Pneumonia incidence and determinants in South Punjab, Pakistan (2016–2020): a spatial epidemiological study at Tehsil-level

Abstract

Background

Pneumonia remains a major cause of morbidity and mortality, particularly in low- and middle-income countries, such as Pakistan. In this study, we aimed to examine the spatial and temporal patterns of pneumonia incidence in South Punjab, Pakistan, and to analyze their association with socio-ecological factors.

Methods

We used case report data from the district health information system (DHIS) over the years 2016 to 2020 and applied global and local Moran’s I to identify spatial autocorrelation. Furthermore, we employed hot and cold spot analysis to identify significant areas with high and low pneumonia incidence. We used Emerging Hot Spot Analysis (EHSA) and time series clustering to examine shifting and temporal patterns of incidence, respectively. In addition, Generalized Linear Regression (GLR) and Multiscale Geographically Weighted Regression (MGWR) models were used to analyze geographic variation in the association of socio-ecological factors and pneumonia incidence.

Results

Our results showed no significant global clustering of pneumonia incidence. Local Moran’s I identified a low-low cluster in DG Khan, while Hot Spot Analysis detected one hot spot in Rajanpur. Multan City showed higher case counts, but this reflected population concentration rather than elevated incidence rates. The temporal analysis confirmed a significant seasonal variation, as well as a decrease in certain Tehsils and an increase in others. Our MGWR model revealed that better female literacy reduced incidence rates of pneumonia, whereas poor housing quality increased incidence rates of pneumonia, particularly in the southwestern areas of South Punjab.

Conclusions

We conclude that socio-ecological variables significantly influenced the incidence of pneumonia in South Punjab, and this association varies substantially over time and space. Our results emphasize the need for locally specific public health interventions to minimize pneumonia incidence in vulnerable populations in Pakistan. Our spatial epidemiological approach can be adapted to other regions of Pakistan and similar socio-ecological contexts in low- and middle-income countries.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12942-025-00420-y.

Introduction

Pneumonia is an infectious disease that causes hundreds of thousands of deaths worldwide, especially among children and elderly [54, 69, 105]. As a respiratory disease, it largely affects the lungs, caused by various pathogens, including bacteria, fungi, parasites, and viruses [9, 92]. It damages the functions of the lungs by causing inflammation and accumulation of mucus and fluid in the small air sacs called alveoli, resulting in difficulty breathing [20]. Pneumonia is an airborne disease and transmissible through coughing or sneezing, and also through the blood, particularly during and immediately following childbirth [128]. Important risk factors include smoking, the use of biofuels such as wood or dung for cooking and heating, high room density, insufficient ventilation, outdoor air pollution, and lack of access to health care [47, 61, 106, 128].

The estimated incidence of pneumonia is 1·5 to 14·0 cases per 1000 person-years [97]. The incidence of pneumonia varies according to place, season, and demographic characteristics. It is more frequent in young children (< 5 years) and adults older than 65 years [8, 41]. In addition, people with existing medical conditions are more vulnerable to pneumonia than those without [128]. It is estimated that three children worldwide die from pneumonia every two minutes [126]. WHO estimates that in 2017, pneumonia-related deaths among children under five years of age reached almost 808,000, resulting in around 15% of all deaths within this specific age group [128]. Based on the current patterns, it is predictable that there will be a total of 735,000 deaths caused by pneumonia in the year 2030 [126, 127]. Furthermore, less developed countries, especially those in Southern Asia and Africa, show a higher rate of childhood pneumonia than developed countries [71]. The greatest proportions of pneumonia were estimated in the Southeast Asian (39%) and African regions (30%) [123].

In Pakistan, pneumonia caused 5.72% of all deaths during 2020 [129]. In 2017, pneumonia became the third leading cause of death among children under five in Pakistan, accounting for 14% of child mortality [119]. Like other parts of Pakistan, it is also a leading respiratory infection in South Punjab, Pakistan. According to the estimates of district health informatics, pneumonia constitutes 1.1% of all diseases and 4% of respiratory diseases during 2016–2020 in this region [18]. In 2021, Punjab, Pakistan, reported an average of 1,043 pneumonia cases per day among children under five years of age, and 884 daily cases among individuals aged five years and above [19]. Bahawalpur and Multan, two major districts from South Punjab, ranked among the top ten districts in terms of pneumonia incidence across the province of Punjab [18].

Spatial epidemiological approaches, such as spatial and temporal clustering methods, play a crucial role in analyzing patterns and trends for various diseases over space and time. These methods allow researchers to spot clusters of disease incidents that show significant proximity in both space and time [34, 42]. Spatial epidemiological approaches may also provide a significant understanding of the root sources for infectious disease and likely causal factors [38, 68]. Similarly, spatial epidemiological approaches facilitate the analysis of time-series data, the integration of spatial and temporal patterns, and the utilization of both 2D and 3D visualization techniques [7, 23]. Spatial clustering methods have been used widely throughout the world to study various infectious and non-infectious diseases, for example, COVID-19 [14, 31], acute respiratory infections [37], lower tract respiratory infections [2, 112], tuberculosis [78], leishmaniasis [42], malaria [17, 32] and diarrhea [33, 39]. Spatial modeling techniques have been employed to assess the relationship between various variables and the incidence of diseases [84, 88]. Geographically weighted regression (GWR) and Multiscale Geography Weighted Regression (MGWR) have evolved into standard and valuable methods in the investigation of spatial associations between potential causes and diseases [55, 72, 81, 111]. In contrast to conventional ‘global’ regression models, MGWR allows associations to vary across various spatial scales, thereby minimizing overfitting and errors. It reduces collinearity by allowing for non-varying, local, or regional relationships between the predictors and the response. MGWR is recommended for investigating spatial heterogeneity and scale in GWR analyses [25, 95].

