Located in the heart of West Africa, Burkina Faso is a landlocked country that shares its borders with six other countries: Mali to the north and west, Niger to the east, Côte d'Ivoire, Togo, Ghana and Benin to the south. Burkina Faso inherits a dry, tropical Sudano-Sahelian climate, characterized by highly variable rainfall ranging from 350 mm in the north of the country to over 1,000 mm in the southwest. This climatic variability, more or less erratic from one geographical area to another, has an impact both on the availability of water resources for agricultural production and on children’s health. Demographically, with a high fertility rate, the country has always been characterized by a young population. The under-15 age group has consistently accounted for over 45% of the total population [12]. The fact that the population is so young places a heavy burden on the region in terms of meeting basic social needs. On the socio-economic front, enormous challenges remain. In 2019, only 29.7% of Burkina Faso’s population was literate [12]. As for the incidence of income poverty, it has not fallen significantly for decades. It fell from 44.5% in 1994 to 47% in 2009, then to 43.2% in 2021 [31].

In Burkina Faso, child mortality is falling steadily. Mortality among children under five has fallen from 187 deaths per 1,000 births in 1993 to 129 deaths per 1,000 births in 2010, and to 48 deaths per thousand in 2021 [11, 12]. As for neonatal mortality, it was eighteen (18) newborn deaths per thousand in 2021 compared with 43‰ in 1993. Although these mortality rates are steadily falling, they remain high and far from the target of 25‰ (for under-five mortality) aimed for in the SDGs by 2030. Malaria and respiratory infections are in fact the main causes of death in children under five, both in routine data from the Ministry of Health and in population observatories [32–34]. These two types of disease are responsible for more than half of all child deaths in the Ouagadougou Health and Demographic Surveillance System (HDSS) [32].

The healthcare system in Burkina Faso is organized as a three-tiered pyramid. The first level, known as the community level, comprises the Health and Social Promotion Centers (CSPS), medical centers, and medical centers with surgical units (CMA). This level serves as the main point of entry into the system and provides primary healthcare, including the management of uncomplicated deliveries, preventive and basic curative services, and health promotion activities. The second level consists of regional hospitals, which are equipped with more advanced technical facilities. These hospitals provide specialized care, act as referral centers for district-level facilities, and oversee the technical supervision of frontline structures. The third level includes national and university hospitals (CHU and CHN), located mainly in large cities such as Ouagadougou and Bobo-Dioulasso. This level manages complex and highly specialized cases and also serves as a hub for medical training and research. The pyramid is designed to ensure a progressive continuum of care, from community-based services to specialized treatment, supported by a referral and counter-referral system. However, in practice, several challenges remain, including geographic inequalities in facility distribution, a shortage of qualified health workers, and weaknesses in the referral system, which undermine efficiency, particularly in rural and peripheral areas.

In recent years, Burkina Faso has introduced several important reforms aimed at strengthening primary healthcare, expanding community-based health services, and improving access to maternal and neonatal care. Key measures include the subsidization and later full removal of user fees for deliveries and emergency obstetric and neonatal care (SONU), as well as the free healthcare policy introduced in 2016, which covers children under five and pregnant and breastfeeding women. These initiatives were designed to reduce financial barriers and promote equitable access to essential services, although significant territorial and contextual disparities persist.

Administratively, the country had a total of 13 regions, 45 provinces and 351 communes. The commune is the last and best-organized administrative entity, with local political management bodies. Figure 1 below illustrates the geographical structure of these administrative entities.

The main data source used in this research is the Burkina Faso Demographic and Health Survey (DHS). It was carried out from July 30 to November 30, 2021 by National Institute of Statistics and Demography (INSD) and the DHS Program. The survey sample is nationally representative, as well as representative by region and by place of residence (urban and rural). The DHS uses a two-stage stratified sampling design. First, enumeration areas are stratified by region and urban–rural residence, then clusters are selected within each stratum using probability proportional to size. In the second stage, a fixed number of households is randomly selected from each cluster after a household listing operation. The main sampling unit is the cluster. Each region was subdivided into urban and rural components to create sampling strata, with independent sampling carried out in each stratum. In all, twenty-six strata were established. In the first sampling stage, 514 clusters were selected with probability proportional to size. In the second sampling stage, 26 households were systematically selected with equal probability from each of the clusters identified in the first stage. The survey covered a representative sample of 17,659 women of childbearing age. The collection method was a direct interview with the targets, using questionnaires implemented on CAPI.

