A novel spatial heteroscedastic generalized additive distributed lag model for the spatiotemporal relation between PM_2.5and cardiovascular hospitalization

Abstract

Many studies have examined the impact of air pollution on cardiovascular hospitalization (CVH), but few have looked at the delayed effects of air pollution on CVH. Additionally, there has been no research on the spatial and temporal differences in how environmental pollutants affect CVH. This study seeks to identify spatial heteroscedasticity in the relation between PM2.5 and CVH by developing a Generalized Additive Distributed Lag (GADL) model. Data on hospitalization due to cardiovascular disease were collected from the Hospital Information System (HIS) of Mashhad University of Medical Science from 2017 to 2020. Air pollution data from 22 air quality monitoring (AQM) stations were obtained from the Environmental Pollution Monitoring Center of Mashhad administrates. Markov Random Field (MRF) smoother was utilized in the GADL model to account for spatial heteroscedasticity in the observations. This developed model is a Spatial Heteroscedastic Generalized Additive Distributed Lag (SHGADL) model. Our use of GADL allowed us to discover a significant relationship between PM2.5 exposures and the risk of CVH at lags 0 and 1 in all districts. Our results reveal heteroscedasticity in the Relative Risks (RR) of PM2.5 on CVH across different districts. After accounting for this spatial heteroscedasticity, we found that the RR of PM2.5 on CVH at lags 0 and 1 were 1.0102 (95% CI: 1.0034, 1.0170) and 1.0043 (95% CI: 1.0009, 1.0078) respectively. The central and southeastern districts showed higher RR for CVH. The developed SHGADL model provides evidence of a significant lagged effect of PM2.5 exposures on CVH, and identifies low- and high-risk districts for CVH in Mashhad. This finding can assist decision-makers in allocating resources and planning strategically, with a focus on local interventions to manage ambient air pollution and providing emergency care for CVH.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-024-81036-3.

Introduction

Cardiovascular diseases (CVDs), such as heart disease, stroke, and arrhythmias, are a significant contributor to illness and death, placing substantial financial strain on countries globally¹. In Iran, CVDs are the primary cause of death, leading to millions of disability-adjusted life years (DALYs)². Additionally, CVDs are responsible for 46% of all fatalities and contribute to 20–23% of the overall disease burden in Iran³. Predictions suggest that the DALYs linked to CVDs in Iranians aged 30 years and older will double by 2025 compared to the statistics from 2005².

Air pollution presents a widespread global health concern that heightens the likelihood of cardiovascular disease. Particulate matter (PM), specifically those with a diameter smaller than 2.5 micrometers (PM_2.5), is a significant contributor to the detrimental impact on cardiovascular health⁴. Recent research, including epidemiological and clinical studies, has indicated that exposure to air pollution, particularly PM_2.5, is linked to a rise in both mortality and morbidity related to cardiovascular disease^5–9.

Various models and methodologies have been used to study the impact of air pollution on health outcomes. Time series methods often use moving averages of air pollution concentrations for exposure assessment, but this may underestimate the overall effect due to the failure to capture the complete lag structure¹⁰. Distributed Lag Models (DLMs) have gained prominence in epidemiological and environmental health studies, allowing for the estimation of associations between environmental exposures and health outcomes at various lagged periods. Recent attempts have aimed to move beyond linear relationships by exploring non-linear effects, leading to the development of the Generalized Additive Distributed Lag (GADL) model. This modeling approach has gained popularity in environmental health studies, particularly in the fields of climate change and air pollution research^11–21. Despite significant progress in environmental exposure models, there is limited research on spatial heteroscedasticity and spatial dependency in assessing the delayed effects of environmental exposure on health outcomes. Studies have shown that there is significant spatial heteroscedasticity in air pollution concentrations at a small scale within urban areas. However, most existing studies on cardiovascular disease have only looked at average ambient pollutant concentrations across entire cities, ignoring the within-city spatial variability in air pollutant levels. By considering spatial heteroscedasticity, GADLs can better capture geographical relationships and provide more accurate estimations, especially in scenarios where spatial factors are important^22,23. Therefore, incorporating the complex spatial structure aspects into GADL is crucial for improving the accuracy and precision of the investigated relationships.

