Malaria prevalence and its determinants across 19 sub-Saharan African countries: a spatial and geographically weighted regression analysis

Abstract

Background

Malaria remains a significant public health challenge, particularly in underdeveloped regions like sub-Saharan Africa, where environmental, housing, and socioeconomic factors drive its spread. This study aims to identify spatial patterns and key determinants of malaria infection among households across 19 sub-Saharan African countries to inform targeted interventions and policy strategies.

Methods

A community-based cross-sectional study was conducted using recent Demographic and Health Survey (DHS) data from 19 sub-Saharan African countries, encompassing 126,424 households and 11,594 clusters. Data processing including weighting, cleaning, and analysis was carried out using Microsoft Excel and Stata version 17. Prevalence estimates and 95% confidence intervals were generated in Stata, accounting for the DHS's complex sampling design through the application of weights, clustering, and stratification. Spatial analyses, including cluster detection and Geographically Weighted Regression (GWR), Were conducted using ArcGIS version 10.7 and SaTScan^™ version 10.2.

Results

Malaria prevalence among households in 19 sub-Saharan African countries was 22.47% (95% CI 22.24%, 22.70%), based on weighted estimates that account for the DHS sampling design. This indicates that approximately one in five households is affected by malaria. Spatial autocorrelation was significant (Global Moran’s I = 0.159; Z = 239.1; p < 0.001), confirming geographic clustering. Hot-spot analysis (Getis-Ord Gi*) highlighted hotspot zones in Benin, Burkina Faso, Togo, Uganda, Rwanda, parts of the Republic of the Congo, and Mozambique. SaTScan™ identified 34 statistically significant spatial clusters, with the most prominent situated in Ghana, Burkina Faso, Togo, and Benin; Anselin Local Moran’s I further revealed intermingled high and low-risk areas. Geographically Weighted Regression showed higher malaria prevalence in rural residents; households with rudimentary or natural roofs; younger heads of household; the poorest wealth quintile; no bed-net ownership, homes using treated bed nets, and large household size (6–12 members). Conversely, risk was lower in the richest households, those headed by women, and dwellings with natural or rustic walls.

Conclusion

Malaria remains highly prevalent (22.47%) in sub-Saharan Africa, with significant spatial clustering in countries like Benin, Burkina Faso, Togo, and Uganda. Key risk factors identified include rural residence, poor housing conditions, lack of bed nets, homes using treated bed nets, and lower socioeconomic status. To reduce the burden, targeted interventions such as the distribution of insecticide-treated bed nets, indoor residual spraying, health education and improved housing should focus on identified hotspot areas. Collaboration among governments, NGOs, and local communities is essential to implement these strategies effectively and meet malaria reduction goals by 2030.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12936-025-05573-6.

Background

Malaria continues to pose a significant public health threat worldwide, with its impact most severe in sub-Saharan Africa (SSA) [1]. The disease is endemic in over 100 countries and places nearly half of the global population at risk [2]. It is transmitted through the bite of infected female Anopheles mosquitoes, carrying one of five species of Plasmodium, among which Plasmodium falciparum is the most virulent, often resulting in severe complications and high mortality [3–5]. In 2023 the World Health Organization reported 263 million cases and 597 000 deaths worldwide, with Africa bearing 94% of cases and 95% of deaths [6].

A recent multilevel analysis of 13 national Malaria Indicator Surveys in SSA, involving over 60,000 children aged 6–59 months, reported a pooled malaria prevalence of 27.4%, with national rates ranging from 5% in Senegal to over 62% in Sierra Leone [7]. Furthermore, over half of the region’s malaria burden is concentrated in six countries such as Nigeria, the Democratic Republic of the Congo, Uganda, Mozambique, Angola, and Burkina Faso illustrating the disproportionate and persistent nature of the disease in the region [8]. Despite the progress gained in reducing malarial morbidity and mortality, the disease remains a major public health problem in many countries of SSA. This emphasizes how malaria is a public health concern in SSA.

In SSA, malaria transmission remains elevated in part due to the growing spread of insecticide resistance in major vectors like Anopheles gambiae and Anopheles funestus. Pyrethroid-resistant mosquito populations have been documented across West, East, and Central Africa, with resistance intensity regularly surpassing WHO thresholds (< 90% mortality) in countries such as Kenya [9], Ghana [10], and the Democratic Republic of the Congo [11]. For example, coastal Kenya has reported low mosquito mortality rates to permethrin and deltamethrin in multiple An. gambiae sites [9], and southern regions of Ghana have seen escalating resistance in An. gambiae and An. funestus that compromise the efficacy of long-lasting insecticidal nets [10]. This widespread resistance diminishes the lethality of insecticide-treated nets and indoor residual sprays, leading to more mosquito bites, higher parasite transmission, and persistent malaria prevalence. In addition favourable climate, and environmental factors also increase vector persistence [12, 13].

Global programmes aimed at increasing access to malaria prevention, diagnosis, and treatment have increased during the past 20 years, particularly in SSA countries where malaria is disproportionately prevalent [1, 14, 15]. Funding for malaria control, for instance, more than doubled between 2005 and 2014, going from US$ 960 million to US$ 2.5 billion [15]. 10.6 million malaria deaths and 1.7 billion infections have been prevented between 2000 and 2020 [8]. These improvements in malaria morbidity and mortality, especially in SSA, have been attributed to the increased use of effective malaria prevention and control measures, including insecticide-treated nets (ITNs), increased indoor residual spraying (IRS), the introduction of rapid diagnostic tests (RDTs) at community levels, artemisinin-based combination therapy (ACT), environmental management, larval and vector control techniques [16, 17].

Previous studies identified several key determinants of malaria in SSA. These factors are classified into three main domains. Socioeconomic factors including household wealth and size, the age and sex of the household heads, and maternal or caregiver education strongly influence exposure and preventive practices [18–22]. Environmental determinants such as housing quality (wall, roof, and floor materials), indoor residual spraying, altitude, and local climate conditions affect mosquito breeding and survival [23–26]. Healthcare-access factors such as ownership and consistent use of insecticide-treated nets (ITNs) and physical or financial access to health services further modulate infection risk [27–29]. The persistently high malaria burden in SSA has also been linked to broader systemic challenges, including widespread poverty, growing resistance in parasites and mosquitoes, insufficient political will, inadequate financing mechanisms, and the slow implementation of elimination strategies [30, 31]. Understanding how these factors interact is critical for designing effective policies and programmes to combat malaria across the region.

Achieving the WHO Global Technical Strategy target of a 90% reduction in malaria incidence and mortality by 2030 [31] hinges on understanding how malaria and risk factors vary across space, particularly in SSA. While numerous studies have catalogued household-level determinants, the strength and direction of these associations shift from one locality to another remains underscored. Geographically Weighted Regression (GWR) is well suited to capture such spatial non-stationarity, yet it has rarely been applied in a multi-country context. Therefore, this study aims to identify spatial patterns and local determinants of malaria using spatial and GWR analysis. Demographic and Health Survey data from 19 SSA countries were used to (i) map household malaria prevalence, (ii) its spatial variation and (iii) apply GWR to identify location-specific predictors of malaria.

Methods

Study setting, and period

This study was conducted in 19 SSA countries using the most recent Demographic and Health Survey (DHS) comprising the Household Member Record dataset (PR file). A total of 19 sub-Saharan African countries namely Angola, Benin, Burkina Faso, Burundi, Cameron, Congo Republic, Côte d’Ivoire, Gabon, Gambia, Ghana, Madagascar, Mali, Mauritania, Mozambique, Nigeria, Rwanda, Tanzania, Togo, and Uganda were included (Fig. 1). This region includes savanna, rainforests, and mountain ranges. Approximately forty-nine countries, including Nigeria, Ethiopia, Kenya, and South Africa, make up this region that stretches over 9.3 million square miles (24 million square kilometres) in area [32, 33]. Recently, it was estimated that the population here exceeds 1.1 billion people, out of which, a large percentage is below the age of 20 [34].

Fig. 1.

The study area map, shape file for the study area map was taken from GADM and plotted using ArcGIS V. 10.7 software

Data source and study population

The data was obtained from the DHS program at www. dhsprogram.com. This study utilized nationally representative data from the Demographic and Health Surveys (DHS), conducted in 19 SSA countries. The DHS program collects is to collect a large, representative sample size data using standardized sampling procedures, structured questionnaires, and consistent coding protocols across countries, ensuring comparability and methodological rigor [35]. It measures key indicators that are considered important, which enables countries to generate data useful to inform policymakers. The data used in this study ranges from 2013 to 2023. The study population included all households from the selected enumeration areas (clusters) in both urban and rural settings that were included in the analysis.

