Malaria is a major health burden in the south-eastern part of Bangladesh, especially in the Chittagong Hill Tracts 1. In 13 endemic districts, total malaria prevalence was 3.97% with the rapid diagnosis test (BRAC and ICDDR,B unpublished report). To control malaria in hilly forest areas is a big challenge in many parts of Asia and South America 2.

Disease mapping techniques help to control malaria, especially in high malaria transmission areas. In order to control malaria, application of GIS and GPS was reported in India, Thailand and Madagascar. GIS was used to detect high risk areas and for malaria surveillance in India. In Thailand, GIS has proven to be a significant tool for forecasting malaria epidemics 3. These statistical techniques depend on case event and count data, where geographic locations (x, y coordinates) are represented as points. Recently, Geographic information system (GIS) and remote sensing (RS) technology have enabled scientists and epidemiologists to study vector borne disease, mapping, to explore environmental relations, and to understand spatial and temporal distribution 4.

Mosquito vectors breed near rivers, pools, agricultural lands and forests; they depend on the existence of water and river flow 5. Malaria risk maps were developed in Sri Lanka based on the household distance from streams and rivers that are known as potential vector breeding sites. People living within 750 meters of streams were identified as high-risk groups 6. In southern Ontario, Canada, SaTScan was used to investigate spatial clusters of giardiasis. SaTScan successfully identified approximate locations and spatial clusters 7. Recently GIS, GPS and spatial statistics have also been used in malaria research and control in sub-Saharan African countries. Malaria maps have long been considered an important tool to control malaria. Generally this type of map is used to predict risk 8.

In order to detect aggregation of disease cases, cluster detection is very important. Until now, there is no published report on the micro-geographic distribution of malaria incidence or prevalence in Bangladesh. This study uses GIS, GPS, and SaTScan tools to detect geographic clusters of malaria, to assess association between malaria cases and geographic risk factors.

The average malaria prevalence in Khagrachari district was 15.47% (n = 750). [District is divided into several thanas. Thanas are again divided into unions. Again unions are divided into mauzas the lowest administrative unit in Bangladesh]. Table 1 lists the individual prevalence of malaria for the thanas in Khagrachari. The highest prevalence was found in Dighinala (22%) and lowest prevalence was found in Panchari (5.72%). SaTScan was used to detect malaria clusters in Khagrachari (n = 750). Table 2 presents there were five clusters in total. Among them, one was most likely clustered and four were secondary clusters (figure 2). The most likely malaria case cluster area is in Dighinala thana (RR = 3.381; p = 0.0002). In most of the secondary clusters, the relative risk ratio was high, but none were statistically significant.

Table 3 presents the distribution of environmental and socioeconomic characteristics by malaria positive, negative and total household. Table 4 lists the adjusted odds ratios for malaria based on the distance from each participating household to various locations, including forest, and water sources. A binary variable with 3 km as the cut-off point was created because most mosquitoes cannot fly more than 3 km from their breeding places 14. Results reveal that people living in houses less than 3 km from the forest were at much higher risk for malaria than people living further away. The strong effect of distance from house to forest remained significant after controlling for a number of potentially confounding variables. People living in houses more than 3 km from the water sources were at much higher risk of malaria than people living within 3 km. After controlling for confounding by other variables, the effect remained significant for the distance from house to water sources. Children had a significantly higher risk for malaria than adults. Independent risk factors for malaria infection found significant in multivariate analysis for the entire surveyed area were proximity to forest, precarious house (house made with temporary construction materials like straw, polythene, bamboo etc), and bed net with <3 in house. Only proximity to water came out as an independent and significant protective factor for malaria infection.

Understanding the spatial distribution of malaria, identifying geographic risk factors and the population at risk are important steps toward effective control of malaria. The data presented in this paper are the first step to understanding malaria in south-eastern Bangladesh from a micro-geographic perspective. The study result suggests that there are 'malaria hot-spots' in the study area. The government of Bangladesh and non-governmental organizations involved in malaria control should consider these results when planning malaria control measures. More efforts should be focused on people living in remote areas. BRAC the largest NGO in Bangladesh is currently implementing malaria control programme in Bangladesh. They have deployed health workers in grass root level. They have supplied ITN (Insecticide treated net) in every household. They should consider this result to control malaria and investigate the reasons of cluster.

They should be provided with more bed nets and anti-malarial drugs. One way to accomplish this would be to redistribute health workers to remote regions. The government should also consider allocating additional resources so that more operational research can be carried out in the micro-geographic area. In particular, malaria maps should be updated on a regular basis as new data become available.

The authors declare that they have no competing interests.

UH designed, analysed and prepared the manuscript, AH collected data, MH analysed the data, RH and SMA were responsible for study design and conduct the study.