Although spatio-temporal patterns of pneumonia have been explored globally [70, 125], research in Pakistan remains limited to clinical, behavioral, or childhood-focused studies [58, 87, 91]. Spatio-temporal modelling has been applied to vector-borne diseases in Pakistan, especially dengue [60, 89]; its use in pneumonia research remains limited. Although spatial analyses have been undertaken for other respiratory diseases in South Punjab, including acute respiratory infections [35], tuberculosis [30, 34], and chronic obstructive pulmonary disease [30], pneumonia remains comparatively understudied. Despite the burden of pneumonia, studies employing ecological and spatial approaches to assess its relationship with socio-environmental factors remain scarce in both South Punjab and Pakistan as a whole. This study addresses this gap by applying spatial–temporal clustering and MGWR to analyze Tehsil-level variations in pneumonia incidence and their associations with housing, literacy, and sanitation in South Punjab. This study addresses this gap by employing spatial–temporal clustering and MGWR to examine Tehsil-level variations in pneumonia incidence and their associations with key socio-ecological determinants in South Punjab.

Accordingly, this study was designed with the following objectives: (1) to detect significant spatial and temporal variations in pneumonia incidence in South Punjab and (2) to investigate the relationship between socio-ecological factors and pneumonia incidence.

Methods

Study area

South Punjab encompasses a total area of 99,579 square kilometers, with a population of 34.7 million, as per the 2017 census. The population density of the region is 349 individuals per square kilometer. Administratively, South Punjab is divided into 11 districts and 43 Tehsils across three divisions: Bahawalpur, Dera Ghazi (D.G) Khan, and Multan [104]. The climate in South Punjab is characterized by extreme heat and dryness during the summer and cold, dry conditions in the winter. Dust storms are a frequent occurrence throughout the summer months [49]. High values of particulate matter (30–70PM_2.5 μg/m³) make this region’s air quality unhealthy and sometimes very unhealthy according to the WHO air quality standards [36]. In addition, almost 87.6% of the population relies on solid fuel for cooking and heating [104]. South Punjab is one of the deprived regions of Pakistan with multi-fold environmental and socio-economic issues [120]. The prevalence of multi-dimensional poverty significantly surpasses that observed in other regions of the province, characterized by high rates of stunting growth, extensive undernourishment, incomplete access to clean water and hygiene, and a great number of out-of-school children [116]. The literacy rate in the region is notably deficient, averaging at 46.2% across the entire population [104, 120]. Additionally, the area exhibits a substantial dependency ratio of 79.6%, highlighting a considerable portion of the population dependent on adults for support [48]. All of these make South Punjab a good choice for this study (Fig. 1).

Fig. 1.

Study area Location and the ratio of Pneumonia incidences per ‰ in South Punjab, Pakistan, during 2016–2020

Data

Details of the outcome and explanatory variables are given in Table 1. Explanatory variables were selected based on theoretical relevance, empirical evidence, and data availability at the Tehsil level. Room density reflects household crowding, which increases the risk of airborne infections such as pneumonia [15, 115]. Female literacy serves as a proxy for maternal health awareness and care-seeking behavior, especially for childhood illnesses [40, 63]. Mud houses indicate poor housing quality and indoor air pollution exposure, both of which contribute to respiratory disease risk [29, 35]. Unimproved excreta and solid waste disposal capture environmental sanitation conditions, while the urban population variable accounts for spatial heterogeneity in exposure and healthcare access [1, 52]. The Pneumonia data were spatially aggregated at the Tehsil level (the administrative subunit of districts), covering reported cases from January 2016 to December 2020 at a monthly resolution. Although union councils represent the most granular geographic unit for settlements, Tehsils were selected as the unit of analysis due to the availability of both disease incidence and socio-ecological data at the Tehsil level only. Primary disease data collection was conducted through the District Health Information System (DHIS) of respective districts, utilizing consolidated reports of cases from each health facility within the district. A comprehensive network of healthcare facilities in the South Punjab region, comprising 102 hospitals, 383 dispensaries, 110 rural health centers, 762 basic health centers, 6 TB centers, 56 sub-health centers, and 68 maternity health centers, facilitated the acquisition of disease data. Demographic data, including total population and urban residents, have been sourced from the most recent census conducted in 2017 [96]. Data on socio-ecological variables have been extracted from the Multi-Cluster Indicator Survey (MICS, 2007–2008) reports [48].

Table 1.

Outcome and explanatory variables

Variable category	Variable name	Measure and unit	Mean	Standard deviation	Data source
Outcome	Pneumonia	Number of reported cases per Tehsil	27,116	19,879.31	DHIS (2016–2020)
Socio-Demographic	Population	Total population per Tehsil	808,071.26	453,518.67	Census 2017
	Urban population	Number of individuals living in urban areas per Tehsil	188,903.02	287,850.56
Socio-ecological	Room density	Number of persons per room per household	3.95	0.33	MICS (2007–2008)
	Female literacy rate	Literate percentage females per Tehsil	31.68	12.21
	Mud houses	Percentage of households that are constructed mainly of mud per household	32.73	14.77
	Unimproved excreta disposal facility	Percentage of households that have unimproved excreta disposal facility per Tehsil	45.20	15.06
	Unimproved solid waste disposal	Percentage of households that have unimproved solid waste disposal facility per Tehsil	66.41	21.85

Methodology

In this study, several different methods were used to understand the disease’s spatial concentration and explain its spatial and temporal patterns and related factors. First, classically, Moran’s I methods were used to understand whether the disease clusters spatially, and LISA and OHSA methods were used to comparatively address where and how this clustering, if any, occurs. Then, emerging hot spots, time series cluster, and trend analyses were performed using a space–time cube to understand the spatial and temporal pattern and trend of the disease. Finally, Exploratory Regression, Generalized Linear Regression (GLR), and MGWR were used to understand the factors associated with these clusters and patterns.

