Spatial analysis and mapping of malaria risk in Malawi using pointreferenced prevalence of infection data
 Lawrence N Kazembe^{1, 2}Email author,
 Immo Kleinschmidt^{2},
 Timothy H Holtz^{3} and
 Brian L Sharp^{2}
DOI: 10.1186/1476072X541
© Kazembe et al; licensee BioMed Central Ltd. 2006
Received: 21 July 2006
Accepted: 20 September 2006
Published: 20 September 2006
Abstract
Background
Current malaria control initiatives aim at reducing malaria burden by half by the year 2010. Effective control requires evidencebased utilisation of resources. Characterizing spatial patterns of risk, through maps, is an important tool to guide control programmes. To this end an analysis was carried out to predict and map malaria risk in Malawi using empirical data with the aim of identifying areas where greatest effort should be focussed.
Methods
Pointreferenced prevalence of infection data for children aged 1–10 years were collected from published and grey literature and georeferenced. The modelbased geostatistical methods were applied to analyze and predict malaria risk in areas where data were not observed. Topographical and climatic covariates were added in the model for risk assessment and improved prediction. A Bayesian approach was used for model fitting and prediction.
Results
Bivariate models showed a significant association of malaria risk with elevation, annual maximum temperature, rainfall and potential evapotranspiration (PET). However in the prediction model, the spatial distribution of malaria risk was associated with elevation, and marginally with maximum temperature and PET. The resulting map broadly agreed with expert opinion about the variation of risk in the country, and further showed marked variation even at local level. High risk areas were in the lowlying lake shore regions, while low risk was along the highlands in the country.
Conclusion
The map provided an initial description of the geographic variation of malaria risk in Malawi, and might help in the choice and design of interventions, which is crucial for reducing the burden of malaria in Malawi.
Background
The burden of malaria in Malawi, like other parts of subSaharan Africa, is a major public concern [1, 2]. Recent estimates report that malaria contributes about 35% of all illnesses in children under five years of age in the country [2, 3]. Current malaria control initiatives aim at halving the burden by the year 2010 through integrated control programmes encompassing vector control (via insecticidetreated nets and indoor residual spraying), intermittent preventive treatment for pregnant women and prompt and effective case management [2, 4]. Effective control requires evidencebased utilisation of resources. The type and degree of interventions need to be based on epidemiological patterns of malaria risk. Malaria risk varies in space and time [5]. It is important to describe the spatiotemporal variability of malaria risk to guide control programmes [6–8].
In the last decade, maps have been produced at different geographical scales in subSaharan Africa [9–13], following the Mapping Malaria Risk in Africa (MARA) project [14], with the aim of identifying areas where greatest control effort should be focussed. In this analysis, the objective was to predict and map malaria risk in Malawi using pointreferenced prevalence data. Existing risk maps are based on a theoretical climatic model [15] or expert opinion [2], but these have important limitations as they fail to provide insight into the transmission of malaria in Malawi. It is important to characterise malaria risk based on empirical evidence using a malariaspecific indicator, in this case, malaria prevalence of infection in children, and assess its relationship with environmental risk factors.
Prediction of risk based on pointreferenced data presents some challenges when the data are sparsely distributed. Such data often exhibit autocorrelation, such that locations close to each other have similar risk. Models should allow for spatial correlation, failing which, the significance of risk factors is overstated [16, 17]. Analyses of pointreferenced data have been carried out using geostatistical models [18], for optimal prediction. Recently, a modelbased geostatistical (MBG) approach has been applied [19]. The approach permits simultaneous modelling of related issues such as risk assessment, spatial dependence, prediction and quantification of uncertainty [20, 21]. Accurate prediction of risk can further be achieved by including environmental factors likely to influence malaria transmission [9].
Several studies have shown that malaria infection is influenced by environmental factors such as temperature, rainfall, humidity and elevation. Specifically, temperature and rainfall act as limiting factors on the development of Anopheles mosquitoes which are the intermediate hosts in the transmission of malaria parasites [22]. In tropical settings, temperature and rainfall conditions are nearly always favourable for transmission. Humidity is also suitable for transmission because it affects the survival rate of mosquitoes. Similarly, elevation above sea level (asl) is known to define the ecology of malaria transmission through temperature [23, 24]. At certain altitudes malaria transmission does not occur because of extreme temperatures that inhibit the mosquito and parasite lifecycle. For small countries like Malawi, topography remains a single most important factor that defines largescale differences in malaria risk because climatic variables change little over the limited range of latitude.