A previous study reported significant variations in PM_2.5 levels across Mashhad districts, but existing studies on air pollution and cardiovascular disease use the average of PM_2.5for all districts, neglecting spatial heteroscedasticity²⁴. The goal of present study is to incorporating this spatial heteroscedasticity in air pollution concentrations and its effect on the risk of CVH by a Spatial Heteroscedastic Generalized Additive Distributed Lag (SHGADL) model.

Materials and methods

Study area

This research was carried out in Mashhad, which is the capital of Razavi Khorasan province in the northeastern region of Iran. With a population estimated at nearly 3.8 million, it is the second most populous city in Iran²⁵. As shown in Fig. 1 the city is comprised of 13 administrative districts.

Data source

The Hospital Information System (HIS) of Mashhad University of Medical Science provided hospitalization data for cardiovascular disease from 2017 to 2020, as depicted in Fig. 2. For these patients all methods were performed in accordance with the relevant guidelines and regulations. The selection of cardiovascular diseases was based on the International Classification of Diseases, 10th Revision (ICD-10), using codes I00-I99. Patients’ residential addresses were geocoded to latitude and longitude formats using Google Maps, and only patients residing within Mashhad city boundaries were included in the study. The average 24-hour concentration of PM_2.5 was gathered from 22 AQM stations located across various parts of Mashhad, as shown in Figs. 1 and 2. These stations are managed by the Environmental Pollution Monitoring Center of Mashhad.

Fig. 2.

Time series plot of PM_2.5 concentrations and CVH from 2017 to 2020, Mashhad, Iran.

Missing data imputation

In our original AQM station data, we found that there was an average of 30% missing data in the PM_2.5measurements (Fig. 1 in Appendix S1). Machine learning methods are best to impute missing data to improve accuracy²⁶. Recent research has indicated that machine learning techniques, particularly the Random Forest (RF) and k-Nearest Neighbor (KNN) algorithms, are effective for filling in missing data^27,28. Random Forest (RF) is a powerful machine learning algorithm that constructs an ensemble of decision trees to make predictions. One of its key advantages is its robustness to overfitting, as it averages the results from multiple trees to improve accuracy. RF is particularly effective at capturing complex nonlinear relationships in the data^29,30. KNN is a non-parametric method that imputes missing values based on the values of the k nearest neighbors. KNN is simple to implement and requires minimal parameter tuning, primarily the choice of k, the number of neighbors to consider. A significant advantage of KNN is that it makes no assumptions about the underlying data distribution, making it flexible and broadly applicable³⁰. We used both models to fill in the missing PM_2.5 data for each AQM station, and then compared their performance using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) (refer to Table 1 in Appendix S1). The results showed that the random forest method outperformed the k-Nearest Neighbor algorithm in filling in missing PM_2.5 data, with the exception of station 6. Therefore, we chose the random forest method as the preferred model for filling in missing data. Figure 2 in Appendix S1, illustrates how the imputed PM_2.5 data aligns with the actual information at each station.

Table 1.

Descriptive statistics of daily CVH and PM_2.5 concentrations in Mashhad, from 2017 to 2020.

		Total	Mean	SD	Min	Percentile			Max
						P 0.25	P 0.50	P 0.75
CVH		45,089	31	12	3	22	30	39	77
Gender
	Male	22,851 (54.67%)	17	7	1	12	17	22	42
	Female	18,943 (45.32%)	14	7	1	10	14	18	47
Age(year)
	<65	22,781 (50.52%)	16	7	0	10	15	20	44
	≥65	22,307 (49.47%)	15	6	1	11	15	19	42
PM 2.5 (µg/m 3 )			32.70	14.04	11.87	24.92	30.56	37.88	90.00

Fig. 1.

