Mapping malaria transmission intensity using geographical information systems (GIS): an example from Kenya

Transcription

1 Annals of Tropical Medicine & Parasitology, Vol. 92, No., 7 2 (998) Mapping malaria transmission intensity using geographical information systems (GIS): an example from Kenya BY J. OMUMBO Kenya Medical Research Institute/Wellcome Trust Collaborative Programme, P.O. Box 434, Nairobi, Kenya J. OUMA, B. RAPUODA Division of Vector Borne Diseases, Ministry of Health, P. O. Box 3, Nairobi, Kenya M. H. CRAIG, D. LE SUEUR National Malaria Research Programme, South African Medical Research Council, P.O. Box 72, Congella, Durban 43, South Africa AND R. W. SNOW* Kenya Medical Research Institute/Wellcome Trust Collaborative Programme, P.O. Box 434, Nairobi, Kenya, and Nuf eld Department of Clinical Medicine, University of Oxford, John Radcliffe Hospital, Headington, Oxford OX3 9DU, U.K. Received 2 September 997, Revised 22 September 997, Accepted 23 September 997 That there are so few examples of the use of epidemiological maps in malaria control may be explained by the lack of suitable, spatially de ned data and of an understanding of how epidemiological variables relate to disease outcome. However, recent evidence suggests that the clinical outcomes of infection are determined by the intensity of parasite exposure, and developments in geographical information systems (GIS) provide new ways to represent epidemiological data spatially. In the present study, parasitological data from 82 cross-sectional surveys conducted in Kenya were abstracted and spatially de ned. Risks of infection with Plasmodium falciparum among Kenyan children, estimated from combinations of parasitological, geographical, demographic and climatic data in a GIS platform, appear to be low for 2.9 million, stable but low for another.3 million, moderate for 3. million and high for.8 million. (Estimates were not available for.4 million children.) Whilst the parasitological data were obtained from a variety of sources across different age-groups and times, these markers of endemicity remained relatively stable within the broad de nitions of high, moderate and low transmission intensity. Models relating ecological and climatic features to malaria intensity and improvements in our understanding of the relationships between parasite exposure and disease outcome will hopefully provide a more rational basis for malaria control in the near future. Despite the established tradition of de ning malaria transmission in communities by stability, intensity and seasonality, there are few * Author to whom correspondence should be addressed. Address for correspondence: KEMRI/Wellcome Trust Collaborative Programme, P.O. Box 434, Nairobi, Kenya. bobsnow@ africaonline.co.ke; fax: examples of how these descriptions have been used to guide and select interventions. There are two basic requirements if epidemiology is to be used to guide control activity: () epidemiological descriptions must relate to outcomes of public-health signi cance, such as disease and death; and (2) these markers must be both easy to collect and interpret. Ó /98/ 7-5 $ Liverpool School of Tropical Medicine Carfax Publishing Ltd