Spatial and spatiotemporal analysis

In the initial phase of this research, we structured the monthly pneumonia case ratio data into a space–time cube using ArcGIS Pro 3.2 [21]. A space–time cube is a three-dimensional model that represents spatial and temporal data by dividing it into discrete cells, where each cell corresponds to a specific location and time interval. This approach allowed for the simultaneous analysis of spatial and temporal patterns in the data [26]. This process began with data preparation, where a column in date format was created for each month within the study period (2016–2020). Monthly pneumonia case counts were recorded in one column, while the ratio of pneumonia cases to the population for each month was calculated and recorded in another column. Spatial statistical methods, including Local Indicators of Spatial Association (LISA), Optimized Hot Spot Analysis (OHSA), Emerging Hot Spot Analysis (EHSA), and time series cluster analysis, were applied to assess how these metrics influence the identification of significant spatial clusters. Additionally, a common identifier column was introduced to enable the association of these records with the corresponding geographical data. Subsequently, the prepared data were adapted into the space–time cube structure to facilitate spatiotemporal analysis. These steps ensured that the data were organized and ready for subsequent spatial and temporal analysis. All spatial autocorrelation and clustering analyses (Global Moran’s I, LISA, OHSA, EHSA) were conducted on population-adjusted incidence rates (cases per 100,000) at the Tehsil level. Raw case counts were examined separately for descriptive purposes, such as highlighting the concentration of cases in Multan City.

In the second stage, Global Moran’s Index analysis was conducted using ArcGIS Pro 3.2 software [21]. This analysis utilized the “Tehsil” layer as level of analysis (Tehsil is the term used for district subsets) in geographical data format and the case rates. Moran’s Index analysis aimed to assess the presence and degree of clustering at the Tehsil level based on both the number and rate of pneumonia cases. This method is instrumental in identifying spatial autocorrelation patterns, indicating whether neighboring Tehsils exhibit similar pneumonia incidence rates and the extent to which these patterns are positively (+ 1) or negatively (− 1) correlated. Positive values indicate that there is positive autocorrelation and similar values are spatially clustered; negative values indicate that there is negative autocorrelation and values are spatially dispersed. If the result is zero (0), it means that there is no spatial autocorrelation. In addition, Z score and P value are calculated with this method. If the Z score is greater than 2.58 or less than − 2.58 and the p value is less than 0.01, the result is reached at a confidence level above 99% [46, 51].

In the third stage, Anselin Local Moran’s I analysis was applied to reveal where and how clustering occurs according to the case rates at the Tehsil level. Anselin local Moran’s I statistic (LISA) is a method used to show spatial clusters. If the index value is positive, it is classified as regions with clusters, and if it is negative, it is classified as regions with spatial outliers. To test statistical significance with Anselin’s Local Moran’s I method, a z-score and p-value are calculated. In this way, statistically insignificant objects are identified.

As a result of the analysis, 5 different coding types appear according to the clustering forms and outlier values. These codings are as follows:

• High-High Clustering (HH): The spatial asset with high value is surrounded by high values.

• Low-Low Clustering (LL): Low-value spatial assets are surrounded by low-value spatial assets.

• High-Low Outlier (HL): An object with a high attribute value is surrounded by elements with a low value.

• Low-High Outlier (LH): An object with a low attribute value is surrounded by objects with a high value.

• Statistically Not Significant: It means that the local Moran’s I value is close to zero.

In the fourth stage, an OHSA was performed according to the rate of cases at the Tehsil level. Hot spot analysis was developed by Art Getis and Keith Ord in 1995. This analysis creates a statistically significant hot and cold spot map using the Getis-Ord Gi* statistic [23, 46, 94]. As the Gi* value calculated in the equation approaches zero, it is understood that there are no high or low values in the neighborhood of the calculated object. The z score is also calculated with this equation. As the z score increases, it is understood that the objects with high values in the geographic data are co-located or clustered. In this way, hot spots are found. As the z score decreases, it is understood that objects with low values are located together. Accordingly, cold spots emerge. OHSA is preferred to produce the most appropriate results compared to the classical Hot Spot analysis (Getis-Ord Gi*) method. OHSA was utilized for two distinct purposes; first, it was employed to identify and compare the differences in results produced by OHSA and LISA. Previous studies [75, 108, 121] have shown that OHSA tends to reduce the occurrence of statistically insignificant outcomes compared to LISA. Second, OHSA, which operates exclusively with spatial data, was used to compare its results with those derived from spatiotemporal analysis (EHSA), which incorporates both spatial and temporal dimensions [110]. This comparison highlights how the inclusion of temporal data affects the outcomes. Since the temporal range of the research data (2016–2020) coincides with the beginning of COVID-19, differentiation in the results was expected.