Study area location (A and B) with distribution of 22 AQM stations within 13 administrative districts in Mashhad (C), Iran.

District-level exposurse assessment.

The exposure to PM_2.5at the district level was calculated using inverse distance weighting (IDW) interpolation. IDW is a deterministic interpolation method that uses the inverse distance between points for weighting. IDW is computationally straightforward to implement, which can be advantageous in studies requiring rapid assessments or when computational resources are limited. Various studies used IDW for PM trend evaluation and PM exposure analysis^32–35. Initially, a 500-meter by 500-meter grid was established within the study area. The values at each grid point were estimated by averaging the weighted sum of values from nearby monitoring stations within the study area. The weights assigned to each neighboring value were determined based on the inverse distance between the monitoring station locations and unmeasured locations, raised to a non-negative power³⁶. Finally, daily averages of PM_2.5for each district were calculated based on the average of the interpolated grid values within that specific district. To validate the IDW method, a leave-one-out cross-validation was used to determine the optimal power value for distance measurement and the number of nearest neighbors. The parameter set that minimized the prediction root mean square error (RMSE) was selected for each day³⁷. The power of distance measurement was tested using five values from 1 to 3, and the number of neighbors varied from 3 to the maximum number of stations.

Dealing with confounding factors

Two Confounding factors influenced the daily number of CVHs during the study period. The first factor was the introduction of a health screening program for Iranians aged 30 and above, which aimed to identify high-risk individuals and refer them to hospitals for further evaluation and treatment. The second factor was the onset of the COVID-19 pandemic, which had a significant impact on hospitals (Fig. 2). The pandemic led to healthcare facilities needing to allocate resources and capacity to treat COVID-19 patients, creating challenges in the delivery of healthcare services. To account for the influence of these two factors, we introduced two dummy variables. Days during the COVID-19 pandemic or days during the screening health program were assigned a value of 1, while other days were assigned a value of 0.

Data analysis

The Generalized Additive Distributed Lag (GADL) model is designed to analyze the relationship between environmental exposures and health outcomes over time, allowing for both nonlinear and lagged effects. Its key functions include:

1. Nonlinear Relationships: The GADL model employs smoothing functions to capture complex, nonlinear associations between exposure variables (e.g., PM2.5 levels) and health outcomes (e.g., cardiovascular hospitalizations). This flexibility is crucial in environmental health studies, where relationships are not always linear.

2. Lagged Effects: The model allows for the examination of delayed effects of exposures, enabling researchers to assess how health outcomes may be influenced by past exposure levels. This is particularly important in understanding the cumulative impact of pollutants like PM2.5 over time.

3. Handling Confounding Factors: The GADL can include additional covariates to control for potential confounders, such as demographic factors or seasonal variations, enhancing the robustness of the findings.

A GADL to analyze the association of a time series of outcomes to covariates is given by

Where, , g is a monotonic link function, and Y is assumed to follow a distribution belonging to the exponential family. The functions represent smoothed relationships between the variables and the linear predictor, defined by the parameter vectors . The variables includes additional predictors or confounders with linear effects specified by the related

coefficients . Here, l and Lrepresent the lag number and the maximum lag, respectively¹⁶. Initially, we established a fundamental model based on the original GADL to examine the non-linear lag effect of PM_2.5 on CVH. Suppose is the number of CVH at day t, following a Poisson distribution , where is the expected value of CVH on day t. Due to the extra variation in relative to , we considered a quasi-Poisson model to deal with the over-dispersion problem. So, The global model for the association between PM_2.5 on CVH is defined by the following formula:

Where log(E(CVH_t)) is the canonical link function for count data with Poisson and quasi-Poisson distribution, α represents the constant term; indicates the cross-basis matrix obtained by applying the GADL to the concentration of PM_2.5; with lag representing the number of lag; ns () indicates the smoother of the nature cubic spline for time trend (Fig. 3 in Appendix S1). The “dw” and “holiday” are the dummy variables for the day of the week and public holidays. Also two dummy variables; “screening” and “covid” included in the GADL for health screening program and the COVID-19 pandemic confounders. The offset term is the logarithm of population.

