Bulletin of the World Health Organization

Spatial heterogeneity of haemoglobin concentration in preschool-age children in sub-Saharan Africa

Ricardo J Soares Magalhães a & Archie CA Clements a

a. School of Population Health, University of Queensland, Herston Road, Herston, Qld., 4006, Australia.

Correspondence to Ricardo J Soares Magalhães (e-mail: r.magalhaes@sph.uq.edu.au).

(Submitted: 12 October 2010 – Revised version received: 23 March 2011 – Accepted: 24 March 2011.)

Bulletin of the World Health Organization 2011;89:459-468. doi: 10.2471/BLT.10.083568

Introduction

Over the past five decades many attempts have been made to reduce the burden of anaemia in vulnerable groups, particularly children less than 5 years of age and pregnant women.13 Based on recent estimates from the World Health Organization (WHO),3 the prevalence of anaemia is 24.8% globally and the highest rates are found in preschool-age children (67.6%) and pregnant women (57.1%) in sub-Saharan Africa. Anaemia is a major public health problem in preschool-age children because it is associated with an increased risk of death and impaired cognitive development,4,5 growth6 and immune function.7

About 50% of all anaemia cases are due to iron deficiency.8 Other major contributors include malaria; infection with human immunodeficiency virus (HIV) and with bacteraemia-causing organisms (e.g. Steptococcus pneumoniae, non-typhi Salmonella species and Haemophilus influenzae type b); neglected tropical diseases (especially those caused by Schistosoma haematobium – the cause of urinary schistosomiasis – hookworm and, to a lesser extent, Trichuris trichiura and Schistosoma mansoni), and inherited haemoglobinopathies and thalassemias.1,918

Currently, the planning of resources required for anaemia control is based on prevalence data from field surveys within a country, which are then extrapolated to the country as a whole.3 However, efficient allocation of health interventions to control anaemia may require more targeted approaches based on information on the geographical distribution of high-risk communities and on an understanding of the relative contribution of major causes of anaemia.8 Geographical differences in the causes of anaemia can be partially explained by large-scale variability in environmental drivers, particularly nutritional and infectious causes. The risk of malaria is known to be associated with elevation and land surface temperature.19 Similarly, nutritional iron deficiency20 and anaemia-causing helminthic infections21 are known to be associated with the distance to a perennial water body, land surface temperature and the normalized difference vegetation index (NDVI) – a number derived by remote sensing that indicates the amount of land vegetation and that stands as a proxy for rainfall. Environmental drivers of anaemia tend to show a high degree of spatial dependence (i.e. geographical clustering).2224 We therefore hypothesized that the burden of anaemia and perhaps major contributors to anaemia vary geographically, even within high-burden African regions.

In anaemia control, the use of national prevalence estimates of anaemia in the presence of subnational variability is likely to hamper the efficient delivery of control programmes. For control policies to be cost-effective, the geographical variability of anaemia must be quantified. Maps showing geographical variation in anaemia would also be useful in the control of parasitic infections that are highly endemic in sub-Saharan Africa. The prevalence of anaemia has been used as a measurable indicator for evaluating control programmes for malaria, schistosomiasis and soil-transmitted helminthiasis because interventions for these infections aim at controlling morbidity. In highly endemic populations, micronutrients are being distributed as part of parasite control programmes to reduce the burden of anaemia.2,2531

We have used data collected and georeferenced by the Demographic and Health Surveys (DHS), together with high-resolution continental maps of selected environmental variables, to demonstrate geographical clustering in mean blood haemoglobin concentration (a measure of anaemia) in sub-Saharan Africa. Our aim was to quantify the spatial dependence of blood haemoglobin concentration over and above what is accounted for by environmental variables known to contribute to nutritional iron deficiency and infectious causes of anaemia. We also sought to determine the extent to which haemoglobin concentration is associated with these drivers (and whether this differs in different regions of Africa) and to build the foundations of a spatial decision-support tool to inform decision-makers about the most efficient approaches to geographical targeting of interventions for the prevention and control of anaemia.

Methods

Data sources

We searched data from nationally representative household-level DHS programmes for all countries in sub-Saharan Africa.32 Data for the countries included in the study include remotely-sensed environmental data on distance to perennial water body, elevation, land surface temperature and NDVI (all data available from corresponding author).7

Measuring anaemia

The MEASURE DHS Project32 tests women (15–49 years of age) and children (usually 6 months to 5 years of age) for anaemia. It involves obtaining a capillary blood sample using finger prick or, in the case of young children, a heel prick, and then using the HemoCue® (HemoCue AB, Ängelholm, Sweden) blood haemoglobin testing system.33 Testing is voluntary and respondents receive the results of their test immediately, along with information about how to prevent anaemia. To determine the level of geographical clustering of blood haemoglobin concentration, we used altitude-adjusted haemoglobin concentration values in children < 5 years of age (data available from corresponding author).

