Urban air pollution, green space, and pneumonia risk: a spatial analysis across health zones in Kinshasa, Democratic Republic of Congo

Abstract

Background

Pneumonia remains a leading cause of morbidity and hospitalization in sub-Saharan Africa, yet the contributions of long-term environmental and infrastructural factors to its spatial distribution are poorly understood. Kinshasa, a rapidly expanding African megacity, experiences high levels of air pollution, substantial vegetation loss, and marked urban inequalities, potentially exacerbating respiratory vulnerability.

Methods

We conducted a retrospective ecological spatial study using 484,954 pneumonia hospital admissions recorded across 35 health zones in Kinshasa between 2018 and 2022. Long-term exposure to ambient air pollution (PM₂.₅, NO₂, CO), vegetation cover (NDVI), major road density, climatic indicators, and healthcare infrastructure were assessed via satellite-derived and geospatial data. Bayesian BYM2 negative binomial models were fitted to estimate age-stratified relative risks (RRs), accounting for spatial dependence and overdispersion. Sensitivity analyses were used to evaluate model robustness.

Results

Pneumonia incidence showed pronounced spatial heterogeneity, with persistent high-risk zones concentrated in central and northern districts. Children under five years of age accounted for 41% of the cases and presented a substantially higher cumulative incidence than older individuals did. After adjustment for spatial effects and multicollinearity, no air pollutant demonstrated a consistent positive association with pneumonia risk. In contrast, vegetation cover was a robust protective factor across all age groups (RR range: 0.58–0.85). Higher major road density was also associated with reduced risk, likely reflecting improved accessibility and urban infrastructure. Areas combining low vegetation, high climatic stress, and limited infrastructure experienced the highest pneumonia burden.

Conclusions

Pneumonia risk in Kinshasa is driven primarily by environmental and infrastructural inequalities rather than by the isolated effects of individual air pollutants. Strengthening urban green infrastructure, reducing environmental stressors, and improving equitable access to healthcare should be central to pneumonia prevention strategies in rapidly urbanizing African cities.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-026-26318-3.

Introduction

Air pollution is among the most pressing environmental and public health threats of the 21st century, with far-reaching consequences for ecosystems, economies, and human wellbeing. Prolonged exposure to ambient pollutants, particularly fine particulate matter (PM₂.₅), nitrogen dioxide (NO₂), sulfur dioxide (SO₂), carbon monoxide (CO) and ground-level ozone (O₃), has been strongly associated with increased risk of chronic diseases, reduced life expectancy, and increased healthcare burdens [1–3]. Globally, billions of people are exposed to air that exceeds safety thresholds, making air pollution the leading environmental driver of morbidity and mortality [4].

In Africa, common sources of pollution include vehicular traffic, open waste burning, the use of biomass fuel, and unpaved roads, which are compounded by dust mobilization and weak regulatory enforcement [5, 6]. The respiratory system is vulnerable to airborne contaminants. Chronic exposure increases the risk of infections, including pneumonia, bronchitis, and influenza [7, 8]. In 2017, respiratory infections were the fourth leading cause of mortality worldwide [9]. Pneumonia remains the deadliest infectious disease, disproportionately affecting young children and the elderly [10]. Each year, it accounts for approximately 120 million cases and up to 1.3 million deaths, with the burden falling primarily in low- and middle-income countries (LMICs) [11–13]. In addition to its mortality toll, pneumonia is a leading cause of hospital admissions and healthcare expenditures [14].

In sub-Saharan Africa, the air quality crisis has intensified owing to rapid urbanization [15], population growth, and industrial activity [16], often in the absence of effective environmental regulations. Cities such as Nairobi, Kampala, Addis Ababa, and Dar es Salaam have experienced significant increases in emissions from vehicles, biomass burning, unregulated small-scale industries, and open dumpsites [17]. Ground measurements and satellite data have frequently revealed that PM₂.₅ levels exceed WHO guidelines, with concentrations peaking during dry seasons due to meteorological inversions and regional dust transport [18]. In Nairobi, PM₂.₅ values between 22 and 112 µg/m³ have been recorded in dense settlements far above the recommended daily limit of 15 µg/m³ [19]. In Rwanda, including the capital, Kigali, ambient PM₂.₅ concentrations typically range from approximately 13 to 65 µg/m³, and PM₁₀ levels are generally between approximately 39 and 61 µg/m³, most commonly recorded in dense urban settlements [15, 20]. Weak regulatory enforcement, limited monitoring capacity, and fragmented policy frameworks have further constrained mitigation efforts across the region [21].

The health implications of this environmental burden are significant. Studies across Africa have linked long-term exposure to PM₂.₅ and NO₂ with increased incidences of respiratory infections, asthma exacerbations, chronic obstructive pulmonary disease (COPD), and cardiovascular disease [22, 23]. Vulnerable groups, particularly children, older adults, and low-income residents near busy roads and industrial zones, face heightened risk. Traffic-related pollutants and particles from waste burning not only contribute to acute respiratory symptoms but also accelerate lung function decline and increase hospitalization rates [24–26]. In 2015, the highest mortality from acute respiratory infections was recorded in children under five years of age globally [27]. The prevalence of acute lower respiratory infection (ALRI) surpassed 12% in Côte d’Ivoire and Cameroon and reached 16.5% in Senegal [27]. In the Democratic Republic of the Congo (DRC), pneumonia remains the leading cause of childhood mortality, with an estimated prevalence of 12.9% in rural areas and 14.5% in Kinshasa, the capital [28, 29].