Sample size determination and sampling method

To assure national representativeness, the survey was based on a two-stage sampling technique. In the first stage, the selection of proportional clusters/enumeration areas was performed using each country’s most recent population and housing census as a sampling frame. In the second step of sampling, a predetermined number of homes from the targeted EAs were chosen using the systematic sampling approach. Equal probability systematic sampling was used after the household listing to choose a specific number of homes from the designated cluster [36]. In each selected household, an eligible woman aged 15–49 years was interviewed using a standardized individual questionnaire, following the DHS protocol. The sample was weighted to improve representativeness and yield results that are more precise. Following DHS guidelines, women’s sample weight (v005) were used for the analysis. This weight accounts for the household weight adjusted by the inverse of the household response rate within each sampling stratum. The raw sampling weights (v005) Were normalized by dividing by 1,000,000, as recommended, to produce appropriate estimates [37]. A complete case analysis, where records with missing values for key variables Were excluded from the final analysis to avoid potential bias. The analysis included data from 11,594 clusters and a weighted sample of 126,424 households that underwent malaria rapid diagnostic testing across 19 SSA. In total, a weighted sample of 126,424 individuals underwent malaria rapid diagnostic testing. This figure refers to eligible individuals sampled within households, typically children under the age of five, in accordance with DHS protocols. Depending on the country survey, testing may also have included older children or women of reproductive age. Therefore, the 126,424 figure reflects the number of individuals tested for malaria, not the number of households surveyed (Table 1).

Table 1.

19 Sub-Sahara African countries demographic and Health Surveys with their respective sample size, and survey year

Countries	Year	Total Cluster Number	Sample size	Frequency (%)
Angola	2015–2016	627	8,319	6.58
Benin	2017–2018	555	6,249	4.94
Burkina Faso	2021	600	5,767	4.56
Burundi	2016–2017	554	6,913	5.47
Cameron	2022	469	4,733	3.74
Congo Republic	2013–2014	540	8,219	6.50
Côte d’Ivoire	2021	539	4,995	3.95
Gabon	2019–2021	390	5,982	4.73
Gambia	2019–2020	281	3,675	2.91
Ghana	2022	618	4,546	3.60
Madagascar	2021	657	5,874	4.65
Mali	2018	379	4,348	3.44
Mauritania	2019–2021	1210	10,144	8.02
Mozambique	2022–2023	619	4,283	3.39
Nigeria	2018	1400	11,197	8.86
Rwanda	2019	500	11,069	8.76
Tanzania	2022	629	5,237	4.14
Togo	2013–2014	330	3,868	3.06
Uganda	2016	697	11,006	8.71
Total		11,594	126,424	100

Malaria diagnostic methods in DHS surveys

The analysis used DHS/MIS malaria data from 19 sub-Saharan African countries (Angola, Benin, Burkina Faso, Burundi, Congo Republic/DRC, Côte d’Ivoire, Gambia, Ghana, Madagascar, Mali, Mozambique, Nigeria, Rwanda, Tanzania, Togo, and Uganda). All surveys employed HRP2-based RDTs specific for P. falciparum, with microscopy confirmation (i.e., thick blood smear examination) conducted in all datasets. Combo HRP2/pan-pLDH (Pan-specific Plasmodium Lactate Dehydrogenase) RDTs were used in seven countries (Angola, Burundi, Gambia, Madagascar, Mozambique, Tanzania, and Togo), while the remaining used HRP2-only RDTs. pan-pLDH detects all human malaria species (P. falciparum, Plasmodium vivax, Plasmodium malariae, Plasmodium ovale) and more useful in regions with mixed species infections. Brands included SD Bioline, Paracheck, CareStart, and First Response [38–40] (Table 2).

Table 2.

Malaria diagnostic methods used in DHS surveys across included countries

Country	RDT used	Microscopy used	RDT type	RDT brand (when known)
Angola	Yes	Yes	HRP2/pan-pLDH	SD Bioline
Benin	Yes	Yes	HRP2	Paracheck
Burkina Faso	Yes	Yes	HRP2	Paracheck
Burundi	Yes	Yes	HRP2/pan-pLDH	SD Bioline
Cameroon	Yes	Yes	HRP2/pan-pLDH	SD Bioline
Congo Republic	Yes	Yes	HRP2	SD Bioline
Côte d’Ivoire	Yes	Yes	HRP2	SD Bioline
Gabon	Yes	Yes	HRP2	—
Gambia	Yes	Yes	HRP2	SD Bioline
Ghana	Yes	Yes	HRP2	—
Madagascar	Yes	Yes	HRP2/pan-pLDH	CareStart
Mali	Yes	Yes	HRP2	Paracheck
Mauritania	Yes	Yes	HRP2	—
Mozambique	Yes	Yes	HRP2/pan-pLDH	SD Bioline
Nigeria	Yes	Yes	HRP2	Paracheck
Rwanda	Yes	Yes	HRP2	First Response
Tanzania	Yes	Yes	HRP2/pan-pLDH	SD Bioline
Togo	Yes	Yes	HRP2/pan-pLDH	First Response
Uganda	Yes	Yes	HRP2	Paracheck

Variables of the study

Outcome variable: The outcome variable was malaria. The status was determined using country-specific combinations of RDTs and microscopy per DHS protocols. Malaria infection were defined at the household level, by classifying a household as positive if ≥ 1 individual tested positive for Plasmodium spp. using RDT (or microscopy where applicable). It was measured as “Yes” for households with positive malaria rapid test, was coded as “1” while “No” for those with negative malaria rapid test, and was coded as “0”. This binary classification (positive vs. negative) was used to assess the burden of malaria at the household level and is supported by prior DHS-based research [41, 42].

Explanatory variables: The following factors were taken into consideration as potential independent variables for the spatial regression models. Sociodemographic variables such as HHH's sex (male and female), HHH's age (10–24, 25–34, 35–49, and 50⁺), educational status (no education, primary education, and secondary/higher education), household size (< = 5, 6–12 and > 12), wealth index (poorest, poorer, middle, richer and richest). Housing-related variables such as presence and type of bed net (no bed net, treated bed net, and untreated bed net), housing quality variables: roof, wall, and floor materials were classified as rudimentary/natural, finished, and other. Environmental variables such as water source (improved and unimproved), altitude (< = 1000 and 1000⁺) [43, 44], and place of residence (urban or rural).

Although household heads under the age of 18 are rare in DHS data, a small number of such cases Were identified in the pooled data set across the 19 SSA countries. These may reflect unique household structures, such as child-headed households in settings affected by conflict, orphanhood, or displacement. These cases were retained to maintain data completeness and better represent real-world demographic variation.

For the analysis, the altitude was categorized as a binary variable using a 1000 m elevation cut-off to distinguish between lowland and highland areas. This threshold is supported by evidence from multiple studies in sub-Saharan Africa showing a sharp decline in malaria transmission above 1000 m due to decreased vector survival, lower temperatures, and reduced parasite development rates [43, 45]. A dichotomous classification facilitates comparability with previous DHS-based malaria studies and avoids the risk of over fitting from sparse data in higher altitude ranges, while still capturing the ecologically meaningful impact of elevation on malaria risk.

Data management and analysis

The statistical program Stata version 17 was used to extract data, code, and conduct descriptive analysis. On the other hand, ArcGIS 10.7 was used to do the geographical analysis. Sampling weights provided by the DHS were applied to account for the complex survey design, including unequal probabilities of selection and non-response, ensuring that the results are nationally representative. Weighted proportions for predictor variables and malaria outcomes were computed using Stata and subsequently exported to ArcGIS for spatial analysis. SaTScan™ was used for spatial scan statistical analysis.

Spatial analysis

Spatial autocorrelation refers to the tendency of nearby geographic areas to exhibit similar values. It states that areas that are closer together are more connected than those that are farther apart [46]. Global spatial autocorrelation, measured using Moran’s I, quantifies the overall degree of clustering in the data and display the pattern whether values are dispersed, random, or clustered [47]. The Moran's I value ranges from 1 to − 1 [48]. The event is scattered when the Moran's I value is close to − 1; however, it indicates that the event is clustered when it is close to 1, and a score of zero indicates no clustering. A statistically significant Moran's I (p < 0.05) confirms the existence of spatial autocorrelation [49].

Mathematically, Moran's I can be calculated as: [50, 51]

$$I=\frac{N}{W}\frac{{\sum }_i^n{\sum }_j^n{w}_{\mathit{ij}}\left(y_i,-,\overline{y}\right)\left(y_j,-,\overline{y}\right)}{({\sum }_i^n{\sum }_j^n{w}_{\mathit{ij}}){\sum }_i\left(y_j,-,\overline{y}\right)2}$$

where N is the number of spatial units, y_i and $y_j$ represent the malaria prevalence values in locations i and j, $\overline{y}$ is the mean prevalence, w_ij is the spatial weight between locations i and j, and W is the sum of all spatial weights.

Hot spot and outlier cluster analysis

The Getis-Ord Gi* statistic was used to identify spatial clusters (hot and cold spots) of malaria infection across 19 SSA countries [49]. This method compares the local sum of an attribute and its neighbours to the global sum to determine areas of significant clustering [47]. A Z-score was calculated to assess statistical significance at p < 0.05 (95% CI). High positive Z-scores indicated significant hot spots (high clustering), while high negative Z-scores indicated cold spots (low clustering). Results were interpreted under the null hypothesis that malaria is randomly distributed. A large absolute Z-score with a small p-value led to rejection of the null hypothesis, confirming spatial clustering.

