Tropical Medicine and International Health doi:10.1111/j.1365-3156.2012.03038.x volume 17 no 9 pp 1086–1091 september 2012

Space-time clusters of dengue fever in

Shahera Banu1, Wenbiao Hu2, Cameron Hurst1, Yuming Guo1, Mohammad Zahirul Islam3 and Shilu Tong1

1 School of Public Health and Social Works, Institute of Health and Biomedical Innovation, Queensland University of Technology, Brisbane, Qld, 2 School of Population Health, University of Queensland, Qld, Brisbane, Australia 3 Griffith School of Environment, Griffith University, Brisbane, Qld, Australia

Abstract objective To examine the space-time clustering of dengue fever (DF) transmission in Bangladesh using geographical information system and spatial scan statistics (SaTScan). methods We obtained data on monthly suspected DF cases and deaths by district in Bangladesh for the period of 2000–2009 from Directorate General of Health Services. Population and district boundary data of each district were collected from national census managed by Bangladesh Bureau of Statistics. To identify the space-time clusters of DF transmission a discrete Poisson model was performed using SaTScan software. results Space-time distribution of DF transmission was clustered during three periods 2000–2002, 2003–2005 and 2006–2009. Dhaka was the most likely cluster for DF in all three periods. Several other districts were significant secondary clusters. However, the geographical range of DF transmission appears to have declined in Bangladesh over the last decade. conclusion There were significant space-time clusters of DF in Bangladesh over the last decade. Our results would prompt future studies to explore how social and ecological factors may affect DF transmission and would also be useful for improving DF control and prevention programs in Bangladesh.

keywords dengue fever, space-time cluster, scan statistic, Bangladesh

have been reported every year in all major cities of Introduction Bangladesh. More than 23 872 cases were reported to the Dengue fever (DF) is one of the most important emerging Directorate General of Health Services (DGHS); 233 were arboviral diseases worldwide (WHO 2000). The virus is fatal between 2000 and 2009. The worst outbreak was transmitted through the bite of container breeding mos- in 2002 with 6132 cases and 58 deaths. Both Aedes aegypti quitoes Aedes aegypti and Aedes albopictus, which are and Aedes albopictus were identified as potential vectors present in most tropical and subtropical countries (Rigau- for DF transmission in Bangladesh (Ali et al. 2003). A Perez et al. 1998). Symptoms of DF infections in humans national guideline based on the WHO protocol was vary from mild flu-like DF to life-threatening dengue developed by DGHS in 2000 to control DF transmission hemorrhagic fever (DHF) or dengue shock syndrome (DSS) and reduce its morbidity and mortality (DGHS 2000). (Gubler 2002). About 50 million people worldwide be- In the absence of an effective vaccine and specific come infected with dengue virus annually. Globally, the treatment, vector control is the only way to prevent DF geographical range of DF transmission has increased transmission. Previous studies suggest that the risk of DF dramatically in recent years (WHO 2000). Social and transmission varies over space and time (Tran et al. 2004; demographic changes such as population growth, urban- Mammen et al. 2008; Thai et al. 2010). Therefore, isation, air travel and climate change contribute to the identification of high-risk areas can be useful for prior- increased incidence and geographical expansion of DF itising DF surveillance and vector control efforts in areas transmission (Gubler 2002; Wu et al. 2007). where they are most needed (Ali et al. 2003). We DF has become a serious public health concern in investigated the spatial and temporal distribution of DF at Bangladesh after the first large-scale outbreak in 2000. a district level in Bangladesh during 2000–2009. Our aim However, evidence suggests that sporadic DF outbreaks was to identify high-risk clustering areas for DF trans- occurred in Bangladesh between 1964 and 1999 (Rahman mission using space-time scan statistics and geographical et al. 2002; Hossain et al. 2003). Since 2000, DF cases information system.

