Spatial patterns and determinants of under-five mortality in Sub-Saharan Africa: a spatial and multilevel analysis

Abstract

Background

Sub-Saharan Africa is still the region having the highest burden of under-five mortality rate in the world. Of 4.8 million under-five deaths in 2023, more than 80 percent of under-five death reported from Sub-Saharan Africa and Southern Asia. While previous studies have examined the determinants of under-five mortality in individual countries, there is limited evidence on its spatial distribution and multilevel determinants across the region. This study aimed to examine the spatial pattern and identify significant factors of under-five mortality in Sub-Saharan African countries.

Method

This study explored the Demographic and Health Survey (DHS) data from nine Sub-Saharan African countries conducted between 2016 and 2022, and used a total of 115,335 live births for analysis. The multilevel logistic regression model was considered and several nested models were compared using the likelihood ratio test, AIC and BIC criteria. Significant predictors of under-five mortality were reported using adjusted odds ratios (AOR) with 95% Confidence Intervals (CI) and p-values < 0.05.

Results

The spatial distribution of under-five mortality in Sub-Saharan Africa was significantly clustered, as indicated by Moran’s Index of 0.552 (p < 0.001). High mortality rates were observed in Burundi (58.5 per 1,000 live births) and low rates in Kenya (33.9), with an overall rate of 48.9 per 1,000 live births. Increased risk was associated with low maternal education (AOR = 1.57, 95% CI: 1.28-1.91), large family size (AOR = 2.60, 95% CI: 2.41-2.81), and multiple births (AOR = 5.73, 95% CI: 5.01-6.36). On the other hand, a lower risk was observed among children born to employed mothers (AOR = 0.90, 95% CI: 0.84-0.97), whose mothers used contraceptives (AOR = 0.59, 95% CI: 0.55-0.63), and those delivered at health facilities (AOR = 0.83, 95% CI: 0.77-0.90).

Conclusions

The study reveals a significant spatial clustering of under-five mortality across Sub-Saharan Africa, with an overall rate that exceeds the Sustainable Development Goal (SDG) target of 25 deaths per 1,000 live births by 2030. Family planning, maternal education, and safe delivery practices would be essential in reducing under-five mortality in Sub-Saharan African countries. Empowering women through education and promoting family planning, safe delivery, and income-generating programs are vital to reduce child mortality. Regional collaboration and sharing best practices can further advance child survival across Sub-Saharan Africa.

Text box 1. Contributions to the literature
• The study demonstrates how under-five mortality is not evenly distributed and exhibits clear spatial patterns across Sub-Saharan Africa.
• Women’s empowerment, education, access to healthcare and family planning can play an influential role in child survival.
• Effective interventions from lower-risk countries can be transferred and adapted to improve child survival in higher-risk regions of Sub-Saharan Africa.

Background

Under-five mortality (U5M), remains a critical global health issue [1]. The Sustainable Development Goals (SDGs), set by the UN in 2015, aim to reduce the under-five mortality rate (U5MR) to 25 per 1,000 live births by 2030 [2]. However, UNICEF reports 38 per 1,000 U5MR in 2023, with 13,800 preventable under-five deaths [3]. In Sub-Saharan Africa, mortality rate was 51.32 per 1,000 live births, which significantly exceeds the SDG target of 25 per 1,000 live births [4].

Despite global progress in reducing U5M, significant regional disparities still persist [5]. In 2023, an estimated 4.8 million children under-five died worldwide. Over 80% of these deaths occurred in low and middle-income countries, with Sub‑Saharan African children facing disproportionately higher risks. According to the report by UNICEF and WHO (2024), U5MRs remain high in Subb-Saharan Africa, with most countries exceeding the global average of 38 deaths per 1,000 live births [6]. A child born in Sub-Saharan countries is about 18 times more likely to die before age five compared to one born in high-income regions [6].

The literature shows that U5M has a significant economic impact on Sub-Saharan Africa. It not only deprives children of their potential and leaves families grieving, but also hinders overall development and well-being in the region. It results in a loss of human capital, a decrease in the skilled labor pool, and hampers economic growth. The burden also includes healthcare costs and support for affected families. Therefore, addressing under-five mortality would be crucial for both individuals and the sustainable development of Sub-Saharan African countries [7–9]. Infectious diseases such as acute respiratory infections, diarrhea, and malaria, along with complications from preterm birth, birth asphyxia, trauma, and congenital anomalies, are regarded as the leading causes of death for children under-five [10].