In this study, we applied the modelbased geostatistical (MBG) approach to analyse and predict malaria risk in Malawi, using pointreferenced prevalence data realised from previous mass malariometric surveys carried out in the country. We adjusted for environmental covariates to accurately predict malaria risk.
Methods
Data sources
Observed mean, minimum and maximum prevalence ratios by districts
District^{†}  Mean prevalence  Minimum prevalence  Maximum prevalence  Data points 

Blantyre  0.44  0.08  0.91  31 
Chikwawa  0.59  0.41  0.92  4 
Chiradzulu  0.13  0.05  0.25  2 
Chitipa  0.26  0.09  0.37  1 
Dedza  0.50  0.20  0.68  1 
Dowa  0.87  0.11  0.93  1 
Karonga  0.35  0.24  0.46  2 
Kasungu  0.53  0.20  0.84  1 
Lilongwe  0.59  0.43  0.75  2 
Machinga  0.36  0.24  0.59  3 
Mangochi  0.52  0.16  0.70  1 
Mchinji  0.50  0.04  0.60  1 
Mulanje  0.48  0.25  0.55  2 
Mwanza  0.25  0.08  0.43  3 
Mzimba  0.24  0.18  0.30  2 
Nkhatabay  0.46  0.22  0.74  1 
Nkhotakota  0.44  0.19  0.63  2 
Ntcheu  0.43  0.07  0.71  1 
Rumphi  0.48  0.41  0.51  2 
Salima  0.87  0.78  0.93  3 
Thyolo  0.63  0.39  0.67  2 
Zomba  0.25  0.05  0.57  5 
At each data location we also extracted the values of the following covariates: (i) mean annual maximum temperature, (ii) mean annual rainfall, (iii) mean annual potential evapotranspiration (PET), and (iv) elevation, from the Spatial Characterization Tool (SCT) [26]. The SCT is a suite of geographically referenced environmental information including those used here, for most subSaharan countries, and provides a framework to compile, querry and access the data. The climatic variables are interpolated gridded surfaces generated from longterm monthly means of historical weather station point data, from the period 1920–1980, across Africa. The elevation data were interpolated from digital elevation models of Africa.
Statistical analysis
Variograms based on the empirical logit of observed prevalence were computed to explore the spatial correlation in the data. Bivariate nonspatial logistic regression models (Equation 1), were fitted for each variable of the following variables: elevation in metres asl, mean maximum temperature in degrees Celsius, rainfall in millimetres per annum and PET in millimetres per annum. Because of possible nonlinear relationships, each variable was converted into a categorical variable for further analysis, with cutoffs based on the natural break points, guided by exploratory scatterplots. The highest category of each variable was considered the reference category, and variables significant at < 0.2 were used to fit bivariate spatial logistic models (Equation 2).
Assume that the number found positive for malaria parasitemia at location i is Y_{ i }out of N_{ i }examined, then Y_{ i }is a binomial random variable, Y_{ i }~ Bin (N_{ i }, p_{ i }), where p_{ i }is the proportion infected at each location. The bivariate ordinary logistic model is given by
where β_{0} is the intercept, v_{ i }is a covariate, β_{1} is the corresponding regression parameter. The spatial correlation is modelled by inclusion of a random effect S_{ i }, i.e.
The spatial component, S_{ i }, is assumed to be a zero mean Gaussian process with variance σ^{2} and correlation function ρ (d_{ ij }, φ). The range φ measures the rate of decay of spatial autocorrelation, and d_{ ij }= x_{ i } x_{ j }measures the Euclidean distance between locations x_{ i }and x_{ j }. Under the spatial model, the response Y_{ i }given the random effects S_{ i }and covariates v_{ i }, is assumed to be conditionally independent and distributed as a binomial outcome [19, 21].
All candidate variables identified in the bivariate spatial models were included in the multiple spatial logistic model (Equation 2) for prediction. However, in the multiple model, parameter β_{1} becomes a vector of regression coefficients corresponding to the vector of covariates v_{ i }.