Despite these risks, environmental health research in the region remains limited, hampered by data gaps, underresourced health surveillance, and a lack of spatially resolved exposure metrics.

Kinshasa, a megacity with more than 15 million people, exemplifies these challenges. The air quality in cities consistently fails to meet international safety standards. Satellite and in situ monitoring reported that PM₂.₅ concentrations as high as 63.2 µg/m³ in 2016 were more than 12 times the WHO annual guideline [30]. Average annual concentrations of 43 µg/m³ were recorded in 2019 and 2022 [31, 32]. However, despite this alarming exposure, epidemiological evidence connecting long-term pollution levels to pneumonia outcomes in Kinshasa is lacking. While existing studies have examined asthma, chronic obstructive pulmonary disease (COPD), and all acute lower respiratory infections (ALRIs) in the Democratic Republic of the Congo (DRC) [33–36], specific environmental determinants of pneumonia-related hospital admissions remain largely unexplored.

This situation is exacerbated by the deteriorating green infrastructure of cities. Urban vegetation parks, trees, and green corridors play crucial roles in filtering air pollutants, mitigating urban heat, and promoting respiratory health. Globally, green spaces are associated with lower risks of asthma, respiratory illnesses, and cardiovascular conditions [37]. In Kinshasa, however, uncontrolled urban sprawl, informal development, and a lack of planning have led to the degradation of these natural buffers. Despite the increasing global recognition of green infrastructure in public health, its protective effects against pollution-induced pneumonia have yet to be examined in the context of African megacities.

This study addresses this critical knowledge gap by analysing, for the first time, the spatial associations between long-term exposure to PM₂.₅, NO₂, and CO; green space availability; and hospital admissions for pneumonia across 35 health zones in Kinshasa, DRC, from 2018 to 2022. Using a Bayesian spatial regression approach, we estimated the age-stratified relative risks while adjusting for key sociodemographic variables. To the best of our knowledge, this is the first spatially explicit environmental epidemiology study in Kinshasa to jointly evaluate air pollution risk and the potential protective effects of urban greenery on pneumonia. By integrating high-resolution environmental data with hospitalization records across administrative zones, our findings offer actionable insights to guide targeted air quality interventions, equitable greenspace planning, and public health policy in African megacities and other high-risk LMIC settings.

Materials and methods

Study area description

This study focuses on Kinshasa, the sprawling capital city province of the Democratic Republic of the Congo (DRC), which is located in the western part of the country between latitudes 4° and 5°S and longitudes of 15° and 16°E. Covering an area of approximately 9,965 km² [38], Kinshasa has diverse topography ranging from low-lying riverbanks along the Congo River to undulating plateaus inland, with elevations between 200 and 610 m above sea level. As one of Africa’s largest megacities, Kinshasa’s population was estimated to be nearly 13.9 million inhabitants in 2019, with an average population density exceeding 1,400 persons per km², although densities vary widely between densely settled urban districts and more sparsely populated peripheral zones [39].

From a public health perspective, Kinshasa is administratively structured into 35 health zones, which form the core operational units for healthcare provision, resource allocation, and epidemiological monitoring. These zones encompass a wide range of settlement types, from densely populated urban districts to peri-urban communities and extensive rural areas, reflecting the city’s remarkable sociodemographic heterogeneity. Although the rural health zones of Maluku I and Maluku II account for 82.8% of Kinshasa’s landmass (Fig. 1a), their unusually large size introduces visual distortion on spatial maps. For this reason, they were omitted from the cartographic representations (Fig. 1b) but were systematically included in all the statistical analyses to ensure comprehensive spatial coverage and analytical completeness [38].

Fig. 1.

Spatial distribution of health facilities across Kinshasa’s 35 health zones, including the rural Maluku I and II zones (1a). Panel 1b presents the distribution limited to the 33 urban and peri-urban zones, excluding Maluku I (No. 34) and Maluku II (No. 35), whose extensive land area would distort the cartographic representation

Study design and data

This study employed a retrospective spatial ecological design, integrating multiple secondary data sources, to investigate the associations between long-term environmental exposure and hospital admissions for pneumonia in Kinshasa, Democratic Republic of the Congo (DRC), over a five-year period from January 2018 to December 2022. Data were spatially aggregated at the health zone level, the most operationally detailed unit within the national health system, enabling high-resolution spatial analysis of the disease burden across the city’s 35 health zones. Hospital admissions for pneumonia were stratified into two age groups, namely, children under five years and individuals aged five years and older, reflecting key demographic categories most relevant to respiratory vulnerability.

Health data, including monthly pneumonia cases and demographic information, were obtained from District Health Information Software (DHIS2), the official digital medical reporting platform of the DRC Ministry of Health. All data were routinely collected by health facilities across the study area and centrally transmitted via standard forms from the National Health Information System (SNIS). Pneumonia diagnoses were based on national clinical guidelines [40], which follow the World Health Organization (WHO) recommendations for pneumonia management in low-income settings [41]. According to these criteria, pneumonia is defined as an acute lower respiratory tract infection presenting with cough or difficulty breathing, together with age-specific thresholds of elevated respiratory rates. Severe pneumonia in children was identified by the presence of general danger signs, chest indrawing, or markedly elevated respiratory rates, whereas nonsevere pneumonia included cases with milder symptoms and respiratory rates below the critical thresholds. These definitions ensure consistency in case classification across zones and over time, as there are areas in the city where people cannot access chest radiography.