In the fifth stage, a spatiotemporal cube was created by using the monthly rate of cases at the Tehsil level. When creating the cube, a 1-month time step interval was determined. This value was determined through a systematic sensitivity analysis based on the first time step bias, last time step bias, and where trends and hot spots formed. This interval was also deemed reasonable based on the size of the obtained data set, the temporal distribution of pneumonia cases, and seasonal variations. This allowed the reliability and robustness of the results obtained with the space–time cube to be tested. The trend and emerging hot spots within the cube, defined based on the case rate, were visualized and analyzed. Trend analysis uses Mann–Kendall trend analysis and calculates the regions with an upward or downward trend together with the confidence level [67, 79]. This test, which is an ordinal correlation method, is performed for each location. The bin values of each slice in the space–time cube are compared. Among all pairs of time slices, if the first value is less than the second, the result is + 1, if it is greater, the result is − 1 and if it is equal, the result is 0. In addition, z and p values of these values are calculated. A weighted + 1 result indicates a positive z score, and a p value less than 0.001 indicates that there is an upward trend in that region at a 99% confidence level. A weighted − 1 result, negative z score, a p value less than 0.001 means that there is a downward trend in that region at a 99% confidence level. A value of 0 means that there is no trend over time [22]. EHSA analysis was then applied using the space–time cube. This analysis produces results using the hot spot and Mann–Kendall trend analysis mentioned in the fourth step. In this way, 17 different pattern types (such as new hot spot, persistent hot spot, or sporadic hot spot) can be identified [24]. Then, Time Series Cluster analysis was applied using the space–time cube. This analysis automatically determined the optimum number of clusters using the pseudo-F statistic. In addition, a data clock was created using the number of cases at this stage. In this way, the increase–decrease times of the number of cases were revealed. In other words, we present the temporal distribution of the number of cases using a data clock.

Regression modeling

In the sixth stage, Exploratory Regression, GLR, and MGWR methods were used, respectively, to reveal the variables associated with the case rates at the Tehsil level. Exploratory regression has been used to search for the most appropriate combinations of driving factors and to get rid of multicollinearity and thus find the models with the highest explanatory rate [5, 65, 66]. In this tool, the values assumed in the search criteria (minimum number of explanatory variables 5, minimum acceptable adjusted R2 value 0.5, maximum VIF value 7.5) were used. These values have been suggested by various scientific studies and ESRI [27, 72, 84]. The Variance Inflation Factor (VIF) value helps to determine whether there is a high or low correlation between independent variables. In this way, the multicollinearity problem in the model can be understood. In the literature, in general, when the VIF value of an independent variable in the model is more than 10, it is called a serious multicollinearity problem. For more sensitive analyses, lower thresholds such as 7.5 and 5 are preferred [50, 83]. To reach a wide range of variables or to increase flexibility, the VIF value was set as 7.5 in our study. In addition, it was also taken into consideration that the VIF values of the variables in the models tested were less than 5. As in the VIF criterion, the minimum number of explanatory variables is set as 5 to reach as wide a range of variables as possible. This value can optionally be 4 or 6. However, being flexible is not kept too high. The R² coefficient, which is expressed as the goodness of fit or explanatory power of the model, expresses how much of the variance of the dependent variable in the model can be explained by the independent variables. Setting the R² value as 0.5 means that it explains more than 50% of the dependent variable in the model. When the R² value falls below 0.5, the reliability of the model decreases [74]. However, when a model is created using GWR-based and nonlinear methods with the same variables compared to linear regressions, the R² value can increase by around 20%. In addition, the adjusted R² value was set as 0.4 in the exploratory regression tool in order to see the flexibility in the model such as VIF and number of variables. At this stage, the combination with the highest R² value and the widest range of variables was identified. Then, GLR, which is a global regression analysis method, was used. ArcGIS Pro 3.2 software was used to implement GLR and MGWR methods that will be mentioned below. GLR is a specific type of GLM where the dependent variable is assumed to follow a normal distribution and the identity link function is used [82, 93]. This is essentially ordinary linear regression but framed within the broader GLM framework. Thus, GLR can be viewed as a special case of GLM where the assumptions align with those of standard linear regression. In this tool, the model type was determined as Continuous (Gaussian). This method reveals the direction (positive, negative) of the relationship between the dependent variable and independent variables.

1

where represents the dependent variable observation (pneumonia incidence rate) at the locations (Tehsils), is the estimated intercept and indicates the value of y when x equals to zero, is the parameter estimate for denotes the set of explanatory variables. Also represents the regression coefficients that describe changes in the dependent variable y when x changes by one unit. All the variables were standardized before GLR analysis. Subsequently, the VIF for each variable was computed to assess collinearity. A threshold of VIF less than 5 was set to ensure the absence of collinearity issues in the model [59, 101]. Then, the Spatial Autocorrelation (Global Moran’s I) method mentioned in the second stage was used to test for the presence of statistically significant clustering in the linear regression (GLR) residuals.

If clustering in GLR model residuals are detected, we use the MGWR method to explore local predictions and to better understand regional variation.

MGWR is formulated by Fotheringham et al. [44] as follows:

2

where is the response variable, is the bandwidth used in the location ((,)), is the predictor, and is the error term.

The MGWR method uses a variable bandwidth and a different geographical weighting matrix for each geographical object. In this way, compared to the GWR method, which uses a fixed bandwidth and a differential weighting matrix, there is no deterioration in the representation of the dependent variable and its correlations by causing both over- and under-fitting of the variables. Thanks to these features, a higher success rate is achieved in determining geographical heterogeneity [6, 43–45, 130].

In the MGWR model, Bisquare kernel function is used to determine the spatial weights. This function considers only spatial relationships in the immediate neighborhood, giving zero weight to observations beyond a certain distance. A fixed number of neighbors was chosen as the neighborhood type, so that appropriate and comparable local samples were obtained in areas with different densities. The neighborhood size was optimized with the Golden Search algorithm to minimize the AICc value of the model. Although this method requires longer computation time, it provides more reliable results, especially in studies with a limited sample size. The MGWR method has been used in many studies due to these features [16, 73, 107, 122, 131]. At this stage, the Spatial Autocorrelation (Global Moran’s I) tool was used to confirm no remaining spatial clustering is present in the MGWR residuals.