Cluster and Outlier Analysis using Anselin Local Moran I in ArcGIS was conducted. Household-level data were spatially referenced and analyzed, classifying locations into high-high (HH) clusters, low-low (LL) clusters, high-low (HL) outliers, and low–high (LH) outliers. HH and LL clusters indicated consistent high or low rates, while HL and LH outliers revealed deviations from surrounding patterns [52, 53].

Spatial scan statistical analysis

The possible statistically significant high-rate clusters of malaria were examined using a spatial scan statistical analysis utilizing the Bernoulli model by Kulldorff SaTScan™ version 10.0.2. Individuals who Were positive for malaria Were regarded as cases, while Individuals who were negative for malaria were regarded as controls. To identify both big and small clusters, higher boundary of 50% of the population was set for the spatial cluster size [54]. It was possible to identify both the primary and secondary clusters. 999 Monte Carlo simulations were used to determine each cluster's log-likelihood ratio (LLR) and corresponding P-value; clusters with a p-value of less than 0.05 were considered statistically significant [55].

Spatial regression analysis

Spatial regression incorporates both global and local modelling approaches to examine geographic variation in relationships between predictors and outcomes [56–59]. In this study, first a global spatial regression model was applied to assess overall associations, followed by a local geographically weighted regression (GWR) to capture spatial heterogeneity in predictor effects across enumeration areas [60, 61]. Prior to model estimation, exploratory regression and diagnostic tests was conducted to ensure model appropriateness. The six assumptions suggested for spatial regression such as spatial non-stationarity, independence of observations, normality of residuals, homoscedasticity, low multicollinearity, and optimal bandwidth selection were investigated using the appropriate tests to guarantee the validity of the regression models [58, 62]. For instance, the Jarque–Bera test was used to assess the normality of residuals. Predictor variables with a p-value less than 0.05 were considered statistically significant, and their coefficients were interpreted accordingly.

Ordinary least squares (OLS) regression

Ordinary Least Squares (OLS) regression was used to examine the influence of socioeconomic, housing, and environmental factors on malaria prevalence after identifying spatial hotspots. The reliability of OLS depends on several assumptions: linearity of relationships, independence of errors, homoscedasticity (constant variance of residuals), absence of multicollinearity among explanatory variables (checked via Variance Inflation Factor (VIF) < 7.5), and normally distributed residuals without spatial autocorrelation. The residuals must be free from spatial autocorrelation [63]. The Koenker Breusch-Pagan (BP) test was applied to assess heteroscedasticity and spatial stability, determining whether OLS was appropriate or if a Geographically Weighted Regression (GWR) model was needed. Model fit was evaluated using the adjusted R2, with higher values indicating better explanatory power and assumption compliance [52, 64, 65]. Variables were retained based on both statistical relevance and theoretical importance, as informed by prior literature.

The OLS regression equation is given as: [66]

$$\text{Yi}=\mathrm{\beta }+\sum _{K=1}^p({\beta }_k{X}_{\mathit{ik}})+{ϵ}_i$$

where y_i is the outcome variable for observation i,; β₀, β₁, β₂, …β_p are the model parameters, X_ik are explanatory variables, and ∈ ₁, ∈ ₂, … ∈ _n are the error term/residuals with zero mean and homogenous variance σ².

Geographically weighted regression (GWR)

Geographically Weighted Regression (GWR) was employed to capture spatial non-stationarity in the relationship between malaria prevalence and its predictors. Whereas Ordinary Least Squares (OLS) fits a single global equation, GWR calibrates a separate local regression for each DHS cluster, weighting nearby observations more heavily than distant ones [67]. As a result, the GWR coefficient varies for every cluster [68]. The GWR is used to create maps of the coefficients linked to each explanatory variable [68], which offer recommendations for focused interventions. The GWR model was calibrated using a bi-square kernel function and an adaptive bandwidth selection approach, which adjusts the bandwidth to include a consistent number of neighbouring clusters across the study area. Bandwidth optimization was performed using Akaike’s Information Criterion corrected for small sample sizes (AICc). This method balances bias and variance to yield the most reliable spatially varying estimates. The adaptive bandwidth is appropriate due to the spatial heterogeneity and varying density of DHS clusters across countries. Model diagnostics included local R² values, residual Moran’s I to check for remaining spatial autocorrelation, and local variance-inflation factors to monitor multicollinearity. Global model performance was compared to OLS using adjusted R² and AICc. Categorical predictors included in the GWR model Were coded using dummy variables, with one category designated as the reference group. Specifically, households with heads aged 25–50 years were used as the reference for age comparisons; users of untreated mosquito nets served as the baseline for net utilization; and the middle wealth quintile was the reference for socioeconomic status. Additionally, households with five or fewer members were considered the reference group for family size, while those with finished walls and roofs were used as baselines for housing material. The reported coefficients represent the spatially varying effects of each category relative to these reference groups.

The GWR model can be written as: [69]

$$\text{Yi}={\beta }_o\left(u_i,v_i\right)+\sum _{k=1}^p{\beta }_k\left(u_i,v_i\right)X_{\mathit{ik}}+{\in }_i$$

where Y_i are observations of response y; u_iv_i are geographical points (longitude, latitude), βk (u_iv_i) (k = 0, 1 … n) are location-specific intercept and slope coefficients u_iv_i; X_ik are explanatory variables at location u_iv_i,; i = 1, 2 … n and ∈ _i are error terms/residuals with zero mean and homogenous variance (σ²). Mapping the local coefficients βk (u_iv_i) (k = 0, 1 … n) enables identification of areas where specific risk factors exert stronger or weaker effects, thereby informing geographically targeted malaria interventions.

Ethical approval and consent to participate

This study involved secondary analysis of publicly available household survey data and, therefore, did not require separate ethical approval. Ethical clearance for each survey was obtained from the respective national review boards prior to data collection, with informed consent secured from all participants according to the protocols of each survey. Data was granted through an online request to the Demographic and Health Surveys (DHS) Program and accessible at (http://www.dhsprogram.com) [70]. The dataset used contains no personally identifiable information, ensuring participant confidentiality. Consequently, additional ethical approval and informed consent were not applicable to this secondary analysis.

Results

Socio-demographic, housing, and environmental characteristics of the study participants

A weighted sample of 126,424 homes from 11,594 clusters was used in this study. According to the household head's sex, men accounted for the majority (71.70%), while women constituted 28.30%. The household head's age was divided into four groups: the majority of household heads (HHH) (40.07%) Were between the ages of 35 and 49, while 4.91% Were between the ages of 10 and 24. 42.40% of household heads had no formal education while 22.06% of household heads had secondary or higher education. Although this percentage may seem strikingly low, it accurately reflects regional disparities in educational access, particularly in rural and underserved areas. This limited level of formal education among household heads has critical public health implications, as education is a known determinant of health literacy, influencing awareness and adoption of malaria prevention and treatment strategies. In terms of family size, 8.18% of households have more than twelve members, whilst 52.35 percent have six to twelve members. While 48.06% of families use a treated bed net, 49.17% of homes do not use one. The wealth index divides households into five economic groups: medium (20.82%), richer (17.88%), richest (14.83%), lowest (24.68%), and poorest (21.79%).

While 43.67% of households have walls made from finished materials, the majority of households (55.82%) have walls made from rudimentary or natural materials. The percentage of households with a rudimentary or natural floor was 49.98%, while the percentage of households with a finished floor was 49.74%. 72.25% of homes have a roof constructed from finished materials, whilst 27.45% of households have a roof composed of rudimentary or natural materials. 33.31% of households use unimproved sources of drinking water, compared to 66.69% that use improved sources. According to the site of residence, 33.38% of households live in cities, while 66.62% of families reside in rural areas. Last but not least, 30.89% of homes reside above 1000 m, while 69.11% of households reside below 1000 m (Table 3).

Table 3.

