Application of GIS in Modeling of Dengue Risk Based on Sociocultural Data: Case of Jalore, Rajasthan, India Alpana Bohra* and Haja Andrianasolo** #
by
*Space Technology Applications and Research (STAR) Program, Asian Institute of Technology (AIT), P. O. Box 4 Klong Luang, Pathumthani 12120, Thailand **Institute of Research for Development (IRD, France) Research Center for Emerging Viral Disease (RCEVD), Visiting Faculty of SAT/STAR, Asian Institute of Technology (AIT), P. O. Box 4 Klong Luang, Pathumthani 12120, Thailand Abstract
Prediction of dengue risk based on sociocultural factors and its possible spatial relationships was investigated in a dengue endemic area of Jalore in Rajasthan state, India. Data were collected through personal interviews, from 77 households, randomly selected from both dengue-affected samples (DAS) and unaffected samples (UAS). Findings indicated that out of sixty socioeconomic and sociocultural variables, only sixteen were co-related significantly at 0.5 and 0.1 level. These sixteen variables were used in the stepwise regression model; only eight variables, namely, frequency of days of cleaning of water storage containers, housing pattern, use of evaporation cooler, frequency of cleaning of evaporation cooler, protection of water storage containers, mosquito protection measures, frequency of water supply and frequency of waste disposal made a significant contribution to the incidences of DF/DHF/DSS with a R2 of 0.958. The geographical information system (GIS) has been used to link the spatial and significant sociocultural indicators with the disease data. Using factorial discriminant analysis and spatial modelling with these eight sociocultural indicators, five classes of risk categories ranging from “very low” to “very high” were identified. Validation of these risk categories on individual houses showed that 94.5% of the houses were correctly classified. The nearest neighbourhood method had been used to prepare a spatial extrapolated social risk area map. The paper highlights the statistical and spatial model development based on the analysis of sociocultural practices adopted by DAS and UAS and from the application of GIS. Keywords: Aedes aegypti, DF, DHF, DSS, sociocultural practices, Pearson’s correlation, regression, geographical information system (GIS)
#For correspondence: alpanabohra@yahoo.com
92
Dengue Bulletin – Vol 25, 2001
Application of GIS in Modeling of Dengue Risk Based on Sociocultural Data: Case of Jalore, Rajasthan, India
Introduction Dengue fever (DF) associated with dengue haemorrhagic fever/dengue shock syndrome (DHF/DSS) has emerged as an important public health problem in the countries of the South-East Asia and Western Pacific regions(1). In India dengue fever has been known since the 19th century and epidemics have been reported from almost all part of the country. In Rajasthan, serological studies on outbreaks of dengue fever have been reported from Jaipur and Ajmer(2,3). However, in an arid region like Jalore, an epidemic occurred during 1985 and again in 1990(4,5). The 1990 outbreak of dengue in Jalore occurred in summer (AprilMay) in contrast to other parts of India, where such outbreaks are commonly reported after the rains between August and November. Until now, nearly all research efforts had focused on the biological, entomological and clinical aspects of DF/DHF/DSS separately. Location-specific studies demonstrating an integrated use of sociocultural practices were lacking. This called for an in-depth study of the interrelationship of sociocultural practices and identification of the most significant risk indicators under the influence of local conditions using statistical modelling as an analytical tool. This study, conducted during December 2000 and July 2001, looked into these issues to evaluate and model the relationships between sociocultural practices and the incidences of DF/DHF/DSS.
western zone of the state. It lies between 24o 37’and 25o 49’ latitude and 71o 11’ and 73o 05’ east longitude. The town has a population of about 40,000 and the climate is characterized as dry with extremes of temperature rising as high as 48oC in summer months and going down to 10oC in winter months. The area has a sandy terrain and high wind velocity. It is situated at 736 metres above sea level. The average yearly rainfall is 421.6 mm, occurring mainly in July and August.