Finally, the geographic data used as input were collected in the coordinate system named WGS 1984 Web Mercator (EPSG 3857). This system was preferred because many geographic data used in MGWR experiments are available in this coordinate system at the global level.

Results

Existence and structure of spatial clustering

From 2016 to 2020, the ratio of the number of cases to the population was analyzed by Moran’s Index method; the result showed a statistically insignificant random distribution (Moran’s Index = 0.007, z-score = 0.47, p-value = 0.632).

We found an LL cluster according to the LISA analysis using case rates, that is, low values of case rates formed next to other low rates only in De-Excluded Area D. G. Khan1 and Dera Ghazi Khan Tehsils, while no statistically significant clusters or outlier patterns were found in other Tehsils (Fig. 2a).

Fig. 2.

LISA results (a), hot spot results (b) of Pneumonia incidence rate in South Punjab, Pakistan during 2016–2020

According to the results of Hot and Cold Spot Analysis using the incidence rates, it was found that hot spots occurred only in the De-Excluded Area Rajanpur Tehsil, and no statistically significant hot or cold spots occurred in all other Tehsils (Fig. 2b).

Spatiotemporal clustering

According to the EHSA results using monthly incidence rates, Sporadic Hot Spot was observed in 7 Tehsils such as Bahawalpur Saddar, Jampur, Rojhan Hasilpur and Oscillating Hot Spot was observed in 7 Tehsils such as Yazman, Rajanpur, Mailsi. Sporadic Cold Spot was found in 5 Tehsils like Layyah and Khan Pur and Oscillating Cold Spot was found in Muzaffargarh Tehsil (Fig. 3a).

Fig. 3.

Emerging hot spot (a), trend analysis (b) of Pneumonia Incidence in South Punjab, Pakistan during 2016–2020

The results of the trend analysis using monthly case rates (Fig. 3b) illustrate the spatial variation in pneumonia incidence trends across South Punjab from 2016 to 2020. Significant upward trends were detected primarily in the southern and central Tehsils, such as Yazman and areas surrounding it, with 99% and 95% confidence levels, indicating an increasing burden of pneumonia in these regions. Conversely, notable downward trends were observed in the northern Tehsils, such as Hasilpur, at both 99% and 90% confidence levels, reflecting a reduction in incidence rates in these areas. Several Tehsils showed no significant trend, showing stable incidence rates over the study period.

Time series clustering

The time series cluster analysis of pneumonia incidence rates across South Punjab from 2016 to 2020 revealed considerable regional spatial and temporal heterogeneity. The study identified four distinct clusters with varying trends and geographic distributions. 27 of the 43 tehsils are in the first cluster (mean case ratio = 2.78), 14 tehsils are in the second cluster (mean case ratio = 4.72), and one tehsil each in clusters three (case ratio = 6.42) and four (case ratio = 24) (Table S4). This method creates these clusters according to both case rates and their distribution over time. The first cluster, shown in blue, comprises 27 Tehsils primarily located in the southern and southeastern parts of South Punjab, including Bahawalpur, Rahim Yar Khan, Yazman, Khanpur, and Lodhran. This cluster exhibited relatively stable and low pneumonia incidence rates throughout the study period. The second cluster (red) comprising 14 Tehsils, was situated mostly in the central region, including Multan, Vehari, and Muzaffargarh, and demonstrated moderate incidence rates with occasional rises. The third cluster (green), consisting of only Hasilpur Tehsil, located in Bahawalpur district, displayed consistently low and stable pneumonia rates with minimal variation. The Fourth cluster (orange), located primarily in the western parts of the region, comprising De-excluded Rajanpur, showed the highest and most volatile incidence rates, with significant peaks occurring in mid-2017 and early 2018 (Fig. 4).

Fig. 4.

Time series clustering and medoids of Pneumonia Incidence in South Punjab, Pakistan during 2016–2020

With the generated data, the number of cases was evaluated only with a temporal approach. Accordingly, it is seen that the number of cases increased between autumn and spring (winter months of Pakistan) seasons and peaked especially in January. Furthermore, it was found that between October and December 2017, the caseload was as high as in January. In addition, in April 2018 and July–August 2017, there were as many cases as in January. In the April-August period of 2020, which coincides with the COVID-19 period, the number of cases was much lower compared to the same period of previous years. Between September-December 2020, the number of cases increased slightly, but remained relatively lower compared to previous years (Fig. 5).

Fig. 5.

Temporal distribution of the number of Pneumonia cases in South Punjab, Pakistan, during 2016–2020

The local socio-ecological determinants of pneumonia in South Punjab

Our global regression model (GLR model) indicated an explanatory power of approximately 40% concerning the incidence rate within the study area. Notably, the results demonstrated statistical significance (p < 0.01) for the explanatory variables (Table S1). The fact that the AICc value is the lowest compared to the other models tested (223.78) indicates the presence of model fit in the context of the available data. Joint F-Statistic 5.739, p-value 0.000287 indicates that the model is generally significant, in other words, all independent variables have a significant effect in explaining the dependent variable. Joint Wald value (15.65, p-value 0.0157) confirmed the significance of the dependent variables in the model. The Koenker (BP) statistic of 19.90 (p-value 0.0028) indicates that there is heteroscedasticity in the model. In the absence of heteroscedasticity in the model, the statistical significance of the independent variables should have been taken into account in the Probability column. Therefore, since there is heteroscedasticity in our GLR model, the values in the Robust_Pr column are taken into account to understand the statistical significance of the independent variables. According to GLR analysis results, all variables are statistically significant and VIF value is below 7.5. While the relationship between room density, literacy rate female and unimproved excreta disposal facility (% household) and incidence rate is negative, the relationship between urban population, mud houses (% household) and unimproved solid waste (% household) and incidence rate is positive (Table S1).

