TemporalRisk Characteristics to Improve EnvironmentalHealth Risk Identification: A Case Study of aDengue Epidemic in Taiwan. Sei Total Environ367:631-40.

Yang, K., Peng, S. Y, Xu, Q. L. and Cao, Y. B.,2007, A study on Spatial Decision SupportSystems for Epidemic Disease Prevention Basedon ArcGis. In. Lai, PC. & Mak, A.S.H. (ed). GISfor Health and the Environment, Development inthe Asia-Pacific Region, 30-43. Berlin: SpringerBerlin Heidelberg.

3 8 Spatial-Temporal Analysis for Identification of Vulnerability to Dengue in Seremban District, Malaysia

Copyright of International Journal of Geoinformatics is the property of International Journalof Geoinformatics and its content may not be copied or emailed to multiple sites or posted toa listserv without the copyright holder's express written permission. However, users mayprint, download, or email articles for individual use.