Primary and secondary data Primary data were collected through a field survey. A structured questionnaire composed of sixty variables, all potentially influencing the occurrence of DF/DHF/DSS, were designed to obtain information through personal interviews and discussions with both dengue-affected samples (DAS) and unaffected samples (UAS). The questionnaire collected data about family details, human dwellings, occupational patterns, awareness and knowledge about dengue, mosquito protection practices, sanitation and waste disposal management, cultural practices regarding storage of water containers and health care. Each individual household in the study area was defined as a sampling unit. All available dengue patients were taken as sample and an equal number of randomly-selected unaffected samples in the study area were also interviewed. Secondary data included demographic information about the town of Jalore; its climate; a list of DF/DHF patients and cases of deaths, along with their addresses as registered in Government Hospital, Jalore, during the 1990 outbreak; entomological data of dengue (adult house index, container
Methods Description of study area Jalore town is one of eleven desert districts in Rajasthan and is situated in the south-
Dengue Bulletin – Vol 25, 2001
93
Application of GIS in Modeling of Dengue Risk Based on Sociocultural Data: Case of Jalore, Rajasthan, India
index); physical environment (land use, land cover), and topographical and administrative map. These data were acquired with the help of the census book, government/nongovernment agencies and from published reports.
Results and discussion Statistical methodology All the sociocultural variables were studied through four steps: (1) Based on literature review, an attempt was made to group the number of variables by combining related variables into one group. Data were therefore arranged into the following six groups: socioeconomic; human dwellings; environment management practices; mosquito protection practices; cultural practices of water storage; and technological adoptions; (2) Detection and screening of data outliers to reduce misleading results. In the first screening of the sixty variables, forty-eight variables were selected; (3) A second screening of variables was conducted based on a significance test of Pearson’s correlation coefficient; (4) Development of predictive model. a regressive-
Statistical analysis of social data Correlation, regression and discriminant analysis were the major statistical tools used in this study for investing and testing the statistical significance in the relationship between sociocultural parameters and dengue incidences.
GIS modelling with social data After the risk categories of each household were identified from discriminant analysis, a database was created in the geographical information system (GIS), which was then linked with spatial point data of each house and surfaces were created from point samples. The inverse distance weighing (IDW) interpolation (nearest neighbour technique) was employed to produce the desired results. The IDW interpolator assumes that each input point has a local influence that diminishes with distance. It weighs the points closer to the processing cell as greater than those farther away. A specified number of points, or optionally all points within a specified radius, can be used to determine the output value for each location. IDW interpolation gives values to each cell in the output grid theme by weighing the value of each point by the distance that point is from the one being analysed and then averaging the values.
Correlation of sociocultural practices with dengue incidences Pearson’s correlation coefficient was computed for the forty-eight variables from the six groups with the incidence of DF/DHF/DSS. This statistical technique could identify and isolate sixteen variables that had the strongest positive or negative correlations, tested at 1% and 5% of significance levels. Dengue Bulletin – Vol 25, 2001
94
Application of GIS in Modeling of Dengue Risk Based on Sociocultural Data: Case of Jalore, Rajasthan, India
Regressive-predictive model From the six groups, the sixteen variables that were significantly correlated to the dengue incidence were submitted to multiple regression analysis. The different characteristics of the sociocultural variables interact together to contribute a combined effect on the dengue incidence. The sixteen variables, which were found to correlate significantly to dengue incidence, were only used for multiple regression analysis. Stepwise regression technique was employed to explore and identify statistically significant sociocultural risk indicators and their relative contribution to the occurrence of dengue incidence by eliminating the insignificant variables. Results of stepwise regression analysis revealed that, out of sixteen, only eight independent variables contributed effectively to dengue incidences. The eight variables were used to derive the following regression equation: Y = - 0.07516 + 0.928X1 - 0.819 X2 + 0.757 X3 + 0.006042 X4 + 0.284 X5 – 0.647 X6 - 0.317 X7 - 0.216 X8
X6 = Mosquito protection measures used by the households X7 = Frequency of water supply X8 = Frequency of waste disposal at community level. The results of the stepwise multiple regression analysis indicated that the multiple R and R2 for the final model were 0.979 and 0.958 respectively. Adjusted R2 was 0.938 explaining 93.8% of the total variation in the dengue incidence.
