Geographical association between socioeconomics and age of dengue haemorrhagic fever patients in Surabaya, Indonesia Yoshiro Nagao,a# Esty M. Rachmie,b Shiro Ochi,c Maria M. Padmidewi,d Kuntariantoe & Masato Kawabataa a
International Centre for Medical Research and Treatment, Kobe University, Kusunoki, Chuo-ku, Kobe, Hyogo, Japan.
b c
Public Health Bureau of Surabaya City, Jalan Jemursari, No. 197, Surabaya, Indonesia.
Department of Environmental Management, Faculty of Agriculture, Kinki University, Nara, Japan d
Public Health Laboratory of East Jawa, Jalan Karangmenjangan, No. 18, Surabaya, Indonesia e
Public Health Bureau of East Jawa, Jalan A. Yani, No. 118, Surabaya, Indonesia
Abstract A study was designed to correlate the ages of dengue patients to the geographical and temporal demographic structure in 28 districts in Surabaya, Indonesia, between 1996 and 2005. The geographical distribution of the mean patient age was stable throughout the study period. The mean patient age did not correlate with the demographic structure but was related to the prevalence of poor housing where mosquito density was high. These results suggested that socioeconomic factors which affect mosquito abundance are more important determinants of the mean age of DHF patients than the demographic variables. Keywords: Demographic structure; Geographical distribution; Socioeconomics; Satellite imagery; Geographical information system; Urbanization; Poverty; Islam; Surabaya; Indonesia.
Introduction Although dengue illnesses affected predominantly small children until the 1970s, the mean age of dengue illnesses has been shifting to adult populations in many south-east Asian countries[1] such as Singapore,[2-4] Thailand[5,6] and Indonesia.[7,8] These observations led to a hypothesis that the mean ages of dengue illnesses are a reverse indicator of mosquito abundance and that increases in the mean ages of dengue patients reflect decreases in #
E-mail: in_the_pacific214@yahoo.co.jp
116
Dengue Bulletin – Volume 35, 2011
Dengue in Surabaya, Indonesia
mosquito abundance.[9-11] However, an alternative hypothesis was proposed, which assumed that mean age of patients with dengue illnesses indicates the demographic structure of the population.[12] In a population with a larger proportion of young children, the mean age of patients would be lower. To date, however, no study has examined actual data to explore the determinant(s) of the mean age of patients with dengue illnesses. The present study obtained the mean age of patients of dengue haemorrhagic fever (DHF) from each district in Surabaya, the second largest city in Indonesia. This variable was regressed against the socioeconomic and demographic variables at the district level to identify factors that affected the mean patient age.
Materials and methods Study area Surabaya is approximately 30 km × 20 km (375 km2) in size and is divided into 28 districts (Figure 1). Numerous modern buildings are located in the centre of the city, while poorlyconstructed houses are situated on the banks of rivers, especially in the northern coastal area. Although these houses have been threatened by occasional flooding, social intervention programmes to relocate the residents have not made much progress.[13,14] For this study, a digital map of Surabaya City was generated based on the official map, using PC-Mapping Auto-Tracer (MAPCOM, Tokyo) and Mapinfo 7.0 (New York). Figure 1: Location of Surabaya, Indonesia (Surabaya is divided into 28 districts)
Dengue Bulletin – Volume 35, 2011
117
Dengue in Surabaya, Indonesia
Epidemiological data Individual DHF cases are reported daily to the Public Health Bureau of Surabaya City (PHBSC) by public health stations (puskesmas), to which private and public hospitals are obliged to report DHF cases. Each hospital was instructed by PHBSC to follow the WHO’s diagnostic criteria for DHF.[15,16] Blood samples from ambiguous cases were sent to the Public Health Laboratory of East Java for confirmation of diagnosis. The reports sent to PHBSC were recorded on paper and subsequently compiled into an electronic format. Since age and residential address were not included in the electronic records, only paper records were used in this study. As a result, only the paper records from 1996, 1997, 1998, 2002, 2003 and 2005 were available from the archives. In total, 10 564 cases of DHF were reported during these six years. Ages and residential addresses were available for 10 079 (95%) of the reported cases, and only these cases were included in the analysis. The size of the population in each district was obtained from the annual reports from the Statistics Bureau of Surabaya City. The population of Surabaya city was 2 344 520 in 1996 and 2 629 001 in 2005.