For mapping, we predicted prevalence of infection at 4000 grid locations covering the entire country. Covariates identified as significant in the bivariate spatial model were used in prediction model. The predicted values were posterior medians realised as part of the MCMC simulations from the posterior predictive distribution [21],
P (Y_{0}Y) ∝ P (Y_{0}β, S_{0}) P (S_{0}S, σ^{2}, φ) P (β, S, σ^{2}, φY), (3)
where Y_{0} are predicted values at new locations given the observed data (Y), and S_{0} is the prediction of some functional of spatial process S. Similarly, approximate standard errors were obtained by dividing the 95% credible interval by 4. The estimates, Y_{0}, were then exported to ArcGIS (Version 8.3; ESRI, 2004) for cartographic representation. We overlayed the final predicted prevalence map on a population density map to calculate the population at risk for different endemicity categories.
Results
Spatial analysis of malaria infection
Variables and regression results from the bivariate nonspatial, Bayesian bivariate and multiple spatial logistic models.
Variable  Bivariate nonspatial model  Bivariate spatial model  Multiple spatial model  

OR^{†}  (95% CI^{‡})  OR  (95% CI)  OR  (95% CI)  
Elevation  
β_{1}(elev1: < 650 m)  1.33  (1.21, 1.46)  1.98  (1.29, 3.05)  1.42  (1.13, 1.99) 
β_{2} (elev2: 650–1110 m)  0.82  (0.75, 0.91)  2.08  (1.53, 2.81)  1.89  (1.38, 2.59) 
Reference (elev3 : > 1110 m)  1.00  1.00  1.00  
Max. Temperature  
β_{3} (Tmax1: < 27°C)  0.77  (0.71, 0.83)  1.88  (1.52, 2.34)  1.68  (1.45, 2.73) 
β_{4} (Tmax2: 27–32°C)  0.63  (0.59, 0.68)  0.85  (0.69, 1.02)  0.88  (0.43, 1.08) 
Reference (Tmax3: > 32°C)  1.00  1.00  1.00  
Rainfall  
β_{5} (Rain1: < 880 mm)  1.15  (1.00, 1.31)  0.81  (0.64, 1.14)  
β_{6} (Rain2: 880–1180 mm)  1.20  (1.11, 1.29)  0.76  (0.54, 2.07)  
Reference (Rain3: > 1180 mm)  1.00  1.00  
PET ^{¶}  
β_{7} (PET1: < 1370 mm)  0.72  (0.61, 0.84)  0.49  (0.33, 1.22)  0.69  (0.21, 2.34) 
β_{8} (PET2: 1370–1510 mm)  0.61  (0.58, 0.65)  0.38  (0.17, 0.89)  0.41  (0.14, 0.97) 
Reference (PET3: > 1510 mm)  1.00  1.00  1.00  
Range (φ)  0.54  (0.23, 0.96)  
Variance (σ^{2})  13.74  (8.80,20.16) 
Results from the bivariate spatial logistic models are given in Table 2. Overall, elevation, mean maximum temperature and PET were associated with malaria prevalence after adjusting for spatial correlation. At elevation of < 650 m asl, relative to elevation ≥ 1110 m, the risk of malaria was higher (OR: 1.98, 95% CI: 1.29–3.05). At elevation between 650 m and 1100 m asl, relative to elevation ≥ 1110 m, there was increased malaria risk (OR: 2.08, 95% CI: 1.53–2.81). The risk of malaria was likely to be more at mean annual maximum temperature of < 27°C relative to temperature > 32°C (OR: 1.88, 95% CI: 1.52–2.34). At temperatures between 27–32°C, relative to temperature > 32°C, the risk was less (OR: 0.85, 95% CI: 0.69–1.02). Rainfall of less than 800 mm and between 800–1180 mm per annum relative to rainfall of more than 1180 mm was not associated with malaria prevalence (OR: 0.81, 95% CI: 0.64–1.14 and OR: 0.76, 95% CI: 0.54–2.07 respectively). Similarly, PET of less than 1370 mm compared to PET of > 1510 mm was not associated with malaria risk (OR: 0.49, 95% CI: 0.33–1.22). However, PET between 1370–1510 mm was significantly associated with lower prevalence of infection than PET levels over 1510 mm (OR: 0.38, 95% CI: 0.17–0.89).
