Developing a risk map of malaria transmission for East Africa
Background: The distribution of malaria in sub-Saharan Africa is determined largely by climatic influences on the development and survival of P. falciparum and its Anopheline vectors. This inter-relationship has been exploited in developing a limited number of predictive maps of malaria's distribution but these climate maps have limitations. Climate alone does not fully describe the complex dynamics of transmission and, in particular, human influences such as urbanization and the use of widespread anti-malarial interventions. The improved accuracy and validation of solely climatedriven maps relies on the availability of robust malariometric training data. To date, such data have been scarce. This study redresses several deficiencies of existing malaria maps for Africa through the collation of an extensive database of empirical P. falciparum prevalence data, the investigation of the relationship between prevalence and a widely-used climate-driven map, an assessment of the influence of urbanization on prevalence and finally, through the use of empirical training data to develop an improved malaria map for Kenya, Tanzania and Uganda. Methods: An extensive published and grey-literature search was conducted between 1996 and 2004 and identified 2003 P. falciparum prevalence surveys conducted among childhood populations across East Africa between 1927 and 2003. Stringent criteria were applied to select the best sample data; only randomly sampled community-based surveys, surveys with samples >=50 children, surveys conducted between 1980-2004 and children aged 0-14 years, and surveys which were spatially and temporally unique. The selected data were used to investigate the association between P. falciparum prevalence and a fuzzy logic climatic suitability (PCS) map of malaria transmission, the effect of urbanization on prevalence and to train Fourier-processed multi-temporal climate surrogate data derived from meteorological satellites in order to predict prevalence for un-sampled areas. Using discriminant analysis, the top ten climatic predictor variables that distinguished best between 4 categories of malaria prevalence (0-<5, 5-<25%, 25-<75% and >=75%) were selected and these used to develop a predictive transmission map.