Detection of spatial clustering of epidemiological variables To interpret the geographical distribution of the epidemiological variables quantitatively, we employed the Getis-Ord Gi statistic,[17] which detects ‘positive cluster’ (spatial clustering of large values) and ‘negative cluster’ (clustering of small values). For this and subsequent spatial analyses, a binary distance matrix was required. Each element of the binary distance matrix was coded “1” if a pair of district centres was within a pre-defined neighbourhood cut-off distance or “0” otherwise. Since the longest minimum distance between district centres was 5.8 km and the shortest maximum distance was 12 km, we defined the neighbourhood cutoff distance as 9 km, the average of those two distances.
Demographic/socioeconomic variables The correlation between the mean patient age and demographic/socioeconomic data was investigated at the district level. These socioeconomic/demographic datasets were obtained from the above-mentioned annual reports from the Statistics Bureau of Surabaya City, and are defined in Table 1. Among these variables, the birth rate and primary school attendance, which represents 94% of children aged 7 to 12 years in Indonesia,[18] are reliable indicators of the age structure. When a variable was missing for a specific district in any given year, its value was interpolated by averaging the values from the years before and after the missing year. Two exceptions were ‘poor housing’ and ‘family size’ in Table 1: the former was reported only in 2005, and the latter was reported only in 2002. We used the values in these years for the whole study period.
118
Dengue Bulletin – Volume 35, 2011
Dengue in Surabaya, Indonesia
Table 1: Demographic/socioeconomic/geographic variables employed as explanatory variables Variable name Definition New births per 1000 individuals per year Population per 1 km2 Deaths per 1000 individuals per year Incoming population per 1000 individuals per year Outgoing population per 1000 individuals per year Number of pupils in primary school (state, private, and Islamic) per 1000 individuals Number of state junior high school students per 1000 individuals Number of private junior high school students per 1000 individuals Number of state senior high school students per 1000 individuals Number of private senior high school students per 1000 individuals Number of kindergarten children per 1000 individuals Percentage of Islamic primary school pupils in the total number of primary school pupils Per capita volume of garbage (m3) collected daily Percentage of park area in total size of district Percentage of agricultural area in total size of district Percentage of poorly constructed housings Number of physicians of public health station per 1000 individuals Number of members per family Coded as 1 for a district that faces the sea or 0 for a district that does not face the sea Percentage of high density residential area in the total district area
1. Birth rate 2. Population density 3. Mortality 4. Immigrants 5. Emigrants 6. Primary school pupils 7. State junior high school students students students students
8. Private junior high school 9. State senior high school 10. Private senior high school 11. Kindergarten children 12. Islamic education 13. Garbage 14. Park areas 15. Agricultural areas 16. Poor housing 17. Public physicians 18. Family size 19. Coastal district 20. High-density residential areas
Dengue Bulletin – Volume 35, 2011
119
Dengue in Surabaya, Indonesia
Geographical variable and satellite imagery Geographical heterogeneity, such as the presence of the sea, may affect the district climate, which is an important determinant of the transmission intensity. To adjust for this effect, a dummy variable indicating whether a district faces the sea or not was incorporated into the statistical analysis. The heterogeneity in land use (for example, the degree of aggregation of premises) may also affect the probability of movement of vector mosquitoes from house to house. To consider this possibility and incorporate a variable independent of official publications, we estimated the percentage of ‘high density residential areas’ (Table 1) for each district using satellite imagery data. Briefly, raw data recorded on 11 July 2009 by the Advanced Visible and Near Infrared Radiometer type 2 loaded on the Advanced Land Observation Satellite was segmented.[19] Normalized Difference Vegetation Index (NDVI) was estimated for each segment.[20,21] Land use was classified into ‘factories’, ‘residential areas’ and ‘crop land’ using the standard supervised classification method based on NDVI (Figure 2). With a cut-off NDVI of −0.2, residential areas were divided into high- and lowdensity residential areas (Figure 3). Figure 2: Map of residential density (Using satellite image data, high-density residential areas (green) and low-density residential areas (red) were identified)
120
Dengue Bulletin – Volume 35, 2011
Dengue in Surabaya, Indonesia
Figure 3: Algorithm used to classify satellite imagery objects (Using the algorithm described in this figure, residential areas were identified from the satellite image data and classified into high- and low-density residential areas, based on the Normalized Difference Vegetation Index (NDVI)) Image
Image object NDVI < 0.6 Water NDVI < 0 .2 Supervised classification NDVI > 0 .2 NDVI > 0 .6 Land
Factories
Residential area NDVI < 0.2
Crop land NDVI > 0.2
Forest & Agricultural field
High-density residential area
Normal density residential area
Statistical analysis Stata 9.2 was used for the statistical analyses. As a screening process, we selected the socioeconomic variables that exhibited a significant rank correlation (P<0.05) with the mean age of patients. For this analysis, the overall dataset was prepared in which the mean patient age was estimated from patient records pooled over the six years, while the socioeconomic/ demographic variables were averaged from those six years. The selected variables were then incorporated into the subsequent regression analyses. We employed spatial regression analysis and longitudinal regression analysis. In both analyses, independent variables that did not make a statistically significant contribution to the regression model were eliminated one at a time (Wald’s test).