A few studies have been conducted on under-five mortality in Sub-Saharan Africa [4, 11]. However, most studies are country-specific, and while informative, they rarely consider how shared environmental, demographic, and socioeconomic factors may jointly affect child mortality across neighboring countries. As a result, these isolated country-level studies provide limited insight into regional patterns and overlook spatial interdependencies. However, understanding cross-country connections is essential for developing regionally coordinated interventions.

This study examined the spatial patterns and factors associated with U5M in Sub-Saharan African countries. Nine countries that have conducted Demographic and Health Survey (DHS) between 2016 and 2022 were included in the study. These countries include Ethiopia, Kenya, Uganda, Rwanda, Burundi, Tanzania, Madagascar, Zambia, and Malawi. We evaluate the spatial dependence of the neighbouring countries across the regions, aiming to identify hotspot countries for U5M by mapping clustering observations. This is achieved by analyzing the spatial autocorrelation, highlighting clustering or dispersion patterns [12]. The Moran’s I which is a spatial autocorrelation statistic was to measure the overall degree of spatial dependence in a dataset. Whereas The Getis-Ord Gi was used to identify hot spot and cold spot areas based on GPS coordinates collected at the nearest community centers for enumeration areas [13].

In this study a multilevel analysis was applied to address the variablity at both the individual and community levels [14]. The individual-level variables include a range of demographic, socioeconomic, and maternal healthcare-related factors, while the community-level variables include the place of residence and country of the children. Integrating spatial and multilevel approaches can provide a better exploration of the hotspots of high U5MRs and the determinants of under-five mortality.

The remainder of this paper is organized as follows. The “Methods” section describes the DHS dataset used in this study and provides an overview of the spatial and multilevel models. The “Results” section presents the key findings from the analysis. The “Discussion” section reflects on these findings in line with existing literature. Finally, the “Conclusions” section summarizes the main contributions of the study.

Methods

Study setting and data source

This study used the most recent Demographic and Health Survey (DHS) data from nine Sub-Saharan African countries that conducted the survey. These include DHS data from Ethiopia (2016), Kenya (2022), Uganda (2016), Rwanda (2019), Burundi (2017), Tanzania (2016), Madagascar (2021), Zambia (2018) and Malawi (2016). To ensure the analysis was representative, we applied the women’s sample weight as recommended by DHS guidelines, accounting for the complex sampling design involving various strata and clusters. After weighting, the study included a total of 115,335 births that occurred within five years prior to each country’s respective DHS survey. The DHS follows standardized sampling procedures, questionnaires, and data collection methodologies, enabling consistent cross-country comparisons [15]. A two-stage sampling technique was used: in the first stage, Enumeration Areas (EAs) were randomly selected; in the second stage, a fixed number of households were systematically selected within each EA. In the DHS surveys, women aged 15–49 and men aged 15–64 were interviewed in the selected households as part of the DHS data collection procedure. However, this study utilized the Kids Record (KR) file, which includes data on mothers who gave birth within five years prior to the survey.

The Demographic and Health Surveys (DHS) data are publicly available and contain no personal identifiers; they are accessible from https://dhsprogram.com/. For this study, permission to use the data was obtained through the DHS Program, which had received ethical approval from the Inner City Fund (ICF) Institutional Review Board (IRB) and the National Ethics Committees of each participating country.

Inclusion and exclusion criteria

The DHS includes all women aged 15–49 years who are either permanent residents of the selected households or visitors who stayed in the household the night before the survey was conducted. For this study, we included live births to these women that occurred within five years preceding each country’s respective survey. Children born more than five years before the survey or to women who were not eligible for the DHS women’s interview were excluded.

Variable description

Dependent variable

The outcome variable of this study, denoted by , is a binary variable indicating whether the child was alive or dead. That is,

1

Independent variables

The study considers a range of individual- and community-level variables as independent variables. The individual-level variables include demographic, socioeconomic, and maternal healthcare-related factors.