2 8 OMUMBO ET AL. The diverse ecology of sub-saharan Africa supports a wide range of basic reproduction rates for Plasmodium falciparum malaria. Whilst well described relationships between the parasite, man and vector provided an empirical basis for eradication of this disease, until recently there has been a paucity of data on the relationship between parasite exposure and the clinical outcom e of disease. Recent eld studies have provided direct evidence that the frequency of parasite exposure from birth determ ines the speed with which effective clinical immunity is acquired among the host population, the clinical spectrum of lifethreatening disease in a community and the extent of active immunization early in life during an innate period of clinical protection (Snow et al., 997). There continues to be much debate over the implications of these observations for disease control through reductions in parasite exposure, by use of insecticide-treated bednets or transmissionblocking vaccines (Molineaux, 997; Greenwood, 997). What remains clear is that the age and clinical patterns of disease will be determined by the intensity of transmission. Thus it seems reasonable to consider malaria as a diverse clinical problem and that both rationalisation and targeting of interventions aimed at disease control would be assisted by an understanding of the local epidem iology of parasite transmission. Whilst factors which determine the frequency of contact between infected vectors and the host population are complex, there remain only two methods by which the product of these contacts can be measured: vectoror human-infection studies. Such studies vary in their complexity, ranging from detailed estimates of annual sporozoite inoculation rates, rates of serological and/or parasite conversion, to the simplest measures of single, point-prevalence estimates of spleen or parasite `rates among the host population. Perhaps it is no coincidence that the most widely available source of information on the intensity of transmission in Africa remains the parasite `rate or, more correctly, as it is not expressed in time, the parasite ratio (i.e. the proportion of subjects with detectable parasitaemia) among children. This ratio was developed into the frequently used index of endemicity by Metselaar and van Theil (959), whose de nitions of hypo-, meso-, hyper- and holo-endemic malaria, which were developed without reference to disease outcome, provide arbitrary distinctions between communities in terms of parasite exposure. The idea that transmission intensity is an important determ inant of disease outcom e which should be used to guide control activities formed the basis of the present study. The most readily available source of information (i.e. the parasite ratio) was selected as the marker of intensity, and new de nitions of `endemicity, based on recent epidemiological evidence, were used. Data on malaria in Kenya were used to examine the parasite ratio s robustness over time, across age-groups and within communities and to produce endemicity maps using geographical information systems (GIS). MATERIALS AND METHOD S Identi cation of Parasitological Data Volumes of the East African Medical Journal (923 99), Transactions of the Royal Society of Tropical Medicine and Hygiene (97 99), Bulletin of World Health Organization (959 99), American Journal of Tropical Medicine and Hygiene (952 99) and Annals of Tropical Medicine and Parasitology (957 99) were searched for data pertaining to parasitological surveys conducted in Kenya. Medline (SilverPlatter International, Boston, MA) was used to identify all journal sources of malaria data collected in Kenya after 95. All 9 postgraduate theses from the faculties of medicine (departments of biochemistry, community health, medicine, paediatrics, pharmacy), science and arts (department of sociology) of the University of Nairobi which were registered between 97 and 99 and for which title, abstract or thesis copy was available were reviewed for malaria-related material. Annual reports, conference proceed-

3 MAPPING MALARIA TRANSMISSION 9 ings and direct contact with local non-governmental organizations, research institutes and mission hospitals allowed the identi cation of unpublished malaria data. Since 95, the Division of Vector Borne Diseases, Kenya Ministry of Health, has allocated limited resources to each of its 45 eld stations to conduct community-based entomological and parasitological surveys. The monthly returns summarising these ndings were systematically examined to identify all cross-sectional parasite-survey data. All data were abstracted between April 99 and May 997 using a standard proforma to record dates of surveys, numbers of children examined and parasite ratios by age, location of survey, methods of sample recruitm ent and laboratory procedures. The geographical location of each survey was established using a digitized place-name gazetteer (GeoName; GDE System s, San Diego, CA), ; 5 maps available from the Survey of Kenya (Series Y73, Department of Surveys, Nairobi, Kenya, ), maps provided in reports and occasionally by use of global positioning systems. The longitudes and latitudes of point-source surveys (villages, schools etc) or multiple positions for areawide surveys were positioned on digitized maps of administrative boundaries to identify boundary codes for each survey. Geographical Data Kenya s national administrative boundaries were originally derived for the purposes of local government, resource allocation (including education and health) and population censuses. The main units form a ve-tiered system from province (of which there are eight) as the rst level, to gazetted districts (47), divisions, locations; and ultimately, the smallest unit, the sub-location. The boundaries are reviewed every 8 years and updated for the purposes of national censuses and local government, although these changes more often involve merging or separating existing smaller units (Anon., 992). Although nine new districts have been formed since the last national census in 989, these have not yet been gazetted, have no con rmed digitized boundaries and have not been used within the present study. All malaria data were attributed to the fourth-level administrative unit, location (location boundaries have not changed since 989), and aggregated to district level. There are 92 locations within Kenya, of which 85 are populated. Their boundaries have been digitized by the United Nations Environmental Programme (UNEP) into a vector format that can be used within commonly available GIS platforms. Because locations were de ned crudely on the basis of population size, they vary in size from.5 3 km 2 ; 75% of locations are, however,, 4 km 2. Additional spatially de ned, geographical information was used to identify lakes, perennial rivers, game parks, forests and protected areas from vector-based les in Arcinfo format (UNEP/GRID, Nairobi, Kenya), Africa Data Sampler (World Resources Institute, Washington, DC) and the Ministry of Agriculture (National Agricultural Laboratories, Nairobi, Kenya). Population data were derived from the 989 national census data (Anon., 989) and estimates of the total populations and populations of children ( 9 years of age) within each location were projected to 997 using combinations of the district s intercensal growth rate and the single-year agedistribution within each district. Data Aggregation, Selection and GIS Display Data were entered twice using Foxpro Ò version.2a (Microsoft), and veri cation identi ed errors in data entry. Range checks were perform ed to identify all correctable errors. Surveys conducted within exactly the same geographical area within a 2-month period were combined to form a single estimate. Furthermore, given the duplication of survey results from identical geographical areas, data were sorted on location to identify a single estimate per location in the development of the endemicity map. Selection was made on the basis of the largest sample, most recent survey, as close to the most appropriate age-range as possible ( years) and where