1086 ª 2012 Blackwell Publishing Ltd Tropical Medicine and International Health volume 17 no 9 pp 1086–1091 september 2012

S. Banu et al. Space-time clusters of dengue fever

Methods Statistical analysis Study area A space-time statistical analysis was applied to detect high- risk clusters of DF using SaTScan software (version 9). Bangladesh, a low-lying, riverine country with a large Case files generated using monthly aggregated numbers of marshy jungle coastline of 710 km on the northern DF cases for each district were used in data analysis. littoral of the Bay of Bengal, is one of the world’s most Population and coordinates data were also used as inputs densely populated countries (964.42 ⁄ km2) in the world. in SaTScan. We fitted a discrete Poisson regression model Bangladesh is divided into seven administrative divisions, to identify space-time clusters after adjustment for which are subdivided into 64 districts and further the uneven geographical density of district population subdivided into upazila or thana. Dhaka is the capital and (Kulldorf et al. 1998; Kulldorf 2010). largest city of Bangladesh. Other major cities include For cluster specification in space-time analyses, three Chittagong, Khulna, Rajshahi, Sylhet and Barisal. Ban- parameters were set for cluster size: the maximum circle gladesh has a tropical monsoon climate characterised by radius in the spatial window, maximum temporal window wide seasonal variation in rainfall, temperatures and and the proportion of the population at risk. 99% of the 64 humidity. Regional climatic differences in this flat country districts were within a 40-km radius. Thus, 40 km was are minor. Four seasons are generally recognised: hot chosen as the maximum radius for all analyses. Space-time (21.7–35.5 C), muggy summer from June to August; analyses using a radius of 10 and 20 km returned very humid and rainy autumn from September to November; similar to those using 40 km. The analyses were conducted cool (11.7–26.8 C) and dry winter from December to using a maximum spatial cluster size of 50% of the February; and warm spring from March to May. About population at risk in the spatial window and a maximum 80% of Bangladesh’s rain falls during the wet monsoon of 50% of the study period in the temporal window. season from June to November. Because of the differences in population densities across Bangladesh, it was decided to limit the spatial cluster size Data collection to 50% of the population at risk. The spatial clusters were also defined to cover <25% and 10% of total population at As DF is a notifiable disease in Bangladesh, any DF risk and similar results were obtained to the 50% of the case detected based on the ‘clinical case definitions’ of population limit, which suggests that the maximum cluster national guidelines for clinical management of dengue size is adequately limited by 50% of the population at syndrome must be reported to the DGHS through the risk. The most likely and secondary likely clusters were Civil Surgeon of the district. A DF suspected case is detected through the likelihood ratio test. Significance of defined by the presence of acute fever accompanied by the clusters was evaluated with Monte Carlo simulation any two of the following clinical symptoms such as was set to 9999. MapInfo Professional (version 10.0) was headache, myalgia, arthralgia, rash, positive tourniquet used to display space-time clusters. test and leucopenia, absence of any other febrile illness and high index of suspicion based on period, popula- tion and place (DGHS 2000). We obtained computer- Results ised data sets containing the number of monthly DF epidemics and outbreaks suspected DF cases and deaths by district in Bangladesh for the period of 1 January 2000 to 31 December 2009 The epidemic pattern of DF fluctuated from 2000 to 2009 from DGHS. There has been no significant change in with major outbreaks in 2000 and 2002 (Figure 1). The the dengue reporting system since 2000 in Bangladesh. monthly number of DF cases ranged from 0 to 3281 Relevant population data and electronic boundaries of (mean = 197.7, SD = 454.05), and the highest number of each district were retrieved from the national census cases was reported in August 2002. DF outbreaks in database managed by the Bangladesh Bureau of Sta- Bangladesh after 2002 showed a decreasing trend. tistics (BBS). District information includes district Although DF outbreaks were regular in Bangladesh, a name, code, area (km2), digital boundaries and base 1-year inter-epidemic period between outbreaks was maps. The district population was generated from the apparent. The monthly number of DF deaths ranged from 2001 to 2010 census data and for the remaining years, 0 to 33, and no death was recorded after 2006. Figure 1 estimated based on linear interpolation (Kulldorf also shows a striking variation in monthly numbers of 2010). This study was approved by the Human districts with DF infection from 2000 to 2009. Peaks in DF Research Ethics Committee, Queensland University of cases generally coincided with high monthly numbers of Technology. districts with DF.