This GADL model is applicable in three key areas. First, in the epidemiological domain, it is utilized to investigate the effects of air pollution on health outcomes in urban environments, considering both immediate and delayed impacts. Second, in the context of climate change, this model is effective for assessing how changes in environmental conditions affect public health over time, particularly concerning air pollution. Lastly, in policy evaluation, the GADL model aids in assessing the effectiveness of interventions designed to reduce pollution levels and their subsequent health impacts, offering valuable insights for policymakers.

Spatial heteroscedastic generalized additive distributed lag (SHGADL)

To incorporate spatial heteroscedasticity, we employed a SHGADL by adding the spatial random effect ( ) for the districts in GADL model as follows:

Where and represent the number of CVH and average PM_2.5 concentration on day t in district i respectively. The spatial function was estimated using the Markov Random Field (MRF) smoother to account for spatial autocorrelations and spatial heteroscedasticity. It is implemented through a conditional autoregressive prior following a Normal distribution:

where represents the neighborhood set containing all adjacent districts sharing boundaries with district i, and denotes the total number of districts in . Additionally, is an unknown variance parameter in district i that needs to be estimated³⁸. Therefore, by including a MRF smoother in the GADL, it can be developed into a special type of conditional autoregressive model. By using log link function, the exponential of estimated parameters can be interpreted as the relative risk for CVH. Moreover, the spatial function can identify each district as a low or high-risk area for CVH after controlling other factors, as the 95% confidence interval (CI) of the spatial estimate is significantly greater than zero.

We used the Akaike information criterion for the quasi-Poisson model (Q-AIC) as a measure of goodness-of-fit³⁹ to find the appropriate exposure response association. Based on the minimum Q-AIC, our approach involved utilizing a linear function exposure response function. For the lag-response association, a B-spline function with 2 degrees of freedom was employed, incorporating a maximum lag of 7 days in the creation of the cross-basis.

All statistical analyses were conducted using R software (version 4.3.2) with the “caret” package for missing imputation, “gstat” package for spatial interpolation of PM_2.5, “dlnm”and “mgcv” packages for modeling, and “ggplot2” for creating figures. The R codes for this study are presented in supplementary file.

Results

Table 1 presents the descriptive statistics for the daily Cardiovascular Hospitalization (CVH) and PM_2.5 concentrations in Mashhad, covering the period from 2017 to 2020, in addition of the descriptive statistics for daily PM_2.5 concentrations. A total of 45,089 patients within Mashhad addresses with CVH were selected for study, 22,851 (54.67%) of these patients were male and 22,307 (49.47%) were older than 65 years. During the 4-year study period, the daily mean of CVH and PM_2.5 concentrations was 31.0 ± 12.0 and 32.70 ± 14.04 µg/m³, respectively.

Figure 3 presents an overview of the impact of PM_2.5 on CVH based on the GADL results. The graph in 3-a illustrates the RR across various levels of PM_2.5 concentration and time lags, showing a significant immediate effect of PM_2.5 on CVH, with a delayed effect for higher PM_2.5 values. In plot 3-b, there is a clear demonstration of a substantial and progressively increasing association between cumulative PM_2.5 exposures and CVH, highlighting the cumulative impact of PM_2.5 on CVH and indicating a significant rise in risk with higher PM_2.5 exposures. The plot 3-c shows the lag-response association at a specific PM_2.5 exposure value of 10, indicating how the relationship between PM_2.5 exposures and CVH varies with different time lags. The plot 3-d displays the overall cumulative association, indicating a significant cumulative effect of a 10-unit increase in PM_2.5 on CVH for both the current day and a following day.

Fig. 3.

