Carlos A Guerra; Simon I Hay; Lorena S Lucioparedes; Priscilla W Gikandi; Andrew J Tatem; Abdisalan M Noor; Robert W Snow

“ The most successful approach to identifying community specific estimates of parasite prevalence was through communications and contacts with malaria scientists identified from peer-reviewed literature sources. Assembling a comprehensive database of PR data cannot depend entirely on information available from peer-reviewed sources, therefore, as sources classified as 'journal' alone only provided 26.4% of the total number of records. ”

“ Afghanistan, Tanzania, Kenya, Zimbabwe and Eritrea are the top five countries in terms of number of PR survey records. A more balanced picture of data availability by country was produced by weighting the number of points by the estimated area that is malarious derived from the global spatial limits of malaria (Figure 5) [25] , [26] . According to this ranking, the five countries for which the least PR records are available are Angola, Mexico, Namibia, Chad and Democratic Republic of the Congo.
The global distribution of P. falciparum PR points from the MAP database. ”

“ Locating survey data
The geo-positioning of surveyed communities is a complicated exercise. There are a number of potential problems when locating a survey in space and the level of difficulty depends largely on the amount of information available for the survey area in question. Having only the name of the location is often insufficient because village names are frequently repeated within countries and even within administrative sub-divisions. ”

“ The development of the MAP PR database faced difficulties related to dealing with meta-data that can be considered generic to the assembly of large health spatial databases. It is clear from the methods reviewed here that such data collection cannot rely solely on published information as only one quarter of our PR data were available from sources found in the scientific literature. Although non-trivial, accessing the vast amounts of information in grey literature and unpublished sources is fundamental to maximizing information coverage. ”

“ Open access to databases of information generated by the research community can synergize individual efforts and are epitomized by the genome mapping projects. Open source models for outputs of scientific research funded by tax-payers and charities are becoming the norm. This has yet to be extended to malaria epidemiology and control.
Methods
The exhaustive searches and assembly process for a global database of malaria parasite prevalence as part of the Malaria Atlas Project (MAP) are described. ”

“ Flow chart for the geo-positioning of PR data points. It is not always clear how communities are described in space during PR surveys. In East Africa, it is common to define a community as represented by widely dispersed homesteads, separated sometimes by several hundreds of metres of agricultural land. Conversely, in several West African countries, a community is more spatially constrained with agricultural land surrounding the village unit. PR survey data are reported by community name and generally not spatial dimensions or settlement density. ”

“ Once robust spatial estimates of global malaria endemicity have been developed, the integration of further epidemiological data to exploit the applicability of these maps is anticipated, including monitoring progress towards international malaria control targets and the projection of financial and commodity needs for MECs. Without publicly available information, national and international agencies will continue to assume that entire regions or countries share similar epidemiological characteristics and that they must share one prescribed menu for control. ”

“ The design, financing and implementation of malaria control demand a quantitative approach to the definition of anti-malarial commodity requirements of populations living under varied malaria transmission risks. The geographic extent of effective intervention coverage and how this relates to the dominant local Anopheles vector species and, in turn, their sensitivity to insecticides, is necessary to design appropriate vector control. Information on the prevalence of drug resistance and population spatial access to medicines is required to devise adequate clinical disease management strategies. ”

“ Another significant obstacle confronted during the assembly of the MAP database was the geo-positioning of data. The accuracy of spatial references in a database is wholly dependent on the precision with which data are located in space. This problem was minimized by developing an elaborate system of geo-positioning based on using multiple sources of information and rigorous quality control. ”

“ Less than half (22/49) of the malaria endemic countries (MECs) in Africa provided information for the most recent reporting year, 2003; the rest were older. These data are neither reconciled at sub-national scales, nor against populations at risk of any standard definition of the intensity of malaria transmission.
Recent work has also involved the assemblage of historical and contemporary evidence on the distributions of malaria risk and Anopheles vector species into a compendium of maps and geographic descriptions of various malaria indicators by region and country [19] . ”

“ Thus, any decisions on positioning survey data are, to some extent, arbitrary. Another common description of communities presented in survey sources are small collections of neighbouring villages. In such cases, authors were often contacted to provide specific information for each village. Where these data were not available, it sometimes seemed reasonable to assume that these represented a single community. ”

“ The following rules were implemented in attributing data as representing a 'point', a 'wide-area' or a small or large 'polygon'. A point was considered as such when the community was named specifically and could be located in space unambiguously, or when the information suggested that a study area was no bigger than ~10 km2. A wide-area was defined as any study area that exceeded ~10 km2 but was no bigger than ~25 km2. A small polygon was assigned to an area between ~25 km2 and ~100 km2. ”

“ Information resources on malaria risk, disease incidence, intervention coverage or drug resistance should have several key features to maximize their utility for constituents. Data assembly requires a filtering system in which data inclusion rules are clearly defined and some basic standards are set before archiving. These inclusion criteria and any methods used during data collation and data entry need to be scientifically defensible and transparent. ”

“ Endemic countries for which no data or no high spatial resolution surveys were available are Algeria, Argentina, Armenia, Azerbaijan, Bangladesh, Belize, Bhutan, Comoros, Djibouti, Dominican Republic, Egypt, El Salvador, French Guiana, Georgia, Guatemala, Guyana, Iran, Kyrgyzstan, Mauritania, Mauritius, Morocco, North Korea, Oman, Panama, Paraguay, Rwanda, South Africa, Saudi Arabia, South Korea, Syria, Tajikistan and Turkmenistan. ”

“ A geo-positioning entry form was designed separately that links directly to the relevant section of the main Access form. In this way, locations and geo-positioning can be updated independently without affecting the main PR data table. This form shows descriptions of the location in which the survey was conducted. These include the name of the community studied and its regional, country and sub-national correspondence, as well as the geographic coordinates and related information about the geo-positioning method. ”

“ African Regional Office, AMRO – Americas Regional Office, EMRO – Eastern Mediterranean Regional Office, EURO – European Regional Office, SEARO – South East Asian Regional Office, WPRO – Western Pacific Regional Office. Spatial distribution of PR records
The number of records varies substantially amongst regions and countries, reflecting not necessarily where malaria is a more serious problem, but where more PR estimates are available, either because surveys are conducted more systematically or because data proved easier to access. ”

“ This source of information resulted in only 25 PR records. Smaller national conference proceedings might have yielded more comprehensive data, but these were not easily accessible with the exception of one conference proceeding in India, available from the National Institute of Malaria Research website [41] . By the very nature of national conferences, post-graduate theses, 'grey' literature reports and ministry of health survey data, these will not have been captured comprehensively by the MAP data searches. ”

“ These were excluded from subsequent analyses given the higher sensitivity and lack of direct comparability with either microscopy or RDT. The ease of RDT methods means that these are increasingly used as part of large PR surveys (e.g. Cambodia, Equatorial Guinea, Kenya, Somalia, Tanzania, Thailand, and Zimbabwe) . ”