Discussion of significant variables Frequency of cleaning of water storage containers: The model showed that the variable Frequency of cleaning of water storage containers made a positive contribution to dengue incidence. The Aedes aegypti mosquito is a domestic breeder and breeding can occur in water storage containers, which are not emptied and cleaned for sufficiently long periods. The Aedes aegypti eggs are normally laid on the damp walls of both artificial and natural containers and they could resist desiccation for several weeks to several months. The eggs hatch when submerged in water. Since water is essential during the first 8 days in the life of mosquitoes, therefore if the frequency of cleaning is more than 8 days, this could contribute to an increase in the abundance of adult mosquitoes and the risk of dengue virus transmission. Whereas, changing water and emptying water storage containers once or twice a week will greatly reduce the risk of dengue fever. It was observed in the study area that people cleaned containers daily, but only those
where, Y = Incidence of dengue (dependent variable) X1 = Frequency of cleaning of water storage containers X2 = Housing pattern X3 = Use of water cooler X4 = Frequency of cleaning of the water cooler X5 = Protection/covering of water storage containers Dengue Bulletin – Vol 25, 2001
95
Application of GIS in Modeling of Dengue Risk Based on Sociocultural Data: Case of Jalore, Rajasthan, India
which were used to store water for drinking purpose. However, those containers which were used to store water for other domestic purposes, i.e. washing, bathing, etc., were cleaned infrequently. These containers were normally cleaned after 10-15 days, or even after one month, thus providing ideal oviposition sites for mosquitoes and subsequent sticking of the eggs. The eggs would then hatch and develop into mosquitoes when inundated. Use of water evaporation coolers: The use of water evaporation coolers generally starts with the onset of summer months. In Jalore, the use of coolers was observed to start in the middle of March or early April. Coolers were used until the end of July. Most coolers were found fitted in openings, initially used as windows, whereas some coolers were of the portable type. Coolers and other containers become excellent places for Aedes mosquito breeding and can lead to widespread transmission of dengue fever. The cooler plays an important role in the breeding of secondary foci. It was observed that once the cooler was fixed to windows, it remained there. With the onset of the monsoons, the breeding of Aedes aegypti larvae spreads from its mother foci to secondary foci, which are coolers. The model indicates the positive relationship of dengue incidence to the use of water coolers in Jalore. Studies by Katyal et al., 1996(6) indicate that coolers play an important role in mosquito breeding, which seems to support research results. Uncovered water storage containers: Open water storage containers provide ideal breeding places for Aedes aegypti mosquitoes. During the survey it was
observed that portable cement tanks, metallic/plastic drums and overhead/ underground tanks were used to store clean water within premises. Most domestic water storage containers were kept uncovered except for underground tanks. Different studies indicate that uncovered water containers and pitchers were significantly associated with dengue infection(7,8). The regression model indicates that the presence of uncovered water containers makes a positive contribution to dengue incidence. It is interesting to note that the epidemic occurred in the summer months (April-May) when there was scarcity of water. This scarcity could result in increased storage of water, thereby increasing the risk of dengue incidence and thus holding a positive correlation. Protection measures against mosquitoes: Use of nets, screening of houses, creating smoke with neem leaves, spraying of insecticides and closing of doors and windows were the common protective measures used against mosquitoes. These measures either reduce the number of mosquitoes or provide protection against bites and thus reduce the risk of dengue infection. The model showed that the variable mosquito protection measures had negative association with the dengue incidence, i.e. the more protection measures were used, the less incidence there was of dengue. Housing pattern: A review of the available literature indicated that in a crowded area, many people living within the short flight range of the vector from its breeding source could be exposed to transmission even if the house index was Dengue Bulletin – Vol 25, 2001
96
Application of GIS in Modeling of Dengue Risk Based on Sociocultural Data: Case of Jalore, Rajasthan, India
low. Therefore, higher population density and interconnection of houses could lead to more efficient transmission of the virus and thus increased exposure to infection#. The transmission of the disease is normally limited by the flight distance of Aedes aegypti during its lifetime. The flight distance of Aedes aegypti could range from a few metres to more than 50 metres in a closed urban environment(9,10). In urban environment where interconnections are not very common, the independent nature of houses limits the flight range of Aedes aegypti and reduces the transmission of the disease. The prediction model indicated that the variable connectivity of houses (independent =1, connected = 0) had a negative correlation with dengue incidence. This correlation is in line with the results from available studies. Cleaning of water evaporation coolers: The model showed that the variable frequency of cleaning of water coolers had a negative impact on dengue incidence. Generally, the prolonged stay of water in a cooler permits damp space as well as litter formation, thus providing nutrition to larval habitats. This permits the growth and emergence of Aedes aegypti mosquitoes, thus increasing dengue risk. The negative correlation indicated that if the frequency of cleaning of coolers was high, there would be less chances of dengue infection. This could appear as evidence given the fact that cleaning prevents potential breeding of mosquitoes by removing litter.
