Malaria hotspots and risk factors among children under-five years of age across eight West African countries: A geospatial analysis of DHS data

Sacla Aidé et al. MWJ 2025, 16:19 https://doi.org/10.5281/zenodo.17777419 RESEARCH ARTICLE | OPEN ACCESS! Malaria hotspots and risk factors among children under-five years of age across eight West African countries: A geospatial analysis of DHS data Edmond Sacla Aidé1*, Adama Kazienga2, Oyelola Adegboye3, Paul Sondo2, Halidou Tinto2 1 Laboratoire de Biomathématiques et d’Estimations Forestières, University of Abomey-Calavi, Benin. 2 Institut de Recherche en Sciences de la Santé/ Clinical Research Unit of Nanoro (IRSS-URCN), Burkina Faso. 3 Menzies School of Health Research, Charles Darwin University, Darwin, NT, Australia. *email: [email protected] Background: Malaria remains a significant public health challenge in sub-Saharan Africa, disproportionately affecting children under five years of age. Understanding the spatial distribution of malaria and its associated risk factors is essential for implementing effective, targeted control strategies. In this study, we investigated spatial variation and key determinants of malaria prevalence among children under five in eight West African countries. Methods: The study used the most recent Demographic and Health Surveys Data from eight West African countries where malaria infection status was determined by microscopy. Generalised Linear Mixed Models were first used to explore associations between malaria infection and sociodemographic predictors, accounting for survey design. These models were extended into Generalised Linear Geostatistical Models to incorporate spatial random effects. Malaria prevalence was predicted at a 10 × 10 km resolution, and exceedance probability maps were generated to identify high-burden areas with prevalence exceeding 30%. Model validation was done using empirical variograms, PIT histograms, and residual spatial analyses. Results: The use of insecticide-treated mosquito nets was significantly associated with reduced odds of malaria infection in four of the eight countries included in the study, while younger child age (<2 years) was consistently associated with lower risk across all countries. In addition, marked spatial heterogeneity in malaria prevalence was observed, with high predicted prevalence in Benin and Côte d’Ivoire and lower prevalence in Ghana and Liberia. Conclusion: This study highlights the importance of geospatial approaches for understanding malaria transmission dynamics in order to tailor malaria control measures to local context. The findings underscore the need to strengthen the effective use of insecticide-treated nets and community-level vector control, while improving spatial surveillance and data integration to support context-specific malaria interventions. INTRODUCTION Malaria remains a major public health threat, particularly in sub-Saharan Africa, where children under five are disproportionately affected. In 2023, malaria caused approximately 597,000 deaths globally, with children under five accounting for 76% of these deaths [1,2]. The WHO African Region carries the highest burden, contributing 94% of cases and 95% of malaria-related deaths worldwide [1,2]. Within West Africa, malaria continues to cause substantial morbidity and mortality despite widespread control efforts [3,4]. Factors such as climate change, increasing insecticide and drug resistance, and disruptions from the COVID-19 pandemic have contributed to the persistence of malaria in the region [3,4]. Malaria burden varies significantly across West African countries due to differences in socioeconomics characteristics, ecological conditions, health system capacity, and intervention coverage. For instance, in 2023, the highest malaria incidence rates were reported in Benin (363 per 1,000), Burkina Faso (353 per 1,000), Mali (346 per 1,000), and Guinea (307 per 1,000), whereas Senegal recorded the lowest (66 per 1,000) [5]. Understanding such national variations is essential for implementing targeted control strategies. Malaria control relies on a combination of interventions, including insecticide-treated nets MalariaWorld Journal | ISSN 2214-4374 1 November 2025, Vol. 16, No. 19 Sacla Aidé et al. MWJ 2025, 16:19 https://doi.org/10.5281/zenodo.17777419 (ITNs), indoor residual spraying (IRS), chemoprophylaxis, preventive chemotherapies, vaccination, and early diagnosis with prompt treatment [1,6]. However, in many West African countries, weak health information systems limit routine surveillance and evaluation, underscoring the need for reliable alternative data sources. Population-based surveys, such as the Demographic and Health Surveys (DHS), provide standardised national