ª 2012 Blackwell Publishing Ltd 1087 Tropical Medicine and International Health volume 17 no 9 pp 1086–1091 september 2012

S. Banu et al. Space-time clusters of dengue fever

3500 35 Dengue cases 3000 Districts with dengue cases 30 Dengue deaths 2500 25

2000 20

1500 15

1000 10 Number of dengue cases Number of dengue deaths/ 500 5 Number of districst with dengue cases 0 0 Jul-00 Jul-01 Jul-02 Jul-03 Jul-04 Jul-05 Jul-06 Jul-07 Jul-08 Jul-09 Oct-00 Apr-01 Oct-01 Apr-02 Oct-02 Apr-03 Oct-03 Apr-04 Oct-04 Apr-05 Oct-05 Apr-06 Oct-06 Apr-07 Oct-07 Apr-08 Oct-08 Apr-09 Oct-09 Apr-00 Jan-01 Jan-02 Jan-03 Jan-04 Jan-05 Jan-06 Jan-07 Jan-08 Jan-09 Jan-00

Figure 1 Monthly number of DF cases, deaths and districts with DF notification between January 2000 and December .

clusters were identified between June and November, Disease clusters which represents a rainy monsoon season in Bangladesh. The cluster analyses show that the space-time distribution Table 1 shows the names of districts included in each of DF was clustered during three periods: 2000–2002, cluster, radius (km), observed cases, expected cases, rela- 2003–2005 and 2006–2009. Using the maximum spatial tive risk (RR) and log-likelihood ratio. The highest risk of cluster size of 50% of the population at risk and a circle the most likely cluster (Dhaka) was found in 2004–2005 radius of 40 km, Dhaka district was the most likely cluster, during June–November (RR = 233.3, P < 0.05), and the and Khulna and Chittagong were secondary clusters risk was attenuated in 2006 during June–November (Figure 2). Another two districts (Barisal and Jhenaidah) (RR = 119.8, P < 0.05) and was lowest in 2000–2002 were secondary clusters in August to October 2000 but (RR = 43.48, P < 0.05). For secondary clusters, the RR those clusters were not found in later years (Figure 2). All ranged from 2.56 to 21.63, and the highest risk was

N N N WE WE WE S S S

DHAKA DHAKA DHAKA JHENAIDAH

CHITTAGONG KHULNA CHITTAGONG KHULNA

BARISAL

2006–2009 2000–2002 2003–2005 Most likely cluster (1) Most likely cluster (1) Most likely cluster (1) Secondary cluster (1) Secondary cluster (4) Secondary cluster (1)

Figure 2 Space-time clusters of DF identified in three different periods in Bangladesh.

1088 ª 2012 Blackwell Publishing Ltd Tropical Medicine and International Health volume 17 no 9 pp 1086–1091 september 2012

S. Banu et al. Space-time clusters of dengue fever

Table 1 Space-time clusters of DF in Bangladesh, 2000–2009

Name of Radius Area Relative Cluster district (km) (Sq km) Time frame No. Obs. No. Exp. risk LLR*