The demographic variables are maternal age (regrouped following [4] as 0 = 15–24, 1 = 25–34, 2 = 35+ years), sex of a child (0 = male, 1 = female), sex of the household head (0 = male, 1 = female), marital status (0 = single, 1 = married, 2 = divorced, 3 = widowed), birth order (1 = 1^st, 2 = 2^nd- 4^th, 3 = 5^th), birth result (coded as 0 = single, 1 = multiple), and family size (0 = 1–3, 1 = 4–6, 2 = 7+ members). The socioeconomic variables include maternal education (0 = illiterate, 1 = primary, 2 = secondary, 3 = higher), maternal occupation (0 = working, 1 = not working), and wealth index (0 = poorest, 1 = poorer, 2 = middle, 3 = richer, 4 = richest). The variables related to maternal healthcare services include the mode of delivery by caesarean section (0 = Yes, 1 = No), the place of delivery (0 = Home, 1 = Health facility), and the use of contraceptive methods (0 = Yes, 1 = No).

The community-level variables are the place of residence (0 = Urban, 1 = Rural) and country, coded as 1 = Ethiopia (ET), 2 = Burundi (BU), 3 = Kenya (KE), 4 = Rwanda (RW), 5 = Tanzania (TZ), 6 = Uganda (UG), 7 = Zambia (ZM), 8 = Malawi (MW), and 9 = Madagascar (MD).

Spatial data analysis and autocorrelation

Spatial data analysis

Spatial statistical analysis involves examination of geographic phenomena, focusing on location, distance, spatial arrangement, and interactions. It incorporates both attribute and spatial information, aiming to determine whether the event patterns are random or clustered. In general, spatial data are identified at specific spatial locations, , represented in two or three dimensions, often defined by latitude and longitude.

The spatial autocorrelation analysis was applied to test whether the observed values of variables in specific locations were independent of the values in neighboring locations. This analysis provides insight into the spatial patterns of geographical features, enabling the detection of clustering, randomness, or dispersion [16]. Positive spatial autocorrelation indicates clustering of similar values, while negative spatial autocorrelation suggests dispersion. In the absence of significant spatial autocorrelation, the distribution is random. The Moran’s I index was used to measure spatial autocorrelation, which determines if the risk of under-five mortality in a particular cluster is similar to that of neighboring clusters. The Getis-Ord Gi, a local spatial statistic, was used to identify hot spot and cold spot areas based on GPS coordinates collected at the nearest community centers for enumeration areas [13]. Furthermore, we used the kriging interpolation to estimate under-five mortality in areas of the countries that were not sampled, using measurements from sampled EAs.

Intra-class correlation (ICC)

The proportion of variance explained by clustering was measured using the ICC, which is computed by:

2

where is cluster-level variance, and 3.29 is the individual-level residual variance [17]. ICC values close to one indicate strong clustering, while zero suggests minimal clustering.

The multilevel logistic regression model

The multilevel logistic regression model is an extension of the ordinary logistic regression model for hierarchical data, explicitly accounting for clustering, such as individuals nested within clusters [18]. A two-level binary logistic regression model was used to analyze the child mortality, where the outcome variable is as defined in Dependent variable section. Let denotes the value of the child in the cluster, then the probability of death, , is modeled using a logit link function, incorporating random cluster effects [19]. Specifically, the two-level model can be specified as:

3

where is the overall average intercept and represents level-1 (individual-level) covariates for individual i in cluster j. denotes level-2 (cluster-level) covariates for cluster j and and are fixed-effect coefficients for level-1 and level-2 predictors, respectively. is a random intercept capturing the deviation of cluster j from the average intercept and is the random slope for covariate in cluster j, representing how the effect of varies across clusters.

In the analysis of this study, a two-level binary logistic regression model without any explanatory variables (null model) was initially fitted to examine the cluster-level variance in child mortality [19]. Note that the null model is also called the empty model or the intercept-only model. The log-odds of for the empty model specified as:

4

The next step in our analysis was to consider the random intercept model which includes individual-level predictors and community-level predictors , modeling log-odds as [19]:

5

where: is the fixed intercept, and are fixed-effect coefficients for individual- and community-level covariates, respectively, represents random intercept for clusters and measures the variability among the cluster. Finally, the model is extended by incorporating random slopes for individual-level predictors, allowing the effects of these predictors to vary across clusters as in Eq. 3.