Socio-demographic, housing, and environmental characteristics of households in 19 SSA

Variable	Category	Sample population (%)	Malaria (%)
			Negative	Positive
Sex of HHH	Male	96,296 (76.17)	73,540 (58.17)	22,756 (18.00)
	Female	30,128 (23.83)	24,477 (19.36)	5651 (4.47)
Age of HHH	10–24	6209 (4.91)	4630 (3.66)	1579 (1.25)
	25–34	34,563 (27.34)	26,372 (20.86)	8191 (6.48)
	35–49	50,656 (40.07)	39,686 (31.39)	10,970 (8.68)
	50 +	34,996 (27.68)	27,329 (21.62)	7667 (6.06)
Educational attainment of HHH	No education	53,604 (42.40)	40,953 (32.39)	12,651 (10.01)
	Primary	44,931 (35.54)	41,453 (32.79)	3478 (2.75)
	Secondary/higher	27,889 (22.06)	27,599 (21.83)	290 (0.23)
Wealth index	Poorest	31,197 (24.68)	20,858 (16.50)	10,339 (8.18)
	Poorer	27,549 (21.79)	19,811 (15.67)	7738 (6.12)
	Middle	26,318 (20.82)	20,499 (16.21)	5819 (4.60)
	Richer	22,606 (17.88)	19,183 (15.17)	3423 (2.71)
	Richest	18,754 (14.83)	17,666 (13.97)	1088 (0.86)
Family size	<  = 5	49,907 (39.48)	39,620 (31.34)	10,287 (8.14)
	6–12	66,177 (52.35)	50,261 (39.76)	15,916 (12.59)
	> 12	10,340 (8.18)	8136 (6.44)	2204 (1.74)
mosquito bed net type	No bed net	62,164 (49.17)	48,308 (38.21)	13,856 (10.96)
	Treated bed net	60,759 (48.06)	46,720 (36.96)	14,039 (11.10)
	Untreated bed net	3501 (2.77)	2989 (2.36)	512 (0.40)
Wall material	Rudimentary/natural wall	70,572 (55.82)	52,721 (41.70)	17,851 (14.12)
	Finished wall	55,205 (43.67)	44,767 (35.41)	10,438 (8.26)
	Other	647 (0.51)	529 (0.42)	118 (0.09)
Floor material	Rudimentary/natural floor	63,188 (49.98)	46,030 (36.41)	17,158 (13.57)
	Finished floor	62,878 (49.74)	51,700 (40.89)	11,178 (8.84)
	Other	358 (0.30)	287 (0.23)	71 (0.06)
Roof material	Rudimentary/natural roof	34,700 (27.45)	24,334 (19.25)	10,366 (8.20)
	Finished roof	91,349 (72.25)	73,381 (58.04)	17,968 (14.21)
	Other	375 (0.30)	302 (0.24)	73 (0.06)
Source of Drinking water	Improved	84,317 (66.69)	67,400 (53.31)	16,917 (13.38)
	Unimproved	42,107 (33.31)	30,617 (24.22)	11,490 (9.09)
Place of residence	Urban	42,202 (33.38)	36,880 (29.17)	5,322 (4.21)
	Rural	84,222 (66.62)	61,137 (48.36)	23,085 (18.26)
Altitude	<  = 1000 m	87,371 (69.11)	66,404 (52.52)	20,967 (16.58)
	> 1000 m	39,053 (30.89)	31,613 (25.01)	7,440 (5.88)

A small number of household heads Were reported as aged below 18 years. These rare cases, potentially representing child-headed households, were retained in the analysis due to their presence in the original DHS dataset

The magnitude of malaria among households in 19 sub-Sahara African countries

The findings of this study revealed that the prevalence of malaria among households in 19 SSA was 22.47% (95% CI (22.24%, 22.70%)) (Fig. 2). Togo had the highest prevalence (41.16%), followed by Côte d'Ivoire (40.12%), and the smallest prevalence Gambia (0.33%), followed by Mauritania (0.85%) (Fig. 3).

Fig. 2.

Prevalence of malaria among households in 19 sub-Saharan African countries, DHS (2013–2023)

Fig. 3.

Malaria prevalence distribution across countries in sub-Saharan Africa, DHS (2013–2023)

Graduated colour (choropleth) density map of malaria

The map depicts the distribution of malaria among households in 19 SSA based on the proportion value. Areas coloured red indicate a higher proportion of malaria with a value between 0.765 and 1.00, while areas coloured blue represent the lowest proportion with a value between 0 and 0.117 (Fig. 4).

Fig. 4.

The graduated colour map of malaria among households in 19 sub-Saharan Africa, DHS 2013–2023

Spatial autocorrelation

Global spatial autocorrelation analysis using Moran’s I revealed a statistically significant and weak-to-moderate clustering of malaria cases in SSA (Moran’s I = 0.159, Z = 239.06, p < 0.001). This indicates that malaria distribution is spatially clustered rather than random. Euclidean distance was used to construct the spatial weight matrix due to the irregular and point-based distribution of DHS clusters across countries. This approach allows for consistent distance-based relationships between clusters, unlike contiguity-based methods that assume uniform adjacency and are sensitive to administrative boundary definitions not applicable to DHS sampling. The analysis applied the inverse distance weighting conceptualization, where nearer features exert greater influence. A fixed distance threshold of approximately 1.95 kms was used to define neighbourhood relationships. Note that this value reflects clustering behaviour within the scale of DHS sampling clusters, though it may be considered narrow for broader spatial patterns (Fig. 5).

Fig. 5.

The spatial autocorrelation result of malaria among households in 19 sub-Saharan African countries, DHS 2013–2023

Hotspot and cold spot cluster analysis

Hotspot areas of malaria Were identified in Benin, Burkina Faso, Togo, Uganda, Rwanda, parts of the Congo Republic, and regions within Mozambique, as indicated by dark red and orange shading corresponding to 95% and 90% confidence intervals, respectively. Conversely, cold spots were observed in Mauritania, Côte d’Ivoire, Madagascar, Tanzania, portions of Angola, and other distinct regions of Mozambique, marked by dark and light blue areas. The appearance of some countries, such as Mozambique, in both hotspot and cold spot categories reflects spatial heterogeneity within national boundaries where certain regions experience significantly higher clustering of malaria, while others show lower-than-expected case concentrations. This underscores the need for geographically targeted interventions rather than uniform national strategies (Fig. 6).

Fig. 6.

The hotspot and cold spot clusters of malaria among households in 19 sub-Saharan African countries, DHS 2013–2023

Cluster and outlier analysis of malaria

The Cluster and Outlier Analysis using Anselin Local Moran’s I identified spatial outlier patterns of malaria prevalence across the study area. High-high (HH) clusters, representing areas with elevated malaria rates surrounded by similarly high-prevalence regions, highlight zones requiring urgent and sustained intervention. In contrast, low-low (LL) clusters denote areas with consistently low malaria prevalence, suggesting effective malaria control in both the focal and neighbouring regions.

High-low (HL) outliers where low-prevalence neighbours surround high-prevalence areas were detected in Mozambique, Madagascar, Angola, Nigeria, Benin, Togo, Ghana, Gabon, and Cameroon. These may reflect localized drivers of transmission that are not present in adjacent areas. Conversely, low–high (LH) outliers, found in the Democratic Republic of the Congo, Mozambique, Mali, Côte d’Ivoire, Burkina Faso, and Nigeria, indicate relatively successful malaria control efforts within these focal areas despite being surrounded by higher-prevalence regions (Fig. 7).

Fig. 7.

Cluster and Outlier analysis of malaria among households in 19 sub-Saharan African countries, DHS 2013–2023

SaTScan clusters analysis of malaria

A total of 34 statistically significant spatial clusters (windows) of malaria Were detected across 19 sub-Saharan African (SSA) countries using SaTScan analysis, based on data from 126,464 households. The clusters varied in statistical strength, as indicated by their Log Likelihood Ratio (LLR) values, which reflect the degree of malaria clustering. The most prominent clusters those with the highest LLR values ranging from 535.998 to 1588.401 Were identified in Ghana, Burkina Faso, Togo, and Benin, signifying regions with the highest concentration of malaria prevalence. The second-tier clusters, with LLR values between 136.437 and 535.997, Were primarily located in Mozambique, Nigeria, and the Republic of Congo, indicating areas with a very high but comparatively lower intensity of clustering. Clusters with lower statistical significance and LLR values between 6.538 and 28.961 were observed in parts of Mozambique and Angola, suggesting relatively lower but still notable malaria prevalence. The spatial distribution of these clusters was visualized using colour coding, with deeper red shades representing stronger statistical significance (Fig. 8).

Fig. 8.

SaTScan cluster analysis of malaria among households in 19 sub-Saharan African Countries, DHS 2013-2023

Spatial regression analysis

Ordinary least squares analysis results

The Ordinary Least Squares (OLS) regression analysis was conducted to examine pertinent predictors of malaria among households. The adjusted R² of the OLS model was 0.342, suggesting moderate explanatory power. The joint F-statistic (367.626, p < 0.001) indicates that the overall model is statistically significant. Similarly, the Joint Wald statistics test (821.919, p < 0.001) further confirms the robustness of the model. The Variance Inflation Factor (VIF) values among some independent variables ranged from 1.282 to 5.278, indicating moderate correlation (below 7.5).

The Koenker BP test (χ² = 33.327, p < 0.001) confirms the presence of heteroscedasticity or non-stationarity in the data. A statistically significant result suggests that the relationships between predictors and malaria prevalence vary across geographical space, justifying the use of Geographically Weighted Regression (GWR), which accounts for spatial non-stationarity. The Jarque–Bera test (χ² = 73.758, p < 0.001) indicates that the residuals are not normally distributed. Although the Jarque–Bera test indicated a deviation from normality in the residuals, this does not invalidate the OLS estimates due to the large sample size used in this study. According to the Central Limit Theorem, the sampling distribution of the OLS estimators tends toward normality as the sample size increases, even when the residuals are not normally distributed [71]. Therefore, since the Koenker (BP) statistic indicated statistical significance, robust probability values were used to assess the significance of the coefficients.