Dengue Bulletin – Volume 35, 2011
121
Dengue in Surabaya, Indonesia
Spatial regression analysis The bias from spatial autocorrelation was adjusted for, using the spatial regression analysis with lag model.[22] The neighbourhood cut-off distance was set to 9 km as mentioned above. The above-mentioned overall dataset aggregated from the 6 years was used.
Longitudinal regression analysis To consider the inter-annual variation, we employed the random effect linear regression model. The data recorded for the individual years were used for this analysis. Year was incorporated as an independent variable to represent the temporal trend.
Results Epidemiology of DHF Table 2 summarizes the epidemiological data used in the analysis. As shown in Table 2, the annual incidence was highly unstable. The geographical distributions of this variable supported this observation, showing apparently unpredictable patterns (Figure 4). On the other hand, Figure 4: Geographical distribution of the incidence of DHF (Districts are classified based on the annual incidence of DHF cases (per 100 000 individuals))
122
Dengue Bulletin – Volume 35, 2011
Table 2: DHF statistics for Surabaya, Indonesia Surabaya as a whole Range, at the district level
Dengue Bulletin – Volume 35, 2011
Population
Number of DHF patients Population 1875 1328 2280 1569 835 2677 10 564 18.0 71.6 15.2 102 18.5 32.7 32 761−211 686 34 687−214 062 33 597−208 333 18.2 64.5 25 580−215 833 19.9 96.8 29 473−206 479 10−235 5−169 3−81 32−236 96−960 20.4 56.4 21 196−205 414 5−125 17.7 80.0 20 834−203 749 9−212 8.06−24.0 11.4−30.0 12.3−27.2 6.10−23.7 6.33−26.7 9.25−20.7 11.4−23.8
Mean age of DHF patients (years) Number of DHF patients
Annual incidence of DHF (per 100 000) Mean age of DHF patients (years)
Annual incidence of DHF (per 100,000) 22.8−211 12.1−120 14.8−182 27.4−177 7.33−89.5 42.5−194 45.8−124
1996
2 344 520
1997
2 356 386
1998
2 431 348
2002
2 431 501
2003
2 553 022
2005
2 629 001
Overall period†
2 457 630
†
Cases from 6 years (1996, 1997, 198, 2002, 2003, and 2005) are aggregated.
Dengue in Surabaya, Indonesia
123
Dengue in Surabaya, Indonesia
the geographical distribution of the mean age of patients was stable: consistently high in the south-eastern districts and low in the north-western districts (Figure 5). Furthermore, this was supported by the geographical clusters of the mean patient age: a positive cluster persisted in the south-eastern region while a negative cluster existed in the north-western region throughout the study period (Figure 6).