Parameter estimation and goodness of fit test

The maximum likelihood estimation (MLE) method was applied to estimate regression coefficients and variance components of a multilevel model [14]. As this study involves a relatively low-dimensional random effects structure, the Laplace approximation was used to maximize the numerically integrated marginal likelihood [14]. The asymptotic chi-square mixture distribution test proposed by Stram et al. [20] was used to assess the variance of the random intercept, which was found to be significantly different from zero, i.e., versus . The deviance statistic from the maximum likelihood procedure was computed to evaluate model fit; this statistic indicates how well the model aligns with the data. This approach is commonly used to assess goodness of fit in generalized linear mixed models (GLMMs).

All data analyses were done using R statistical software [21], but the data management was handled using SPSS Version 27.0 [22]. ArcGIS [23] was used to visualize the spatial distribution of under-five mortality patterns and identify high- and low-risk areas through mapping and hot-spot analysis.

Results

Descriptive summary

Sociodemographic characteristics

Of the total 115,335 children under the age of five, the majority, i.e., 91,207 (79.1%), resided in rural areas. Around 54,331 (47.1%) of the mothers were between the ages of 25 and 34, and 58,655 (50.9%) had completed primary education. The majority of respondents, 97,048 (84.1%), were married, and 27,390 (23.7%) were in the poorest wealth index category. Approximately 90,382 (78.4%) of household heads were male, and 59,893 (51.9%) had households consisting of four to six members (see Table 1).

Table 1.

The demographic, socioeconomic, and maternal healthcare-related characteristics among selected countries in Sub-Saharan Africa ()

Variables	Live	Death	Total	Percent (%)	U5M (%)
Maternal Age
15–24	32, 480	1, 801	34, 281	29.7	5.3
25–34	51, 883	2, 449	54, 331	47.1	4.5
35–49	25, 326	1, 399	26, 724	23.2	5.2
Residence
Urban	23, 043	1, 087	24, 130	20.9	4.5
Rural	86, 645	4, 562	91, 207	79.1	5.0
Maternal Educational Level
No Education	24, 881	1, 477	26, 357	22.9	5.6
Primary	55, 680	2, 975	58, 655	50.9	5.1
Secondary	23, 386	1, 023	24, 409	21.2	4.2
Higher	5, 741	174	5, 915	5.1	2.9
Family Size
1–3	15, 520	1, 572	17, 092	14.8	9.2
4–6	57, 224	2, 670	59, 893	51.9	4.5
7 and More	36, 945	1, 407	38, 352	33.3	3.7
Sex of Household Head
Male	85, 921	4, 462	90, 382	78.4	4.9
Female	23, 768	1, 187	24, 955	21.6	4.8
Wealth Index
Poorest	25, 916	1, 474	27, 390	23.7	5.4
Poorer	23, 017	1, 256	24, 273	21.0	5.2
Middle	21, 243	1, 036	22, 279	19.3	4.7
Richer	20, 354	1, 079	21, 433	18.6	5.0
Richest	19, 157	804	19, 961	17.3	4.0
Birth Outcome
Single	106, 751	5, 008	111, 759	96.9	5.4
Multiple	2, 938	640	3, 578	3.1	17.9
Place of Delivery
Home	28, 461	1, 759	30, 221	26.2	5.8
Facility	81, 227	3, 889	85, 117	73.8	4.6
Sex of Child
Male	55, 245	3, 181	58, 427	50.7	5.4
Female	54, 443	2, 468	56, 911	49.3	4.3
Mode of Delivery by CS
No	101, 672	5, 183	106, 855	92.6	4.9
Yes	8, 017	466	8, 483	7.4	5.5
Marital Status
Single	6, 308	320	6, 628	5.7	4.8
Married	92, 390	4, 658	97, 048	84.1	4.8
Widowed	1, 640	109	1, 749	1.5	6.2
Divorced/Separated	9, 350	562	9, 912	8.6	5.7
Birth Order
1 st Order	26, 773	1, 488	28, 261	24.5	5.3
2nd - 4th Order	53, 237	2, 474	55, 711	48.3	4.4
5th Order	29, 679	1, 687	31, 366	27.2	5.4
Maternal Occupation
Had Working	72, 604	3, 898	76, 502	66.3	4.5
No Working	37, 084	1, 751	38, 835	33.7	5.1
Contraceptive Method
Yes	52, 246	1, 956	54, 203	47.0	3.6
No	57, 442	3, 693	61, 135	53.0	6.0

Abbreviations: CS Cesarean section, N total births, U5M under-five mortality

The vast majority of children, 111,759 (96.9%), were born as singletons, and 58,427 (50.7%) were male. Regarding birth order, 55,711 (48.3%) of the children were - order. More than three-fifths of the respondents, 85,117 (73.8%), gave birth at health facility. Approximately 8,483 (7.4%) of the respondents underwent a cesarean section (CS) for delivering their child.