Female Household heads (β = − 0.154, p < 0.001), age of HHH (10–24) (β = 0.150, p = 0.009), poorest households (β = 0.111, p < 0.001) and richest households (β = − 0.111, p = p < 0.001), Were significantly associated with malaria. Additionally, HH with 6–12 members (β = 0.105, p < 0.001), HH with no bed net (β = 0.058, p < 0.002), HH with treated bed net (β = 0.187, p < 0.001), rudimentary/natural wall (β = − 0.072, p < 0.001), rudimentary/natural roof (β = 0.055, p = 0.007), and rural households (β = 0.090, p < 0.001) Were all significantly associated with malaria in 19 SSA countries. Although some explanatory variables were not statistically significant, they were retained in the model due to their theoretical and contextual relevance based on prior literature. Their inclusion ensures a more comprehensive understanding of malaria risk factors and helps control for potential confounding (Table 4).

Table 4.

Summary and model diagnostic outputs of OLS model fitting results for malaria among households in 19 SSA

Variable		Coefficient	Standard error	t-statistics	Probability	Robust SE	Robust Probability P-value	VIF
Intercept		0.371	0.048	7.661	0.000*	0.069	0.000*	–
Female		– 0.154	0.011	– 13.278	0.000*	0.023	0.000*	1.820
HHH age between 10–24		0.150	0.027	5.424	0.000*	0.057	0.009*	1.282
HHH age above 50		0.011	0.011	1.013	0.311	0.021	0.596	1.769
Secondary/higher education		– 0.314	0.031	– 1.011	0.234	0.051	0.103	1.654
Poorest HH		0.111	0.009	11.177	0.000*	0.028	0.000*	2.072
Richest HH		– 0.111	0.010	– 10.575	0.000*	0.014	0.000*	2.021
HH with 6–12 member		0.105	0.009	10.767	0.000*	0.017	0.000*	3.708
HH with no bed net		0.058	0.010	5.820	0.000*	0.019	0.002*	2.825
HH with treated bed net		0.187	0.009	19.451	0.000*	0.018	0.000*	3.526
HH with untreated bed net		0.016	0.029	0.558	0.576	0.050	0.745	1.370
Rudimentary/natural wall		– 0.072	0.008	– 8.633	0.000*	0.016	0.000*	5.278
Rudimentary/natural floor		0.020	0.008	2.310	0.020*	0.019	0.296	4.692
Rudimentary/natural roof		0.055	0.007	7.208	0.000*	0.021	0.007*	2.711
Unimproved water source		0.008	0.006	1.241	0.214	0.016	0.619	1.755
Rural		0.09	0.006	13.751	0.000*	0.011	0.000*	3.725
1000+ m Altitude		– 0.011	0.006	– 1.968	0.049	0.014	0.414	2.272
Ordinary Least Square regression model diagnostic test
Number of observations	11,594		Adjusted R-squared				0.342
Joint F-statistics	367.626		Prob(> F), (16,11242) degree of freedom				< 0.001*
Joint Wald statistics	821.919		Prob (> chi-squared), (16) degree of freedom				< 0.001*
Koenker (BP) statistics	33.327		Prob (> chi-squared), (16) degree of freedom				< 0.001*
Jarque–Bera statistics	73.758		Prob (> chi-squared), (2) degree of freedom				< 0.001*
Akaike’s Information Criterion (AICc)	59864.00

SE: Standard error; VIF: Variance Inflation Factor

^*Significant independent variable at p-value less than 0.05

Geographical weighted regression analysis results

The local spatial regression model, Geographically Weighted Regression (GWR), demonstrated a statistically significant improvement in model performance over the global Ordinary Least Squares (OLS) regression explaining the spatial variation of malaria among households across 19 SSA. The adjusted R² increased notably from 0.342 in the OLS model to 0.531 in the GWR model, indicating a better fit and enhanced explanatory power in the local model. Furthermore, model comparison using the Akaike Information Criterion (AIC) showed a substantial reduction from 59,864.00 in the OLS model to 56,176.82 in the GWR model. The lower AIC value in the GWR model reflects improved model parsimony and better performance in capturing spatial heterogeneity. These improvements suggest that GWR more effectively accounts for local spatial heterogeneity in the relationship between malaria and household-level predictors (Table 5).

Table 5.

Summary of geographically weighted regression (GWR) analysis of malaria among households in 19 sub-Saharan African countries

Parameter	Value
AIC	56,176.82
R2	0.537
Adjusted R2	0.531
Sigma	2.915

When interpreting the GWR coefficients, it is essential to consider that categorical variables Were analysed in reference to baseline categories. Specifically, households with heads aged 25–50 years were used as the reference for age comparisons; users of untreated mosquito nets served as the baseline for net utilization; and the middle wealth quintile was the reference for socioeconomic status. Additionally, households with five or fewer members were considered the reference group for family size, while those with finished walls and roofs were used as baselines for housing material. The reported coefficients represent the spatially varying effects of each category relative to these reference groups.

The GWR model showed a spatially varying association between female-headed households and malaria prevalence. Negative coefficients indicating lower malaria prevalence in areas with higher proportions of female household heads were observed predominantly in Mali, northern Nigeria, parts of Cameroon, northern Tanzania, and the Republic of Congo. The strength of this association varied across regions, with GWR coefficient values ranging from approximately − 0.374 to − 0.189. Darker blue shades indicate stronger negative associations (lower malaria prevalence with higher proportions of female-headed households). Positive associations were not observed. This suggests that in certain contexts, female household leadership may be associated with improved malaria outcomes, possibly due to differences in household health practices or care-seeking behaviours. These spatial variations underscore the need for localized, gender-sensitive malaria interventions rather than uniform strategies (Fig. 9).

Fig. 9.

Spatial distribution of GWR coefficients for the association between female household heads and malaria prevalence across 19 Sub-Saharan African countries DHS 2013–2023

The GWR model revealed a positive and spatially heterogeneous association between household head age 10–24 and malaria prevalence. In particular, northern Nigeria, western Cameroon, parts of central Tanzania, and central Côte d'Ivoire exhibited the strongest positive GWR coefficients, with values ranging from approximately + 0.336 to + 0.656. Positive values (shaded in red) indicate regions where younger household headship is associated with increased malaria risk. This implies that malaria prevalence is higher in regions where younger individuals in this age range head a larger proportion of households. The elevated risk in these areas may reflect challenges faced by young household heads, including limited access to health information, reduced decision-making power, or socioeconomic vulnerability. These findings highlight the need for targeted malaria prevention strategies and youth-specific support services, particularly in regions with a higher prevalence of young-headed households (Fig. 10).

Fig. 10.

Geographically weighted regression coefficients for the association between malaria prevalence and household head age 10–24 across 19 Sub-Saharan African countries, DHS 2013–2023

The geographically weighted regression (GWR) analysis identified a strong positive association between malaria prevalence and the proportion of households classified in the poorest wealth quintile, though the strength of this relationship varied spatially. Regions shaded in deep red indicate stronger positive associations notably, northern Nigeria and central Togo exhibited the highest positive GWR coefficients, ranging from + 0.344 to + 0.624, indicating that in these regions; higher poverty levels significantly increase malaria risk. These spatial patterns emphasize that economic deprivation is a localized but powerful driver of malaria transmission, likely due to poor housing conditions, limited access to treated bed nets, and lower healthcare utilization in these communities. Policymakers should prioritize poverty-targeted malaria control efforts such as subsidized preventive tools and outreach services in regions where socioeconomic vulnerability and malaria overlap most severely (Fig. 11).

Fig. 11.

GWR coefficients for the relationship between malaria prevalence and the poorest wealth quintile across 19 Sub-Saharan African countries, DHS 2013–2023

The geographically weighted regression (GWR) analysis revealed a strong negative spatial association between malaria prevalence and the proportion of households in the richest wealth quintile, though the strength of the relationship varied geographically. Darker blue shades indicate stronger negative associations. The largest negative GWR coefficients, ranging from − 0.288 to − 0.191, were observed in Côte d’Ivoire, Mali, Benin, southern Nigeria, and central Uganda, suggesting that in these areas, higher concentrations of wealthiest households significantly reduce malaria risk. This pattern likely reflects enhanced housing quality, access to treated bed nets, healthcare services, and greater health awareness among wealthier populations. These findings reinforce the role of socioeconomic protection as a buffer against malaria transmission and highlight the need to expand malaria prevention programmes to underserved, lower-income areas where the protective effects of wealth are absent (Fig. 12).

Fig. 12.

GWR coefficients for the relationship between malaria prevalence and the wealthiest quintile across 19 Sub-Saharan African countries, DHS 2013–2023

The geographically weighted regression (GWR) analysis identified a spatially heterogeneous positive association between malaria prevalence and medium-sized households (6–12 members). Local GWR coefficient values ranged from + 0.191 to + 0.288, indicating a moderate to strong correlation in several regions. Notably, the strongest positive spatial associations were observed in Burkina Faso, Benin, Togo, Ghana, Uganda, Mozambique, and Madagascar, where larger household size may intensify malaria risk. This pattern suggests that increased household crowding may elevate exposure to malaria vectors, likely due to shared sleeping arrangements, limited bed net availability, and increased indoor vector contact. The results highlight the importance of incorporating household composition into malaria prevention strategies, particularly in high-burden zones. Public health policies should prioritize bed net coverage proportional to household size, and targeted health education on malaria risk in crowded setting (Fig. 13).