Maps and population at risk
Predicted percentage of atrisk population of children aged 1–10 years, by district and risk category.
District  0–15%  15–30%  30–45%  45–60%  60–100%  Population^{† §} 

Likoma  0  0  100  0  0  1,173 
Chitipa  30  40  20  10  0  21,098 
Karonga  0  20  10  10  60  31,267 
Rumphi  10  60  20  5  5  20,531 
Mzimba  50  20  15  15  0  97,263 
Nkhatabay  20  10  20  20  30  24,380 
Mzuzu City  0  70  30  0  0  12,627 
Nkhotakota  0  15  15  20  50  38,184 
Ntchisi  0  0  0  5  95  27,166 
Dowa  0  0  0  5  95  67,357 
Salima  0  0  0  5  95  41,126 
Kasungu  0  10  30  30  30  78,357 
Mchinji  0  0  20  20  60  53,843 
Lilongwe rural  20  20  15  15  30  214,472 
Lilongwe City  0  25  25  25  25  63,557 
Dedza  15  20  60  5  0  81,928 
Ntcheu  40  40  15  5  0  58,214 
Mangochi  40  25  25  5  5  99,604 
Balaka  30  30  30  5  5  39,273 
Machinga  25  20  25  10  20  60,757 
Chiradzulu  30  30  20  20  0  34,236 
Phalombe  0  0  15  15  75  35,682 
Mulanje  5  15  30  20  30  61,682 
Blantyre rural  20  40  30  10  0  112,095 
Blantyre city  0  85  15  0  0  67,009 
Zomba  0  10  70  10  10  81,298 
Zomba city  0  10  10  15  65  9,143 
Mwanza  40  25  15  20  0  22,341 
Chikwawa  15  15  20  20  30  57,355 
Thyolo  0  5  35  10  50  68,442 
Nsanje  10  30  60  0  0  32,248 
Total  14  22  25  12  27  2,732,434 
Discussion
In this analysis, a map showing the spatial variation of malaria risk in children aged 1–10 years in Malawi was produced using pointreferenced prevalence of infection data. The map is a first attempt towards the empirical description of malaria risk in Malawi, and differs from the climatic suitability model map [15] or expert opinion [2]. In contrast to expert's broad classification of malaria risk, this map (Figure 4) shows that malaria risk varies widely in the country even within districts. Nevertheless, in agreement with expert opinion [2, 31, 33], our map identifies highest risk along the lakeshore, Shire river valley and central plain areas, and lowest in the highland areas of Rumphi, Mzimba, Chitipa and the Kirk range.
The map will be useful for focussed malaria control activities. Currently, for example, there are plans to scaleup the coverage of insecticide treated nets (ITN) as part of the strategy to reduce the burden of malaria [34]. It is important, therefore, to identify and carefully plan prior to scalingup the ITN program. Among other things, this map can advise areas to be targeted, for instance all areas at highest risk. Considering the atrisk population estimates (Table 3), and the recent cost estimates of net delivery [35], the cost of scalingup can be calculated. Since estimates are available up to the local level, it is possible to plan up to that scale. Furthermore, the estimates may provide baseline information against which the success of an intervention programme could be assessed, through followup surveys in the future. A case in point is the Innovative Vector Control Consortium which plans to extend insecticide residual spraying, currently implemented in Mozambique, to southern Malawi [36]. The effectiveness of their tools can be compared against this map. Sentinel points for spraying can be selected using this product. Another control initiative that may find the map useful is the President's Malaria Initiative (PMI), which is a U.S. government initiative designed to cut malaria deaths in half in target countries in subSaharan Africa, including Malawi [37].
The analysis shows the importance of integrating risk factors in the spatial prediction of malaria risk. Generally, the results showed that elevation plays an important role in defining malaria risk in Malawi, a fact recognized by experts in the country [31, 33], and has been confirmed in several other studies carried out in the continent [23, 24, 38]. The fact that malaria prevalence was only marginally associated with climatic variables can be explained by the limited range of latitudes within small countries like Malawi, and hence the minimal variation in climate. The results, therefore, do not contradict the importance of climatic variables in predicting malaria risk in general [15, 22].