and subnational health indicators, including malaria infection confirmed by microscopy, and serve as valuable tools for evidence-based planning [7,8]. Previous studies in sub-Saharan Africa have identified several individualand household-level factors associated with malaria infection among children under five. These include child age, sex, maternal education, socioeconomic status, residence, and ITN usage [9,10]. Region-specific studies in West Africa confirm these associations: in Benin, high household density, low socioeconomic status, and poor bednet conditions increased infection risk [11]; in Burkina Faso, household wealth, maternal education, and ITN use were significant determinants [12,13]; in Côte d’Ivoire, infection was associated with child age, sex, and household socioeconomic status [14]; and in Ghana, residential area, child age, and mosquito net ownership influenced risk [15,16]. Multi-country analyses further highlight the consistent role of these determinants in shaping malaria burden [17]. Traditionally, these factors have been studied using logistic regression, multilevel, or mixed-effects models [14,17,18]. While informative, such approaches often ignore spatial dependence, limiting their ability to identify high-risk areas and guide geographically targeted interventions. Spatial correlation is particularly important in regions with heterogeneous ecological and sociodemographic conditions, where malaria transmission dynamics vary across small geographic scales. Geospatial modelling approaches, particularly model-based geostatistics, address this limitation by incorporating spatial correlation, environmental covariates, and uncertainty to generate high-resolution risk maps [19,20]. These models not only reveal spatial heterogeneity but also facilitate subnational evaluation of intervention impact, supporting more efficient and equitable malaria control strategies [21]. Previous applications of geostatistical models in malaria research have successfully identified high-risk areas and associations with environmental, demographic, and socioeconomic factors [22-24]. Despite these advances, few multi-country geostatistical studies using microscopy-confirmed DHS data have been conducted in West Africa. This gap limits understanding of spatial heterogeneity in malaria prevalence among children under five and constrains the targeting of interventions in the region. This study addressed this gap by applying model-based geostatistics to identify high-risk areas and determinants of malaria infection across eight West African countries. The objectives of this study were to: (i) estimate malaria prevalence among children under five years of age; (ii) identify individualand household-level determinants of infection; and (iii) map spatial heterogeneity and detect high-risk areas to guide targeted malaria control efforts. METHODOLOGY Study design and population The latest Demographic and Health Survey (DHS) data for thirteen West African countries was used. These DHS surveys employed a two-stage stratified cluster sampling design to ensure national representativeness. In the first stage, enumeration areas (clusters) were randomly selected from national census sampling frames. In the second stage, households were systematically sampled from the selected clusters. Household and individual-level weights were applied to adjust for survey design effects and ensure the sample's representativeness at national and sub-national levels. Also, countries were divided into strata based on urban and rural areas, with further stratification. The study population was children under five living in sampled households in the DHS survey of the selected countries. This study initially considered the most recent Demographic and Health Surveys (DHS) data from thirteen West African countries: Benin, Burkina Faso, Côte d’Ivoire, Ghana, Gambia, Guinea, Liberia, Sierra Leone, Senegal, Togo, Niger, Nigeria, and Mali. However, only eight countries were retained in the final analysis (Benin, Burkina Faso, Côte d’Ivoire, Ghana, Guinea, Liberia, Nigeria, and Togo). The selection was based on the availability of malaria infection status MalariaWorld Journal | ISSN 2214-4374 2 November 2025, Vol. 16, No. 19 Sacla Aidé et al. MWJ 2025, 16:19 https://doi.org/10.5281/zenodo.17777419 determined by microscopy. Using microscopybased malaria infection data provides a standardised and highly specific measure of Plasmodium falciparum infection. This approach is particularly important in multi-country analyses, where consistency in diagnostic methods enhances the comparability and validity of spatial and statistical modelling results. The five excluded countries (Gambia, Sierra Leone, Senegal, Niger, and Mali) were omitted