Study sites
The study was conducted in Antananarivo, the capital of Madagascar. This city extends over 80 sq km and its population was estimated at 1,697,000 inhabitants in 2007 [19]. The urban area of Antananarivo is located on a vast alluvial plain, extending from 18°48'S and 47°24'E in the North West, to 19°00'S and 47°42'E in the South East. The urban area consists of administrative, commercial, industrial and residential areas, with patches of agricultural land, most of which are used to grow rice. Antananarivo is crossed by the river Ikopa and canals (irrigation, drainage). Farmers make use of rainwater-fed irrigation systems, which are carefully managed, both communally and individually, to ensure that the rice fields receive adequate water supplies. The rest of the plain contains rice fields, which display dynamic changes during the growing season. Uncultivated rice fields are left flooded, or are allowed to become overgrown with weeds or reeds.
This area has a tropical, high-altitude climate, with a cold, dry season from May to October and a hot and wet season from November to April. The mean annual temperature is 18°C, with a maximum in November (26°C) and a minimum in July (10°C). Annual rainfall ranges from 1,000 mm to 1,600 mm per year.
The study area included the dense urban core of Antananarivo, consisting of six districts, known as the urban municipality, and the surrounding suburbs. The suburbs of Antananarivo contain 28 municipalities in the public office of inter municipal cooperation. The suburbs and the six districts constitute the "Great Antananarivo". The urban municipality has 16 public health centres. Five were selected to participate in the study: one in the centre and three located about 5 kilometres from the city centre. Each of the considered suburbs has a health centre, and four centres were selected for study (10-15 kilometres from the city centre; Figure 1). Health centres were selected on the basis of their geographical location and their inclusion in the public health system, making them accessible to the general population (health care costs lower than those charged by private health centres). One of the centres is run by a religious group. It participated in previous study that one observed positive malaria case without stay outside Antananarivo.
Geographical data
The geographical data collected included an Enhanced Thematic Mapper plus (ETM+) image from Landsat 7, with bands 1 to 7. The ETM+ image from Landsat 7, provided by USAID, was acquired in May 2000, at the beginning of the dry season. It is a georeferenced image, with a pre-processing level of 1G and a resolution of 30 m. Two Advanced Synthetic Aperture Image Mode Precision (ASA_IMP_1P) images from Envisat, Digital Elevation Model (DEM) topographic maps and data from the development office of Antananarivo were used to update data concerning the location and extent of rice fields. The images from Envisat were acquired in January 2004, early in the development of the rice crop, and in July 2004, when rice was mature or had already been harvested. These images were multi-look, C-Band, digital images, with vertical/vertical (VV) polarisation and a resolution of 25 m. These images were acquired from the CAT1-2320 Project of the European Space Agency. The DEM was provided by the European Space Agency in 2004, as part of the Epidemio Project, and was obtained from Advanced Synthetic Aperture Radar data from Envisat with a resolution of 50 m. Nine topographic maps from the national institute of geography and hydrography were used for fieldwork, to complete the visual interpretation of images for the location of training and test sites. The scale of these maps was between 1:10,000 and 1:50,000. The geographical coordinates of these sites were determined with a handheld global positioning system (GPS). The development office of Antananarivo provided population data, hydrological networks and data concerning the administrative boundaries of the "fokontany" (neighbourhood, the smallest administrative district).
Climatic data
Climatic data were available from the National Meteorological Service and included minimum, maximum and mean temperatures and rainfall data. These data present normal monthly temperatures and rainfall for a period of 30 years (from 1971 to 2000) and constitute the reference values for climatic data. Average climate predictions, by 10-day period, for 2006 to 2007, were obtained from the African Centre of Meteorological Application for Development.
Location of rice fields
The maximum likelihood method was used for classification purposes, with the Landsat 7 image. This method was adopted because the training sites were well chosen and corresponded to highly homogeneous sites. Classification accuracy was estimated by calculating the Kappa coefficient and the confusion matrix for the training class pixel. Radar images taken on two different dates were used to complete the results. The backscatter coefficient was calculated as follows: σ0 = 10 × log 10 (pixel value). The image from the wet season (January) was compared with the image from the beginning of the dry season (July), to assess the differences between the two periods. Thresholds were identified for the assessment of changes, by matching the values of backscatter coefficients with the type of landscape. Changes were identified on the basis of the combination of colours from the two images. The classification generated with the optical data was improved by results from radar images, which removed ambiguities relating to whether certain areas were covered by rice fields or other types of vegetation.
Field surveys were carried out on two occasions, for environmental interpretation: i) Training and test sites were identified. The co-ordinates of the rice fields were recorded with a handheld global positioning system (GPS). ii) Checking, on the ground, that areas identified as rice fields really were rice fields.