The modelling was based on spatial statistical methods. These offer an attractive and better alternative to the GIS mapping approach which incorporates the spatial correlation inherent in the data [9, 17, 19, 20]. Furthermore, the method allows errors of estimations to be quantified making it possible to assess the precision of the map and significance of covariates (Figure 5). In addition, adjusting for spatial correlation avoids overstating the significance of covariates (Table 2) [16, 17]. Spatial correlation may arise because of omitted or unobserved covariates, and incorporating the spatial random effect in the model further allows these to be accounted for [39].
The results presented here have some limitations. First, the data points used for analysis were available at 73 sites, but were sparsely distributed in the northern and central region (Figure 1). This has potential to bias the estimates. However, the inclusion of risk factors at predicted sites may have reduced this bias. Second, the data used here span a period of 20 years, and the risk may have not been constant with time. For example, increase in population density, urbanization, agricultural and socioeconomic development have brought change over this period, which may have affected the pattern of malaria risk. Be that as it may, the high endemicity and the absence of sustainable and effective interventions in the country suggest that malaria risk has changed little during this time. Accordingly, such data can still be used to generate reliable and informative malaria risk estimates. Another limitation is that the age range of 1–10 years may not be ideal as the level of immunity may not be homogenous in this age group. Gemperli et al [11] provide a method of converting a set of heterogeneous agespecific prevalence onto a common scale of transmission intensity for prediction and mapping purposes.
Malaria transmission is very complex and prediction based on few covariates may compromise the accuracy of the map. Malaria transmission drivers go beyond topographical and climatic variables, and may include sociodemographic factors and include urbanization, population growth and local variation in vector habitat [40–42]. In practice, a wide range of environmental covariates have been used [12]. When as many independent covariates as possible are added to the model, the accuracy of the predicted map may be maximized and it may be worthwhile to explore how the map would change when relevant covariates become available. Updating malaria maps should therefore be carried out on a regular basis as new data become accessible.
Despite these limitations, the map of predicted risk of infection provides a much needed characterization of geographical variation of malaria risk in Malawi. It is the only one that provides estimates at all locations, and therefore offers much needed evidencebased stratification of malaria risk. Through the map it is possible to determine which areas require greatest control effort. More important, it provides a baseline against which the effectiveness of current control efforts can be assessed.
Abbreviations
 CI:

Confidence Interval
 Credible :

Interval
 GIS:

Geographical Information Systems
 ITN :

nsecticide treated nets
 MARA:

Mapping Malaria Risk in Africa
 MBG:

Model based geostatistics
 MCMC:

Markov Chain Monte Carlo
 OR:

Odds Ratio
 SCT:

Spatial Characterisation Tool
Declarations
Acknowledgements
LNK would like to acknowledge the research training grant received from WHO/TDR and suppor from Medical Research Council, Durban, South Africa during his PhD training.
Authors’ Affiliations
References
 WHO/UNICEF: Africa malaria report. 2003, World Health Organization/UNICEF Report series: WHO/CDS/MAL/2003, 1093Google Scholar
 Government of Malawi: Malaria policy. 2002, Lilongwe: National Malaria Control Programme Community Health Sciences Unit Government of MalawiGoogle Scholar