Geopositioning

The geopositioning methods used by MEASURE DHS32 include the identification of all households in enumeration areas or clusters.34 The information assembled in the current database provides geographical coverage of 35% of administrative areas in sub-Saharan Africa. Coverage was good in western Africa (45% of administrative areas, corresponding to 65% of total area) and eastern Africa (45% of administrative areas, corresponding to 81% of total area). For large areas of central and southern Africa, the only data were from Cameroon and the Democratic Republic of the Congo in central Africa, and Lesotho and Swaziland in southern Africa (data available from corresponding author). Therefore, for analysis, the survey results from Cameroon were included in the western African region and the results from the Democratic Republic of the Congo, Lesotho and Swaziland were included in the eastern African region.

Environmental properties

Relationships between haemoglobin concentration and the environmental variables at each cluster site were investigated using locally-weighted least squares smoothing curves with the gplot package of the R software (R Foundation for Statistical Computing, Vienna, Austria) (data available from corresponding author).35 Statistical associations between haemoglobin concentration measurements at each cluster site and the environmental variables distance to a perennial water body, elevation, land surface temperature and NDVI were tested using fixed-effects multivariable linear regression models in Stata version 11 (StataCorp. LP, College Station, United States of America).

Analysis of clustering

The extent of geographical clustering in haemoglobin data can be quantified using a semivariogram.36 This is a plot of the semivariance of all pairs of locations at a series of defined separating distances. It can be characterized by the semivariance due to spatial structure (sill or spatially structured variance, which represents the tendency for geographical clustering), the spatially unstructured semivariance (nugget, which represents natural random variation, very small-scale spatial variability or measurement error) and the distance at which locations can be considered independent (range, which represents the size of geographical clusters).36 Semivariograms are particularly important in the assessment of spatial variation of spatial point data because they allow for the quantification of the cluster size, the tendency for geographical clustering within a region and the relative contribution for clustering that is explained by a particular modifiable factor.

To investigate geographical clustering in blood haemoglobin concentration measurements, we used empirical semivariograms of the raw haemoglobin data and the residuals of multivariable models, using the geoR package of the R software.35 The proportion of spatially structured variance in the raw haemoglobin concentration data that is accounted for by environmental covariates was estimated by dividing the partial sill of multivariable models by the partial sill of the raw data; this estimate indicates how well geographical clustering of anaemia is explained by environmental covariates. The proportion of variance that is spatially structured was estimated by dividing the partial sill by the sum of the partial sill and nugget; this measure indicates the role of location in explaining variation in haemoglobin concentration in regions of sub-Saharan Africa.

Results

Haemoglobin concentrations

We included data from 2862 cluster sites in western Africa and 2999 cluster sites in eastern Africa. This included 24 277 children < 5 years of age in western Africa and 25 343 in eastern Africa. Most children (76%) were residing in rural areas and both sexes were equally represented. In both regions, haemoglobin declined towards the end of the first year of life and then increased towards 5 years of age (data available from corresponding author). The mean haemoglobin concentration was significantly lower in western than in eastern Africa (Table 1) and the geographical distribution varied between regions (Fig. 1 and Fig. 2). In western Africa, the geographical distribution of severe anaemia (haemoglobin < 70 g/l) was heterogeneous and was present in a large cluster straddling the border between Burkina Faso and Mali. In eastern Africa, haemoglobin was homogenously low (100–110 g/l) to moderate (70–100 g/l) across the region, and haemoglobin concentrations < 70 g/l were localized in small clusters. The proportion of children with haemoglobin < 110 g/l was highest in western Africa (Table 1). All countries in sub-Saharan Africa had a prevalence of anaemia > 40% (data available from corresponding author); the lowest prevalence was in Swaziland (42%) and the highest was in Burkina Faso (91%).

Fig. 1. The spatial distribution of haemoglobin concentration for cluster sites included for western Africaa
Fig. 1. The spatial distribution of haemoglobin concentration for cluster sites included for western Africa<sup>a</sup>
a Includes Cameroon.Map produced using ArcGIS version 10 (ESRI, Redlands, CA, United States of America).
Fig. 2. Spatial distribution of haemoglobin concentration for cluster sites included for eastern Africaa
Fig. 2. Spatial distribution of haemoglobin concentration for cluster sites included for eastern Africa<sup>a</sup>
a Includes the Democratic Republic of the Congo, Lesotho and Swaziland.Map produced using ArcGIS version 10 (ESRI, Redlands, CA, United States of America).