Under-five mortality

The overall rate of under-five mortality in the selected Sub-Saharan Africa countries was found to be 48.9 (95% CI : 43.2, 54.9) per 1000 live births. We observed variation in mortality rates across the countries ranging from 33.9 (95% CI : 31.2, 36.7) per 1000 live births in Kenya to 58.6 (95% CI : 54.7, 62.6) per 1000 live births in Burundi (see Fig. 1). We can observe that under-five mortality rate was higher for women aged 15–24 (5.3%), followed by those aged 35–49 (5.2%) and 25–34 (4.5%). Of the total, 640 children (17.9%) who were born as multiples died. Similarly, the rate of under-five mortality is higher in rural areas than in urban areas. In rural areas, 5.0% of children under the age of 5 die before their fifth birthday, compared to 4.5% in urban areas. The rate of under-five mortality is also higher among children of mothers with no education than among children of mothers with higher levels of education. 5.6% of children of mothers with no education die before their fifth birthday, compared to 5.1% of children of mothers with primary education, 4.2% of children of mothers with secondary education, and 2.9% of children of mothers with higher education. Moreover, the rate of under-five mortality was higher among male children (5.4%) and women who underwent cesarean section for delivery (5.5%).

Fig. 1.

Forest plot showing the rate of U5M across countries in Sub-Saharan Africa

Spatial analysis of under-five mortality

From our analysis, the estimated Global Moran’s Index was 0.552 with a p-value (see Fig. 2). This indicates a statistically significant spatial clustering pattern. The corresponding normal distribution curve shows that the observed z-score of 5.01 lies far in the right tail, providing strong evidence of positive spatial autocorrelation. The lower panel further supports this, showing that nearby regions share similar values, highlighting spatial clustering in under-five mortality across Sub-Saharan Africa.

Fig. 2.

Spatial autocorrelation of under-five mortality among selected countries in Sub-Saharan Africa

A total of 6736 clusters were included in this study to analyze the spatial distribution of under-five mortality. The blue color indicates the areas with a low proportion of under-five mortality, whereas the red color indicates enumeration areas with a high proportion of under-five mortality. The highest proportions occurred in Ethiopia, Tanzania, and Madagascar. Whereas the low proportion of under-five mortality was registered in Kenya, Zambia, and Malawi (see Fig. 3).

Fig. 3.

Spatial distribution of under-five mortality among selected countries in Sub-Saharan Africa

The local Getis-Ord Gi statistics have identified areas with significant hot spots and cold spots in terms of under-five mortality (see Fig. 4). The color red represents the presence of significant hot spots, indicating high-risk areas for under-five mortality in Burundi, Madagascar, and Ethiopia. On the other hand, the color blue represents the cold spot areas, indicating low-risk areas for under-five mortality. Specifically, these cold spot areas were observed in Kenya and Rwanda.

Fig. 4.

Hot spot analysis of under-five mortality among selected countries in Sub-Saharan Africa

Based on the ordinary kriging interpolation, Burundi, Madagascar, and Ethiopia are predicted to be more risky compared to other countries (see Fig. 5), red color indicates high-risk areas). In the figure, low-risk areas were indicated by the green color; therefore, Kenya, Rwanda, Zambia, and Malawi are predicted to be lower-risk areas for under-five mortality.

Fig. 5.

Ordinary kriging interpolation of under-five mortality among selected countries in Sub-Saharan Africa

The multilevel logistic regression analysis

Before performing a multivariable multilevel logistic analysis, we have performed a univariable multilevel regression, i.e., using only one independent variable. This helps to screen potential predictor variables that can be included in the multivariable analysis. Based on the Hosmer–Lemeshow guideline [24], we retained those variables that had a p-value less than 0.2. Accordingly, the community-level variables, such as place of residence and country, along with individual-level variables, including birth outcome, child’s sex, birth order, maternal education, maternal age, contraceptive use, family size, place and mode of delivery, wealth index, and respondent’s employment status, were incorporated into the model.