Our analysis sample covers live births to women aged 15 to 49 surveyed during the DHS 2021. In all, 7,225 births were involved. These are live births to 6,627 women aged 15 to 49. These are births that took place in the last 3 years prior to the survey. A total of 137 neonatal deaths were recorded over the period.

The variable of interest in this analysis is death of the newborn within 28 days of birth. It is coded as 1 if the newborn died before reaching 28 days of life, and 0 otherwise. In addition, the independent variables taken into account include the child’s sex, mother’s age at birth, place of delivery, birth order, intergenital interval, type of pregnancy (single or multiple), mother’s level of education, cluster distance from the nearest health and social promotion center, place of residence, region of residence, mother’s marital status and household standard of living. The measure of household living standards used in this analysis is the DHS wealth index, obtained directly from the DHS database. This variable is constructed using a principal component analysis (PCA) applied to a set of household assets and amenities (e.g., housing characteristics, access to water and sanitation, ownership of durable goods). The first component of the PCA, which captures the largest share of variance in asset ownership, is retained to generate a continuous wealth score for each household. These scores are then standardized (mean = 0, standard deviation = 1) to allow comparability across the sample. Finally, households are ranked and typically grouped into five quintiles of wealth (poorest, poorer, middle, richer, richest), which serve as a widely used proxy for socioeconomic status in demographic and health research.

Two analytical approaches are employed in this research. First, the determinants of perinatal mortality are examined using a log-binomial regression model, which directly estimates risk ratios. To assess the validity of the model, a likelihood ratio test is performed by comparing the complete model with the null model, thereby providing an overall test of model significance. In addition, the predictive quality of the model is evaluated using the Receiver Operating Characteristic (ROC) curve, a method widely recognized as one of the most reliable tools for assessing predictive performance in regression models [35]. A model with an Area Under the Curve (AUC) greater than 70% is generally considered acceptable, reflecting a fairly good predictive ability.

The regression model is expressed as follows: y i ∼ B e r n o u l l i   p i log p i = β 0 + ∑ p = 1 P β p x p i (1) p i = exp β 0 + ∑ p = 1 P β p x p i , 0   < p i < 1 . (2)

Where p i is the probability that a child dies before 28 days after birth; x p i are the predictor variables; the β p are the coefficients; and β 0 is the intercept. Under the log-binomial model, exp( β k ) is a risk ratio for a one-unit increase in x k (See Equations 1, 2).

In the second part, we use geospatial analysis to explore communal inequalities in neonatal mortality. This analysis is carried out in two stages.

As a first step, we assessed the spatial autocorrelation of neonatal death proportions between communes using the global Moran index, in order to verify the relevance of a spatial approach. Geospatial weighting was performed by identifying neighboring communes according to a proximity criterion: two communes are considered neighbors if they share a border (Tobler’s law). This principle is based on the law of geography formulated by Waldo Tobler, according to which “everything is connected to everything, but things that are close are more so” [36, 37]. It highlights the importance of geographical distance in structuring spatial relationships. A matrix of weights was thus constructed, attributing to each commune an influence proportional to the number of its neighbors. This weighting was used to calculate the Moran index and identify any significant spatial organization of neonatal mortality rates on a communal scale.

Secondly, to analyze spatial inequalities in neonatal mortality between communes, we used a Bayesian model with a binomial distribution, thanks to the Integrated Nested Laplace Approximation (INLA) package. This type of model makes it possible to simultaneously take into account the specific characteristics of each commune, as well as the spatial dependency effects linked to their vicinity. To this end, we have incorporated a spatial effect based on the Besag–York–Mollié (BYM) 2 model, which combines local variations with a neighborhood structure. For the model parameters, a priori distributions were specified. For the spatial effect, a relatively flexible configuration was chosen (parameters: 0.5 and 0.01) so as not to impose too strong an assumption from the outset. For the regression coefficients, an a priori distribution centered on 0 with a large uncertainty was used, corresponding to a cautious approach letting the data guide the results.