 Holtz TH, Marum LH, Mkandala C, Chizani N, Roberts JM, Macheso A, Parise ME, Kachur SP: Insecticide treated bednet use, anaemia, and malaria parasitaemia in Blantyre district, Malawi. Trop Med Int Health. 2002, 7: 220230. 10.1046/j.13653156.2002.00846.x.PubMedView ArticleGoogle Scholar
 WHO: Roll Back Malaria. 1998, Geneva: World Health OrganizationGoogle Scholar
 Kleinschmidt I, Sharp B, Mueller I, Vounatsou P: Rise in malaria incidence rates in South Africa: small area spatial analysis of variation in time trends. Am J Epidemiol. 2002, 155: 257264. 10.1093/aje/155.3.257.PubMedView ArticleGoogle Scholar
 Carter R, Mendis KN, Roberts D: Spatial targeting of interventions against malaria. Bull WHO. 2000, 78: 14011411.PubMedPubMed CentralGoogle Scholar
 Le Sueur D, Binka F, Lengeler C, De Savigny D, Snow B, Teuscher T, Toure Y: An atlas of malaria in Africa. Africa Health. 1997, 19: 2324.PubMedGoogle Scholar
 Snow RW, Marsh K, leSueur D: The need for maps of transmission intensity to guide malaria control in Africa. Parasitol Today. 1996, 12: 455457. 10.1016/S01694758(96)30032X.View ArticleGoogle Scholar
 Kleinschmidt I, Bagayoko M, Clarke GP, Craig M, Le Sueur D: A spatial statistical approach to malaria mapping. Int J Epidemiol. 2000, 29: 355361. 10.1093/ije/29.2.355.PubMedView ArticleGoogle Scholar
 Kleinschmidt I, Omumbo J, Briet O, van de Giesen N, Sogoba N, Mensah NK, Windmeijer P, Moussa M, Teuscher T: An empirical malaria distribution map for West Africa. Trop Med Int Health. 2001, 6: 779786. 10.1046/j.13653156.2001.00790.x.PubMedView ArticleGoogle Scholar
 Gemperli A, Vounatsou P, Sogoba N, Smith T: Malaria mapping using transmission models: applications to survey data from Mali. Am J Epidemiol. 2006, 163: 289297. 10.1093/aje/kwj026.PubMedView ArticleGoogle Scholar
 Omumbo JA, Hay SI, Snow RW, Tatem AJ, Rogers DJ: Modelling malaria risk in East Africa at highspatial resolution. Trop Med Int Health. 2005, 10: 557566. 10.1111/j.13653156.2005.01424.x.PubMedPubMed CentralView ArticleGoogle Scholar
 Snow RW, Gouws E, Omumbo J, Rapuoda B, Craig MH, Tanser FC, le Sueur D, Ouma J: Models to predict the intensity of Plasmodium falciparum transmission: applications to the burden of disease in Kenya. Trans R Soc Trop Med Hyg. 1998, 92: 601606. 10.1016/S00359203(98)907817.PubMedView ArticleGoogle Scholar
 MARA/ARMA: Towards an atlas of malaria in Africa First Technical Report Durban: MARA. 1998, [http://www.mara.org.za.]Google Scholar
 Craig MH, Snow RW, le Sueur D: A climatebased distribution model of malaria transmission in subSaharan Africa. Parasitol Today. 1999, 15: 10511. 10.1016/S01694758(99)013964.PubMedView ArticleGoogle Scholar
 Thomson MC, Connor SJ, D'Alessandro U, Rowlingson B, Diggle P, Cresswell M, Greenwood B: Predicting malaria infection in Gambian children from satellite data and bed net use surveys: the importance of spatial correlation in the interpretation of results. Am J Trop Hyg Med. 1999, 61: 28.Google Scholar
 Boyd HA, Flanders WD, Addiss DG, Waller LA: A residual spatial correlation between geographically referenced observations: a Bayesian hierarchical modeling approach. Epidemiol. 2005, 16: 532541. 10.1097/01.ede.0000164558.73773.9c.View ArticleGoogle Scholar
 Cressie NAC: Statistics for spatial data (rev. ed.). 1993, John Wiley and Sons: New YorkGoogle Scholar
 Diggle PJ, Tawn JA, Moyeed RA: Modelbased geostatistics (with discussion). Appl Statist. 1998, 47: 299350.Google Scholar
 Diggle PJ, Moyeed R, Rowlingson B, Thomson MC: Childhood malaria in the Gambia: a case study in modelbased geostatistics. Appl Statist. 2002, 51: 493506.Google Scholar
 Diggle PJ, Ribeiro PJ, Christensten OF: An introduction to modelbased geostatistics. Spatial statistics and computational methods. Edited by: Möller, J. 2003, New York: SpringerVerlag, 4386.View ArticleGoogle Scholar
 Cox J, Craig MH, LeSueur D, Sharp BL: Mapping malaria risk in the highlands of Africa. 1999, MARA/HIMAL Technical Report: London/DurbanGoogle Scholar