4 OMUMBO ET AL. Fig.. Kenya s district boundaries and the malaria-exclusion areas based upon climatic restrictions (yellow) and unpopulated national parks and forests (green).

5 MAPPING MALARIA TRANSMISSION Fig. 2. Map showing fourth-level administrative boundaries (of locations), estimates of high (dark red), moderate (medium red) and low (light red) Plasmodium falciparum endemicity, the complete mask layer for low probability of malaria (yellow), and locations for which no relevant data were available (white).

6 2 OMUMBO ET AL. samples were randomly selected within the community. Data were re-coded where possible to a single-point estimate covering age-ranges between and years. The three arbitrary levels of stable endemicity used were `high ( $ % of the sample showed evidence of infection), `moderate (2% 59% of the sample were infected) and `low (, 2% of sample were infected), and were based upon observations of disease risk between communities with different parasite ratios in childhood (Snow et al., 997). These de nitions assume that where less than one in ve of the childhood population is found to have evidence of infection on cross-sectional survey, parasite exposure in childhood will be so low that disease incidence may simply be a function of parasite encounters and not greatly modi ed by acquired immune responses. In these areas, disease risk in childhood has been shown to be low and spread evenly across all age-groups. Conversely, communities which experience a high frequency of infection (% or greater) will represent a high rate of parasite exposure from birth, early acquisition of immunity, a concentration of disease risk within the rst 2 years of life and a paradoxically lower risk of disease throughout childhood compared with settings with moderate transmission (i.e. with parasite ratios of 2% 59%). Among the latter, disease risk is spread over the rst 5 years of life and more commonly involves pathologies with cerebral involvement. Survey data per location were therefore re-coded as high, moderate or low and linked to spatial co-ordinates of each location boundary using the GIS platform (Mapinfo Ò Professional version 4.; Mapinfo Corporation, Troy, NY). In an attempt to de ne a mask layer for the endemicity maps produced, two levels were derived. Firstly, uninhabited areas such as national parks, forests and unpopulated administrative regions were overlaid using the available data sources described above and Mapinfo Ò. Secondly, a climate model was developed using IDRISI software (IDRISI Project, Clarkson University, Worcester, M A) based upon temperature-dependent rates of sporozoite development and upon rainfall which would limit breeding-site availability. Both temperature and rainfall data for Kenya were abstracted from the Australian National University s Centre for Resource and Environm ental Studies rasterized database on African climate (Hutchinson et al., 995). A date-less, 5-month, continuous, fuzzy-logic model was used to identify areas where temperature, rainfall or both resulted in a, 5% probability of transmission occurring (M. H. Craig, R. W. Snow and D. le Sueur, unpubl. obs.). These raster images were overlaid on administrative boundaries to identify locations where. 5% of the geographical area was within this zone of lowprobability transmission. These areas represented both unstable situations, with the potential for epidemic malaria, and areas where malaria transmission cannot be sustained. These locations were stored in the mask layer. RESULTS General Geographical Features and Exclusion Layers Using the climatic probability model of transmission, 37 locations were identi ed as representing areas which had, 5% probability of sustaining transmission on the basis of temperature or rainfall. These areas are shown in Fig. together with the unpopulated national parks and forests. [The climatic mask layer encompasses 95% of the locations which lie at. 24 m above sea level, an altitude which Garnham (948) previously used to de ne the highlands and mountainous regions of Kenya where malaria was absent. It also includes large areas of districts (Nandi, Kericho, Uasin Gishu and Trans Nzoia) traditionally regarded as susceptible to epidem ic malaria (Garnham, 948; Roberts, 94; Ngindu et al. 989; Some, 994). The excluded areas in the north and north eastern regions of Kenya encompass the arid zones, including the districts of Garissa and Mandera.] Overall, 373 locations were judged to be within the exclusion layer shown in Fig.,