Several nested multilevel logistic regression models were fitted to the data and compared for their efficiency. A clustering effect due to variations across countries was also assessed. Results in Table 2 show that applying a two-level model, including the clustering effect, yielded a higher log-likelihood, logLik = −20902, than a single-level model, logLik = −22765. The likelihood ratio test for these models was statistically significant at the 0.05 level of significance, implying that there is strong evidence that the under-five mortality varies among clusters. Of all considered two-lvel models, the random intercept model had the lowest AIC (39501) and BIC (39791) and the highest logLik (= −19721) values (see Table 2). These show that integrating both individual-level and community-level variables provided the best fit for the data. Similar variables that were significant in the univariable analysis were considered when comparing the candidate models.

Table 2.

Comparison of models: Null (with a single level and two levels), random intercept and random coefficients models

Model	AIC	BIC	logLik
Intercept-only a single level	45533	45543	−22765
Intercept-only two levels	41807	41826	−20902
Random intercept	39501	39791	−19721
Random coefficient model	39518	39803	−19765

Abbreviations: AIC Akaike Information Criterion, BIC Bayesian Information Criterion, logLik log-likelihood

Based on the random intercept model, maternal education, birth outcome, sex of a child, place of delivery, family size, contraceptive method, birth order, mode of delivery, and respondent working are significantly associated with the odds of under-five mortality at the 0.05 level of significance (see Table 3). More specifically, the odds of experiencing under-five mortality were 17% lower (AOR = 0.83, 95% CI : 0.77, 0.90) for children delivered in a health center compared to those delivered at home. Children who were part of multiple births had 5.7 times higher odds of under-five mortality (AOR = 5.73, 95% CI : 5.01, 6.36) compared to singletons. Children born to mothers who used a contraceptive method had 41% lower odds of under-five mortality (AOR = 0.59, 95% CI : 0.55, 0.63) compared to children born to mothers who did not use contraception. Similarly, the odds of U5M death are 10% lower among children born from employed mother (AOR=0.90, 95% 0.84, 0.97) compared to children born to non-employed mother.

Table 3.

Results of multilevel logistic regression with random intercept analysis for factors associated with under-five mortality among selected countries in Sub-Saharan Africa

Variables	Estimate	SE	AOR (95% CI)	P-value
Intercept	1.05	0.10	2.88 (2.35, 3.53)	< 0.001
Place of delivery
Home (Ref)			1.00
Facility	−0.19	0.04	0.83 (0.77, 0.90)	< 0.001
Birth outcome
Single (Ref)			1.00
Multiple	1.75	0.05	5.73 (5.16, 6.36)	< 0.001
Contraceptive method
No (Ref)			1.00
Yes	−0.53	0.03	0.59 (0.55, 0.63)	< 0.001
Maternal occupation
No working (Ref)			1.00
Had working	−0.11	0.04	0.90 (0.84, 0.97)	0.004
Sex of child
Female (Ref)			1.00
Male	0.26	0.03	1.29 (1.22, 1.37)	< 0.001
Maternal education
Higher education (Ref)			1.00
Secondary	−0.01	0.04	0.99 (0.91, 1.07)	0.835
Primary	0.16	0.06	1.17 (1.05, 1.31)	0.006
No education	0.45	0.10	1.57 (1.28, 1.91)	< 0.001
Birth order
1 st order (Ref)			1.00
2nd–4th order	−0.01	0.044	0.99 (0.91, 1.08)	0.794
5	−0.32	0.06	0.72 (0.64, 0.82)	< 0.001
Family size
1–3 (Ref)			1.00
4–6	0.96	0.04	2.60 (2.41, 2.81)	< 0.001
7 and more	1.48	0.050	4.38 (3.98, 4.83)	< 0.001
Mode of delivery by CS
No (Ref)			1.00
Yes	−0.21	0.06	0.81 (0.72, 0.91)	< 0.001
Country
Ethiopia (Ref)			1.00
Burundi	0.22	0.09	1.25 (1.05, 1.48)	0.014
Madagascar	0.19	0.10	1.21 (1.00, 1.48)	0.054
Malawi	−0.04	0.09	0.96 (0.81, 1.14)	0.628
Zambia	−0.19	0.09	0.83 (0.69, 0.99)	0.049
Tanzania	−0.11	0.09	0.90 (0.75, 1.07)	0.222
Uganda	−0.04	0.09	0.96 (0.81, 1.13)	0.614
Rwanda	−0.17	0.09	0.85 (0.72, 1.00)	0.053
Kenya	−0.14	0.09	0.87 (0.74, 1.03)	0.113
0.41	ICC	0.11	N Cluster	6736