The association between air pollution exposure and CVH in Mashhad, Iran, from 2017 to 2020. (a): Exposure–lag–response surface risk. (b): cumulative exposure-response association. (c): lag- response association at 10-unit exposure. (d): overall cumulative association.

The result of GADL fitting reveals a significant immediate effect of PM_2.5 on CVH (RR = 1.019, CI_95%: (1.005, 1.033)).

Results of SHGADL model

Table 2. displays district-specific average daily PM_2.5 concentrations and the rate of CVH per 100,000 of population. The average daily PM_2.5 concentration varied between 29.8 ± 10.1 and 34.6 ± 13.3, and the highest level was in District 7. Additionally, the average daily CVH varied across districts, ranging from 0.8 to 3.6, and the highest value was in District 13.

Table 2.

Descriptive statistics of daily PM_2.5 concentration and daily rate of CVH in districts of Mashhad, Iran, during 2017 to 2020.

Districts	PM2.5 (µg/m3)		Daily CVH per 100,000 population
	Mean	SD	Mean	SD
1	31.5	11.6	1.5	1.1
2	31.1	11.7	1.2	0.6
3	30.9	11.6	0.9	0.6
4	30.9	11.6	1.3	0.9
5	32.3	11.9	0.8	0.8
6	33.8	12.5	0.8	0.7
7	33.9	10.7	1.4	1.0
8	30.2	10.3	1.8	1.5
9	31.0	10.9	0.8	0.6
10	30.6	11.4	0.8	0.6
11	31.9	10.7	0.9	0.7
12	29.7	11.5	0.9	1.2
13	30.3	10.4	3.6	6.0

Figure 4 illustrates the variation in CVH rates among the 13 districts of Mashhad. Over the course of our four-year study, five districts surpassed the average CVH rate. Three of these districts had particularly high rates, with crude rates ranging from 1000 to 5200 cases per 100,000 of population.

Fig. 4.

Geographic distribution of total CVH per 100,000 of population from 2017 to 2020, Mashhad, Iran.

The spatially adjusted effect of PM_2.5on CVH is depicted in Fig. 5 using the SHGADL model. The exposure-lag-response surface in the plot 5-a of this figure shows the lag response at different exposure intensities and the exposure-response at various lags. By integrating spatial heteroscedasticity into the GADL model, the results indicate a significant spatially adjusted association between PM_2.5 exposures and CVH. The plot 5-b the figure illustrates a direct association between cumulative PM_2.5 exposures and CVH. The plot 5-c demonstrates the lag response association at a specific PM_2.5exposure value of 10 across the lag space. It was found that a 10-unit increase in PM_2.5 exposures significantly raised the risk of CVH on the same day as well as in the subsequent two days. The plot 5-d in the figure shows a significant cumulative effect of a 10-unit increase in PM_2.5 on CVH for both the current day and the following day. The covariate estimations of SHGADL model are summarized in Table 2 of Appendix S1. As shown in this table the main effects of COVID-19 and PM2.5 were statistically significant, indicating that each had a notable impact on CVH independently. However, the interaction term between COVID-19 and PM2.5 was not statistically significant, demonstrating that the relationship between PM2.5 and CVH did not significantly change during the COVID-19 pandemic.

Fig. 5.

The spatial adjusted association between air pollution exposure and CVH in Mashhad, Iran, from 2017 to 2020. (a): Exposure–lag–response surface risk. (b): cumulative exposure-response association. (c): lag- response association at 10-unit exposure. (d): overall cumulative association.

Table 3 displays the RRs of lagged and cumulative exposure to PM_2.5 on CVH using the SHGADL model. The results indicate significant positive associations at lag 0 and 1 for the lagged effect. A 10 µg/m³ increase in PM_2.5 is shown to significantly elevate the risk of CVH at lag 0 (RR = 1.0102, CI95%= (1.0034, 1.0170)) and lag 1 (RR = 1.0043, CI95% =(1.0009, 1.0078)). Moreover, cumulative exposure to PM_2.5 is found to significantly impact the risk of CVH at lag 0–3 days post-exposure.