Frequency of water supply: The model showed that the variable frequency of water supply was negatively correlated to incidences of dengue. Water supply in most houses, especially during summer (March to June), was inadequate and not reliable. Water scarcity, resulting in increased and prolonged storage of water for domestic use in various types of containers, subsequently becomes the cause of breeding of Aedes aegypti. Water storage practices in the area, due to irregular water supplies, were a possible cause for higher vector concentration in the sampled houses, thus increasing dengue transmission. It means more infrequent the supply of water, more the practice of water storage and more the presence of vectors, thus increasing the growth, transmission and risk of dengue infection. Frequency of solid waste removal: Frequency of garbage removal was the eighth contributing factor, which influenced in a negative direction. The presence of solid wastes around the households, such as cans, car parts, bottles, old used tyres and other junk material found in several houses, created potential breeding sites. Dumping of solid waste for long periods of time such as 15-20 days supported the breeding of Aedes aegypti and increased the transmission of disease. If the frequency of collection and disposal of solid waste by local bodies increases, it would control the Aedes breeding and thus would reduce transmission.
Development of spatial model # In dengue epidemiology, there is a multiple-case syndrome, as the vector is a day-biting species. Since the mosquito has to bite a number of persons in a single house to get one blood meal, that is the reason more people in a house get infected. – Editor
Both spatial and sociocultural parameters could be important in determining disease emergence and transmission. GIS could create possible links between spatial data 97
Dengue Bulletin – Vol 25, 2001
Application of GIS in Modeling of Dengue Risk Based on Sociocultural Data: Case of Jalore, Rajasthan, India
and their related descriptive information, which could include socioeconomic and sociocultural parameters. The objective of spatial modelling was to determine the applicability of GIS as a tool to identify varying degrees of spatial social risks in Jalore related to dengue incidence and transmission.
Table 1. Social risk indicators and their weightage Social risk indicators Frequency of cleaning of water storage containers a. 1-4 days b. 5-15 days c. >16 days Housing pattern a. Independent house b. Mixed c. Interconnections Use of water coolers a. 5 days/month b. 6-10 days/month c. >15 days/month Frequency of cleaning of water cooler a. 1-4 days b. 5-15 days c. >16 days Protection of water storage container a. Fully covered b. Sometimes c. Mostly uncovered Mosquito protection measures a. Screens b. Insecticides c. close windows d. Smoke/burning herbs (neem) e. Mosquito-net Frequency of water supply a. Everyday b. Alternate c. Every 3 days d. 4-7 days Frequency of waste removal a. Everyday b. Weekly c. >15 days Risk scores
1 2 3 1 2 3 1 2 3
Weighting of sociocultural practices Method of “weights” was observed to be a suitable technique which would have a combined effect of various social risk factors contributing to the incidence of dengue. To develop a combined social risk category, the eight social risk indicators identified from stepwise regression analysis were selected. Based on a review of literature, weights were then assigned to their associated parameters indicating the degree of an individual risk indicator. In order to maintain uniformity among all social risk indicators, an equal weighing method was used. Weights of 1-3 were assigned to associated practices. For a given social risk indicator, a higher weight (3) was given to the practices with a higher risk of dengue incidence, medium weight (2) was assigned to the one contributing medium risk in the incidence of dengue, and a low score (1) was given to the practice with low risk of dengue incidence. For example, in the case of the risk indicator - “Frequency of solid waste removal”, the lowest value of 1 was assigned to the practice of short duration (once in 1-4 days), 2 to medium duration (once in 5-15 days) and 3 to long duration (more than 15 days). The detailed weighting for all the eight risk indicators is presented in Table 1.