Owing to the lack of ground-based monitoring data, environmental exposures were extracted from the Google Earth Engine via satellite-derived products that capture key atmospheric pollutants and land cover indicators relevant to respiratory health. The concentrations of fine particulate matter (PM₂.₅), nitrogen dioxide (NO₂) and carbon monoxide (CO) were used as primary indicators of ambient air pollution. PM₂.₅ data were sourced from the Moderate Resolution Imaging Spectroradiometer (MODIS) satellite platform, while CO and NO₂ data were also retrieved from Sentinel-5P satellite products [42, 43].

The surface density (mol/m²) of NO₂ and CO was converted into a volumetric concentration (µg/m³). The conversion was performed as follows:

where X is the tropospheric column density (mol/m²), M is the molecular weight (g/mol), and H is the assumed mixing layer height (m). This approach assumes a uniform vertical distribution of the gas within the boundary layer. In the absence of vertically resolved NO₂ and CO profiles, we adopted a constant mixing-layer height (250 m) as a representative average for our study region and period, which is consistent with previous satellite-based air quality studies [44, 45].

Greenness exposure was assessed via the normalized difference vegetation index (NDVI), also derived from the Moderate Resolution Imaging Spectroradiometer (MODIS), which provides a proxy for the vegetation cover surrounding residential areas. NDVI values range from − 1 to + 1, with higher values indicating greater vegetative density, which has been linked to respiratory and thermal health benefits [46].

Key infrastructural and environmental covariates were incorporated into the analysis to address potential confounding and spatial variations in health outcomes. These included the density of major roads, the number of health facilities per health zone, and the distance to the nearest health facility. All infrastructure data were obtained from GRID3 (Geo-Referenced Infrastructure and Demographic Data for Development), a high-resolution geospatial dataset compiled by the Center for International Earth Science Information Network (CIESIN), Columbia University (2024). These covariates allowed for the adjustment of exposure models to account for disparities in healthcare access and built environment factors that may influence pneumonia hospitalization rates, particularly among vulnerable populations.

By harmonizing hospital admission data, satellite-based environmental exposures, and infrastructure indicators within a consistent spatial framework, this study provides a robust platform for evaluating spatial risk patterns and potential environmental drivers of pneumonia in a rapidly urbanizing megacity in Africa. With respect to the environmental data, multiyear average exposures (2018-2022) were used to reflect chronic exposure levels rather than short-term effects.

Data analysis

To describe the pattern of pneumonia in the health zones of Kinshasa, we calculated the cumulative incidence (CI) per 10,000 people (Eq. (1)) [47], which represents the ratio of new cases during a given period to the population at risk during the same period. The results are presented in tables and maps.

1

where CI_it is the cumulative incidence per year in the health zone, O_it is the number of cases observed per year in the health zone, and pop_it is the population of the health zone.

The cumulative incidence over five years was then calculated by dividing the number of people who developed the event during the five years by the population for the entire study period [48]. These results are presented in the form of a geographical map. To construct health indicators by health zone standardized to their average for the entire study area, the standardized incidence ratio (SIR) for hospital admissions was calculated and presented graphically. The calculation of the SIR allows the number of hospital admissions for pneumonia observed in a health zone to be compared with the number expected if the cumulative incidence in that health zone is identical to that of the reference population (in this case, the cumulative incidence of pneumonia in all 35 health zones in Kinshasa). The SIRs were reported in all health zones for the five years of the study, providing an overview of the population’s health status [49].

The SIR is calculated for each health zone as follows:

2

where O is the number of cases observed per year in the health zone and E is the expected number of cases.

In practice, for the calculation of the expected cases (Ei), since we performed the analysis separately for each age category (the whole population, children under 5 years, and individuals aged 5 years and above), the computation was simplified. This is because the reference cumulative incidence is already age-adjusted through the use of age-specific groups:

3

where reference cumulative incidence_j is the overall (regional, i.e., Kinshasa) cumulative incidence for age stratum j, and Population_ij is the corresponding population in health zone_i.

Notably, for SIRs greater than 1, the cumulative incidence in the health zone was greater than that in the reference zone. If the SIR is 1, the cumulative incidence in the health zone is not different from that in the reference zone. If it is less than one, the cumulative incidence in the health zone is lower than that in the reference zone. The confidence intervals were interpreted with a 5% risk of error.

A correlation matrix was created to examine the relationships between the climate and environmental variables. A correlation coefficient greater than 0.5 was considered significant. A spatial analysis tool was employed to detect clusters, hotspots, and cold spots associated with pneumonia across health zones in the city of Kinshasa. Moran’s I statistic was used to evaluate the spatial dependence of the pneumonia cases. Hotspots and coldspots were further identified via the local indicator of spatial autocorrelation (LISA). The results of the local Moran’s I were classified into four categories: high-incidence clusters (high-high), low-incidence clusters (low-low), high-incidence clusters surrounded by low-incidence neighborhoods (high-low), and low-incidence clusters surrounded by high-incidence neighborhoods (low-high) [50].

Statistical methods for assessing the relationships between various environmental factors and the number of reported pneumonia cases per health zone were implemented via a negative binomial model to account for overdispersion, with the expected denominator (E) included as an offset.

For a negative binomial log-linear regression model, the log risk for area i is defined as:

4

where denotes the global spatial component.

The incorporation of the expected counts is achieved through the offset term , yielding the full model specification:

5

where:

• where represents the estimated incidence of pneumonia,

• is the structured spatial effect (ICAR/BYM2 formulation), and.

• where is the unstructured spatial effect (IID component).