Explanatory variables Table 3 summarizes the individual explanatory variables and their rank correlations with mean age of DHF patients. The following variables exhibited significant rank correlation with mean age of patients: emigrants (P=0.0320), private junior high school students (P=0.0472), private senior high school students (P=0.0118), kindergarten children (P=0.0007), Islamic education (P=0.0001), garbage (P=0.0354), agricultural area (P=0.0395), and poor housing (P=0.0021). Figure 5: Geographical distribution of the mean age of DHF patients (Districts are classified based on the mean age of DHF patients (in years) in individual study years or over all 6 years combined)
124
Dengue Bulletin – Volume 35, 2011
Dengue in Surabaya, Indonesia
Figure 6: Geogaphical clusters of the mean age of DHF patients (Districts are classified based on statistical significance of the Getis-Ord Gi statistic, estimated from the mean age of DHF patients. Positive cluster represents clustering of large values, while negative cluster represents clustering of small values)
Spatial and longitudinal regression analyses From these eight variables, only those exhibiting significant contribution to the statistical model were selected by Wald’s test. Private senior high school students, Islamic education and poor housing were selected by spatial regression analysis (column (a) of Table 4). The longitudinal regression analysis selected private senior high school students, poor housing and year (column (b) of Table 4).
Geographical distribution of selected variables Table 4 indicates that in both spatial and random-effect regression analyses, two factors showed statistically significant contribution to the mean age of patients: private senior high school students as a positive contributor and poor housing as a negative contributor. Therefore, we selected these two variables as the most robust predictors of the mean age of patients. The relationship between these variables and mean age of patients is shown in Figure 7. Figure 8 shows the geographical distribution of these socioeconomic/demographic variables.
Dengue Bulletin – Volume 35, 2011
125
Dengue in Surabaya, Indonesia
Table 3: Summary of district attribute variables averaged through the study period and rank correlations with mean age of DHF patients Variable name Mean 12 12 057 3.6 20 16 111 19 29 11 21 28 10 0.0030 0.30 41 17 0.042 3.68 0.36 31 Range 9.5 − 19 819 − 40,333 2.6 − 5.1 8.6 − 53 9.4 − 22 57 − 149 0 − 67 7.6 − 61 0 − 98 0 − 85 12 − 49 0.26 − 34 0.00137 − 0.00562 0 − 1.8 0 – 94 5.1 – 41 0.019 − 0.074 3.15 – 3.93 0–1 0.75 − 85 Rank correlation with mean patient age and (P) −0.096 (P=0.6278) −0.091 (P=0.6437) −0.15 (P=0.4496) 0.22 (P=0.2618) 0.41 (P=0.0320) 0.23 (P=0.2302) −0.0082 (P=0.9668) 0.38 (P=0.0472) −0.021 (P=0.9165) 0.47 (P=0.0118) 0.60 (P=0.0007) −0.66 (P=0.0001) 0.40 (P=0.0354) 0.31 (P=0.1116) 0.39 (P=0.0395) −0.56 (P=0.0021) 0.24 (P=0.2106) −0.37 (P=0.0520) −0.19 (P=0.3231) −0.039 (P=0.8422)
1. Birthrate (per 1000) 2. Population density (per square kilometer)
3. Mortality (per 1000) 4. Immigrants (per 1000) 5. Emigrants (per 1000) 6. Primary school pupils (per 1000) 7. State junior high school students (per 1000) (per 1000) (per 1000) (per 1000)
8. Private junior high school students 9. State senior high school students 10. Private senior high school students 11. Kindergarten children (per 1000) 12. Islamic education (%) 13. Garbage (cubic meters per person) 14. Park areas (%) 15. Agricultural areas (%) 16. Poor housing (%) 17. Public physicians (per 1000) 18. Family size (per family) 19. Coastal district (binary) 20. High density residential areas (%) Bold text indicates statistical significance.