because microscopy results were either unavailable, incomplete, or based solely on RDTs, preventing harmonised data analysis across all countries (Table 1). The outcome variable was the microscopy-confirmed Plasmodium falciparum infection status among children aged 6–59 months living in sampled households. For each survey cluster i, the number of malaria-positive children (yi) out of the total examined (ni) was recorded, with N denoting the total number of clusters. Thus, malaria prevalence at each cluster location (xi) was defined as yi/ni. The selection of covariates was informed by previous studies on malaria risk factors among children under five and by the availability of comparable variables across all eight DHS datasets [19,33]. Only individualand household-level variables consistently measured in the surveys were included, namely: child age, child sex, maternal education, maternal age, household head sex, and bednet ownership or use. These variables have been widely reported as major sociodemographic determinants of malaria risk. Other relevant factors such as housing characteristics, environmental, and climatic covariates were not incorporated due to their absence or inconsistency in the DHS datasets. Future analyses integrating external geospatial datasets could help capture these ecological influences on malaria transmission dynamics more comprehensively. In addition to these sociodemographic data, geographical coordinates were recorded for each survey cluster, allowing for spatial analysis of malaria prevalence and associated risk factors. To maintain confidentiality, DHS applies a displacement procedure to the cluster coordinates, ensuring anonymity while preserving spatial accuracy for large-scale epidemiological studies. An overview of the data processing and analytical workflow, from DHS data acquisition to geostatistical modelling and visualisation, is presented in Figure 1. Geostatistical model A Generalised Linear Mixed Model (GLMM) using the package survey [4] was first employed to investigate the factors associated with malaria infection in children under five. The objective was to determine which covariates, within a given set, showed a statistically significant association with malaria infection. This model accounted for both fixed and random effects to ensure robust estimates in the presence of hierarchical survey data. This survey design included stratification, clustering, and sampling weights to ensure unbiased population-level inference. Let Yi denote the number of microscopy-confirmed malaria-positive children out of ni examined at location xi, where i=1,…,N. The malaria prevalence at location xi is defined as p(xi)=Yi/ni. The number of positive cases is assumed to follow a binomial distribution: Yi∼Binomial(ni,p(xi)) The linear predictor is specified as: MalariaWorld Journal | ISSN 2214-4374 3 November 2025, Vol. 16, No. 19 Table 1. Reasons for omitting countries in the analysis. Sacla Aidé et al. MWJ 2025, 16:19 https://doi.org/10.5281/zenodo.17777419 logit(p(xi))=d(xi)⊤ β + S(xi) + Zi where d(xi) is the vector of covariates at location xi, β represents fixed-effect coefficients, S(xi) is a spatially structured Gaussian process capturing spatial correlation, and Zi is an independent random effect (nugget effect) accounting for small-scale variability. The spatial process S(xi) is assumed to follow a zero-mean stationary Gaussian process with covariance function: Cov(S(xi),S(xj)) = σ2 exp(-∥xi-xj∥/ϕ) where σ2 is the spatial variance and ϕ is the range parameter that controls the rate of spatial correlation decay. Model parameters β, σ2, ϕ and τ2 were estimated using the Monte Carlo Maximum Likelihood (MCML) method as implemented in the PrevMap R package [35,36]. Unlike Bayesian approaches, MCML does not require prior specification for these parameters, treating them as unknown constants estimated directly from the data. Predictions were performed on a regular grid with a spatial resolution of 10 × 10 km to generate spatially continuous malaria prevalence estimates. This resolution was chosen as a pragmatic balance between spatial detail and the positional uncertainty of DHS cluster coordinates (DHS implements random displacement of cluster locations urban up to 2 km, rural up to 5 km, with a small proportion displaced up to 10 km), which limits the meaningful spatial precision of survey-based prevalence estimates. A 10-km pixel therefore avoids over-interpreting fine-scale patterns that the input data cannot support, while remaining sufficiently fine to identify subnational hotspots relevant for programmatic decision-making. In addition, the 10-km grid offered a computationally tractable number of prediction locations for Monte Carlo maximumlikelihood