Fig. 13.

Geographically weighted regression coefficients for the association between malaria prevalence and family size of 6 to 12 across 19 Sub-Saharan African countries, DHS 2013–2023

The geographically weighted regression (GWR) analysis identified a strong positive spatial association between malaria prevalence and the proportion of households not using insecticide-treated bed nets. Local GWR coefficients ranged from + 0.241 to + 0.423, with the highest values concentrated in Ghana, Benin, Togo, Côte d'Ivoire, and parts of Tanzania. These elevated coefficients indicate that in regions with low bed net coverage, the risk of malaria was substantially higher, highlighting localized gaps in vector control efforts.

This spatially varying relationship suggests that universal net distribution alone may not be sufficient; localized behavioural, logistic, or accessibility challenges may hinder effective usage. Policymakers should prioritize geographically tailored interventions, including intensified distribution efforts in high-risk zones and community-based education on proper net usage. These findings also support the allocation of malaria control resources based on fine-scale risk maps, ensuring high-impact deployment of preventive measures (Fig. 14).

Fig. 14.

Geographically weighted regression coefficients for the association between malaria prevalence and household with no bed net across 19 Sub-Saharan African countries, DHS 2013–2023

The geographically weighted regression (GWR) analysis revealed a counterintuitive positive spatial association between malaria prevalence and the proportion of households reporting the use of treated bed nets, with local GWR coefficients ranging from + 0.215 to + 0.337. This positive association was most prominent in the Republic of Congo, Mali, and parts of Côte d'Ivoire, as illustrated by the darker red shading on the map. These findings suggest that, in these regions, higher reported use of treated bed nets was still linked to elevated malaria risk. This pattern may point to inconsistent or improper usage, such as damaged nets, incorrect installation, or seasonal use only. It may also reflect underlying ecological or behavioural factors that reduce the protective effect of treated nets, such as high mosquito resistance or households living in highly endemic environments. Public health strategies in these high-risk areas should, therefore, go beyond distribution metrics. Policy efforts must incorporate community education campaigns on proper and consistent use of treated nets, assessment of net condition, and resistance monitoring. These actions would ensure that net usage translates into real reductions in malaria transmission (Fig. 15).

Fig. 15.

GWR coefficients for the relationship between malaria prevalence and households with treated bed net across 19 Sub-Saharan African countries, DHS 2013–2023

The geographically weighted regression (GWR) analysis identified a strong negative spatial association between malaria prevalence and households with walls made of natural or rudimentary materials, although the relationship varied across countries. Local GWR coefficients ranged from − 0.365 to − 0.217, with the strongest negative associations observed in Madagascar, Mozambique, and the Republic of Congo, as indicated by the darker blue regions in the map. This inverse association suggests that in these specific contexts, such housing structures may paradoxically correlate with lower malaria risk. Several hypotheses could explain this pattern: rudimentary materials may allow better airflow and reduce indoor humidity conditions less favourable for mosquito survival or communities living in such dwellings may also reside in regions with lower vector density due to environmental or elevation-related factors.

However, these findings should be interpreted cautiously. In other regions, rudimentary housing is often associated with higher vulnerability to malaria due to exposure. Thus, this spatial variation highlights the need for context-specific assessments of housing interventions. Public health programmes should avoid one-size-fits-all assumptions about housing quality and instead integrate local entomological, ecological, and cultural factors into malaria risk assessment (Fig. 16).

Fig. 16.

Geographically weighted regression coefficients for the association between malaria prevalence and household with wall built from rudimentary/natural material across 19 Sub-Saharan African countries, DHS 2013–2023

The geographically weighted regression (GWR) analysis showed a positive spatial association between malaria prevalence and households with roofs made of natural or rudimentary materials, though the strength of this relationship varied across regions. Local GWR coefficients ranged from + 0.184 to + 0.309, with the strongest positive associations found in Rwanda, Tanzania, Uganda, and Mozambique, as indicated by deep red shading on the coefficient map. This pattern suggests that in these countries, poorer roof quality may contribute to higher malaria risk by enabling easier mosquito entry and inadequate protection from environmental exposure. Natural and rudimentary roofing materials such as thatch, grass, or palm leaves often lack structural integrity and may not support insecticide spraying or net hanging, thereby compounding exposure risk. These findings emphasize the importance of integrating housing improvement strategies including roofing upgrades into malaria control programmes, especially in high-risk zones identified by the GWR. Interventions could focus on promoting affordable and protective roofing alternatives in vulnerable communities, particularly where positive spatial correlations are strongest (Fig. 17).

Fig. 17.

GWR coefficients for the relationship between malaria prevalence and households with a roof built from rudimentary/natural material across 19 Sub-Saharan African countries, DHS 2013–2023

The geographically weighted regression (GWR) analysis revealed a positive spatial association between rural residence and malaria prevalence, with the local GWR coefficients ranging from + 0.198 to + 0.342. This indicates that in certain regions, households located in rural areas face a significantly higher risk of malaria compared to their urban counterparts. The strongest positive correlations were observed in Nigeria, Togo, Benin, Ghana, Tanzania and part of Gabon, where rural geography coincides with environmental and socioeconomic conditions conducive to malaria transmission. These findings suggest that rural settings often lack essential malaria prevention infrastructure, such as effective drainage, consistent access to insecticide-treated bed nets, and proximity to health services. Moreover, rural households may be closer to breeding grounds (e.g., stagnant water or agricultural irrigation systems), further amplifying the transmission risk. This spatial heterogeneity underscores the importance of geographically tailored malaria interventions. Public health authorities should prioritize rural-focused strategies in identified area, including improving access to health education, expanding bed net distribution, and enhancing environmental vector control measures (Fig. 18).

Fig. 18.

Geographically weighted regression coefficients for the association between malaria prevalence and rural residents across 19 Sub-Saharan African countries, DHS 2013–2023

Discussion

This study confirms that malaria remains a significant public health burden across 19 sub-Saharan African countries, with a pooled household prevalence estimate of 22.47% (95% CI 22.24%, 22.70%). This finding aligns with a study conducted in Ghana with 22.3% [72]. The finding of this study is higher than the study conducted among under 5 children in sub-Saharan African countries with 18.8% [73]. The result of this finding is smaller than the study conducted in 11 sub-Saharan African countries with 24.2% [16], Uganda 23.3% [74], Republic of Congo 29.3% [75], Ghana 39.1% [76] and Malaysia 33.6% [77]. These discrepancies may stem from several contextual and methodological differences, including the target population, geographic scope, survey timing, diagnostic methods, and definitions of malaria positivity. Additionally, factors such as socioeconomic development, vector ecology, climatic conditions, healthcare accessibility, and implementation of malaria control policies vary widely between and within countries, influencing transmission dynamics and prevalence rates. Rather than indicating contradictory evidence, these variations reflect the spatial and demographic heterogeneity of malaria risk across the region. They underscore the need for localized and adaptive malaria control strategies that account for geographic, environmental, and systemic differences.

The spatial autocorrelation analysis demonstrated that malaria prevalence in 19 sub-Saharan African countries is not randomly distributed but exhibits significant spatial clustering. Specifically, the Getis-Ord Gi* analysis revealed statistically significant clusters, reflecting areas with unusually high or low malaria burdens. While hotspot regions were identified in countries such as Benin, Burkina Faso, Togo, Uganda, Rwanda, and parts of the Republic of Congo and Mozambique, contrasting cold spots emerged in Mauritania, Côte d'Ivoire, Madagascar, Tanzania, and parts of Angola and Mozambique.

These spatial disparities highlight the influence of a complex interplay of environmental, social, and health system factors. Hotspot areas are often characterized by higher exposure to vector breeding conditions such as standing water, high humidity, and proximity to bodies of water as well as socioeconomic challenges including poverty, poor housing, and limited access to healthcare [78–80]. Cultural practices, ineffective or inconsistent use of insecticide-treated nets (ITNs), and limited indoor residual spraying (IRS) coverage can further exacerbate malaria risk in these areas [81].

In contrast, cold spot regions may benefit from stronger public health infrastructure, better access to preventive tools, improved socioeconomic conditions, and more urbanized living environments that reduce mosquito breeding and human-vector contact. Successful intervention programmes such as widespread IRS, effective ITN usage, and community health outreach may contribute to the lower malaria burden in these regions. Understanding the spatial distribution of malaria is essential for optimizing resource allocation. Tailored, location-specific interventions are needed to target hotspots more aggressively while maintaining preventive strategies in emerging cold spot regions to sustain gains and avoid resurgence.