7 MAPPING MALARIA TRANSMISSION 3 representing 8.5 million people or 3% of the total projected population of Kenya in 997. Parasite Surveys from Locations outside the Mask Layer Data for 82 independent cross-sectional parasitological surveys conducted on children in Kenya between 92 and 99 were identi ed following a 2-month search. Ten surveys could not be geographically positioned to the location level because the inform ation provided in the report was inadequate. The remaining 72 surveys covered 24 (3%) of the 72 locations not within the mask layer. Over half (28) of the locations covered were represented by more than one survey and the data from the most precise, recent and appropriate age-range survey were selected for each of these. The resulting surveys covered 33% of the total potentially exposed population within Kenya or 2.7 million children aged, years. A summary of these data by district is given in Table. The distribution of high-, moderate- and low-transmission locations is shown in Fig. 2. Whilst there remains a large part of the country for which no empirical data were available, several general observations can be made. Firstly, within a district there exists marked variation in the level of endemicity. Despite the diversity in the types and timing of the surveys (described below), a more accurate picture of the within-district heterogeneity is shown for selected surveys in Kili district (Fig. 3). Here, only the 2 surveys conducted on community rather than school-based surveys among children aged 9 years after 987 demonstrate the extent of this local heterogeneity. Secondly, despite localized variations, the data do provide a general impression about the regional variations in endemicity. Figure 4 represents district-based parasite-ratio categories derived from the median ratios from all geographically independent surveys within the district (Table ). Whilst for some districts these median estimates derive from only a few surveys across a wide geographical area, a general pattern emerges. Districts which include the highest surveyed parasite ratios are located mainly around Lake Victoria (Siaya, Homa Bay and Kisumu districts), moderate, stable, transmission intensity is a predominant feature of the southern part of the coastal region (Kwale and Kili districts) and the lowest intensities of stable transmission are within the central regions of the country. Survey Sources and Inherent Variations in the Parasite Ratio Differences in the sizes of the samples selected produce inherent variation in the precision of the estimates of parasite ratio: 2 surveys did not provide a denominator, 44 (22%) were conducted on samples of 2 subjects each and 92 (4%) were conducted on samples of. 2. Furthermore, the surveys were abstracted from a variety of sources and sampled different age categories with different sampling strategies over a wide time period. The stability of the parasite ratio, between different age-groups, within a location over time and by different sampling strategies, was therefore of particular interest. Sixty-three (29%) of the unique-location studies were conducted among subjects aged 9 years, 29 (4%) among subjects aged years, 28 (3%) among subjects aged 2 9 years, 23 (%) among subjects aged 5 9 years, 3 (%) among subjects aged 5 years, among subjects aged 9 years, and the remaining 47 (22%) surveys among other age categories. Thirty surveys were abstracted where data were presented in a variety of age-groupings among large community-based samples and the re-aggregation of these data allowed an examination of the parasite ratio according to a variety of age-classes. Plots of parasite ratios by different age groupings consistently provided correlation coef cients of..94 and misclassi cations of endemicity class of, 8%. The worst examples were the comparisons of parasite ratios among those aged 5 years with those of subjects aged 5 9 years (r 5.84), where 7% of comparisons resulted in a different endemicity classi cation As mentioned above, the abstracted location-speci c survey results related to studies conducted over a 7-year period: 2 (9.3%) were conducted before 9 whilst 24 (58%)

8 4 OMUMBO ET AL. Fig. 3. Map plotting the results of community-based surveys of Plasmodium falciparum prevalence in Kili district after 988, showing areas of high (dark red), moderate (medium red) and low (light red) endemicity, and locations for which no relevant data were available (white).

9 MAPPING MALARIA TRANSMISSION 5 Fig. 4. Map of district-based estimates of Plasmodium falciparum endemicity based on the median parasite ratios observed in location-specific surveys. Areas of high (dark red), moderate (medium red) and low (light red) endemicity, areas where this is no or very low probability of malaria transmission (yellow), and locations for which no relevant data were available (white) are indicated.