126
Dengue Bulletin – Volume 35, 2011
Dengue in Surabaya, Indonesia
Table 4: Regression models to explain mean age of DHF patients (a) Spatial regression with selected variables (n=28) Coefficient (P) Private senior high school students Islamic education Poor housing Year 0.048 (P=0.005) −0.13 (P=0.002) −0.099 (P=0.008) not used R2=0.74 −0.17 (P=0.001) −0.25 (P<0.001) R2=0.28 (b) Random-effect regression with selected variables (n=168) Coefficient (P) 0.069 (P=0.006)
Figure 7: Mean age of DHF patients plotted over socioeconomic variables (Mean age of DHF patients estimated over all 6 years was plotted over socioeconomic variables. Please note the remarkable dependence of mean age of patients upon poor housing (A), private senior high school (B), and Islamic education (C). Three demographic variables [birth rate (D), primary school attendance (E), and mortality (F)] did not show relationship to the mean age of patients)
Dengue Bulletin – Volume 35, 2011
127
Dengue in Surabaya, Indonesia
Figure 8: Geographical distribution of socioeconomic variables (Districts were classified based on the socioeconomic or demographic variables. The geographical distributions of poor housing (A) and Islamic education (B) overlapped with that of low mean age of patients (Figure 5 and Figure 6). On the contrary, the geographical distribution of private senior high school attendance (C) overlapped with that of high mean age of patients. None of the demographic variables [birthrate (D), primary school attendance (E) and mortality (F)] showed geographical distribution related to mean patient age)
Discussion The geographical distribution of the mean age of patients in Surabaya was stable during the study period (Figure 5 and Figure 6). We subsequently found that the mean patient age was related negatively to the prevalence of poor housing, but positively to the use of private senior high schools. These results may be interpreted as follows: in developing countries, poor premises are not equipped with window screens or air-conditioners, which hinder entry of mosquitoes.[23,24] In addition, poor premises are not supplied with piped water and they rely on household water containers. Therefore, poor housing provides ideal breeding places for Aedes.[25-27] In contrast, private senior high school attendance may indicate economic wealth, which affords window screens, air-conditioners and piped water supply. Alternatively, private senior high school attendance may reflect the educational level and awareness important for mosquito reduction. Therefore, the findings from this study are consistent with the hypothesis that the mean age of DHF patients is a reverse indicator of Aedes abundance. On the other hand, in 128 Dengue Bulletin – Volume 35, 2011
Dengue in Surabaya, Indonesia
Surabaya, the geographical distribution of the mean age of DHF patients was not associated with any demographic variables examined (Table 3, Table 4, Figure 7). This implies that the socioeconomic factors that surrogated Aedes abundance were more influential determinants of the mean age of DHF patients than demographic variables. The present study will not negate the importance of socioeconomic factors other than those which remained in the final statistical models (i.e. private senior high schools and poor housing). For example, family size, which showed a non-significant but considerable rank correlation with mean patient age (Table 3), may indicate vulnerability to dengue transmission among family members. The small sample size and dependence on the official publications may have blunted the statistical power of the present study. Further study with a wider spatio-temporal spread as well as diverse sources of information is warranted.
Acknowledgements We are grateful to Atsushi Yamanaka for collecting official publications from Surabaya, and to Tony Pilkington for assistance with geographical data preparation.
References [1] Guha-Sapir D, Schimmer B. Dengue fever: new paradigms for a changing epidemiology. Emerg Themes Epidemiol. 2005; 2:1. [2] Ooi EE, Hart TJ, Tan HC, Chan SH. Dengue seroepidemiology in Singapore. Lancet. 2001; 357: 685686. [3] Ooi EE, Goh KT, Chee Wang DN. Effect of increasing age on the trend of dengue and dengue hemorrhagic fever in Singapore. International Journal of Infectious Diseases. 2003; 7: 231-232. [4] Ooi EE, Goh KT, Gubler DJ. Dengue prevention and 35 years of vector control in Singapore. Emerging Infectious Diseases. 2006;12: 887-893. [5] Chareonsook O, Foy HM, Teeraratkul A, Silarug N. Changing epidemiology of dengue hemorrhagic fever in Thailand. Epidemiology and Infection. 1999;122: 161-166. [6] Nagao Y, Svasti P , Tawatsin A, Thavara U. Geographical structure of dengue transmission and its determinants in Thailand. Epidemiology and Infection. 2008; 136: 843-851. [7] Sumarmo. Dengue haemorrhagic fever in Indonesia. Southeast Asian Journal of Tropical Medicine and Public Health. 1987; 18: 269-274. [8] Setiati T, Wagenaar, JFP ., Kruif, MD., Mairuhu ATA., Gorp ECM., Soemantri A. Changing epidemiology of dengue haemorrhagic fever in Indonesia. Dengue Bulletin. 2006; 30: 1-14. [9] Egger JR, Ooi EE, Kelly DW, Woolhouse ME, Davies CR, Coleman PG. Reconstructing historical changes in the force of infection of dengue fever in Singapore: implications for surveillance and control. Bulletin of the World Health Organization. 2008; 86: 187-196.