The exposure variables included the atmospheric concentrations of PM2.5, NO₂, and CO (expressed per 10 µg·m⁻³), the normalized difference vegetation index (NDVI, per 0.1 unit), the density of major roads (MR), and healthcare accessibility indicators (number of facilities, NHF; distances, DHF). All continuous covariates were centered and scaled (z scores) to allow comparison of effects (RR per 1 standard deviation).

The spatial structure was modelled via a Bayesian hierarchical BYM2 model, which combines a spatially structured component (ICAR) and an unstructured component (IID) with a queen-contiguity adjacency matrix. The hyperparameters were specified via weakly informative PC priors. Inference was performed via the Laplace approximation (INLA).

Model comparisons were based on DIC and WAIC, and sensitivity analyses included (i) single-pollutant models, (ii) principal component analysis (PCA) for pollutant variables, and (iii) explicit separation of the spatial components (ICAR + IID). Predictive diagnostic methods include Conditional Predictive Ordinates (CPO), the Probability Integral Transform (PIT), and the identification of influential zones. All analyses were performed via the spdep package in R statistical software (version 4.5.2).

Results

Spatial distribution of environmental exposures and Climatic conditions

Marked spatial heterogeneity was observed in ambient air pollution, climatic variables, and vegetation cover across Kinshasa’s health zones (Figs. 2 and 3). The annual mean concentrations of NO₂ ranged from 11.8 to 15.7 µg/m³, with the highest levels concentrated in central and northern health zones characterized by dense urbanization and heavy traffic. The PM₂.₅ concentrations varied between 32 and 45 µg/m³ and exhibited similar spatial patterns, with elevated levels in the north-central and northwestern zones. The carbon monoxide concentrations exhibited a narrow spatial range (5.98–6.35 mg/m³), with high values observed primarily in the northern and northeastern areas.

Fig. 2.

Spatial variation in key ambient air pollutants (NO₂, PM₂.₅, and CO) across health zones in Kinshasa

Fig. 3.

Spatial variation in climatic variables and vegetation cover across health zones in Kinshasa. TMAX = maximum temperature; PPT = precipitation; VPD = vapour pressure deficit. Spatial variation in key climatic variables and vegetation cover across health zones in Kinshasa

The vegetation cover, measured via the NDVI, exhibited a strong spatial gradient. The NDVI values exceeded 0.5 in the southern and peri-urban zones, whereas the central urban corridors presented very low vegetation cover (NDVI = 0.1). Climatic indicators also vary spatially, with zones showing low NDVI values often coinciding with relatively high maximum temperatures and elevated vapour pressure deficits (VPDs). These patterns indicate the coexistence of high pollution, low greenness, and increased climatic stress in the central and northern districts.

Infrastructure and healthcare accessibility

Substantial disparities in infrastructure and healthcare accessibility were observed across health zones (Fig. 4). The travel time to the nearest health facility was shortest in central urban zones but exceeded 30 min in several eastern and peripheral zones. The health facility density per 10 000 inhabitants was highest in urban centers and markedly lower in peri-urban and rural areas.

Fig. 4.

Spatial variation in infrastructure indicators across health zones in Kinshasa. DHF distance to health facilities, NHF number of health facilities per 10,000 inhabitants, MR  main road density, LR local road density

The major road density (MR) was concentrated in central and highly urbanized districts, whereas peripheral zones presented limited connectivity. The local road density followed a similar spatial pattern. These infrastructural inequalities highlight marked differences in mobility and access to healthcare services across the city.

Correlation structure among the environmental and infrastructure variables

The Spearman correlation matrix revealed strong interrelationships among the environmental variables (Fig. 5). PM₂.₅, NO₂, and CO were highly correlated, indicating shared emission sources or overlapping atmospheric processes. The NDVI was negatively correlated with all air pollutants and positively correlated with precipitation, reflecting an ecological gradient from greener, less polluted zones to densely built, highly polluted areas.

Fig. 5.

Spearman correlation matrix of climatic, air pollution, vegetation, and infrastructure variables. The variables include climatic indicators (TMAX maximum temperature, TMIN minimum temperature, PPT precipitation, VPD  vapour pressure deficit), air pollution indicators (NO₂ nitrogen dioxide, CO carbon monoxide, PM₂.₅ fine particulate matter), vegetation indicators (NDVI normalized difference vegetation index), and infrastructure indicators (DHF distance to health facilities, NSc number of schools, NHF number of health facilities, MR main road density, LR local road density). Positive and negative correlations illustrate key environmental and infrastructural interrelationships across health zones in Kinshasa

The temperature indicators (TMAX and TMIN) were strongly positively correlated with the VPD, whereas the precipitation was inversely correlated with both the temperature and VPD. The infrastructure-related variables exhibited strong internal correlations but showed limited direct associations with the pollution indicators.

Spatial and Temporal distributions of pneumonia incidence

Between 2018 and 2022, a total of 484 954 pneumonia hospital admissions were reported across Kinshasa’s 35 health zones. Spatial mapping of cumulative case counts and standardized incidence ratios (SIRs) revealed pronounced geographic disparities (Fig. 6). High SIR values (approximately 3) were concentrated in north-central health zones, whereas southern and southwestern zones generally presented SIRs below 1.

Fig. 6.

Spatial distribution of pneumonia burden across Kinshasa’s health zones. (a) Total number of pneumonia cases reported between 2018 and 2022.(b) Standardized incidence ratio (SIR) for pneumonia during the same period, adjusted for the underlying population. Darker shades indicate areas with higher case counts (panel a) or elevated relative risk (panel b), highlighting localized high-burden clusters

Temporal analysis revealed persistent spatial patterns over the study period (Fig. 7). High-incidence clusters in north-central zones remained stable from 2018 to 2022, whereas several eastern zones gradually increased until 2021, followed by stabilization in 2022. Most southern zones exhibited consistently low incidence rates throughout the study period.

