Peter W Gething; Anand P Patil; David L Smith; Carlos A Guerra; Iqbal RF Elyazar; Geoffrey L Johnston; Andrew J Tatem; Simon I Hay

“ The most commonly measured metric of malaria transmission is the parasite rate: the proportion of individuals infected at a given point in time. In 2009, the Malaria Atlas Project (MAP) assembled all available data from Plasmodium falciparumparasite rate (PfPR) surveys, and used model-based geostatistics (MBG) to generate a global map of estimated PfPR for the year 2007 [3] . That map provided new insights into global patterns of malaria endemicity and, through the careful handling of uncertainty, a framework for assessing those areas where knowledge of endemicity is inadequate. ”

“ The global ubiquity of PfPR surveys means that they are the only feasible data source for large-scale malaria mapping [1] , [2] . Other metrics of malaria transmission, however, have distinct and crucial roles in informing control decisions. The basic reproductive number for malaria, PfR0, quantifies the potential for the disease to spread within a naive population [6] , [7] . The same metric for scenarios moderated by malaria control has been termed PfRc [8] . ”

“ Whilst the spatial variation in risk mandates the generation of robust maps that can guide disease control, the dynamic nature of malaria endemicity means that such maps must be continually updated if they are to remain relevant. The ongoing scale-up of major malaria control initiatives represents the largest potential perturbation to local and regional malaria transmission systems for many decades, and heightens the requirement for regular assessments of risk. ”

“ These metrics underpin mathematical models of transmission that are central to contemporary questions in malaria control [9] : identifying optimal intervention suites and coverage levels, predicting timelines of declining endemicity, and assessing the regional feasibility of elimination [2] , [10] - [11] . If these values exceed one, infection prevalence increases to a steady state, and if less than one, prevalence declines. Thus, if sustained disease control reduces transmission intensity by a factor that exceeds PfR0, the parasite will eventually be eliminated. ”

“ The remaining aspect of uncertainty arises from spatial or temporal variation in transmission intensity that occurs over short spatial or temporal scales. The early cartography of malaria risk aimed to classify wide areas into risk strata [65] , and this has led to a tendency to think of endemicity as a smoothly varying phenomenon. In reality, however, an area considered to belong to a particular endemicity 'class' will likely display a huge amount of variation, with parasite rates sampled at nearby villages often differing dramatically regardless of sample size. ”

“ However, whilst these high stable endemic areas of Africa present the most serious challenges to control, it is vital to avoid the simplistic notion that this level of risk characterizes Africa as a whole when, in reality, the continent displays highly diverse endemicity within its limits of transmission. Some 203 million people live in regions at intermediate stable risk (between 5% and 40% PfPR2-10) , where the interruption of malaria transmission has been proposed as a realistic objective if universal ITN coverage can be achieved [23] . ”

“ Second, a malaria transmission model was utilized to describe the relationship between PfPR and PfRc [8] , [11] , [13] . The transmission model assumes that infections by different parasite types can accumulate in a single human host (super-infection) , and that they clear independently. The model also assumes that exposure risk is distributed unevenly in the population (heterogeneous biting) but that this heterogeneity can be represented by a simple statistical distribution model (a one-parameter family of Gammadistributions) . ”

“ To map endemicity within the boundaries of stable transmission, the global assembly of georeferenced PfPR surveys held by MAP was first updated. Data assembly has been a continuous activity of MAP since 2005 [30] and the volume of malariometric data now available to inform estimates of risk worldwide has grown markedly in the last two years, driven in part by national sample surveys that include malaria biomarkers. ”

“ As in Africa, the epidemiological feasibility of significant reductions in transmission in these lowest endemicity regions is established, but the technical, logistical and economic challenges associated with scaling up intervention coverage across more than a billion people at risk are self-evident. ”

“ The distinction between areas of high and low transmission intensity is even more pronounced for PfRc than for PfEIR. The significant majority of mapped pixels (82%) have a predicted median value of less than two. Of the remaining 18% of higher value pixels, nearly all are in Africa. Around 10% exceed a PfRc of 10, and a tiny handful (< 1%) exceed 100. As with PfEIR, these median values represent only the central tendency of predictions at each location, and the associated uncertainty is an equally important component of the prediction. ”

“ The spatial density of surveys required varies from place to place as a function of the degree of spatial heterogeneity in underlying transmission, with highly diverse regions needing more surveys. An equivalent rule applies in the temporal dimension: where endemicity has remained relatively constant through time, or has changed in a predictable way, then older surveys are more useful for contemporary predictions than in those places experiencing rapid or unpredictable changes in transmission intensity. ”

“ In the MBG framework, PfPR2-10 values were modelled as a function of nearby survey data - which were weighted in each prediction according to their spatial and temporal proximity - and of a large suite of environmental covariates. Candidate spatial covariates were chosen based on factors known to interact with, and influence, the epidemiology of P. falciparumincluding climatology surfaces interpolated from networks of meteorological stations [32] and remotely sensed data from Earth observation satellites in their raw form and used as input into categorical global land cover products [33] . ”