2000–2002 1 Dhaka 0 307.42 1 ⁄ 7 ⁄ 2001 to 31 ⁄ 10 ⁄ 2002 8037 415.15 43.48 18865.32 2 Khulna 0 56.83 1 ⁄ 8 ⁄ 2000 to 30 ⁄ 11 ⁄ 2000 590 28.39 21.63 1239.71 3 Chittagong 0 773.57 1 ⁄ 8 ⁄ 2000 to 30 ⁄ 9 ⁄ 2000 590 39.40 15.58 1057.01 4 Jhenaidah 0 227.23 1 ⁄ 8 ⁄ 2000 to 31 ⁄ 8 ⁄ 2000 24 4.80 5.00 19.44 5 Barisal 0 515.38 1 ⁄ 8 ⁄ 2000 to 3 ⁄ 10 ⁄ 2000 2.63 21.33 2.63 19.43 2003–2005 1 Dhaka 0 307.42 1 ⁄ 6 ⁄ 2004 to 30 ⁄ 11 ⁄ 2005 4887 181.22 233.31 14773.74 2 Chittagong 0 773.57 1 ⁄ 10 ⁄ 2003 to 30 ⁄ 11 ⁄ 2003 39 15.32 2.56 12.81 2006–2009 1 Dhaka 0 307.42 1 ⁄ 6 ⁄ 2006 to 30 ⁄ 11 ⁄ 2006 2144 35.44 119.80 7326.35 2 Khulna 0 56.83 1 ⁄ 7 ⁄ 2006 to 30 ⁄ 9 ⁄ 2006 46 4.87 9.53 62.32

No. Obs, number of observed cases; No. Exp, number of expected cases; LLR, Log-likelihood ratio. *P < 0.05. Most likely cluster.

identified for Khulna in 2000 during August–November DF transmission, perhaps, it is the largest city in Bangla- (RR = 21.64, P < 0.05) which reduced to 9.53 in 2006– desh with a high population density. DF can be easily 2009. transmitted by mosquitoes in this area. Chittagong and Khulna in the south were secondary clusters (Figure 2); they are the fastest growing cities close to a seaport and Spatial dispersion of DF airport. Apart from population density, movement of We attempted to identify whether changes in DF trans- infected individuals and ⁄ or mosquitoes into this region mission varied with latitude and longitude of district from overseas by air or ship may increase the chance of DF centroids in the periods 2000–2002, 2003–2005 and 2006– transmission (Sutherst 2004). Although rainwater collec- 2009. A logistic regression model was constructed with the tion for domestic purposes is not a common practice in dichotomous outcome variable defined as whether or not Bangladesh, 35.5% of households in the southwest coastal an increase of DF cases occurred in each district between region use rainwater because of arsenic contamination and the three periods. Longitude and latitude of district high salinity of the ground water (Ferdausi & Bolkland centroids were entered as explanatory variables. The 2000). Storage of rainwater in uncovered containers in this results indicate that changes of DF transmission were not area might increase the risk of DF transmission by significantly associated with geographical variation (i.e. providing a suitable mosquito breeding habitat (Beebe latitude and longitude) during 2000–2002, 2003–2005 and et al. 2009). Climatic variation and socio-economic factors 2006–2009. may also play an important role in promoting DF trans- mission (Hu et al. 2011, 2012). Further investigation in these high-risk areas is required to understand the Discussion dynamics of DF transmission and to discover the role of The results of this study indicate a significant variation in biological, social and environmental factors in the trans- spatiotemporal distribution of DF in Bangladesh. The mission of DF. geographical extent of notified DF cases has declined in The geographical range of DF clusters in Bangladesh Bangladesh over the study period. Dhaka was the highest declined over the last decade (Figure 2). We speculate that risk area for DF transmission. effective management of DF patients according to the Space-time cluster analysis is a valuable tool to examine national dengue guidelines and huge public awareness how spatial patterns change over time. This study shows after the first outbreak in 2000 helped reduce DF that DF transmission in Bangladesh was clustered in three transmission (Rahman et al. 2002). There is no routine periods and provides a clear pattern of DF clustering dengue vector control programme in Bangladesh, within this country. Dhaka was the most likely cluster for although irregular mosquito control activities exist in

ª 2012 Blackwell Publishing Ltd 1089 Tropical Medicine and International Health volume 17 no 9 pp 1086–1091 september 2012