Fig. 7.

Spatiotemporal variation in annual pneumonia incidence across Kinshasa, 2018–2022. Maps display yearly incidence rates (per 100,000 population) across health zones, illustrating persistent high-incidence hotspots in the north-central area and regions showing emerging or stabilizing trends over time

Age-specific distribution of pneumonia burden

Children under five years of age accounted for 41% of all pneumonia cases (n = 199 751), whereas individuals aged five years and older accounted for 59% (n = 285 203). The five-year cumulative incidence was substantially greater among children under five years of age (994 per 10 000 person-years) than among older individuals (330 per 10 000 person-years) (Additional files 1–4 A).

Pronounced geographic disparities were observed in both age groups. Gombe and Lingwala consistently presented the highest incidence rates in children under five years of age and in older individuals. Several health zones presented elevated risk across both age groups, whereas others presented age-specific deviations (Fig. 8).

Fig. 8.

Age-stratified standardized incidence ratios (SIRs) for pneumonia across health zones in Kinshasa. SIR values greater than 1 indicate a higher-than-expected incidence relative to the citywide reference population. The results highlight persistent high-risk clusters most notably in Gombe and Lingwala across both age groups, as well as zones with comparatively lower risk

Spatial cluster analysis

Global Moran’s I analysis indicated no significant spatial autocorrelation in pneumonia incidence across health zones (Moran’s I = -0.0347, p = 0.52). Local indicators of spatial association (LISA) identified a limited number of statistically significant local clusters (Fig. 9). Ten high-high clusters and seven low-low clusters were detected, alongside spatial outliers classified as high-low and low-high zones. Overall, these results suggest localized clustering within an otherwise spatially heterogeneous distribution.

Fig. 9.

Spatial clusters of pneumonia incidence in Kinshasa from 1 January 2018 to 31 December 2022 are based on Anselin Local Moran’s I. High-high and low-low clusters indicate statistically significant hotspots and coldspots, respectively, whereas high-low and low-high clusters represent spatial outliers. Only a limited number of zones show significant local autocorrelation, reflecting predominantly random spatial patterns

Bayesian Spatial regression analysis

Bayesian hierarchical negative binomial models with a BYM2 spatial specification were fitted to estimate the posterior distribution of pneumonia relative risks across health zones. The models jointly accounted for spatially structured dependence and unstructured heterogeneity, while adjusting for environmental and infrastructural covariates. Relative risks (RRs) were derived by exponentiating the posterior distributions of the regression coefficients. Uncertainty was quantified using 95% posterior credible intervals (CrI).

Main model –total population

In the BYM2 negative binomial model fitted to the total population (Table 1), the posterior distributions indicated clear spatial heterogeneity in pneumonia risk across health zones.Vegetation cover, measured by the normalized difference vegetation index (NDVI), showed consistent posterior support for a protective association. A one-standard deviation increase in NDVI was associated with a lower posterior relative risk of pneumonia (posterior median RR = 0.65; 95% CrI: 0.42-0.99), indicating that higher vegetation cover tended to correspond to lower estimated pneumonia risk. Major road density (MR) also displayed posterior evidence of a negative association with pneumonia risk (RR = 0.80; 95% CrI: 0.66-0.97), suggesting that zones with greater road connectivity generally exhibited lower estimated risks, likely reflecting infrastructural and accessibility characteristics. For ambient air pollutants, the posterior distributions for PM₂.₅ and CO were centered close to unity and exhibited wide credible intervals that included 1, indicating substantial uncertainty and limited posterior support for an association with pneumonia risk. Nitrogen dioxide (NO₂) presented a posterior distribution with a median RR below 1 (RR = 0.51; 95% CrI: 0.30-0.88). However, this inverse association should be interpreted cautiously, as strong multicollinearity among pollutants limits causal interpretation and may reflect residual spatial confounding rather than a true protective effect. Healthcare facility density (NHF) showed no clear posterior evidence of association with pneumonia risk, with credible intervals largely overlapping unity.

Table 1.

Posterior distribution of environmental and infrastructural effects on pneumonia relative risk - Bayesian spatial BYM2 model (total population)

Variable	Coefficient β (95% CrI)	RR (95% CrI)
PM2.5 (per 1 SD)	β = -0.077 (-0.466, 0.313)	RR = 0.926 (0.627–1.368)
NO₂ (per 1 SD)	β = -0.665 (-1.204, -0.127)	RR = 0.514 (0.300-0.881)
CO (per 1 SD)	β = 0.336 (-0.170, 0.843)	RR = 1.400 (0.843–2.323)
NDVI (per 1 SD, 0.1 unit)	β = -0.435 (-0.867, -0.005)	RR = 0.647 (0.420–0.995)
MR (major road density, 1 SD)	β = -0.218 (-0.410, -0.026)	RR = 0.804 (0.663–0.974)
NHF (health facilities, 1 SD)	β = -0.015 (-0.247, 0.216)	RR = 0.985 (0.782–1.241)

Sensitivity analyses

A series of sensitivity analyses was conducted to assess the robustness of posterior estimates to alternative model specifications (Tables 2 and 3). These included single-pollutant models, principal component-based representations of air pollution, alternative spatial prior specifications, and changes in variance and mixing priors.Across all sensitivity analyses, the posterior distributions for NDVI and major road density remained stable in both direction and magnitude, providing consistent posterior support for their associations with lower estimated pneumonia risk. In contrast, pollutant-specific posterior estimates varied considerably across model formulations and showed limited consistency, with wide credible intervals and posterior medians that were sensitive to specification choices.Model comparison metrics (DIC and WAIC) indicated that inclusion of individual pollutants did not meaningfully improve model fit relative to models excluding pollutant terms, further supporting the limited explanatory contribution of pollutant-specific effects at the health-zone scale.