Entomological study
Entomological studies were conducted at six sites (two in the centre, two in neighbouring districts and two in suburb municipality). At each study site, we surveyed temporary, permanent, and semi-permanent breeding sites and mosquito larval populations at the beginning, middle and end of the wet season (November 2006, January and March 2007, respectively). Samples of mosquito larvae were collected from aquatic habitats at each site. The larvae were collected in a white dish [20] if the breeding site was shallow, or with a small nylon gauze net mounted on a circular frame (15 cm in diameter) and attached to a wooden handle (1 m) if the breeding site was deeper. The use of a gauze net ensured that we did not lose the small larvae, and this approach was particularly useful for rice fields [21]. The larvae were transferred to 250 ml plastic bottles containing water from the breeding site, for transfer to the laboratory at the Institute Pasteur of Madagascar, where they were housed in insectariums until emergence.
Seven houses located within 200 m of the nearest large breeding site were randomly selected and sampled for adult mosquitoes by pyrethrum spray catches (WHO 1975). The collections were conducted between 6.00 and 10.00 am on three separate days. Adult anophelines from pyrethrum spray catches and larval surveys were identified morphologically to species [22] and preserved dry in vials containing a desiccant (silica gel). Sibiling species of Anopheles gambiae s.l. were identified to species using rDNA-PCR method and the protocol described by Scott et al. [23]. DNA was extracted from the legs of mosquitoes as described by Collins et al [24]. The heads and thoraces of all female anophelines were screened by Enzyme-linked immunosorbent assays (ELISA) for Plasmodium falciparum circumsporozoïte proteins [25].
Epidemiological study
The epidemiological survey began in November 2006 and continued until December 2007. Clinical examinations were carried out by clinicians, on the basis of the presumed malaria case definition of the National Malaria Control Programme: temperature ≥37°5C in the absence of symptoms associated with other diseases. Thick blood smears were collected on blotting paper discs the size pieces of confetti, from patients with presumed malaria who had given informed consent. Additional information was collected by questionnaire: patient identification, social and economic status (type of dwelling, means of water supply, use of mosquito nets etc.), travel details (number of journeys, date and place) and history of malaria (number of clinical episodes, treatment received, source of treatment) in the previous three months. The travel report provided information on periods away from Antananarivo in the three months before the consultation. Blood samples were collected and questionnaires were completed each month by a local consultant at each site. PCR identified genuine cases of malaria from the blood samples supplied by patients. Real-time PCR based on the Mongold technique was performed in the Rotor-Gene 3000 machine, using the intercalating agent SYBR Green I as the fluorescent marker [26].
A school survey was carried out in March 2007 at the sites studied (centre, immediate vicinity and suburbs). We investigated one school in each area, selecting those nearest the health centres. We examined 200 school children between the ages of six and ten years at each school, with their parents' consent. Rapid diagnostic tests were used to test for malaria in febrile school children (temperature ≥37.5°C). Children testing positive were treated according to national guidelines.
This study was approved by the national ethics committee (No. 005-SANPF, delivered by the ministry on January, 8th 2007).
Determination of risk factors
Antananarivo is a city geomorphologicaly contrasted; one part the hilly zones, often with steep slope liable to erosion and the other part the low zone liable to flooding. The lower area liable to flooding is defined as an area with an altitude not exceeding 1,250 m. This value corresponds approximately to the alert status of the disaster contingency plan in case of flood [27]. Part of these areas consists of rice fields or wetlands. These characteristics led us to consider the altitude, population density and rice fields as factors influencing malaria in our input data. The population density was calculated in relation to habitable areas because of large tracts of farmland.
A geographical information system was constructed for the integration and manipulation of geographical data (altitude, rice field surface area), field data (entomological and epidemiological) and other spatial referenced data (population density, health centres). Kruskal-Wallis tests were used to assess the effects of these factors on the occurrence of confirmed malaria cases. Correlations between the factors studied and malaria cases (with and without stay outside Antananarivo) were studied at 15-day intervals over the year-long study period. Risk analysis was carried out at two levels: at the level of the individual, based on social and economic status, and at district level, based on the presence of confirmed malaria cases in the neighbourhood, with or without reported travel. Chi square tests were used to assess the significance of differences between positive and negative cases of malaria, as a function of the factor considered. McNemar matched pairs tests, taking the neighbourhood as its own control, were performed to assess risk at the level of the neighbourhood.
Monthly averages were calculated for temperature. Total were calculated for monthly rainfall. Temperature and rainfall during the study period were compared with normal values, in Spearman's rank correlation analysis. We then analysed the correlations between temperature and confirmed malaria cases and between rainfall and confirmed malaria cases. Climate data were studied separately from the other factors because of difference of spatial scale. Data on altitude, population density and rice fields are available by neighbourhood while climate data concern the entire study area.