10 OMUMBO ET AL. TABLE Summary of available data from parasitological surveys, selected for mapping by location within each district No. of locations District Total Attributed to mask Others for which empirical data were available Median and (range) of parasite ratios (%) Nairobi ( ) ( ) 24.2 (8.3 3.) 33. (3. 9.) (2.3 5.) ( ( ( ( ( ) 3.8 ( ) 8.5 (2. 2.) 2.2 ( ) 2.2 ( ) 2.2 ( ) ( ). ( ) 58.7 ( ) 37.3 ( ) 3.5 ( ) (.3 4.4) 2.9 ( ) 2. (. 44.4) RIFT VALLEY PROVINCE Baringo Bomet Elegeyo Marakwet Kajiado Kericho Laikipia Nakuru Trans Nzoia Turkana Uasin Gishu Nandi Narok West Pokot W ESTERN PROVINCE Bungoma Busia Kakamega Vihiga Garissa Mandera Wajir 84.) 47.2) 77.8) 4.) COAST PROVINCE Kwale Kili Lamu Mombasa Samburu Taita Taveta Tana River NYANZA PROVINCE Homa Bay Kisii Kisumu Migori Nyamira Siaya EASTERN PROVINCE Embu Isiolo Kitui Machakos

11 MAPPING MALARIA TRANSMISSION 7 (Table Continued) No. of locations District Makueni Marsabit Meru Total Attributed to mask Others for which empirical data were available Median and (range) of parasite ratios (%) ( ).4 (. 45.3) / ( ) 33.2 (. 94.5) CENTRAL PROVINCE Nyeri Kiambu Kirinyaga Muranga Nyandarua Nithi All districts were conducted within the last years. Using the entire data-set, those surveys which were conducted within four identical geographical areas at different times (over a wide time-frame) using comm unity-based sampling procedures among similar age-groups were examined (Table 2). Although the results of these surveys represent only a few geographical areas, they suggest that areas of low or high endem icity remain constant over a 3-year time-period. How ever, those areas which are at the boundaries of an endemicity class (e.g. Area 3 in Table 2) may change class. The location-speci c surveys presented in Table derive from a variety of sources: 29 (3.%) from journal sources, six (2.8% ) from postgraduate theses, 3 (4%) from the reports of non-governmental organizations or of community surveys by mission hospitals, or personal comm unications, and the majority [49 (7%)] from Ministry of Health surveys conducted by the Division of Vector Borne Diseases. Of the latter, 3 (42%) were conducted in schools whilst the remaining studies were conducted on samples selected from the community. There were too few surveys conducted within the same geographical area at the same time to test form ally whether schoolor comm unity-based sampling approaches al- tered the perceived endem icity class for a given area. However, school and comm unity surveys among 5 9-year-olds at a lowtransmission setting at Kili Township gave similar parasite ratios (4% and %, respectively), and similar parasite ratios were obtained within the same age classes from school and comm unity surveys among a hightransmission community in western Kenya (95% and 89%, respectively). Whilst sampling from schools may introduce an inherent bias toward aparasitaemic subjects at the extrem es of transmission, it seems to provide similar results to those obtained by sampling within the comm unity. DISCUSSION Ironically, despite the existence of highresolution maps of even the rarest of cancers in the Western world (Gardner et al., 983), there are few detailed maps of either the risks of exposure or disease for malaria, Africa s single largest cause of mortality. This may have a historical basis, best summarised by the conference on malaria eradication in Equatorial Africa held in Kampala in 95, which recom mended the following: `¼ to govern-