According to the MGWR model, the adjusted R² value of pneumonia incidence rate is 0.67. The MGWR bandwidth range of pneumonia incidence rate varies between 0 and 43 (see Table S2, and Table S3). To ensure that the spatial autocorrelation in the GLR residuals is eliminated in the MGWR analysis, Moran’s I test was applied to the MGWR model. Accordingly, Moran’s I value of the scaled standardized residuals in the pneumonia incidence rate MGWR model was − 0.17, p = 0.07, and z-score = − 1.77. These values indicate that the residual values in the pneumonia prevalence of the under-five MGWR model are statistically dispersed at a 90% confidence level, i.e. not clustered.

The relationship between the total incidence rate and urban population in 2016–2020 is positive statewide but not statistically significant. Similarly, the relationship between the variable unimproved unimproved solid waste (% household) and case rate is positive across the region but not statistically significant. The relationship between room density and case rate is negative but statistically significant only in Sadiqabad Tehsil. The relationship between case rate and female literacy is negative across the state. In 26 Tehsils this relationship is statistically significant. These are mostly clustered in the north-east (Hasilpur and its neighbourhood) and south-west (Rojhan and its neighbourhood). The relationship between mud houses (% household) and case rate is positive across the state and is stronger in the south-west. This relationship is statistically significant in 24 of the Tehsils. The relationship between unimproved excreta disposal facility (% household) and the incidence rate is negative and statistically significant state wide. This relationship is stronger in the southwest (Fig. 6).

Fig. 6.

Spatial distribution of coefficient and significance of explanatory variables for Pneumonia in South Punjab, Pakistan: urban population (a), room density (b), literacy rate female (c), mud houses (% household) (d), unimproved excreta disposal facility (% household) (e), unimproved solid waste (% household) (f)

Discussion

This study is the first to examine spatiotemporal clusters and spatial relationships between demographic and socio-ecological characteristics and pneumonia in South Punjab, Pakistan (2016–2020). Spatial and spatiotemporal analyses revealed clustering patterns, with central areas showing higher case counts, while emerging hotspots appeared in parts of the east. A negative association between female literacy and pneumonia incidence was observed, particularly in the southwest, whereas poor housing conditions were positively associated with incidence in the same area.

Spatial–temporal trends of pneumonia

The Moran’s I analysis indicated clustering when applied to raw case counts; however, such clustering likely reflects population concentration in larger urban Tehsils rather than elevated risk. In contrast, incidence-based analyses (population-adjusted) revealed no significant global clustering, suggesting that pneumonia risk was more evenly distributed across the region. This distinction highlights the importance of incidence rates over raw counts in assessing spatial epidemiological patterns [11, 124].

LISA identified a LL cluster in the DG Khan area (De-Excluded Area D.G. Khan and D.G. Khan Tehsils) (Fig. 2a), while Optimized Hot Spot Analysis detected one hot spot in Rajanpur (De-Excluded Area Rajanpur) (Fig. 2b). Although Multan City and surrounding Tehsils reported high numbers of cases, this primarily reflected their large populations rather than a statistically significant incidence cluster. For instance, the Tehsils showing hotspots of pneumonia include mainly high-density urban Tehsils such as Multan city (7903 persons per km²), comparable to LL Tehsils with low-density regions D.G. Khan (547 persons per km² and De-Excluded Area D.G. (34 persons per km²) [104]. Similarly, the LH Tehsils of Jahanian and Duniapur are characterized by relatively low population density. Overall, the eastern side of South Punjab is more densely populated compared to the western side. Previous studies have shown that higher contact rates of infectious diseases are generally associated with greater population density [56, 76, 85]. In contrast, our analysis identified Rajanpur as a hotspot despite being a low-density region (Fig. 3b), suggesting that factors beyond population density may contribute to pneumonia incidence patterns in this area.

Our spatial epidemiological approach provided further insights into these clustering patterns. The results are consistent with EHSA (Fig. 3a) and time-series clustering (Fig. 4), which identified Rajanpur as a sporadic hotspot and a member of the fourth cluster. This suggests that Rajanpur exhibited an elevated incidence intermittently during the study period, but for less than 90% of the time, and was never classified as a cold spot [24]. In all years, the De-Excluded Area Rajanpur Tehsil has seen a case rate of at least 20 per thousand. Although the case rate generally declined in 2020, corresponding to the early stages of the COVID-19 process, this value never fell below 20 per thousand in Rajanpur Tehsil (Fig. 1). This is consistent with the hot spot in Fig. 2b and the absence of a significant trend in Fig. 3b. Multan City and its neighboring Tehsils displayed a downward trend in case ratios (Fig. 3b), a finding supported by EHSA, which classified Multan City as a diminishing hotspot surrounded by sporadic and oscillating hotspots (Fig. 3a). The western part of the region was characterized by persistent, intensified, and sporadic cold spots based on case ratios, whereas the south–southwestern Tehsils were identified as sporadic and oscillating hotspots. In contrast, several eastern Tehsils exhibited both upward trends and oscillating hotspot patterns, suggesting that some of these areas had transitioned from previous cold spots [24]. These geographic differences may be linked to population density, as the western region is sparsely populated due to rough terrain, while the central plains and eastern Tehsils are more densely settled, which may contribute to observed disparities in clustering patterns.