1 2 3
1 2 3 1 1 2 1 2 1 1 2 2 1 2 3
98
Dengue Bulletin – Vol 25, 2001
Application of GIS in Modeling of Dengue Risk Based on Sociocultural Data: Case of Jalore, Rajasthan, India
Since the sociocultural practices related to the eight social risk indicators varied from household to household, these scores were assigned separately for each of the 77 households belonging to DAS and UAS groups.
Development of social risk levels The discriminant analysis approach was used to obtain household-wise social risk scores, and by using the histogram/box plot technique, social risk scores were translated into social risk levels. The histogram depicts the mean, minimum and maximum values and standard deviation of discriminant scores of the 77 households. To derive risk levels, percentiles technique was used. For example, if five possible risk levels are to be identified, discriminant scores at 20 percentile, 40 percentile, 60 percentile and 80 percentile could be used as the cut-off points as shown in Table 2 and Figure 1. Table 2. Percentile households and discriminant scores Percentile 10 20 25 30 40 50 60 70 75 80 90 100 Discriminant score -2.5360433 -2.0483010 -1.8912598 -1.6407418 -0.9233625 -0.1139570 0.4975759 1.4409430 1.7847998 2.3510595 2.9958430 3.6073759
These risk levels were termed as very low, low, medium, high and very high risk levels respectively with scores of 1, 2, 3, 4 and 5 for the assigned risk levels respectively. For each household, it is necessary to check the predicted risk levels from discriminant scores to their actual class. For this, estimates of the classification function coefficients were used. It was observed that the application resulted in 94.8% correct classification under the risk level categories, which was very positive from statistical considerations. Figure 1. Histogram of discriminant scores 14 12 10
Frequency
8 6 4 2 0 Std. Dev = 2.00 Mean = 0.00 N = 77.00 -3.00 -2.00 -1.00 0.00 1.00 2.00 3.00 -2.50 -1.50 -.50 .50 1.50 2.50 3.50
Discriminant Scores
Development of a spatial social risk model As the sample households were spatially distributed and social risk information with regard to these households was collected and analysed, it could provide spatiallydistributed social risk levels. The spatial (point data) with their attributes were input into GIS and a spatial point-wise risk-level map was developed. This was achieved by digitizing spatial locations of houses of DAS (37 households) and UAS (40 households) samples located in the Jalore administrative map as shown in Figure 2. On the
Dengue Bulletin – Vol 25, 2001
99
Application of GIS in Modeling of Dengue Risk Based on Sociocultural Data: Case of Jalore, Rajasthan, India
administrative map of Jalore, locations of houses of DAS and UAS were overlayed. GIS databases were developed separately for DAS as well as UAS groups having information on social risk levels. Nearest
neighbourhood technique of extrapolation was used to develop a social risk map of the area. This provided location-wise social risks of dengue incidence. Analysed results are presented in Figure 3.
Figure 2. Spatial location of dengue affected and unaffected houses in Jalore
Figure 3. Dengue risk levels associated with social and cultural parameters in Jalore
100
Dengue Bulletin – Vol 25, 2001
Application of GIS in Modeling of Dengue Risk Based on Sociocultural Data: Case of Jalore, Rajasthan, India
This figure (map) shows the spatial distribution of the five social risk levels, which were identified through discriminate analysis. The analysis indicated that a large percentage of the area (61.09%) had very low social risk, whereas 16.90% of the area had high risk, 12.35% had low risk, followed by 6.58% of the area with medium risk. Only a very small area, 3.09%, had very high risk. Overall, little more than 26% had medium to high social risk.