“ For the remaining high uncertainty nations however, there is less immediate cause for optimism and the mandate for substantial new investment to support national malaria surveys in these countries is clear. In contrast, some countries are generating abundant parasite rate data and have a growing appetite to generate bespoke national-level maps tailored to meet local control planning needs. ”

“ One illustration of this uncertainty is provided by the two smaller maps in Figure 5: Figure 5B shows areas where the predicted posterior median PfEIR value is less than one, but the 90th percentile value exceeds 10. Such areas are widespread, and include large tracts of malaria endemic Asia. In a similar way, Figure 5C shows areas where median PfEIR is less than 10, but where there is at least a 10% chance that PfEIR exceeds 100. Such areas are widespread in Africa, and are also found in high-transmission regions of Asia including parts of India, Myanmar, and the island of New Guinea. ”

“ The presented maps of PfEIR and PfRc rely on models that link these metrics to the underlying PfPR2-10 predictions. Independent analysis of transmission using this same MAP database but with different mathematical models [62] , [14] , [63] , [15] , [44] , [64] would inevitably lead to different estimates. Differences among models are often difficult to resolve because of the intrinsic problems with identifiability and the difficulty of obtaining the right sorts of data, and independent modelling studies are urgently needed for external cross-model validation. ”

“ The spatial distribution of Plasmodium falciparumbasic reproductive number under control (PfRc) in 2010. Panel A shows the point estimate (posterior median) PfRc prediction for each pixel within the stable limits of transmission in 2010. The colour scale is logarithmic to allow better differentiation across the heavily positively skewed distribution of values. Areas of unstable transmission (medium grey areas, where PfAPI < 0.1 per 1,000 pa) or no risk (light grey, where PfAPI = 0 per 1,000 pa) are also demarked. ”

“ To remain useful, however, these maps must remain contemporary. The year 2010 has a particular significance as an evaluation milestone for malaria global health policy [4] - [5] and a huge expansion in the availability of parasite rate surveys since 2007, as well as ongoing refinement in spatial modelling techniques, including the use of environmental covariates, has provided an opportunity to carry out a major revision of the map for this benchmark year. ”

“ The steady state assumption implies that a population has been exposed for some time, so it is consistent with and most suitable for describing malaria prevalence in older children, i.e. for PfPR2-10 [8] . The model can be written so that each of the three transmission metrics can be predicted as a function of the other two (Additional File 7) . This, combined with the different candidate models linking PfPR with PfEIR, means numerous formulations can be defined for predicting PfRc [22] , [45] , [24] . ”

“ In parallel with efforts to assemble these databases, work has been underway to engineer an online infrastructure that will allow users to visualize the location of all survey data available for export and download all data used in the models for which appropriate permissions are available. This data explorer can also be found on the MAP website. ”

“ A spatially tailored approach to optimising national control strategies is at odds with aspects of current guidelines promoting universal coverage, but may become increasingly important as international financing for control comes under pressure. ”

“ Where remotely sensed imagery was available as multi-temporal data, temporal Fourier analysis (TFA) was used to ordinate the data by decomposing the temporal signal into an additive series of harmonics of different seasonal frequencies [34] , [35] . ”

“ The median predicted prevalence was 6.7% with the lowest and highest predicted PfPR2-10 values 0.8% and 21.0%, respectively. These summary statistics are indicative of higher endemicity predictions in some regions compared to the 2007 map, which largely resulted from the doubling of input data for the Americas region, including much better coverage in the more intense transmission foci of northern Amazonia. ”

“ GLJ is supported by a National Science Foundation Graduate Research Fellowship. PWG, APP, DLS, AJT and SIH also acknowledge support from the RAPIDD program of the Science & Technology Directorate, Department of Homeland Security, and the Fogarty International Center, National Institutes of Health (http://www.fic.nih.gov) . ”

“ Asia (46%) , with a much smaller proportion in the Americas (2%) (Table 1) . In America and CSE Asia, children under 15 years approached a third (30%, in both regions) of the total PAR, whilst in Africa this proportion rose to 43% (Table 1) .
Table 1 "Populations at risk of Plasmodium falciparummalaria in 2010 (millions) " Unstable risk (PfAPI < 0.1 per 1,000 people pa) and stable risk (PfAPI ≥0.1 per 1,000 people pa) . Stable risk is sub-divided into three age-standardized [31] and control related PfPR2-10 endemicity classes [2] . ”

“ Africa+: Africa, Yemen and Saudi Arabia; CSE Asia: Central and South and East Asia; DRC: Democratic Republic of Congo; GMAP: Global Malaria Action Plan; GRUMP: Global Rural-Urban Mapping Project; ITN: Insecticide treated net; MAP: Malaria Atlas Project; MBG: Model-based geostatistics; MCMC: Markov chain Monte Carlo; PAR: Population at risk; PfAPI: P. falciparumannual parasite incidence; PfMEC: P. falciparummalaria endemic country; PfPR: P. falciparumparasite rate; PfPR2-10: P. falciparumparasite rate in two-10 year olds; PfEIR: P. falciparumentomological inoculation rate ”