estimation and yielded a spatial support compatible with commonly used environmental covariates. This predictive prevalence was obtained as the posterior mean of p(x) at each grid location, incorporating both spatially structured and non-structured effects. To this end, exceedance probability (EP) represents the probability that malaria prevalence exceeds a predefined threshold to identify high-risk areas. This threshold was fixed at 30%, indicating that areas where the predicted malaria prevalence had a probability of exceeding 30% were classified as high-risk regions. This prevalence threshold was used to delineate areas with a high malaria burden, warranting intensive and sustained vector control interventions, as supported by previous studies [25,26]. These predicted prevalence and exceedance probability maps were generated to highlight regions with a high burden of malaria visually. These maps provide valuable insights for targeted malaria control interventions, allowing public health officials to allocate resources more effectively and implement strategic prevention measures. To ensure comparability across MalariaWorld Journal | ISSN 2214-4374 4 November 2025, Vol. 16, No. 19 Figure 1. Data processing and workflow analysis for malaria risk mapping across eight West African countries. The flowchart summarises the sequential steps undertaken in this study. The workflow highlights the integration of nationally representative microscopy-based DHS data with spatial modelling to identify malaria hotspots and support evidence-based targeting by national malaria control programmes. Sacla Aidé et al. MWJ 2025, 16:19 https://doi.org/10.5281/zenodo.17777419 countries, a consistent legend scale was applied to all maps. The performance of the fitted geostatistical models was evaluated using three complementary diagnostics: (i) the empirical variogram of residuals to assess whether residual spatial correlation was adequately captured; (ii) the probability integral transform (PIT), which compares predicted probabilities with observed outcomes to evaluate model calibration; and (iii) the spatial distribution of standardised residuals to identify any remaining spatial patterns unexplained by the model. A schematic overview of the validation workflow is provided in Figure 2 for clarity. The empirical variogram of these residuals was then computed and compared against a 95% confidence envelope derived from permutation tests. If residual spatial correlation remained, it suggested that important spatially structured factors influencing malaria prevalence had not been accounted for, warranting further model refinement. Software Data management and variable recording were conducted using Stata 17 (StataCorp, 2021). All statistical analyses were performed in R version 4.3.1 (R Core Team, 2023), with statistical significance set at p < 0.05. Generalised Linear Mixed Models (GLMMs), accounting for complex survey design, were implemented using the survey package [4]. Parameter estimation, spatial prediction, and model validation for the geostatistical models were conducted using the PrevMap package [27,28]. Ethical considerations This study utilised publicly available secondary data from the Demographic and Health Surveys (DHS) Programme. The DHS surveys follow rigorous ethical protocols approved by the ICF Institutional Review Board and by the national ethics committees of each participating country. Informed consent was obtained from all participants prior to data collection. The datasets are fully anonymised and contain no identifiable information on survey respondents. Prior to accessing the data, the corresponding author registered on the DHS Program website and was granted permission to download and use the datasets for this research. Data access and ethical guidelines are publicly available at https://dhsprogram.com/data/. RESULTS Malaria infection risk factors for children < 5 yrs Out of the 13 countries surveyed, only eight had complete malaria infection status based on microscopy and were therefore included in the analysis. Table 2 presents the estimated associations between selected demographic and household-level factors and malaria infection among children under five across these eight countries in West Africa. Child sex, maternal education, and the mother's age were not significantly associated with malaria MalariaWorld Journal | ISSN 2214-4374 5 November 2025, Vol. 16, No. 19 Figure 2. Schematic overview of the model validation workflow. The diagram illustrates key validation steps used to assess the performance of the geostatistical model: (1) computation of empirical variograms of residuals; (2) PIT-based model calibration check; (3) examination of standardised residuals for