The Anselin Local Moran’s I analysis identified significant spatial outlier clusters of malaria in sub-Saharan Africa. High–low (HL) clusters areas with high malaria prevalence surrounded by areas of low prevalence were observed in Mozambique, Madagascar, Angola, Nigeria, Benin, Togo, Ghana, Gabon, and Cameroon. These patterns suggest that certain local conditions may be intensifying malaria risk within these specific areas, even though their surrounding regions report lower transmission levels. Such conditions could include inadequate malaria interventions, poor access to healthcare, gaps in vector control coverage, or unique environmental and housing characteristics that support mosquito breeding.

Conversely, low–high (LH) outliers regions with relatively low malaria prevalence surrounded by high-prevalence areas were identified in parts of Nigeria, Burkina Faso, Côte d’Ivoire, Mali, Mozambique, and the Democratic Republic of the Congo. These regions may be benefitting from more effective malaria control efforts or stronger health systems compared to their neighbours. Such areas can serve as models for best practices in malaria prevention. Their success may be attributed to better community engagement, improved healthcare infrastructure, higher education levels, or successful deployment of interventions like IRS and ITNs. Understanding these LH clusters provides valuable insight into what works in malaria control and where such approaches could be scaled up or replicated in neighbouring high-risk areas. Integrating these spatial insights into policy planning can help prioritize resource allocation. HL clusters require urgent investigation to identify and address local drivers of transmission, while LH areas should be studied to understand and sustain their relative success.

The SatScan cluster analysis among 126,464 households identified 34 windows containing significant clusters in 19 Sub-Saharan African countries. However, Areas with the most statistically significant clusters with the highest LLR values between 535.998 and 1588.401, indicating regions with the highest prevalence of malaria Were identified in Ghana, Burkina Faso, Togo, and Benin. A recent study in Ghana stated that malaria is an endemic disease and accounts for 38.0% of all outpatient attendance with the most vulnerable groups being children under 5 years of age and the result of their study suggests the prevalence of malaria was 20.63% [82]. Approximately 85% of ITNs were freely received from the public and private sectors, according to the GDHS survey. However, based on budgetary expenses and technological viability, the government and non-governmental groups only implemented the IRS in a few districts of Ghana [78, 83]. As a result, ITN became the primary preventive strategy for a sizable percentage of homes that Were not covered by the IRS. According to a study using Burkina Faso representative statistics, the combined prevalence of malaria was 44% [84]. A WHO research states that 384,000 children under the age of five died from malaria in sub-Saharan Africa in 2019, with Burkina Faso accounting for 4% of these deaths. Malaria is one of the main causes of these deaths [85]. This could be because a large portion (43.8%) of Burkinabe people live in conditions marked by extreme poverty (less than 1.90 USD a day) and they and their children become vulnerable to several health consequences including malaria [86, 87]. This in turn becomes the primary barrier to the use of various malaria prevention strategies. The Togo Ministry of Health report from 2018 states that malaria is an endemic disease in the country, with 2,002,877 cases and 905 deaths reported among the general population, which is approximately 7 million people [88]. Another study revealed a prevalence of 32.1% and in Togo, the IRS is an intervention included in the strategic plan, but it has not been implemented yet [89]. But the preventive effect of ITNs depends on their insecticidal properties and physical integrity [90].

The geographically weighted regression (GWR) analysis revealed a counterintuitive positive spatial association between malaria prevalence and the proportion of households reporting the use of treated bed nets, with local GWR coefficients ranging from + 0.215 to + 0.337. This positive association was most prominent in the Republic of Congo, Mali, and parts of Côte d'Ivoire. Catching malaria despite using bed nets was also perceived to be associated with spending evenings outdoors [91] often for social activities like resting outdoors, communal television watching, and public drinking venues simply because these are common activities people do when they are outside. Furthermore, even when households utilize treated bed nets, the number of malaria cases increases due to other environmental factors like temperature, altitude, stagnant water sources, and housing conditions. The insecticidal qualities and physical integrity of ITNs determine their preventative impact [90]. Frequent washing, Soap type, and drying under direct sunlight can degrade both the physical and chemical integrity of these nets reducing their effectiveness [92, 93]. Additionally studies identified that despite high net ownership, consistent use was hindered by factors such as discomfort, difficulties in hanging nets, poor net condition (holes, fraying), and misconceptions about their efficacy leading to paradoxical high malaria rates in areas with reported high bed-net usage [94, 95]. This study provides empirical support that net availability does not necessarily translate into effective protection and underscores the importance of educational and behavioural components in malaria prevention strategies.

The GWR analysis revealed a spatially heterogeneous negative association between female-headed households and malaria prevalence, particularly in Mali, northern Nigeria, and parts of Cameroon, northern Tanzania, and the Republic of Congo, where local coefficients ranged from − 0.374 to − 0.189. This spatial pattern implies that regions with a higher concentration of female-headed households tend to have lower malaria burdens. These findings are consistent with studies from Ethiopia, where male-headed households reported higher malaria incidence [96]. One possible explanation is that female heads may be more engaged in household health management practices, such as consistent use of insecticide-treated nets (ITNs), environmental sanitation, and prompt care-seeking behaviour [97]. Moreover, women in these roles may prioritize children's and household health more consistently, which can indirectly contribute to reduced exposure to malaria vectors. This emphasizes the importance of empowering women in household decision-making and targeting health promotion strategies that leverage the proactive role of female household heads in malaria prevention.

The GWR model identified a positive and spatially variable association between household heads aged 10–24 and malaria prevalence. This association was strongest in northern Nigeria, western Cameroon, central Tanzania, and central Côte d'Ivoire, where local GWR coefficients ranged from + 0.336 to + 0.656. These findings highlight a potentially vulnerable demographic: younger household heads who may face structural disadvantages such as limited financial resources and inadequate knowledge of disease prevention practices [98]. Additionally, this age group especially males often represents the most economically active segment of the population, engaging in outdoor work during evening hours, when mosquito exposure is highest [99]. This behavioural pattern, coupled with inconsistent use of personal protective measures (e.g., repellents, protective clothing), increases their susceptibility to malaria transmission. Public health interventions should therefore consider tailored messaging and targeted resource allocation for younger household heads, particularly in high-risk regions identified through spatial modelling.

In addition, the geographically weighted regression (GWR) analysis revealed a spatially heterogeneous yet consistently strong positive association between malaria prevalence and households in the poorest wealth quintile. Notably, northern Nigeria and central Togo recorded the highest GWR coefficients, ranging from + 0.344 to + 0.624, indicating a significant concentration of malaria burden in regions with high poverty levels. This relationship points to the structural vulnerabilities of impoverished communities, where financial constraints limit access to effective malaria prevention tools, such as insecticide-treated bed nets (ITNs), indoor residual spraying (IRS), and antimalarial vaccines. Moreover, residents in these areas often resort to substandard or unregulated medications due to cost, which may not only fail to treat the infection effectively but also contribute to drug resistance and adverse health outcomes [100]. Additionally, the most impoverished households are less able to afford various malaria prevention measures, such as treated bed nets, IRS, and vaccinations. Furthermore, the nutritional status of the majority of these households is somewhat compromised due to financial constraints, which lowers their immunity and makes them more susceptible to the disease [101]. Addressing malaria in these high-burden, low-resource settings, therefore, requires integrated interventions that combine economic support, community education, and improved access to proven preventive and curative measures.

This study identified a strong negative spatial association between malaria prevalence and the proportion of households in the richest wealth quintile, with notable geographic variation in effect size. The lowest GWR coefficients, ranging from − 0.288 to − 0.191, were concentrated in Côte d’Ivoire, Mali, Benin, southern Nigeria, and central Uganda. These findings highlight the protective role of socioeconomic advantage in mitigating malaria risk. Wealthier households are more likely to invest in comprehensive malaria prevention strategies, including consistent use of insecticide-treated bed nets (ITNs), improved housing structures that reduce mosquito entry, and timely access to healthcare services for early diagnosis and treatment [74]. Additionally, financial capacity allows these households to bypass overcrowded and resource-limited public health facilities by seeking private care, which often ensures quicker and more reliable treatment. Moreover, the ability to maintain clean household environments and access to piped water or better sanitation infrastructure may further reduce exposure to malaria vectors [102]. Improved nutrition and general wellbeing among wealthier populations also contribute to stronger immune responses, further reducing susceptibility to infection. These spatial patterns underscore the importance of poverty reduction and structural improvements as integral components of malaria control strategies.

Furthermore, the geographically weighted regression (GWR) analysis revealed a spatially heterogeneous and positive association between malaria prevalence and medium-sized households (6–12 members), with local GWR coefficients ranging from + 0.191 to + 0.288. The strongest associations were concentrated in Burkina Faso, Benin, Togo, Ghana, Uganda, Mozambique, and Madagascar. These findings highlight how household crowding can exacerbate malaria transmission in specific geographic settings. Larger households may increase the likelihood of intra-household transmission, as a single infected member can serve as a persistent source of exposure for others [74]. Additionally, the financial strain associated with supporting a larger family may limit the household's ability to invest in effective prevention strategies such as insecticide-treated nets (ITNs), indoor residual spraying (IRS), and timely access to antimalarial treatment or vaccinations. This resource dilution effect is particularly concerning in low-income settings where malaria burden is already high [103]. Moreover, limited sleeping space in overcrowded homes may reduce the protective effectiveness of available bed nets, further increasing the vulnerability of all members [104]. These patterns underscore the need for household-level interventions that prioritize resource allocation and health education in densely populated, low-resource communities.