The mathematical formulation is as follows:

We consider children i living in communes j = 1 , … , 351 .

The log-binomial model is specified as: log p i j = β 0 + ∑ p = 1 P β p   x p , i j + u j

β 0 : intercept (basic risk log).

Before examining neonatal mortality according to individual and contextual variables, we first present the raw estimates at the regional level (cf. Figure 2). At the national level, the neonatal mortality rate is estimated at 18.05 deaths per 1,000 live births (95% CI: 14.6–21.5). However, major regional disparities are apparent, even if they are not appreciably significant. For example, while the Centre region has a neonatal mortality quotient of around 12.1 (95% CI: 2.8–21.4), the Cascades region has almost 51 (95% CI: 18.0–83.0) neonatal deaths per 1,000 live births.

Furthermore, the descriptive analysis reveals significant disparities in the occurrence of neonatal deaths, linked to the socio-economic, demographic and contextual characteristics of mothers and births (Table 1). Analysis by the length of the interbirth interval reveals marked differences (p < 0.001). Children born after an interval of less than 2 years have a neonatal mortality incidence of 5.46%, compared with 0.88% when the interval is between two and 3 years. Sex is another dimension where differences are observed: boys have a neonatal mortality incidence of 2.36%, significantly higher than that of girls (1.41%). According to the nature of the birth, the results show that multiple births have a neonatal mortality incidence of 6.37%, significantly higher than that of single births (1.69%, p < 0.001).

With regard to the mother’s age at delivery, children born to mothers aged 35 to 49 have a neonatal mortality incidence of 3.21%, higher than that observed among children born to younger mothers (18–34 years), with a statistically significant difference (p < 0.001).

Differences were also observed according to ethnicity. Children born to mothers belonging to the Bobo (2.89%) and Peulh (2.69%) groups had higher neonatal mortality rates than those of other ethnic groups (p < 0.001). Likewise, the results relating to the means of transport used to reach the place of delivery indicate a marked disparity: the use of non-conventional means or public transport is associated with very high incidences of neonatal mortality, reaching 14.29%, which reflects significant difficulties in accessing obstetric services.

Finally, the presence of a child under five living in the household significantly influences the incidence of neonatal mortality. Indeed, the incidence of neonatal mortality is 22.14% when there is at least one child under 1 year in the household.

The set of variables included in the analysis model gives an AUC of 0.81 (as shown by the ROC curve in Supplementary Figure S1), indicating good discriminating ability and providing non-random predictions. Thus, the holistic analysis of neonatal mortality highlights a set of sociodemographic, biological and contextual factors significantly associated with an increased probability of death within the first 28 days of life (See Table 2).

Among child characteristics, male sex stands out as a factor strongly associated with excess neonatal mortality. Compared to girls, boys show a significant excess risk (RR = 1.60; p = 0.007), confirming a male biological vulnerability during this critical period. The type of birth is also an important differentiating factor. Children from multiple births are almost three times more likely to die within the first 28 days of life than those from single births (RR = 3.63; p < 0.001).

With regard to ethnicity, children born to mothers belonging to Peulh groups were 2.11 times more likely to die during the first 28 days of life, compared with those born to mothers belonging to Mossi groups (p = 0.031), reflecting possible cultural, territorial or healthcare access disparities. The inter-genetic interval has a strong protective effect on neonatal mortality. Compared with children born less than 2 years after a previous birth, those born after a birth interval of 3 years or more have a 72% lower risk of dying within the first 28 days of life (p < 0.001). The risk is also by 81% (p < 0.001) among children born after an interval of two to 3 years, and by 54% (p < 0.021) among those from a first pregnancy.

With regard to mother-related factors, the mother’s level of education also plays a role in neonatal mortality. Children born to mothers with primary education are 2,27 more likely to die within the first 28 days of life, compared with children born to mothers with secondary education and above (p = 0.005).