Environmental properties

Mean haemoglobin concentration in preschool-age children in both regions was negatively associated with land surface temperature and NDVI and positively associated with elevation (Table 2). However, the effect of distance to a perennial water body differed between western and eastern Africa, with mean haemoglobin concentration being negatively associated with it in western Africa and positively associated with it in eastern Africa.

Analysis of clustering

In the raw data on haemoglobin concentration, western Africa showed a greater tendency for geographical clustering (partial sill = 16.33) than eastern Africa (partial sill = 8.42). After taking into account the effect of environmental variables (residual variance), the tendency for clustering of mean haemoglobin values was more pronounced in western Africa than in eastern Africa (Fig. 3 and Fig. 4). Also, clusters of mean haemoglobin concentration were larger in western Africa than in eastern Africa (Fig. 3 and Fig. 4). Our results indicate that environmental variables could account for 13% and 27% of geographical clustering in western and eastern Africa, respectively. In eastern Africa, 100% of the residual variance in haemoglobin concentration can be explained by location (i.e. the separating distance between cluster sites); in western Africa, only 91% of residual variance in haemoglobin concentration can be explained by location.

Fig. 3. Residual geographical clustering of haemoglobin concentration in western Africa,a based on a multivariable linear regression model
Fig. 3. Residual geographical clustering of haemoglobin concentration in western Africa,<sup>a</sup> based on a multivariable linear regression model
a Includes Cameroon.b One decimal degree at the equator is approximately 111 km.
Fig. 4. Residual geographical clustering of haemoglobin concentration in eastern Africa,a based on a multivariable linear regression model
Fig. 4. Residual geographical clustering of haemoglobin concentration in eastern Africa,<sup>a</sup> based on a multivariable linear regression model
a Includes the Democratic Republic of the Congo, Lesotho and Swaziland.b One decimal degree at the equator is approximately 111 km.

Discussion

The raw DHS data indicate that anaemia in preschool-age children is a severe public health problem in sub-Saharan Africa countries, with most national prevalence estimates exceeding 40% (data available from corresponding author). These statistics underline the failure of national programmes and interventions to reduce the burden of anaemia in young children. One reason for the lack of success is that anaemia-control interventions are designed on the assumption that nutritional iron deficiency is the major cause of anaemia.1 In targeting communities with the highest prevalence of childhood anaemia within sub-Saharan Africa, it is useful to identify and geographically position anaemia-control resources, based on the relative importance of different anaemia contributors. Our approach addresses important operational constraints for anaemia control in the African continent by shedding new light on the distribution of anaemia severity within the countries studied. It also highlights the role of known environmental drivers of anaemia related to nutrition and infection, thus adding value to national-level summary statistics of the DHS data. We found considerable geographical variation in haemoglobin concentration in sub-Saharan Africa. Western Africa should receive priority, particularly those areas straddling the border between Burkina Faso and Mali, where most anaemia cases are moderate to severe.

Most DHS surveys include the collection of empirical information on factors that may contribute to childhood nutritional anaemia (e.g. micronutrient measurements, maternal haemoglobin concentration and place of residence). However, information on infectious causes of anaemia is not routinely collected. The collection of stool, urine and blood samples in the design of DHS surveys would make it possible to gather important epidemiological information on infectious and hereditary causes of anaemia. In the absence of comparable individual-level clinical data on these contributors, we adopted an ecological approach, using existing remotely-sensed environmental data as proxies of anaemia contributors. Nevertheless, we found that haemoglobin concentration is significantly associated with known environmental drivers of anaemia-causing parasitic infections and nutritional iron deficiency, such as distance to a perennial water body, elevation, land surface temperature and NDVI (Table 2). These effects were estimated to be greater in western Africa for NDVI and land surface temperature, and in eastern Africa for elevation. The results suggest that different environmental factors play varying roles in the anaemia burden in different geographical regions. This situation is exemplified by the different relationships between haemoglobin concentrations and the environmental variables distance to a perennial water body and elevation in western and eastern Africa (data available from corresponding author). These findings suggest that strategies for anaemia control should be tailored to local conditions while taking into account the specific etiology in a given location.

Previous approaches to describing the anaemia burden in Africa have typically been made at the national level with haemoglobin data from the field surveys available within a country, which are then extrapolated to the country as a whole. While such estimates are useful for advocacy and resource estimation at the national level, they are of limited practical relevance to the targeting of control efforts. Our results demonstrate that haemoglobin concentration is highly clustered geographically in both western and eastern Africa (Fig. 3 and Fig. 4). The size of the clusters and the tendency to cluster differ considerably between western (Fig. 3) and eastern Africa (Fig. 4), after taking into account the effect of environmental covariates. These findings highlight the non-stationary nature (i.e. spatial variation in spatial dependence) of the spatial processes leading to anaemia in preschool-age children throughout sub-Saharan Africa. Non-stationary spatial variation may occur because of human-induced environmental transformations, geographical variation of climate or topography, implementation of disease control, or the presence of different species or strains of parasites, intermediate hosts and vectors. Our results suggest that environmental drivers of anaemia-causing factors play differing roles in different regions of Africa.