Seasonal variations, particularly the marked increase in pneumonia cases during the winter months (especially January; Fig. 5), likely explain the temporal clustering observed in South Punjab. The region experiences extreme climatic conditions, with December and January being the coldest months, when temperatures often fall below 15 ℃ [32]. Similar seasonal associations have been reported elsewhere; Lin et al. [77] found that a 1 ℃ decrease in ambient temperature corresponded to a 0.03 increase in monthly pneumonia admissions per 10,000 people in Taiwan. Likewise, Cilloniz et al. [13] and Herrera-Lara et al. [53] noted a higher incidence of polymicrobial pneumonia during winter compared to other seasons. Other acute respiratory infections, such as influenza, the most prevalent infection in Pakistan, also peak in spring and winter [92]. Supporting this pattern, Fatima et al. [32] documented widespread acute respiratory infections in Bahawalpur, South Punjab, during the winters of 2010–2012. Environmental conditions likely compound this risk, such as fog events, which increase aerosol inhalation [108], occur on average seven days per month in December and January [32], while frequent dust storms contribute further to respiratory vulnerability through direct particle inhalation [102].

In addition, pneumonia cases were also observed during the early summer months. Previous research has shown that dust storms are associated with a higher incidence of respiratory infections in this region [35]. While dust storms occur across South Punjab, they are particularly intense in the southeastern zone encompassing the Cholistan Desert [32]. This aligns with our findings, where eastern Tehsils exhibited upward trends in pneumonia incidence (Figs. 3 and 4), supporting the role of dust exposure in shaping spatial patterns. Notably, in 2020, reported pneumonia cases declined, likely reflecting the impact of the COVID-19 pandemic, during which reduced healthcare access and reporting disruptions occurred due to government-imposed mobility restrictions [99].

Taken together, these results suggest that climatic, environmental, and socio-economic factors jointly contribute to the observed spatial–temporal disparities in pneumonia incidence across South Punjab.

Spatial variability in the socio-ecological determinants of pneumonia

Our model revealed a positive but insignificant relation among the urban population, unimproved solid waste, and pneumonia (Fig. 6a and f). Studies support this relation, as dense and poorly sanitized environments facilitate the transmission of pneumonia infections [1, 52, 62]. Previous studies have shown that household crowding can contribute to respiratory infections, but the relationship may vary by setting [15]. In South Punjab, households are typically large, with an average of 6.6 persons per household and a room density of 3.9 persons per room [118]. Previous studies have shown that sharing a bedroom with more than two persons can significantly increase the likelihood of severe pneumonia [114]. In contrast, our analysis revealed a negative and statistically insignificant association between room density and pneumonia incidence across the region, suggesting that other contextual factors may influence this relationship (Fig. 6b). Furthermore, inadequate housing structures have been consistently linked with higher vulnerability to respiratory diseases in similar settings [28, 29, 35].

Our MGWR model indicated a significant positive association between the proportion of mud houses and pneumonia incidence, with this relationship being particularly pronounced in the southern and southwestern Tehsils. These areas are predominantly rural and characterized by lower socio-economic conditions, with approximately 32% of households using mud as the primary building material [90].Poor housing quality may exacerbate exposure to environmental and indoor risk factors. In particular, indoor air pollution is a critical concern, as nearly 87% of households in the study area rely on solid fuels such as wood and coal for cooking and heating, while only 8.4% have access to clean fuel sources such as natural gas [48]. Previous studies have linked such exposures, along with tobacco smoking, to increased susceptibility to respiratory diseases, including pneumonia [3, 4, 86, 100, 109].

Our MGWR results showed a significant negative association between female literacy and pneumonia incidence, with the relationship being particularly strong in southwestern Tehsils (Fig. 6c). Overall literacy rates in South Punjab remain low, averaging 46.6% [48], which may limit health awareness and care-seeking behavior. Previous studies indicate that mothers with limited education often lack awareness of the severity of childhood pneumonia [40], and similar associations between illiteracy and higher susceptibility to respiratory infections, including COVID-19, have been reported [84]. Maternal ability to recognize symptoms and seek timely care is critical for reducing pneumonia-related mortality in children under five [63], however, this capacity is often constrained in settings with low female literacy, as documented in Pakistan [87].

South Punjab is also characterized by high multidimensional poverty and low overall population well-being [90]. Evidence from other settings, such as Brazil, shows that poorer regions tend to have higher pneumonia incidence [113], and a review by Chisti et al. [12] highlighted that factors like severe malnutrition amplify pneumonia burden in disadvantaged populations. In South Punjab, more than 20% of children are underweight [117], further compounding vulnerability. Additionally, the region experiences high levels of air pollution, nearly double the multidimensional poverty burden of other parts of Punjab, and limited access to specialized healthcare facilities, medicines, clean water, and sanitation [120]. Collectively, these socio-economic and environmental disadvantages appear to be associated with higher pneumonia incidence in the region.