Table 3.

The RRs of exposure to PM_2.5 on CVH by fitting the SHGADL model.

Lag	Lagged effect	Cumulative effect
	RR(CI95%)	RR(CI95%)
0	1.0102 (1.0034, 1.0170)	1.0102 (1.0034, 1.0171)
1	1.0043 (1.0009, 1.0078)	1.0146 (1.0047, 1.0246)
2	1.0004 (0.9973, 1.0034)	1.0149 (1.0040, 1.0260)
3	0.9982 (0.9945, 1.0020)	1.0132 (1.0017, 1.0247)
4	0.9978 (0.9942, 1.0017)	1.0111 (0.9989, 1.0234)
5	0.9996 (0.9965, 1.0026)	1.0107 (0.9979, 1.0236)
6	1.0030 (0.9996, 1.0064)	1.0137 (1.0010, 1.0266)
7	1.0083 (1.0015, 1.0152)	1.0222 (1.0088, 1.0358)

Figure 6 depicts the estimated spatial function, which illustrates the spatial heteroscedasticity of the relative risk of CVH. The map reveals a distinct and disparate pattern within our study area. Specifically, districts in the central and southeastern regions exhibit higher relative risks for CVH than other areas. Among the 13 districts, six have RRs greater than 1. So, when the PM_2.5 concentrations are elevated, these districts are at a heightened risk for CVH in Mashhad.

Fig. 6.

Spatial heteroscedasticity of the relative risk of CVH in Mashhad, Iran.

Discussion

Air pollutants, particularly PM_2.5, have been suggested as a potential factor affecting human health and have been identified as a major global public health threat according to the WHO’s report⁴⁰. This study utilized a novel SHGADL model to investigate the spatial correlation between the time series of PM_2.5 concentrations and cardiovascular health in Mashhad from 2017 to 2021. Our approach goes beyond previous research by incorporating spatial variability into the analysis, addressing an important aspect that had not been explored in earlier studies. Initially, we employed a non-spatial GADL model, which revealed a significant link between PM_2.5 exposures and an increased risk of hospitalizations related to cardiovascular disease on the same day. Specifically, for every 10 µg/m³ increase in PM_2.5 concentrations, the relative risk of daily hospitalizations for cardiovascular disease increased by RR = 1.019, CI 95%= (1.005, 1.033).

Our results are consistent with a study conducted by Xu et al., in which they noted a significant increase in CVD hospital admissions (0.30, CI 95%= (0.20, 0.39) for every 10 µg/m³ rise in PM_2.5concentration from the previous day to the current (lag 0–1)⁴¹. Similarly, a study carried out in Wuhan, China, found that a 10 µg/m³ increase in PM_2.5(lag 0–2 days) was linked to a 1.23% (CI 95% = (1.01–1.45)) increase in the risk of hospital admissions for CVD⁴². In Iran, a systematic review and meta-analysis showed that for every 10 µg/m³increase in the combined concentrations of all air pollutants, there is a 0.7% (CI 95% = (0.6–0.9%)) increase in hospitalizations due to cardiovascular problems⁴³.

However, many studies have presented evidence connecting air pollutants to adverse health effects by assuming uniform exposure within each region or city. This assumption may result in inaccuracies in estimating exposure⁴⁴. By integrating a spatially varying MRF process to account for spatial heteroscedasticity among districts, we were also able to address the spatial autocorrelation between observations. This approach is likely to generate more precise estimates compared to using citywide average concentrations. Our analysis showed that a 10 µg/m³ increase in PM_2.5exposures was significantly linked to an increase in CVH at lag 0 and lag 1. Consistent with prior research^45,46, our findings indicate that the effect estimates derived from district-level PM_2.5 (using SHGADL) were higher than those obtained from averaged PM_2.5 values across multiple stations (using GADL). This suggests that previous time-series studies relying on average levels may have underestimated the impact of PM_2.5⁴⁵.