Table 2.

Sensitivity analyses of single-pollutant and PCA models assessing the robustness of environmental and infrastructural predictors of pneumonia risk

Variable	Model A (PM2.5)	Model B (NO₂)	Model C (PCA)
Pollutant	+ 0.04 [-0.20 ; +0.20]	-0.32 [-0.78 ; +0.15]	0 (ns)
NDVI	-0.44 [-0.86 ; -0.05]	-0.16 [-0.58 ; -0.20]	Stable
MR	-0.31 [-0.53 ; -0.15]	-0.26 [-0.47 ; -0.09]	Stable
NHF	-0.04 [-0.28 ; +0.20]	-0.03 [-0.29 ; +0.22]	Variable
DIC	719.54	717.99	718.70
WAIC	720.42	719.21	720.31
Model without pollutant	DIC = 717.93/WAIC = 718.98	Equivalent to NO₂ model	-

Table 3.

Robustness of posterior relative risk distributions across Bayesian spatial model specifications

Sensitivity Category	Model Tested	PM₂.₅ (RR [95% CrI])	NO₂ (RR [95% CrI])	NDVI (RR [95% CrI])
Reference Model	Loose Negative Binomial	0.905 [0.631–1.301]	0.514 [0.300-0.881]	0.647 [0.420–0.995]
1. Variance Prior	Strict Variance (σ)	0.905 [0.631–1.301]	0.514 [0.300-0.881]	0.647 [0.420–0.995]
2. Mixing Prior	Uniform φ (Sensitivity 2)	0.880 [0.590–1.350]	0.490 [0.290–0.800]	0.693 [0.490–0.914]
3. Spatial Structure	Tower Type (Sensitivity 3)	1.050 [0.750–1.450]	0.450 [0.250–0.720]	0.579 [0.338–0.992]
4. Distribution	Poisson Model (Sensitivity 4)	0.905 [0.631–1.301]	0.514 [0.300-0.881]	0.647 [0.420–0.995]

Age-stratified models

Age-stratified BYM2 models were fitted separately for children under five years of age and for individuals aged five years and older. Among children under five years of age (Table 4), vegetation cover exhibited posterior support for a protective association (posterior median RR = 0.85; 95% CrI: 0.56-0.82), indicating that higher NDVI values were associated with lower estimated pneumonia risk in this age group. Major road density also showed posterior evidence of a modest negative association (RR = 0.83; 95% CrI: 0.70-1.00).For air pollutants, the posterior distributions of PM₂.₅, NO₂, and CO were characterized by substantial uncertainty, with wide credible intervals. Although the posterior median for NO₂ was below unity, this pattern was considered biologically implausible and likely reflects residual confounding due to strong collinearity among pollutants rather than a protective effect. Among individuals aged five years and older (Table 5), vegetation cover again demonstrated consistent posterior support for lower estimated pneumonia risk (RR = 0.58; 95% CrI: 0.34-0.99). For the remaining covariates, including air pollutants, the posterior distributions showed limited evidence of association, with credible intervals largely overlapping unity. Overall, the age-stratified analyses indicated that vegetation cover was the most stable determinant across demographic groups, whereas pollutant-specific effects were not robust to model specification.

Table 4.

Posterior distributions of pneumonia relative risks among children under five years of age - Bayesian spatial BYM2 model

Variable	Beta (95% CrI)	RR (95% CrI)
PM2.5 (per SD)	-0.098 (-0.465, 0.27)	0.907 (0.628, 1.310)
NO₂ (per SD)	-0.640 (-1.146, -0.133)	0.527 (0.318, 0.875)
CO (per SD)	0.352 (-0.125, 0.828)	1.422 (0.882, 2.289)
NDVI (per SD)	-0.16 (-0.58, -0.20)	0.852 (0.560, 0.816)
MR (per SD)	-0.182 (-0.361, -0.003)	0.834 (0.697, 0.997)
NHF (per SD)	-0.066 (-0.284, 0.152)	0.936 (0.753, 1.164)

Table 5.

Posterior distributions of pneumonia relative risks among individuals aged five years and older - Bayesian spatial BYM2 model

Variable	Beta (95% CrI)	RR (95% CrI)
PM2.5 (per SD)	-0.105 (-0.579, 0.378)	0.90 (0.56–1.46)
NO₂ (per SD)	-0.836 (-1.50, -0.17)	0.43 (0.22–0.84)
CO (per SD)	0.432 (-0.199, 1.061)	1.54 (0.82–2.89)
NDVI (per SD)	-0.546 (-1.09, − 0.01)	0.58 (0.34–0.99)
MR (per SD)	-0.211 (-0.45, 0.03)	0.81 (0.64–1.03)
NHF (per SD)	0.049 (-0.236, 0.336)	1.05 (0.79–1.40)

Adjusted spatial distribution of relative risk

The adjusted spatial distribution of posterior median relative risks derived from the BYM2 models is presented in Figure 10. Elevated estimated risks were concentrated in central and northern health zones across all age groups, whereas lower posterior relative risks were generally observed in zones characterized by higher vegetation cover and greater infrastructural connectivity. These maps represent smoothed posterior estimates that account for spatial dependence and random heterogeneity, thereby reducing instability in zones with small population sizes and emphasizing broader spatial patterns rather than local random variation.