Recommendations for socio-environment and public health interventions

Based on our findings, several recommendations can be made. Though notable strategies have already been implemented in areas of social and health development in South Punjab, such as the Punjab Poverty Graduation Initiative, the establishment of 252 healthcare schemes, and 2 new tertiary care hospitals [120], there is a need to improve specialized health care facilities, especially in hotspot Tehsils. In addition, healthcare facilities must be prepared for an increase in demand during peak pneumonia months, leading to seasonal planning. Lastly, strengthening overall public health infrastructure, including vaccination programs and disease surveillance systems, is necessary for early detection and response to pneumonia. Hussain et al. [57] found a lack of public health measures, such as vaccines to fight pneumonia, and demanded the expansion of existing public health interventions in the northern part of Pakistan. Though the provision of pneumococcal vaccine (PCV10) was introduced in 2002 by the Expanded Program on Immunization (EPI) in Punjab [98], however, the situation can be improved through community mobilization approaches for prompt recognition and care seeking for pneumonia, especially for children [103]. Females, particularly mothers, need to be addressed in awareness campaigns. Targeted awareness campaigns focusing on pneumonia prevention and early symptom recognition should be prioritized, especially for women and caregivers in low-literacy regions. Public investment is urgently needed to improve housing infrastructure and environmental sanitation in rural and underserved Tehsils. Thus, the study highlights clear spatial disparities in pneumonia incidence across South Punjab, calling for seasonal preparedness plans in high-burden tehsils, especially during winter months. Targeted literacy campaigns, particularly for women, and housing and sanitation improvements should be prioritized in vulnerable areas. Incorporating geospatial analysis tools into local health systems can further support real-time monitoring, effective resource allocation, and informed decision-making for pneumonia control.

Strengths and limitations of the study

The key strength of our study is that it is the first comprehensive investigation of spatiotemporal clusters and spatial relationships of socio-ecological factors with pneumonia in South Punjab, filling a significant gap in the existing literature. We use advanced spatial analysis, including Moran’s Index, LISA, hot and cold spot analysis, spatiotemporal analysis, and time series clustering, which not only provide a solid understanding of pneumonia distribution but also improve the reliability and validity of the results. By incorporating socio-ecological factors into GLR and MGWR models, our study offers insights into the underlying determinants of pneumonia in this region. Our findings highlight the need for actionable public health measures and enhanced socio-ecological interventions and health resource allocation in hotspots and cluster areas. This study can be replicated not only for other regions of Pakistan but also for other infectious diseases. We had to use ArcGIS Pro’s Exploratory Regression tool because it works with the GLR method. Since the main topics we targeted in our research were obtained using the MGWR method, we ignored the problems of the GLR method. Methods such as the recommended log-linked Poisson GLM, which produce more meaningful estimates than linear regression models (such as OLS, GLR) with counts and ratios, can be considered in future research.

Several limitations should be acknowledged in this research. First, the data utilized in the study are limited to reported cases of pneumonia without including detailed demographic information such as the age and sex of the affected individuals. The absence of this demographic description could potentially limit the depth of analysis and understanding of specific population groups that may be more susceptible to pneumonia. Subsequently, it is crucial to evaluate the potential influence of the COVID-19 pandemic on the reporting of pneumonia cases, particularly in 2020. The pandemic may have disrupted healthcare services, leading to underreporting or delays in reporting pneumonia cases. As a consequence, the accuracy and reliability of the study’s results may have been impacted by this underreporting or potential data bias. Besides this, the absence of recent socio-economic data at the Tehsil level for the South is another limiting factor. Other data, such as socio-economic data, are not produced in sufficient detail or frequency, particularly in less developed and developing countries. Even if the best options are evaluated for data that is not available for consistent dates, situations that affect the results, albeit to a small extent, may arise. However, we do not believe that this will be significant enough to completely alter the overall findings. In addition, it can be assumed that social variables will not change significantly over periods of 4–5 years, except for geopolitical, migration, and other reasons. We also did not incorporate climatic variables in our MGWR model. Distance from healthcare facilities, dust storms, and various pollution data were not used in this study.

Although MGWR has been applied in studies with small samples [10, 80], some caution that fewer than several hundred observations may undermine reliability [64]. Our analysis, based on 43 Tehsils in South Punjab, reflects the most detailed administrative data available but carries risks of overfitting, as suggested by the 12-point gap between R² and adjusted R² (Table S1). While MGWR improved model fit (adjusted R² from 0.40 to 0.67) and reduced spatial autocorrelation in residuals, many variables lacked significance or were modeled with constant bandwidths, limiting interpretability. Residual analysis (Fig. S1) confirmed localized over- and under-predictions, though without significant clustering, supporting the model’s capacity to capture spatial patterns. The effect of potential confounding factors on these relationships could be more rigorously assessed by testing them across different scales and study areas. Recognizing these limitations is essential, as they may introduce biases and restrict the generalizability of the findings; the variables considered for model testing are summarized in Table S5.

Conclusion

This study is the first to investigate the spatiotemporal clusters and spatial relationships between demographic and socio-ecological factors and pneumonia in South Punjab, Pakistan, from 2016 to 2020. We found that central high-density regions were hotspots of pneumonia, but they were diminishing and showed a downward trend. Furthermore, new emerging hotspots were being identified in the eastern region with an upward trend. However, western low-density regions are persistent, intensifying, and sporadic cold spots. We further concluded that with increasing female literacy, pneumonia incidence decreases, while with increasing mud houses, pneumonia incidence increases, and this relationship becomes more pronounced towards the southwest. Our findings emphasize the necessity for targeted socio-ecological and public health interventions, improved healthcare facilities, and strengthened public health infrastructure, especially in identified hotspots. Future research should incorporate age-specific and climatic variables to refine predictive modeling and enhance evidence-based planning for pneumonia control.

Author contributions

Ömer Ünsal: Methodology, Formal analysis, Visualization; Writing—review and editing. Oliver Gruebner: Supervision, Methodology, writing—review and editing. Munazza Fatima: Conceptualization, investigation, methodology, formal analysis, visualization, and writing—review and editing.

Funding

No funds, grants, or other support were received.

Data availability

The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

We obtained Pneumonia reported cases data through data provision letter No: 554 (MIS) dated 5–06-2023 from the Directorate General Health Services, Punjab, Pakistan.

Consent for publication

Not applicable, as we use aggregated data.

Competing interests

The authors declare no competing interests.