Abbreviations: AOR Adjusted odds ratio, CI Confidence Interval, CS Cesarean section, Ref Reference category, SE Standard Error

Being male had 29% higher odds of under-five mortality (AOR = 1.29, 95% CI: 1.22, 1.37) than being a female child. Children born to mothers who had no formal education had 57% higher odds of under-five mortality (AOR = 1.57, 95% CI: 1.28–1.91), while children born to mothers with primary education had 17% higher odds (AOR = 1.17, 95% CI: 1.05–1.31) of under-five mortality, respectively, compared to children born to mothers who attained higher education. The odds of under-five mortality are 28% lower among children born fifth or higher in their family (AOR = 0.72, 95% CI: 0.64–0.82.64.82) compared to children born first in their family. The odds of under-five mortality were 160% higher (AOR = 2.60, 95% CI: 2.41, 2.81) among children born into family sizes of four to six, and 338% higher (AOR = 4.38, 95% CI: 3.88, 4.94) among children born into family sizes of seven or more, compared to those born into family sizes of one to three. Finally, the odds of under-five mortality are reduced by 19% for children delivered by cesarean section (AOR = 0.81, 95% CI: 0.72, 0.91) compared to those delivered by non-cesarean methods.

The country of the child (level-2 variable) showed a statistically significant effect on the likelihood of experiencing under-five mortality (p-value < 0.05). This implies that differences between countries contributed to the variation in child mortality rates, even after accounting for the individual-level factors in the model. Children in Burundi had 25% (AOR=1.25, 95%CI: 1.05 - 1.48) higher odds of under-five mortality than children in Ethiopia. Conversely, children in Zambia have the lowest odds of dying before the age of five, which is 17% (AOR = 0.83, 95% CI: 0.61 - 0.91) lower odds of under-five mortality than children in Ethiopia. The estimated intra-class correlation (ICC) based on the two-level logistic regression model is 0.11, which suggests that 11% of the total variability in the under-five mortality was ascribed to the differences between communities (see Table 3).

Discussion

This study contributes to a better understanding of how socioeconomic, demographic, and maternal health factors affect child survival across Sub-Saharan Africa. The findings suggest that improvements in maternal education, reproductive health practices, and access to healthcare services are crucial in reducing under-five mortality. The results are consistent with the idea that maternal knowledge, household resources, and healthcare utilization play vital roles in determining child health outcomes. They also emphasize the need for policies that strengthen maternal and child health programs, expand family planning initiatives, and improve the quality and accessibility of institutional delivery services, particularly in high-risk countries. Such actions are essential for accelerating progress toward achieving the Sustainable Development Goal of reducing under-five mortality to below 25 deaths per 1,000 live births by 2030.

In this study, the Demographic and Health Survey (DHS) data across nine Sub-Saharan African countries, which were conducted between 2016 and 2022, were examined to assess the spatial patterns and associated factors of under-five mortality. The overall rate of under-five mortality was 48.9 per 1,000 live births. Variation in mortality rates was observed across these countries, ranging from 33.9% in Kenya to 58.5% in Burundi. These values are higher than the Sustainable Development Goal (SDG) target by 2030, which is 25 deaths per 1,000 live births. These results signify that extensive interventions to reduce child mortality in Sub-Saharan Africa are needed [25]. Burundi, Madagascar, and Ethiopia are among the high-risk countries for under-five mortality. The family size, birth order, multiple births, maternal education, respondent working, child’s gender, contraceptive method, mode of delivery, and institutional delivery are significantly associated with under-five mortality.