Geographically, some regions have significantly higher mortality risk levels. This is particularly true of the Cascades (RR = 3.87; p = 0.009), Centre-Est (2.36; p = 0.053) and Central Plateau (RR = 2.67; p = 0.033) regions, where children have a significantly higher probability of death than in other regions. Access to care emerges as a critical dimension. Children who seek care in secondary health centers are 44% (p = 0.011) less likely to die before their 28th birthday than those who seek care at home or elsewhere than in a health center. In addition, the means of transport used to access care had a marked influence on newborn survival. The use of public transport is associated with an extremely high risk of neonatal mortality (RR = 8.57; p < 0.001).

The estimate of Moran’s I at the commune scale yields I = −0.0128 (p = 0.561), indicating that there is no significant spatial autocorrelation, which means that neonatal deaths are randomly distributed across communes. In addition, estimating the probability for each commune of achieving a neonatal mortality level less than or equal to the 12‰ threshold by 2030 under SDG 3.2.2 revealed that all communes in Burkina Faso have a low level of risk (predicted probability <0.80). This means that the 351 communes of Burkina Faso will meet SDG 3.2.2, all else being equal. However, 34 communes fall within a range that warrants vigilance (Cf. Supplementary Table S2). All of these communes have probabilities between 0.20 and 0.80, which represents a moderate risk. These include, among others, Kpuéré, Madouba, Legmoin, Boussou-Koula, Niankôrôdougou, Botou, and Batié (Urban commune).

In addition, analysis of relative risk predictions at the commune level, based on Jenks’ discretization, reveals spatial inequalities with a dual center–periphery structure, where neonatal mortality is much higher the farther one moves away from the center of the country, a low-risk zone relative to the country’s peripheries (cf. Figure 3). Peripheral communes such as those mentioned above should be monitored closely. In contrast, the communes of Ziniaré, Nanoro, Ténado, and Boken, among others, present a relatively low risk of neonatal deaths.

This study has a number of strong points that reinforce the reliability of the results obtained to inform public policy. Firstly, it is based on recent, representative and disaggregated national data from the EDS 2021, enabling a detailed analysis at the scale of Burkina Faso’s 351 communes. In addition, the joint use of multivariate statistical models and spatial Bayesian approaches provides a sound methodology for identifying determinants and highlighting spatial inequalities in neonatal mortality.

However, certain limitations need to be highlighted. Indeed, the analysis is based on declarative data, susceptible to recall bias, and on a cross-sectional approach, limiting the causal interpretation of the associations observed. In addition, certain relevant contextual variables, such as the actual quality of care or local conflict dynamics, are not directly integrated. A further limitation relates to the security situation during data collection, which led to incomplete coverage in some regions, including the Sahel, East, Centre-North and Boucle du Mouhoun. This may have contributed to an underestimation of neonatal deaths in certain communes, such as Djibo, where the estimated risk appears unexpectedly low compared to known regional trends.

Despite these limitations, the results obtained remain sufficiently robust and relevant to support the implementation of targeted, territorialized and evidence-based strategies within the framework of neonatal health policies.

The results of this study show that neonatal mortality in Burkina Faso remains a major public health challenge, shaped by strong biological, social, and territorial disparities. While individual factors such as the child’s sex, birth interval, and type of delivery significantly influence the risk of death, structural inequalities related to access to care, socio-economic status, and socio-cultural barriers play an equally decisive role. Spatial analysis revealed a core of priority communes with particularly high risks including both rural and urban areas challenging the assumption that urban environments systematically guarantee better health outcomes. Moreover, the Jenks discretization highlights that medium-risk communes are often concentrated around high-risk ones, pointing to a territorial contiguity of vulnerabilities.

In this context, achieving the Sustainable Development Goals (SDGs) for reducing neonatal mortality requires the implementation of multi-sectoral, territorially differentiated strategies. These must be informed by disaggregated, spatially explicit data, and adapted to the realities of both rural peripheries and underserved urban zones. Only a contextualized and equitable approach, grounded in scientific evidence and sensitive to local specificities, will make it possible to achieve a sustained reduction in neonatal mortality in Burkina Faso [20, 34, 43–45].