 Bødker R, Akida J, Shayo D, Kisinza W, Msangeni HA, Pedersen EM, Lindsay SW: Relationship between altitude and intensity of malaria transmission in the Usambara mountains, Tanzania. J Med Entomol. 2003, 40: 706717.PubMedView ArticleGoogle Scholar
 Maxwell CA, Chambo W, Mwaimu M, Magogo F, Carneiro IA, Curtis CF: Variation of malaria transmission and morbidity with altitude in Tanzania and with introduction of alphacypermethrin treated nets. Malaria J. 2003, 2: 2810.1186/14752875228.View ArticleGoogle Scholar
 Omumbo J, Ouma J, Rapuoda B, Craig MH, le Sueur D, Snow RW: Mapping malaria transmission intensity using geographical information systems: an example from Kenya. Ann Trop Med Parasitol. 1998, 92: 721. 10.1080/00034989860120.PubMedView ArticleGoogle Scholar
 Corbett JD, Collis SN, Bush BR, Muchugu EI, O'Brien RF, Burton RA, Stone CM, Martinez RE, Jeske RQ: Texas Almanac Characterization Tool A resource base for characterizing the agricultural natural and human environments of Texas. 1999, Texas Agricultural Experiment Station Texas A&M University System, Blackland Research Center Report No. 9905, December 1999, documentation and CDROMGoogle Scholar
 Gelman A, Carlin BP, Stern H, Rubin D: Bayesian data analysis. 1995, Chapman and Hall: LondonGoogle Scholar
 Christensten OF, Ribeiro PJ: GeoRglm: a package for generalized linear spatial models. RNEWS. 2002, 2: 2628.Google Scholar
 R Development Core Team: R: A language and environment for statistical computing. 2004, R Foundation for Statistical Computing Vienna Austria, [http://www.rproject.org.]Google Scholar
 Anselin L: Local indicators of spatial association. Geograph Analys. 27: 93115.
 Coombes L, Ngwale M, Chavasse D: BITNET: Blantyre insecticide treated net project Baseline KAP study report. 1998, Blantyre: Population Services InternationalGoogle Scholar
 National Statistical Office: Malawi population projections. 2003, Zomba: NSOGoogle Scholar
 Tambala P, Macheso A, Ziba C: Malaria vector assessment. 1992, Lilongwe: Ministry of Health Government of MalawiGoogle Scholar
 The Global Fund to Fight AIDS, Tuberculosis and Malaria. [http://www.theglobalfund.org]
 Stevens W, Wiseman V, Ortiz J, Chavasse D: The costs and effects of a nationwide insecticidetreated net programme: the case of Malawi. Malaria J. 2005, 4: 2210.1186/14752875422.View ArticleGoogle Scholar
 Kuehn BM: Group to revamp vectorcontrol methods. JAMA. 295: 22382239. 10.1001/jama.295.19.2238.
 Rowe A: Options for evaluating the impact of control efforts on mortality in African countries with high malaria burden: an analysis of U.S. President's Malaria Initiative. Malaria Branch Centre for Disease Control and Prevention. Atlanta Georgia USA, (Accessed on June 2, 2006)., [http://www.cdc.gov/malaria/cdcactivities/options_for_evaluating.htm]
 BruceChwatt LJ: Lessons learned from applied filed research activities in Africa during the malaria eradication era. Bull WHO. 1984, 62S: 1929.Google Scholar
 Ver Hoef JM, Cressie N, Fisher RN, Case TJ: Uncertainty and spatial linear models for ecological data. Spatial Uncertainty for Ecology: Implications for Remote Sensing and GIS Applications. Edited by: Hunsaker C, Goodchild M, Friedl M, Case T. 2001, New York: SpringerVerlag, 214237.View ArticleGoogle Scholar
 Omumbo JA, Guerra CA, Hay SI, Snow RW: The influence of urbanisation on measures of Plasmodium falciparum infection prevalence in East Africa. Acta Tropica. 2005, 93: 1121. 10.1016/j.actatropica.2004.08.010.PubMedPubMed CentralView ArticleGoogle Scholar
 Hay SI, Noor AM, Nelson A, Tatem AJ: The accuracy of human population maps for public health application. Trop Med Int Health. 2005, 10: 10731086. 10.1111/j.13653156.2005.01487.x.PubMedPubMed CentralView ArticleGoogle Scholar
 Hay SI, Guerra CA, Tatem AJ, Atkinson PM, Snow RW: Urbanization, malaria transmission and disease burden in Africa. Nature Rev Microbiol. 2005, 3: 8190. 10.1038/nrmicro1069.View ArticleGoogle Scholar
Copyright
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.