Dengue Bulletin – Volume 35, 2011
129
Dengue in Surabaya, Indonesia
[10] Nagao Y, Koelle K. Decreases in dengue transmission may act to increase the incidence of dengue hemorrhagic fever. Proceedings of the National Academy of Sciences of the United States of America. 2008;105: 2238-2243. [11] Thammapalo S, Nagao Y, Sakamoto W, Saengtharatip S, Tsujitani M, Nakamura Y, Coleman PG, Davies C. Relationship between Transmission Intensity and Incidence of Dengue Hemorrhagic Fever in Thailand. PLoS Negl Trop Dis. 2008; 2: e263. [12] Cummings DA, Iamsirithaworn S, Lessler JT, McDermott A, Prasanthong R, Nisalak A, Jarman RG, Burke DS, Gibbons RV. The impact of the demographic transition on dengue in Thailand: insights from a statistical analysis and mathematical modeling. PLoS Med. 2009; 6: e1000139. [13] Santosa H. Community participation in the upgrading of informal settlement and housing at the river bank of Surabaya. CIB World Building Congress. 2007. 2007:1964-1971. [14] Wibowo A. Segmental development design for Wonokromo waterfront settlements at Surabaya. Informal Settlements and Affordable Housing; 2007. Semarang, 2007. [15] WHO. Dengue haemorrhagic fever: diagnosis, treatment and control. Geneva: 1986. [16] World Health Organization. Dengue haemorrhagic fever: diagnosis, treatment, prevention and control. 2nd edition. Geneva: WHO, 1997. [17] Getis A, Ord JK. The analysis of spatial association by use of distance statistics. Geographical Analysis. 1992; 24: 189-206. [18] UNICEF. Basic education for all. Indonesia. 2009. http://www.unicef.org/indonesia/education.html [19] Earth Observation Research Center JAEA. Advanced land observation satellite, advanced visible and near infrared radiometer type 2 (AVNIR-2). http://www.eorc.jaxa.jp/ALOS/en/about/avnir2.htm - 12 January 2012. [20] Wang L, Sousa W, Gong P . Integration of object-oriented and pixel-based classification for mapping mangroves with IKONOS imagery. International Journal of Remote Sensing. 2004; 25: 5655-5668. [21] Tucker C. Red and photographic infrared linear combinations for monitoring vegetation. Remote Sensing of Environment. 1979; 8: 127-150. [22] Anselin L, Hudak S. Spatial econometrics in practice. A review of software options. Regional Science and Urban Economics. 1992; 22: 509-536. [23] Waterman SH, Novak RJ, Sather GE, Bailey RE, Rios I, Gubler DJ. Dengue transmission in two Puerto Rican communities in 1982. American Journal of Tropical Medicine and Hygiene. 1985; 34: 625632. [24] Thammapalo S, Chongsuwiwatwong V, Geater A, Lim A, Choomalee K. Socio-demographic and environmental factors associated with Aedes breeding places in Phuket, Thailand. Southeast Asian Journal of Tropical Medicine and Public Health. 2005; 36: 426-433. [25] Tun-Lin W, Kay BH, Barnes A. The Premise Condition Index: a tool for streamlining surveys of Aedes aegypti. American Journal of Tropical Medicine and Hygiene. 1995; 53: 591-594. [26] Barrera R, Navarro JC, Mora JD, Dominguez D, Gonzalez J. Public service deficiencies and Aedes aegypti breeding sites in Venezuela. Bulletin of the Pan American Health Organization. 1995; 29: 193-205. [27] Sharma K, Angel B, Singh H, Purohit A, Joshi V. Entomological studies for surveillance and prevention of dengue in arid and semi-arid districts of Rajasthan, India. J Vector Borne Dis. 2008; 45:124-132.
130
Dengue Bulletin – Volume 35, 2011