S. Banu et al. Space-time clusters of dengue fever

some cities, not specifically for Aedes mosquitoes factors associated with clustering. However, our future (Chepesiuk 2003). According to Dhaka City Corporation research will focus on the investigation of various (DCC), adulticide and larvicide are sprayed every summer climatic, ecological and socio-demographic determinants in areas with a high mosquito population. DCC also of DF in clustered areas. sprays insecticide indoors and outdoors in notified DF outbreak areas. Besides vector control measures, socio- Conclusion economic changes, population immunity and changes in viral strain and fitness may also contribute to the In summary, this study demonstrates that DF transmission decreased DF incidence in Bangladesh (Podder et al. in Bangladesh was clustered in different spatial and 2006; Cummings et al. 2009). Further research is needed temporal settings and the geographical distribution of to identify possible reasons for declining trend of DF in DF appears to have contracted over recent years. The Bangladesh. impact of socio-demographic changes and climatic factors Dhaka was constantly identified as the most likely on DF transmission in clustered areas remains to be cluster, and very few districts were identified as determined. Our findings can be useful for the Bangladesh secondary clusters for DF transmission in Bangladesh in health authority to further improve DF control and recent years. We believe that patients detected with DF prevention strategies. Additionally, the cluster methods in other districts might have acquired infections during developed in this study may have wider applications in the their visits to Dhaka and symptoms manifested when field of disease surveillance and risk management. they returned to their home districts. Although virus was transmitted to other districts through infected Acknowledgements patients, it might not continue its transmission because of the reduced number of vectors and the reduced We thank The Director General of Health Services, Dhaka number of virus populations in the dry season (Cum- for providing DF case records and the Bangladesh Bureau mings et al. 2004; Aaskov et al. 2006; Williams et al. of Statistics for census data. The work was supported by 2010). However, we were unable to identify the origin QUT postgraduate scholarships (SB, YG), NMHRC post- of DF transmission in Bangladesh and its spreading doctoral training fellowship (WH) and NMHRC research direction owing to the lack of information on patient’s fellowship (ST). location, their movement, demographics and mosquito control activity. This should be a priority for future DF References research in Bangladesh. This is the first study to examine the spatiotemporal Aaskov J, Buzacott K, Thu HM, Lowry K & Holmes EC (2006) pattern of DF in Bangladesh. It has clearly demonstrated Long-term transmission of defective RNA viruses in humans and the heterogeneity of DF risk at the district level in Aedes mosquitoes. Science 311, 236–238. Bangladesh and revealed the spatiotemporal pattern of DF Ali M, Wagatsuma Y, Emch M & Breiman RF (2003) Use of a across the country. As cluster analysis could be an geographic information system for defining spatial risk for dengue transmission in Bangladesh: role for Aedes albopictus in important tool for decision-makers to prioritise areas an urban outbreak. American Journal of Tropical Medicine and where more surveillance and disease prevention efforts are Hygiene 69, 634–640. required, our findings can be useful for the Bangladesh Beebe NW, Cooper RD, Mottram P & Sweeney AW (2009) health authority to further improve DF control and Australia’s dengue risk driven by human adaptation to climate prevention strategies. The method developed in this study change. PLoS Neglected Tropical Diseases 3, e429. may have wider applications in the field of disease Chepesiuk R (2003) Mosquito mismanagement? Environmental surveillance and risk management. Health Perspectives 111, A636. Our study has three key limitations. First, in our Cummings DAT, Irizarry RA, Huang NE et al. (2004) Travelling analysis, we used reported cases aggregated at the district waves in the occurrence of dengue haemorrhagic fever in level, which prohibits analysis at a higher spatial resolu- Thailand. Nature 427, 344–347. tion and may lead to important local clusters being Cummings DAT, Iamsirithaworn S, Lessler JT et al. (2009) The impact of the demographic transition on dengue in Thailand: missed out. Second, there is likely variation in the quality insights from a statistical analysis and mathematical modeling. of the DGHS dengue surveillance data. Underreporting is PLoS Medicine 6, e1000139. possible in the DGHS data when people infected by DF DGHS (2000) National Guidelines for Clinical Management of have subclinical infection and did not seek for medical Dengue Syndrome. Government of Bangladesh, Dhaka, attention. Finally, we only identified potential DF clusters Bangladesh. in this preliminary study, but did not explore possible risk