Our research reveals that there is a heteroscedasticity in the estimated risk of cardiovascular health issues across the 13 districts of Mashhad, with higher relative risk values observed in the central and southeastern regions compared to other areas. District 13 of Mashhad is the oldest district of the city, which due to the presence of a pilgrimage destination (the Holy shrine of Imam Reza) is a crowded and heavily trafficked area. During certain times of each year, 1 to 3 million pilgrims from different regions of Iran come to Mashhad and stay in the numerous accommodations in District 13. Most of these pilgrims are elderly and are among the cardiovascular hospitalization cases in this district, although they are not part of the district’s population. During these special periods, the air pollution in District 13 increases significantly due to heavy vehicle traffic. As a result, the highest risk of cardiovascular hospitalization is observed in this district.

It is important to take into account the spatial heteroscedasticity of ambient air pollutants when identifying populations at high risk. Similar study in Beijing, China investigated the spatial heteroscedasticity in the relationship between particulate matter less than 10 μm (PM₁₀) concentrations and cardiovascular mortality. The results indicated that a 2.46% (CI95%= (1.22, 3.72)) increase in daily cardiovascular mortality was associated with an interquartile rise in PM₁₀ concentration. These findings emphasize that the risk of mortality related to air pollutants differs across geographic districts.

To the best of our knowledge, this is the first study to investigate the spatiotemporal association between CVH and air pollution that simultaneously accounts for spatial heteroscedasticity and nonlinear lag effects. Indeed, nonlinear lag analysis and spatial analysis have been treated as separate entities in previous research studies. However, it is worth noting that one study has demonstrated the importance of incorporating a spatial function into the GADL⁴⁶. By doing so, this approach acknowledges and accounts for spatial dependencies, offering a more comprehensive and accurate representation of the complex relationships between variables. The integration of complex spatial structure aspects into the GADL can increase the robustness of the analysis, and capturing additional variances that may be missed when these aspects are not considered.

There are some limitations to this study. First, data for PM10 and other air pollutants were not available, so it was not possible to adjust for these variables. Second, our study only considered exposure within residential areas and did not take into account important factors influencing individual exposure, such as daily commuting and working in different areas. Third, some factors including personal and socioeconomics information that could affect the effects were not controlled for in this study.

Conclusion

In conclusion, this study shows the complex relationship between PM_2.5 exposures and CVH. The results show a significant immediate and delayed effect of PM_2.5 on CVH. Spatial analysis explored distinct patterns, indicating higher relative risks of PM_2.5 for CVH in the southern and central regions of Mashhad. These significant spatiotemporal relationships provide evidence about the causal effect of PM_2.5 on CVH. Our findings underscore the need for targeted interventions in specific high-risk areas to mitigate the health impact of air pollution.

The developed SHGADL model contributes to the understanding of the spatial and temporal dynamics of air pollution’s impact on cardiovascular health, providing crucial information for public health planning and policy implementation to reduce the burden of CVH in Mashhad and similar urban settings. Moreover, our analysis reveals analytically that considering spatial heteroscedasticity improves the precision of estimating the delayed effect of PM_2.5 on CVH.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

H. A: Conducted statistical analysis and manuscript writing. K. H: Provided scientific advice on data analysis, contributed to manuscript development. M.S: Participated in data collection, data pre-processing, and execution of geocoding tasks. S.A: Conceptualized the research, supervised the project, and scientifically revised and editing the manuscript. All authors read and approved the final manuscript.

Funding

None.

Data availability

The datasets generated and analyzed during the current study are available upon request to the corresponding author upon permission from the deputy of research of Mashhad University of Medical Sciences.

Declarations

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

The study protocol was approved by the Ethics Committee of Mashhad University of Medical Science with number: IR.MUMS.FHMPM.REC.1401.026 and the informed consent was obtained from all subjects and/or their legal guardian(s).

Consent for publication

All authors approved the manuscript and this submission.