12 8 OMUMBO ET AL. TABLE 2 Data from community-based surveys showing variations in parasite ratios over time within each of four geographical areas Area Year Parasite ratio (%) Level of endemicity (3/) 5.5 (22/42). (7/9) 2.2 (8/358).3 (/53).7 (27/253) 73.3 (2/288) 45.5 (77/7).9 (459/8) 8.2 (57/758) 89. (49/527) 94.5 (95/958) Low Low Low Low Low Low High Moderate High High High High ments responsible for the administration of African territories that malaria should be controlled by modern methods as soon as feasible, whatever the original degree of endem icity and without awaiting the outcom e of further experim ents (WHO, 95). During a recent meeting in Brazzaville at the World Health Organization s regional of ce, a statement was prepared recomm ending the integration of insecticide-treated bednets (ITBN) into malaria-control activities across Africa (WHO, 99) and many contend that this intervention should be implemented in every endemic community in Africa (Lengeler et al., 997; Molineaux, 997). Thus, 4 years after the Kampala conference, policy-makers still prefer not to commit themselves as to whether levels of malaria endemicity should in uence the choice or expected outcom e of a particular malaria-control activity. This view is expressed despite the wide variation, with level of endemicity, in protective ef cacies shown for ITBN against childhood mortality, the lowest ef ciacies occurring in communities exposed to the highest levels of P. falciparum transmission (Binka et al., 99; Habluetzel et al., 997). Furtherm ore, the results of several studies have indicated that the basic epidemiology of disease and mortality is determined by transmission inten- sity. Age, clinical syndromes and rates of disease have been shown to be dependent upon the level of endemicity within a given community (Greenw ood et al., 99; Slutsker et al., 994; Snow et al., 994, 997; Trape and Rogier, 99). If the risks of malariaattributable morbidity and mortality were not a function of the frequency with which a host has previously encountered infection, and hence of active immunization, then it is conceivable that interventions aimed at reducing parasite exposure would work equally well everywhere. If this were true, then maps of malaria endem icity need only show where malaria transmission could occur. However, it does not seem to be true. An understanding of the level of endemicity would allow the managers of control programmes to de ne the extent of naturally acquired immunity within a community, to appreciate the age windows of disease risk and how these windows (and the clinical spectrum of disease) may change with targeted intervention, and to use interventions rationally, where they are most likely to succeed (Snow and Marsh, 995; Snow et al., 99, 997). Consequently, one of the most basic of tools for national malaria-control programmes is the endem icity map. Whilst many countries in sub-saharan Africa will have such maps, these are often not derived

13 MAPPING MALARIA TRANSMISSION from empirical evidence but instead are based upon expert-opinion of the length of transmission seasons determ ined by climate (Anon., 959, 95, 92). The rst attempts to develop an empirical map of malaria endemicity for Kenya are presented here. The present mask layer does not allow for a distinction between unsuitable and unstable transmission potential but merely provides geographical areas where the probability of transmission occurring is low. In Nandi district, for example, the mask layer does not cover several locations where `epidemics of clinical malaria among all age-groups have been described (Roberts, 94) and consequently the information shown in Fig. 4 may be distorted. The use of climatic probabilities to de ne transmission risk still requires further developm ent but has been hampered by the lack of a consensus view on what constitutes an epidem ic and the lack of empirical longitudinal data on changes in vector and parasite dynamics in relation to changes in the incidence of disease. One of the prim ary intentions of the present study was to establish how much epidemiological data related to malariatransmission intensity already existed within Kenya. The 2-m onth, intensive search revealed a surprisingly large amount of previously unused inform ation within published and unpublished archived sources. Overall, 82, cross-sectional, malarial-parasite surveys conducted over the last 7 years across the country were identi ed. There is little doubt that these surveys vary in quality, coverage and sample size, and some may argue that the data retrieved cannot be used to provide national estimates of malaria endemicity. Nevertheless, these data represent all the inform ation that is available and examination of them indicates that, whilst the parasite ratio does vary within a single setting according to the date of the survey, the selected ages of the sample and sampling frame, the broad categories of high, moderate and low transmission vary little. Despite the robustness of this marker of transmission intensity within broad classi cations, these classi cations remain arbitrary and require further examination using 9 empirical data and mathematical models of exposure v. disease outcome. Furthermore, there remain large parts of Kenya for which no relevant data are available (Figs 3 and 4). Models to predict endemicity, which will allow the estim ation of endem icity based upon ecological determinants and thus provide both higher resolution and more complete coverage of infection ratios within a country, are required. In the present study, a GIS was used to present data along nationally recognized administrative boundaries. GIS technology allows the spatial representation of endem icity data in conjunction with other disparate sources of data such as population and topography. What emerged was a very heterogeneous pattern, both across the country and within districts (Figs 2 and 3). This was not surprising given the diverse ecologies within districts, which would either support or restrict vector abundance, depending upon the local variations in breeding-site availability, altitude and other topographical and demographic features. Small-area variation in malaria endemicity has been well described (Bradley et al., 98; Cattani et al., 98; Jambulingham et al., 99; Sharp and le Sueur, 99). Although a generally accepted feature of malaria, such variations have implications for how control program mes de ne resource allocation. Although administrative boundaries are not recognized by malaria vectors and thus represent arbitrary, geographical, transmission units, they provide a means by which data on population density, other demographic and disease data and local resources can be similarly attributed for healthplanning purposes. As most national health-resource allocation is presently based upon district-level requirements, a districtlevel endemicity map, based upon the best available, geographically independent, survey data, was the goal of the present investigation (Fig. 4). It remains unclear whether such district-level maps adequately allow for the disparity in transmission intensity within a district for district health-planners to de ne universal or targeted intervention policy. Nevertheless, the map in Fig. 4 does generally