The environmental variables in the haemoglobin concentration model account for only 13% of the geographical clustering in western Africa, but for 27% in eastern Africa. This result supports the suggestion that drivers of anaemia differ in these regions and that haemoglobin concentration, particularly in western Africa, is being driven by factors not accounted for in our models. Our results also indicate that, in western Africa, 9% of haemoglobin variance unexplained by environmental covariates is not related to location. Haemoglobinopathies and thalassemias are important inherited haematological conditions, particularly in western Africa,37 and could, in part, account for the remainder of haemoglobin variability. In addition, the difference in spatial effects presented in this study potentially reflects differences in food systems or possibly deterioration in food production driven by socioeconomic factors at smaller spatial scales. Overall, our findings reinforce the need for further studies to understand how different factors (hereditary, nutritional or infectious) affect anaemia burden at smaller spatial scales.

Our approach generated new knowledge of use for the design and implementation of more cost-effective control programmes for childhood anaemia, including nutrient supplementation and infectious disease control. First, we identified significant geographical variability in the severity of anaemia. This information will inform resource allocation for control of severe forms of anaemia, which requires strategies different from those needed for milder cases.3841 Second, we found that the effect of environmental drivers (e.g. anaemia-causing parasite infections and nutritional iron deficiency) on the burden of anaemia varies by region. This information will allow the identification of areas where micronutrient supplementation is likely to have side-effects.4244 An example is the increased severity of infectious disease linked to the delivery of iron supplementation in areas where parasitic infection is highly endemic. Third, we quantified geographical clustering within regions of sub-Saharan Africa; this is paramount for the development of modern cartographic resources that could be used as operational tools for targeting anaemia control. This information could be incorporated into anaemia risk maps that control for the major contributors to anaemia, to predict haemoglobin concentration (and possibly the prevalence of anaemia) in unsurveyed areas, potentially across the continent. To date, such maps have been created at subnational, national, regional and continental scales for malaria;22 at subnational, national and regional scales for neglected tropical diseases;23 at national level for malnutrition;45 and at continental level for thalassemias,24 but have yet to be produced for anaemia.

Our findings should be viewed in the light of the study’s assumptions and limitations. Environmental covariates were used as proxies for contributors to anaemia in preschool-age children. This approach provides a somewhat imprecise measurement of exposure to possible anaemia contributors and may therefore result in regression dilution bias, which can lead to underestimation of the observed effects.46 Although the observed relationships are biologically plausible, in the absence of individually collected data it is not possible to know to what extent the magnitude of the relationships represent an artefact introduced by ecological fallacy. Next, although the collated information on anaemia in preschool-age children is extensive in western (65% coverage) and eastern Africa (80% coverage), our maps suggest that for many areas of the continent little or no georeferenced data are available via the DHS. This is particularly the case in the central and southern African regions, including 5 countries (Kenya, Mozambique, Nigeria, South Africa and the Sudan) that are among the 10 most populated countries in sub-Saharan Africa (Fig. 1 and Fig. 2). As a way to provide meaningful estimates of geographical clustering across sub-Saharan Africa, we allocated available DHS data for central and southern Africa into the western and eastern African regions. This means that the estimates of geographical clustering are not necessarily representative of administrative divisions within the central and southern African regions. Nevertheless, those countries for which DHS data are currently unavailable constitute priority countries for obtaining more haemoglobin concentration data in future iterations of our approach, which could include literature searches. To facilitate studies such as this one, all future DHS should include georeferencing of communities.

The quantification of geographical variation in anaemia burden and in region-specific relationships with known drivers of major contributors to anaemia has allowed us to review the rationale underpinning the design and implementation of programmes for reducing anaemia in preschool-age children. Knowledge about the relative contribution of nutritional, infectious and hereditary causes of anaemia in different regions can help in the design of more cost-effective delivery of programmes that target these causes. Such programmes might include micronutrient supplementation, provision of fortified food, infectious disease control and transfusion services in sub-Saharan Africa.


Acknowledgements

We thank MEASURE DHS for granting permission to use the African DHS data sets under the project Spatial heterogeneity of anaemia in Sub-Saharan Africa.

Funding:

RJSM is funded by an International Research Award from the University of Queensland (#41795457). ACC is funded by an Australian National Health and Medical Research Council Career Development Award (#631619).

Competing interests:

None declared.

References

Share