remaining spatial structure; and (4) interpretation of diagnostics for model adequacy. Sacla Aidé et al. MWJ 2025, 16:19 https://doi.org/10.5281/zenodo.17777419 infection in any of the countries included. Similarly, sex of household head showed no association with malaria risk, except in Côte d’Ivoire, where children living in male-headed households had significantly higher odds of malaria infection (OR = 1.30, p = 0.029). In addition, the use of insecticide-treated bednets was significantly associated with reduced malaria risk in four of the eight countries (BurkinaFaso, Benin, Ivory Coast, and Ghana). For instance, children who slept under treated nets (such as MalariaWorld Journal | ISSN 2214-4374 6 November 2025, Vol. 16, No. 19 Table 2. Factors associated with the prevalence of malaria in children < 5 yrs of age in eight West African countries (CI = 95% confidence interval; HH = Head of Household). Table 3. National prevalence of malaria according to DHS surveys (CI = 95% confidence interval). Sacla Aidé et al. MWJ 2025, 16:19 https://doi.org/10.5281/zenodo.17777419 long-lasting insecticidal nets) had 28% lower odds of malaria infection (OR = 0.72, p < 0.001) compared to those who did not use a net. In contrast, the age of the child was significantly associated with malaria infection across all eight countries. For example, children under the age of 2 years had 40 to 63% lower odds of malaria infection compared to those above 2 years old. National prevalence of malaria infection National-level malaria prevalence based on microscopy was estimated for each of the eight included countries, with corresponding 95% confidence intervals as shown in Table 3. Overall, there was significant geographical variation in malaria prevalence across the eight countries, reflecting differences in transmission intensity, environmental conditions, and control efforts. For instance, the highest prevalence was recorded in Benin (39.73%, 95% CI: 38.49%–41.00%), followed by Togo (29.83%, 95% CI: 28.25%–31.46%), whereas Ghana had the lowest prevalence (9.96%, 95% CI: 9.10%–10.89%). Malaria prevalence at sampled cluster locations The prevalence at sampled cluster location was estimated among children under five years across the eight countries, as shown in Figure 3. The prevalence at the sampled location exhibits substantial spatial heterogeneity. In countries such as Burkina Faso, Ghana, and Guinea, most sampled clusters exhibited low to moderate malaria prevalence levels, with prevalence frequently below the 30% MalariaWorld Journal | ISSN 2214-4374 7 November 2025, Vol. 16, No. 19 Figure 3. Observed malaria prevalence at DHS cluster locations across eight West African countries. Each point represents the microscopy-confirmed malaria prevalence among children under five at the DHS cluster level. Source: Demographic and Health Surveys (DHS), 2017–2022; authors’ analysis. Burkina Faso N 0 100 200 300 km Benin N 0 50 100 km Cote d'Ivoire N 0 100 200 300 km Ghana N 0 50 100 150 km Guinea Malaria coverage (%) 0−20 % 20−40 % 40−60 % 60−80 % 80−100 % N0 100 200 km Liberia N 0 50 100 150 200 km Nigeria N 0 200 400 600 km Togo N 0 10 20 30 40 50 60 km Sacla Aidé et al. MWJ 2025, 16:19 https://doi.org/10.5281/zenodo.17777419 MalariaWorld Journal | ISSN 2214-4374 8 November 2025, Vol. 16, No. 19 Figure 4. Test for spatial autocorrelation of malaria prevalence among children under five across eight West African countries. The figure shows the results of spatial autocorrelation analysis using Moran’s I statistic applied to cluster-level malaria prevalence derived from DHS data. A positive and statistically significant Moran’s I indicates spatial clustering of malaria risk across the study area. Source: Demographic and Health Surveys (DHS), 2017–2022; authors’ analysis. Sacla Aidé et al. MWJ 2025, 16:19 https://doi.org/10.5281/zenodo.17777419 threshold. In contrast, Benin, Côte d'Ivoire, Nigeria, and Togo present wider variations, with several clusters showing higher malaria prevalence (above 40%), particularly in southern Benin and parts of Nigeria and Togo. Meanwhile, Liberia exhibits relatively moderate malaria prevalence across most of its regions, with limited pockets of very high prevalence. MalariaWorld Journal | ISSN 2214-4374 9 November 2025, Vol. 16, No. 19 Figure 5. Model validation of Bayesian geostatistical predictions of malaria prevalence. Posterior predictive checks comparing observed and predicted malaria prevalence at DHS cluster locations. The figure demonstrates the model’s goodness of fit and reliability of spatial predictions, indicating that the Bayesian geostatistical model accurately captures observed spatial patterns. Source: Demographic and Health Surveys (DHS), 2017–2022; authors’ analysis.