14 2 OMUMBO ET AL. correspond to previous, expert-opinion, seasonality maps derived for Kenya during the 95s (Anon., 959). It differs from previous maps in showing that, despite a long transmission season on the Kenyan coast and around Lake Victoria, the latter is able to sustain a much higher level of P. falciparum endem icity. This low-resolution map and projected estim ates of the population in 997 indicate that, of Kenyan children aged, years, about.33 million will be exposed to a low intensity of transmission, 3.2 million to a moderate intensity of transmission and.8 million to high intensity of transmission (.4 million children remained unclassi ed). The present results provide evidence that inform ation on malaria endem icity can be obtained from existing archived sources and that these can be developed, through the use of GIS, into useful tools for the managers of national programmes for disease control. Through the use of such basic GIS platforms, models of P. falciparum endem icity can be () developed to provide high-resolution maps of malaria risk and (2) used in combination with data on population distribution to identify target control stratagems which combine resource availability and an appreciation of local epidemiology. Despite the infancy of this area of epidem iology, such approaches should offer a more rational basis for malaria control. ACKNOWLEDGEMENT S. This study received nancial support from the Wellcom e Trust, U.K., the International Development Research Centre, Canada, the South African Medical Research Council and the Kenya Medical Research Institute (KEMRI). R.W.S. is a Senior Wellcom e Trust Research Fellow (grant 3334). The authors wish to acknowledge the technical assistance provided by R. Kruska and O. Okello and the support of the MARA/A RMA collaboration. This paper is published with the permission of the director of KEMRI. REFERENCES A NON. (959). Atlas of Kenya. Nairobi: Surveys of Kenya A NON. (95). Atlas of Tanganyika, 3rd Edn. Dar es Salaam: Department of Lands and Surveys. A NON. (92). Atlas of Uganda. Kampala: Department of Lands and Surveys. A NON. (989). Kenya Population Census, 989, Vol.. Nairobi: Central Bureau of Statistics. A NON. (992). Constitution of Kenya, Chapter III, Part I, Section 42. Nairobi: Electoral Commission. B INKA, F. N., K UJABE, A., A DJUIK, M., W ILLIAMS, L. A., L ENGELER, C., M AUDE, G. H., A RMAH, G. E., K AJIHARA, B., A DHIAMAH, J. H. & S MITH, P. G. (99). Impact of permethrin impregnated bednets on child mortality in Kassena-Nankana district, Ghana: a randomised controlled trial. Tropical Medicine and International Health,, B RADLEY, A. K., G REENWOOD, B. M., G REENWOOD, A. M., M ARSH, K., B YASS, P., T ULLOCK, S. & H AYES, R. (98). Bed nets (mosquito nets) and morbidity from malaria. Lancet, ii, C ATTANI, J. A., M OIR, J. S., G IBSON, F. D., G INNY, M., PAINO, J., D AVIDSON, W. & A LPERS, M. P. (98). Small area variations in the epidemiology of malaria in Madang Province. Papua New Guinea Medical Journal, 29, 7. G ARDNER, M. J., W INTER, P. D., T AYLOR, C. P. & A CHESON, E. D. (983). Atlas of Cancer Mortality in England and Wales Chichester, U.K.: John Wiley. G ARNHAM, P. C. C. (948). The incidence of malaria at high altitudes. Journal of the National Malaria Society, 7, G REENWOOD, B. M. (997). Malaria transmission and vector control. Parasitology Today, 3, G REENWOOD, B. M., M ARSH, K. & SNOW, R. W. (99). Why do some African children develop severe malaria? Parasitology Today, 7, H ABLUETZEL, A., D IALLO, D. A., E SPOSITO, F., L AMIZANA, L., PAGNONI, F., L ENGELER, C., T RAORE, C. & C OUSENS, S. N. (997). Do insecticide-treated curtains reduce all-cause child mortality in Burkina Faso? Tropical Medicine and International Health, in press.