1090 ª 2012 Blackwell Publishing Ltd Tropical Medicine and International Health volume 17 no 9 pp 1086–1091 september 2012

S. Banu et al. Space-time clusters of dengue fever

Ferdausi SA & Bolkland MW (2000) Rainwater harvesting for Podder G, Breiman RF, Azim T et al. (2006) Short report: origin of application in rural Bangladesh. 26th WEDC conference, dengue type-3 viruses associated with the dengue outbreak in Water,Sanitation and Hygiene:Challenges of the Millennium. Dhaka, Bangladesh,in 2000 and 2001. American Journal of Dhaka, Bangladesh. Tropical Medicine and Hygiene 74, 263–265. Gubler DJ (2002) Epidemic dengue ⁄ dengue hemorrhagic fever as a Rahman M, Rahman K, Siddque AK et al. (2002) First outbreak of public health, social and economic problem in the 21st century. dengue hemorrhagic fever, Bangladesh. Emerging Infectious Trends in Microbiology 10, 100–103. Diseases 8, 738–740. Hossain MA, Khatun M, Arjumand F, Nisaluk A & Breiman RF Rigau-Perez JG, Clark GG, Gubler DJ, Reiter P, Sanders EJ & (2003) Serologic evidence of dengue infection before onset Vorndam AV (1998) Dengue and dengue haemorrhagic fever. of epidemic, Bangladesh. Emerging Infectious Diseases 9, Lancet 352, 971–977. 1411–1414. Sutherst RW (2004) Global change and human vulnerability to Hu W, Clements A, Williams G & Tong S (2011) Spatial analysis vector-borne diseases. Clinical Microbiology Reviews 17, of notified dengue fever infections. Epidemiology and Infection 136–173. 139, 391–399. Thai KTD, Nagelkerke N, Phuong HL et al. (2010) Geographical Hu W, Clements A, Williams G, Tong S & Mengersen K (2012) heterogeneity of dengue transmission in two villages in southern Spatial patterns and socioecological drivers of dengue fever Vietnam. Epidemiology and Infection 138, 585–591. transmission in Queensland, Australia. Environmental Health Tran A, Deparis X, Dussart P et al. (2004) Dengue spatial and Perspectives 120, 260–266. temporal patterns, French Guiana, 2001. Emerging Infectious Kulldorf M (2010) SaTScanTM user guide for version 9.0. http:// Diseases 10, 615–621. www.satscan.org/. WHO (2000) Dengue Hemorrhagic Fever: Diagnosis, Treat- Kulldorf M, Athas W, Feuer E, Miller B & Key C (1998) Evalu- ment,Prevention and Control. WHO, Geneva, Switzerland. ating cluster alarms: a space-time scan statistic and brain cancer Williams CR, Bader CA, Kearney MR, Ritchie SA & Russell RC in Los Alamos, New Mexico. American Journal of Public (2010) The extinction of dengue through natural vulnerability of Health 88, 1377–1380. its vectors. PLoS Neglected Tropical Diseases 4, e922. Mammen MP, Pimgate C, Koenraadt CJM et al. (2008) Spatial Wu PC, Guo HR, Lung SC, Lin CY & Su HJ (2007) Weather as an and temporal clustering of dengue virus transmission in Thai effective predictor for occurrence of dengue fever in Taiwan. Villages. PLoS Medicine 5, e205. Acta Tropica 103, 50–57.

Corresponding Author Shilu Tong, School of Public Health and Social Works, Institute of Health and Biomedical Innovation, Queensland University of Technology, Kelvin Grove, Qld 4059, Australia. Tel.: +61 7 3138 9745; Fax: +61 7 3138 3369; E-mail: [email protected]

ª 2012 Blackwell Publishing Ltd 1091