Fig. 10.

Spatial distribution of posterior pneumonia relative risks in Kinshasa estimated using Bayesian BYM2 spatial models and adjusted for environmental and infrastructural factors, stratified by age group

Discussion

This study provides a comprehensive spatial assessment of pneumonia risk in Kinshasa by jointly examining long-term ambient air pollution, vegetation cover, climatic stress, and infrastructural characteristics across 35 health zones. Using five years of hospitalization data and Bayesian spatial modelling, we demonstrate pronounced spatial inequalities in the burden of pneumonia that persist across age groups. Our findings indicate that pneumonia risk in Kinshasa is shaped less by isolated pollutant-specific effects and more by combined environmental and infrastructural conditions, particularly vegetation loss, climatic stress, and urban structure. These inequalities mirror the patterns of environmental injustice documented in other rapidly urbanizing regions of Sub-Saharan Africa (SSA) [23, 27, 51–53].

Principal findings

Three key findings emerge from this analysis. First, pneumonia incidence exhibits strong spatial heterogeneity across Kinshasa, with persistent high-risk zones concentrated in the central and northern districts. Second, although ambient air pollution levels are uniformly high across cities, no pollutant (PM₂.₅, NO₂, or CO) has a robust positive association with pneumonia risk after accounting for spatial correlation and multicollinearity. Third, vegetation cover (NDVI) and, to a lesser extent, major road density (MR) consistently emerged as protective factors across all model specifications and age groups. Children under five years of age bear the highest burden of pneumonia, yet the spatial patterns of risk remain broadly similar across age strata, highlighting shared contextual determinants.

Air pollution, multicollinearity, and the absence of pollutant-specific effects

Kinshasa experiences persistently high ambient air pollution, with mean PM₂.₅ concentrations far exceeding WHO guideline values. However, despite this high exposure, the pollutant-specific associations with pneumonia were not robust in our spatial models. This finding does not contradict the established biological plausibility of air pollution as a respiratory risk factor. Instead, it reflects several methodological and contextual challenges common in environmental epidemiology in low- and middle-income urban settings.

First, strong multicollinearity among PM₂.₅, NO₂, and CO limits the ability to disentangle the independent effects of individual pollutants. These pollutants share common emission sources, including traffic, biomass combustion, and informal industrial activities, resulting in overlapping spatial distributions. Second, the use of multiyear average exposures may obscure short-term pollution peaks that are more directly linked to acute respiratory infections such as pneumonia. Third, spatial contrasts in pollution levels across health zones were relatively modest compared with the overall magnitude of exposure, reducing the statistical power to detect zone-level differences.

The inverse association observed for NO₂ should therefore be interpreted cautiously. Rather than reflecting a protective effect, it likely represents residual confounding or the role of NO₂ as a proxy for unmeasured urban structural factors correlated with better infrastructure or healthcare access. In contrast, some regional studies have shown associations between PM₂.₅ or NO₂ and respiratory infections in Senegal, Kenya, Uganda, and South Africa [22, 23, 51, 52, 54, 55].

Vegetation cover as a robust protective factor

Vegetation cover emerged as the most consistent and robust protective determinant of pneumonia risk in this study. Across all the models and age groups, a higher NDVI was associated with substantially lower pneumonia risk. This finding aligns with a growing body of international literature demonstrating the protective role of urban green spaces on respiratory health.

Several mechanisms may explain this association. Vegetation can reduce exposure to air pollutants by intercepting particulate matter and absorbing gaseous pollutants. Green spaces also mitigate urban heat and reduce vapour pressure deficits, both of which are linked to respiratory vulnerability and immune stress. In addition, greener environments may improve ventilation, reduce population density pressures, and promote healthier living conditions [46, 56–62].

In Kinshasa, areas with relatively high NDVIs were predominantly located in southern and peri-urban zones, which also presented relatively low climatic stress and low pneumonia incidence. Conversely, central districts characterized by severe vegetation loss presented the highest disease burden. These findings highlight the critical role of green infrastructure in buffering against environmental stressors in rapidly urbanizing African megacities.

Infrastructure, mobility, and access to healthcare

Major road density was consistently associated with a reduced risk of pneumonia, a finding that may initially appear counterintuitive given the well-established role of road traffic as a source of air pollution. However, in the urban context of Kinshasa, major road density does not primarily reflect traffic-related pollution intensity but rather serves as a proxy for broader infrastructural development and urban organization. Indeed, health zones with higher road density are typically characterized by better connectivity, improved access to healthcare facilities, and more efficient emergency response systems. Importantly, in our data, major road density showed no strong correlation with ambient air pollution indicators (PM₂.₅, NO₂, and CO), supporting the interpretation that MR captures infrastructural advantages rather than pollution exposure.

Improved mobility and accessibility may facilitate earlier healthcare-seeking behavior, timely diagnosis, and prompt initiation of treatment, thereby reducing the severity and duration of pneumonia episodes. Similar findings have been reported in other urban health studies, where infrastructure development mitigated health risks despite increased urban activity [63, 64].

Environmental inequalities and spatial injustice

The spatial distribution of pneumonia risk in Kinshasa reflects pronounced environmental and infrastructural inequalities. Central and northern health zones experience a convergence of adverse conditions, including high pollution levels, low vegetation cover, elevated vapour pressure deficit, dense populations, and limited environmental buffering capacity. These cumulative exposures create conditions of heightened vulnerability, which is consistent with frameworks of environmental injustice.