15 MAPPING MALARIA TRANSMISSION 2 H UTCHINSON, M. F., N IX, H. A. & M C M AHON, J. P. (995). AfricaÐ a Topographic and Climatic Database. Canberra: Australian National University. J AMBULINGHAM, P., M OHAPATRA, S. S. S., G OVARDHINI, P., K UMAR, L. D., M ANOHARAN, A., P ANI, S. P. & D AS, P. K. (99). Microlevel epidemiological variations in malaria and implications for control strategies. Indian Journal of Medical Research, 93, L ENGELER, C., S MITH, T. A. & SCHELLENBERG, J. A. (997). Focus on the effect of bednets on malaria morbidity and mortality. Parasitology Today, 3, M ETSELAAR, D. & VAN T HEIL, P. M. (959). Classi cation of malaria. Tropical and Geographical Medicine,, 57. M OLINEAUX, L. (997). Nature s experiment: what implications for malaria prevention? Lancet, i, N GINDU, A. M., K ABIRU, E. W., M BAABU, D. A. N., O DERO, W. O. O. & S IONGOK, T. K. A. (989). Outbreak of epidemic malaria in Uasin Gishu DistrictÐ 988. In Proceedings of the 8th KEMRI/KE TRI Conference, p. 5. Nairobi: Kenya Medical Research Institute. R OBERTS, J. M. D. (94). The control of epidemic malaria in the highlands of Western Kenya. Part I. Before the campaign. Journal of Tropical Medicine and Hygiene, 7, 7 8. S HARP, D. L. & LE SUEUR, D. (99). Malaria in South AfricaÐ the past, the present and selected implications for the future. South African Medical Journal, 8, S LUTSKER, L., T AYLOR, T. E., W IRIMA, J. & STEKETEE, R. W. (994). In-hospital morbidity and mortality due to malaria-associated severe anaemia in two areas of Malawi with different patterns of malaria infection. Transactions of Royal Society Tropical Medicine and Hygiene, 88, S NOW, R. W. & M ARSH, K. (995). Will reducing Plasmodium falciparum transmission alter malaria mortality among African children? Parasitology Today,, S NOW, R. W., B ASTOS DE A ZEVEDO, I., L OWE, B. S., K ABIRU, E. W., N EVILL, C. G., M W ANKUSYE, S., K ASSIGA, G., M ARSH, K. & T EUSCHER, T. (994). Severe childhood malaria in two areas of markedly different falciparum malaria transmission in East Africa. Acta Tropica, 57, S NOW, R. W., M ARSH, K. & LE S UEUR, D. (99). The need for maps of transmission intensity to guide malaria control in Africa. Parasitology Today, 2, S NOW, R. W., O M UMBO, J. A., L OW E, B., M OLYNEUX, C. S., O BIERO, J. O., P ALMER, A., W EBER, M. W., P INDER, M., N AHLEN, B., O BONYO, C., N EWBOLD, C., G UPTA, S. & M ARSH, K. (997). Relation between severe malaria morbidity in children and level of Plasmodium falciparum transmission in Africa. Lancet, i, S OME, E. S. (994). Effects and control of Highland malaria epidemic in Uasin Gishu District, Kenya. East African Medical Journal, 7, 2 8. T RAPE, J. F. & R OGIER, C. (99) Combating malaria morbidity and mortality by reducing transmission. Parasitology Today, 2, W ORLD H EALTH O RGANIZATION (95). Report of the Malaria Conference in Equatorial Africa. Technical Report Series No. 38. Geneva: WHO. W ORLD H EALTH O RGANIZATION (99). Meeting on insecticide-impregnated materials. TDR News, 5, 3.

16