The geographically weighted regression (GWR) analysis identified a strong positive spatial association between malaria prevalence and the proportion of households not using insecticide-treated bed nets. Local GWR coefficients ranged from + 0.241 to + 0.423, with the highest values concentrated in Ghana, Benin, Togo, Côte d'Ivoire, and parts of Tanzania. The costly expense of bed nets, the absence of free nets (omitted during free distribution), and the anxiety about the discomfort and heat that come with sleeping under bed nets were some of the reasons put forth for the lack of bed nets [105]. Another explanation for this could be that bed nets act as a physical barrier for the vectors that cause malaria; however, the absence of usage makes it easier for people to be bitten.

In addition, this study identified a strong and spatially variable negative association between malaria prevalence and households with walls constructed from natural or rudimentary materials. The geographically weighted regression (GWR) model produced local coefficients ranging from − 0.365 to − 0.217, with the most pronounced negative associations observed in Madagascar, Mozambique, and the Republic of Congo. This counterintuitive pattern may reflect complex interactions between housing quality and other malaria prevention strategies. While improved housing is typically associated with lower malaria risk, these findings suggest that households with basic wall materials may be more likely to implement other compensatory preventive measures, such as consistent use of insecticide-treated nets (ITNs), indoor residual spraying (IRS), or targeted environmental management practices. Additionally, in some settings, these materials may allow for greater ventilation or smoke dispersion from indoor cooking fires, which can inadvertently reduce indoor mosquito presence [106, 107]. These insights emphasize that while housing structure is a key determinant of malaria risk; its impact is mediated by complementary interventions. Public health strategies should therefore combine structural housing improvements with ventilation enhancements and vector control measures to maximize protection in high-risk communities.

Furthermore, the geographically weighted regression (GWR) analysis demonstrated a spatially variable but generally positive association between malaria prevalence and the proportion of households with roofs constructed from natural or rudimentary materials. GWR coefficient values ranged from + 0.184 to + 0.309, with the most pronounced associations observed in Rwanda, Tanzania, Uganda, and Mozambique. This pattern highlights a potential structural vulnerability: poorly constructed or thatched roofs may increase exposure to malaria vectors by offering more entry points for mosquitoes, especially in the absence of ceiling barriers or protective netting [108]. In addition, daytime indoor temperature may be higher in houses with metal roofs and no ceilings than in traditional thatched homes [109], which, if exceeding a certain threshold could lower mosquito survival [110] and reduce parasite development within the mosquito [111]. Traditional thatched roofs may maintain favorable microenvironments for mosquito survival, potentially facilitating sustained transmission. These findings suggest that while rudimentary roofing may be a marker of poverty or inadequate housing, the interaction between construction type and indoor climate should be considered in malaria control strategies. Thus, beyond providing insecticide-treated nets, programmes aimed at improving housing quality especially roofing and ceiling insulation may offer sustainable and complementary protection against malaria in high-risk regions.

The geographically weighted regression (GWR) analysis identified a spatially varying but generally positive association between rural residence and malaria prevalence, with local GWR coefficients ranging from + 0.198 to + 0.342. Notably, the strongest associations were observed in Nigeria, Togo, Benin, Ghana, Tanzania, and parts of Gabon. These findings underscore that rural households in specific regions bear a disproportionate burden of malaria, which may reflect a combination of ecological, infrastructural, and service-delivery challenges. One plausible explanation for this spatial trend is the greater presence of mosquito breeding sites in rural settings, particularly due to proximity to agricultural fields, stagnant water bodies, and inadequate environmental management. For example studies in Ghana [112] and Burkina Faso [113] demonstrate that dwellings situated near irrigation sites and water reservoirs are more likely to experience increased malaria morbidity due to persistent exposure to Anopheles mosquitoes. Additionally, rural areas often suffer from weaker healthcare infrastructure and limited access to preventive measures such as insecticide-treated nets (ITNs) and indoor residual spraying (IRS) [114]. Public health messaging and health education campaigns may also have limited reach in these regions due to lower media penetration [115]. The combination of higher environmental exposure and reduced access to timely prevention and treatment increases vulnerability among rural populations. These results highlight the need for geographically targeted malaria control interventions that prioritize rural zones with high GWR coefficients. Beyond vector control, integrated strategies that address infrastructure, healthcare access, and environmental management are critical to reducing malaria transmission in these high-risk rural settings.

Conclusions and implication of the study

Malaria remains a significant public health challenge in Sub-Saharan Africa, with a prevalence of 22.47%, underscoring the urgent need for spatially targeted and context-specific interventions. The spatial analyses conducted in this study including SaTScan identified Benin, Burkina Faso, Togo, and Uganda as high-rate clusters. Geographically weighted regression demonstrate substantial geographic variation in malaria burden and highlight the influence of socio-demographic and environmental factors. Key risk factors identified include rural residence, poor housing conditions, lack of bed nets, homes using treated bed nets, and lower socioeconomic status. To effectively reduce malaria transmission and achieve targets set by the Sustainable Development Goals (SDGs) and the WHO Global Technical Strategy for Malaria (2016–2030), priority should be given to strengthening interventions in high-risk rural areas and geographic hotspots. Expanding access to preventive tools, such as insecticide-treated nets (ITNs) and indoor residual spraying (IRS), especially among low-income and large households, is essential. Additionally, targeted health education and community-based behaviour change programmes must empower households utilizing treated bed net to adopt effective practices. Governments and policymakers should foster strong public–private partnerships to ensure affordable and equitable access to malaria control resources, while community-based organizations and NGOs can support local engagement and intervention delivery. Finally, tailored multi-sectoral strategies focusing on areas with high spatial clustering of malaria will be critical to reducing malaria-related morbidity and mortality across Sub-Saharan Africa.

Strength of the study

This study offers several notable strengths. It is the first study to apply spatial and geographically weighted regression (GWR) analyses to investigate malaria prevalence and its associated factors across 19 Sub-Saharan African countries using Demographic and Health Survey (DHS) data. The analysis benefits from a large, weighted, and nationally representative dataset, enhancing the external validity and generalizability of the findings. Methodologically, the study integrates advanced spatial techniques such as Geographic Information Systems (GIS), SaTScan cluster detection, and GWR. These tools not only enabled the identification of statistically significant malaria hotspots but also provided geographically localized estimates of predictor effects offering greater nuance than traditional global models.

Limitations of the study

Despite its strengths, this study has several limitations. First, the cross-sectional design of the DHS limits the ability to draw causal inferences. Second, the absence of meteorological variables such as rainfall, temperature, and humidity may introduce omitted variable bias, as these are well-known determinants of malaria transmission. Third, not all sub-Saharan African countries had complete malaria data available in the DHS, which limits regional coverage. Fourth, rapid diagnostic test (RDT) results used to estimate malaria prevalence may be subject to misclassification bias, potentially under- or overestimating the true disease burden. Additionally, the study’s reliance on household-level data may obscure within-region heterogeneity due to limitations in spatial resolution. Although the use of GWR accounts for spatial heterogeneity, it does not replace the need to consider hierarchical structure of DHS data. Finally, the use of self-reported and observational variables may introduce reporting bias or residual confounding not accounted for in the models.

Abbreviations

AORs

Adjusted odds ratios

CI

Confidence interval

DHS

Demographic Health survey

EA

Enumeration Area

GWR

Geographical weighted regression

HHH

Household head

IRS

Insecticide residual spray

ITN

Insecticide-treated bed net

LLR

Log-likelihood ratio

NGO

Non-governmental organization

OLS

Ordinary Least Squares

SDG

Sustainable development goal

SSA

Sub-Sahara Africa

STATA

Statistical software for data science

Author contributions

G.Y. participated in conceptualization, data abstraction, selection, data extraction, data analysis, and manuscript writing. L.D., E.A.W., A.A.A., and E.A.A. participated in data extraction and management. A.B., M.N.A., and M.T. participated in data analysis and manuscript writing. G.Y. and L.D. guided the work and manuscript writing. All of the authors have read and approved the final manuscript.

Funding

This study did not receive any funding.

Data availability

The datasets used in this study are available from the DHS website [https://dhsprogram.com/Data/] upon request from the DHS program team. Written permission to use the data was obtained from Measure DHS. All data generated or analyzed during this study are included in this published article [and its supplementary information files].

Declarations

Ethics approval and consent to participate

After submitting an online request for access to sub-Saharan African DHS data, a permission letter from www.dhsprogram.com allowed the study to proceed. The datasets were handled with complete assurance. The sub-Saharan DHS provided the secondary data used in this investigation. Independent consent is not relevant for us because the DHS office handled informed consent, confidentiality, anonymity, and privacy of study participants ethically. The data was not used outside of this study.

Informed consent statement

Informed consent and anonymity do not apply to our study as these were done after the responsible implementing institution collected the data.

Competing interests

The authors declare no competing interests.