The persistence of high-risk clusters over time suggests that these inequalities are structural rather than transient. Our findings are consistent with those of an Ethiopian study that identified marked geographic heterogeneity in acute respiratory infections (ARIs) among children under five years of age, with high-risk clusters concentrated in the northern and central regions [65]. Addressing the pneumonia burden in Kinshasa therefore requires not only clinical interventions but also systemic urban and environmental reforms.

Age-related vulnerability

Children under five years of age accounted for the largest proportion of pneumonia cases and presented the highest incidence rates, which is consistent with global epidemiological evidence [66]. Biological susceptibility, immature immune systems, and increased relative exposure to environmental hazards contribute to this vulnerability. However, the spatial patterns of pneumonia risk were broadly similar across age groups, indicating that contextual environmental determinants operate at the population level.

The absence of strong pollutant-specific effects in children may reflect the dominant role of household-level factors, such as indoor air pollution, overcrowding, and socioeconomic conditions, which were not captured in this ecological analysis. Nevertheless, the strong protective effect of vegetation in this age group suggests that improving the overall environmental context may yield substantial benefits for child health.

Strengths and limitations

This study has several key strengths. It combines a large, multiyear dataset of pneumonia hospitalizations with high-resolution environmental and infrastructural data and applies Bayesian spatial models that account for both structured and unstructured spatial variation. In addition, extensive sensitivity analyses were conducted, which supported the robustness and stability of the main findings.

However, several limitations should be acknowledged. First, the ecological study design does not allow causal inference at the individual level. Second, exposure estimates based on satellite-derived data and multiyear averages may not fully capture the short-term variability in air pollution that could be relevant for pneumonia onset. In addition, strong multicollinearity among air pollutants limits the interpretation of pollutant-specific associations.

Third, the health information system used for data collection aggregates pneumonia cases into only two age categories: children under 5 years and individuals aged 5 years and above. Consequently, older children and adults are grouped together, preventing more detailed age-specific analyses beyond the under-five population. This limitation may have influenced the interpretation of results for the ≥ 5-year age group.

Finally, several potentially important confounding factors, including indoor air pollution exposure, socioeconomic conditions, vaccination coverage, and nutritional status, could not be incorporated into the analysis due to data limitations.

Despite these limitations, the consistent protective associations observed for vegetation cover and infrastructural indicators across multiple model specifications and age groups strengthen confidence in the findings. Overall, the results highlight the importance of the environmental and structural determinants of respiratory health and provide relevant evidence to inform urban public health planning in rapidly urbanizing African settings.

Conclusion

This study provides the first comprehensive spatial assessment of pneumonia risk in Kinshasa, integrating long-term environmental exposure, vegetation cover, climatic stress, and urban infrastructure. Our findings demonstrate that the burden of pneumonia in this rapidly growing megacity is driven primarily by environmental and infrastructural inequalities rather than by the isolated effects of individual air pollutants.

Despite persistently high levels of ambient air pollution across all health zones, pollutant-specific associations with pneumonia were not robust after accounting for spatial dependence and multicollinearity. In contrast, vegetation cover consistently emerged as a strong protective factor across all age groups, while major road density also showed a protective association, likely reflecting improved accessibility and urban connectivity rather than reduced pollution exposure. Children under five years of age bore a disproportionate share of the disease burden, underscoring their heightened vulnerability to adverse environmental conditions.

At the city scale, health zones located in the central and northern districts characterized by low vegetation cover, high climatic stress, dense urbanization, and limited environmental buffering experienced the highest pneumonia risk. These spatial patterns highlight persistent environmental health inequities and point to clear opportunities for targeted intervention. Expanding and preserving urban green infrastructure, particularly in high-burden central and northern zones, may represent a feasible and sustainable strategy to reduce respiratory vulnerability. Integrating health indicators into urban planning decisions, including land-use planning, road development, and climate adaptation strategies, could further enhance the prevention of pneumonia and other environmentally mediated respiratory diseases.

As Kinshasa continues to urbanize, closer integration among public health, environmental management, and urban planning will be essential to reduce preventable respiratory morbidity. Future research incorporating individual-level data, indoor air pollution exposure, and finer spatiotemporal metrics is needed to better elucidate causal pathways and support evidence-based interventions aimed at improving respiratory health equity in rapidly urbanizing African cities.

Authors’ contributions

MKN, FB, DMM, TTT, BOK, EI, GMK, and JMNK contributed to the study conception and design. The material was prepared via MNK, FB, or DMM. Formal data analysis and evaluation were performed via MNK, FB and DMM. A critical data review was performed by MNK, FB, DMM, LA, EI, GMK, and JMNK. The first draft of the manuscript was written by MNK, and all the authors commented on the previous versions of the manuscript. All the authors have read and approved the final manuscript.

Funding

This study did not receive any external funding.

Data availability

All data used in this study are available at [https://docs.google.com/spreadsheets/d/1IB8L8CXBTTCc-X3bKEIC0SFNA3zeL2o1/edit? usp=sharing&ouid=107625050571109857114&rtpof=true&sd=true] .

Declarations

Ethics approval and consent to participate

Administrative and ethical approval for the use of anonymized DHIS2 data was granted by the Scientific Committee of the Ministry of Public Health, DRC (Reference No. MSPHP.1251/SG/DESP/BKA/2146/YRS/2023). As the study used aggregated secondary data with no individual identifiers, patient consent was not needed.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.