Children whose mothers had no formal education had higher odds of experiencing under-five mortality compared to those whose mothers had attended secondary or higher education. This indicates that better education and understanding about the benefits of childcare have a positive impact on child survival [4, 26–28]. Such care may begin with attending prenatal and postpartum visits to reduce the risk of intrapartum and postpartum complications and having vaccinations or vaccine-preventable diseases [29]. The choice of the place of delivery is also important, as there was a reduction in the risk of under-five mortality for delivering in health facilities compared to giving birth at home. This finding is consistent with several studies conducted in Sub-Saharan African countries [4, 27, 30, 31]. According to our findings, use of contraceptives is another significant factor in reducing child mortality, and this is supported by a previous study by Bitew et al [32]. This association could be due to the fact that contraceptive use leads to longer birth intervals, which ultimately enhances the chances of the child’s survival [33, 34]. In addition, the risk of under-five mortality was significantly less for children delivered by caesarean section than those delivered via vaginal birth. Though there are plenty of studies that support our result [35–37], there are a few findings that contradict this conclusion [38, 39], meaning that reported significantly higher under-five mortality rates for children delivered by cesarean section compared to those delivered vaginally. This evidence suggests that caesarean section should be performed only when medically necessary and in care [40].

Our findings revealed that children in larger families had a significantly higher risk of experiencing under-five mortality. This shows the increasing challenges for larger families to meet their basic needs and provide adequate care [38, 41]. It has been indicated that children born as multiple births under the age of five had a significantly higher risk of experiencing under-five mortality compared to those born as single births. This result is in line with conclusions about potential medical complications, both short-term and long-term, that often accompany multiple births [27, 42, 43]. The current study indicated that being the firstborn child increased the chances of under-five mortality compared to those with five or more siblings. One possible explanation for this is that the firstborns are often born to younger and less experienced mothers, which may increase the risk of congenital malformations and perinatal conditions [41, 44].

This study revealed that mothers who were employed had a lower risk of experiencing under-five mortality compared to mothers who were unemployed. This may be because mothers who have their own income can provide sufficient services and quality of life and improve the standard of living to improve their children’s survival rate [45]. Finally, this study indicated that male children were at increased risk of under-five mortality that may be due to biological factors. For example males are more vulnerable to morbidities such as low Apgar score and intrauterine growth restriction (IUGR), respiratory insufficiency, and prematurity than females, which can lead to higher mortality and short-term morbidity [46–49].

Conclusions

The spatial distribution of under-five mortality was found to be significantly clustered among the selected Sub-Saharan African countries. High mortality rates were observed in Burundi, Madagascar, and Ethiopia, whereas Kenya and Rwanda exhibited relatively lower rates. However, all countries remain above the Sustainable Development Goal (SDG) target of 25 deaths per 1,000 live births by 2030, with an overall rate of 48.9 per 1,000 live births. Strengthening women’s access to education in high-risk regions is crucial for improving healthcare decision-making and reducing child mortality. Lessons from lower-risk countries can guide the design of effective interventions. Promoting contraceptive use, family planning, and safe delivery practices, together with income-generating initiatives, can further enhance child survival. Finally, fostering regional collaboration and sharing best practices will be crucial to accelerating progress toward reducing under-five mortality and improving child well-being across Sub-Saharan Africa.

Abbreviations

AIC

Akaike Information Criteria

AOR

Adjusted Odds Ratio

ArcGIS

Aeronautical Reconnaissance Coverage Geographic Information System

BIC

Bayesian Information Criteria

CI

Confidence Interval

DHS

Demographic and Health Survey

EA

Enumeration Area

ICC

Intra-Class Correlation

LISA

Local Indicators of Spatial Autocorrelation

MDG

Millennium Development Goals

NGO

Non Governmental Organization

OK

Ordinary Kriging

SAC

Spatial Autocorrelation

SDG

Sustainable Development Goals

SPSS

Statistical Package for the Social Sciences

SSA

Sub-Saharan Africa

U5M

Under-five Mortality

UN

United Nations

UNICEF

United Nations International Children’s Emergency Fund

WHO

World Health Organization

Authors' contributions

CD conceived that idea. ABT and DK were involved in the conceptualization and design of the study. CD handled the data extraction, while CD, ABT, DK and LKD ensured data verification. Statistical analysis was performed by CD and AY. The manuscript was written by CD, AY, and ABT, with LKD providing a critical review. All authors read and approved the manuscript.

Data availability

The dataset used and analyzed in this study is openly available from DHS website https://dhsprogram.com/.

Declarations

Ethics approval and consent to participate

Ethics approval was not required, as the dataset is publicly available.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.