Conclusion Prediction of dengue risk based on sociocultural factors was investigated in a dengue endemic area of Jalore. The data analysis and modelling revealed that the sociocultural factors such as the housing patterns, limited use of mosquito protection measures, irregular water supplies, poor management of waste disposal, storage of water on the premises due to inadequate water supplies in summer months, and prolonged storage of water for domestic and other purposes significantly affected the incidence of dengue. Storing of water in houses created conditions conducive to the breeding of Aedes aegypti mosquitoes and led to more pronounced vector presence. Stepwise regression analysis was found to be an appropriate technique for identifying significant social risk indicators which contributed to increased transmission of disease. It may, therefore, be concluded that any step taken to improve any of the above social and cultural practices would have favourable effects on reducing dengue cases. Such analysis provides valuable information for the planning of precautionary measures and for controlling the spread of
DF/DHF/DSS. The objective of spatial modelling was to create a linkage between households, their sociocultural practices and dengue incidence. The spatial model is capable of identifying five different levels very low, low, medium, high and very highrisk levels of dengue incidence for the study area. It would contribute significantly to the spatial prediction of social risk levels in Jalore. Furthermore, the approach could assist in focusing and implementing precautionary and preventive strategies to monitor and control the incidence of dengue more effectively.
References 1. World Health Organization, 1975. Technical advisory committee on dengue haemorrhagic fever for the South-East Asia and Western Pacific regions: Technical Guides for Diagnosis, Treatment, Surveillance, Prevention and Control of Dengue Haemorrhagic Fever. Ghosh SN, Pavri KM, Singh KRP, Sheikkh BH, D’Lima LV, Mahadev PVM, Rao TR. Investigations on the outbreak of dengue fever in Ajmer city, Rajasthan state in 1969. Part I. Epidemiological, clinical and virological study of the epidemic. Indian Journal of Medical Research, 1974, 62: 511-522. Padbidri VS, Dandavate CN, Goverdhan MK, Bhatt UKM, Rodrigues FM, Lima VD. An investigation of aetiology of the 1971 outbreak of febrile illness in Jaipur city, India. Indian Journal of Medical Research, 1973, 61: 1737-1743. Chouhan GS, Rodrigues FM, Shaikh BH, IIKal MA, Khangaro SS, Mathur KN, Joshi KR, Vaidhye NK. Clinical and virological study of dengue fever outbreak in Jalore city, Rajasthan, 1985, Indian Journal of Medical Research, 1990, 91: 414-418. Joshi V, Mathur ML, Dixit AK, Singhi M. Entomological studies in a dengue endemic area, Jalore, Rajasthan, Indian Journal of Medical Research, 1996, 104: 161-165.
2.
3.
4.
5.
Dengue Bulletin – Vol 25, 2001
101
Application of GIS in Modeling of Dengue Risk Based on Sociocultural Data: Case of Jalore, Rajasthan, India 6. Katyal R et al. Seasonal variations in Aedes aegypti population in Delhi, India. Dengue Bulletin, 1996, 20: 78-81. Koopam JS, Prevots DR, Marin MAV, Dantes HG, Aquino MLZ, Longini IM Jr, Amor, JS. Determinants and predictors of dengue infection in Mexico. American Journal of Epidemiology, 1991, 13: 1168-1177. Khera A, Sharma RS. Epidemiological investigations of an outbreak of dengue fever in Malikpur village under Najafgarh block, Delhi. Dengue Newsletter, 1993, 18:1.
9.
Morlan HB, Hayes RO. Urban dispersal and activity of Aedes aegypti. Mosquito News, 1958, 18: 137-144.
7.
8.
10. Reiter P, Amador MA, Anderson RA, Clark GG. Short report: Dispersal of Aedes aegypti in urban areas after blood feeding as demonstrated by rubidium-marked eggs. American Journal of Tropical Medicine and Hygiene, 1995, 52: 177179.
102
Dengue Bulletin – Vol 25, 2001