Spatial Association between Air Pollution and Non-communicable Diseases and their Distribution in India

Abstract

Background:

Non-communicable diseases (NCDs) are the fastest-growing global health burden, and air pollution is a major threat to human health. Growing evidence suggests that air pollution especially PM_2.5 is associated with diabetes mellitus and hypertension, which are currently major public health concerns. There has not yet been any study on the association between air pollution and NCDs at a national level in India. We used the National NCD Monitoring Survey (NNMS) surveyed diabetes and hypertension data (2017–2018) of 9,759 individuals in India and superimposed it on a 20-year average concentration of PM_2.5 in ambient air, measured by satellite-based imagery (2000–2019), to find the association of PM_2.5 with diabetes and hypertension.

Methods:

Spatial and geostatistical methods were employed to assess the association and examine the distribution of PM2.5, diabetes, and hypertension rates across geographic regions in India.

Results:

We observed a weak positive correlation between hypertension and PM_2.5 levels, whereas diabetes showed both positive and negative correlations across geographic spaces, albeit with low strength. We did not find any statistically significant results that explain an association, suggesting a lack of association between the two.

Conclusions:

Since we did not find any significant result of the association, which may be evidence of not having any direct association, conducting prospective cohort studies may help find if there is any association at all.

INTRODUCTION

In India, non-communicable diseases (NCDs) are estimated to account for 66% of mortality and a significant proportion of disease burden.[1] Diabetes and hypertension are the most common NCDs of public health concern in India. The prevalence of diabetes has increased in India, from 5.5% in 1990 to 7.7% in 2016.[2] Likewise, there is a reported increase of 89% in the burden of disease in terms of disability adjusted life years (DALYs) attributable to hypertension.[3] While the rising burden of NCDs can be attributed to lifestyle changes and the increasing prevalence of behavioral and metabolic risk factors, recent research has also identified air pollution as a risk factor for most NCDs. It has been strongly argued that PM_2.5 is a risk factor for cardiovascular disease.[4,5,6] One study provided evidence of linking PM_2.5 exposure to diabetes mellitus in the Mexican population.[7] A nationwide cross-sectional study conducted in China concluded that long-term exposure to PM_2.5 may increase the risk of type 2 diabetes mellitus.[8] Findings from a US-based study indicate that PM2.5 may contribute to an increase in diabetes prevalence.[9] A study conducted in the Japanese population found that long-term exposure to PM_2.5 is associated with an increased risk of diabetes.[10] These studies support the evidence of a link between PM_2.5 and diabetes. A few cross-sectional studies conducted in India indicated a link between high exposure to PM_2.5 with diabetes mellitus and hypertension.[11,12,13,14] However, some studies have not found an association between air pollution and diabetes. A study conducted in India has shown no association between particulate matter and diabetes prevalence.[15,16] Another study found no statistically significant association between air pollution and type 2 diabetes.[17] Yoo-Jung Choi et al. stated in their study that the correlation between blood pressure and O₃ and NO₂ was significantly higher in hypertensive patients; however, there was no significant correlation between systolic blood pressure and particles of PM_2.5 or PM₁₀ from 0 to 8 hours before blood pressure measurement.[18] Recently, a secondary data-based cohort study found an increased risk of diabetes associated with PM_2.5 and NO₂ exposure; however, the effects remained when restricting the data to exposure levels below the NAAQS.[19] Another study by Karamanos et al. concluded that the association between air pollutants and blood pressure varies by gender during adolescence and suggests further studies on fine particulate matter and blood pressure levels in young adults.[20] Thus, we are yet to find conclusive evidence of the association of PM_2.5 with common NCDs like type 2 diabetes mellitus and hypertension.

Most of these studies have been conducted in countries where the PM_2.5 concentration in ambient air is within permissible limits. Further, these studies were conducted among small population groups and restricted to specific geographic areas. In contrast, India’s average PM_2.5 concentration in ambient air is significantly higher than the permissible level and exceeds that found in most developed countries. According to the logic of a linear dose-response relationship, given the high concentration of PM_2.5, the prevalence of these two NCDs is expected to be significantly higher in India. At the same time, there is a wide range of variation in PM_2.5 concentrations across the country, which should be reflected in the distribution of diabetes and hypertension in the country. However, to date, no studies have analyzed national-level data to identify an association between the spatial distribution of PM_2.5 and the two NCDs in question.

The PM_2.5 concentrations in ambient air across India are much higher than those observed in high-income countries. It is also widely heterogeneous, given the diversity of land use patterns, weather parameters, and landscapes. Furthermore, there are insufficient monitoring sites in India to provide nationally representative data. However, satellite imagery-based PM_2.5 levels are now available for the entire country, spanning a 20-year period. The[21] robustness of satellite-based estimation of PM_2.5 concentrations has been proven by various researchers.[21]

We conducted this study to investigate whether there is a spatial association between ambient air PM_2.5 concentrations and the prevalence of hypertension and diabetes mellitus.

MATERIALS AND METHODS

Study design

The study is a secondary data analysis to determine the spatial association between PM_2.5 levels in ambient air and the prevalence of Hypertension and Diabetes mellitus in India.

PM_2.5 data

The PM_2.5 data were derived from a national ambient PM_2.5 database created for air quality management in India, based on satellite imagery for two decades (2000–2019). The algorithm used for extracting data is developed from previous works and is based on a scaling factor that incorporates the impacts of local emissions, atmospheric processes, meteorology, and regional transport on the aerosol optical depth (AOD)–PM_2.5 relationship.[21] Details of the algorithm for estimating PM_2.5 from satellite AOD are provided by Dey et al.[21] AOD was converted to surface PM_2.5 using a dynamic conversion factor from MERRA-2 reanalysis. The satellite-derived data showed a strong correlation with ground monitoring data collected at monitoring sites by the Central Pollution Control Board (CPCB). The correlation coefficients for 24-hour average PM_2.5 and annual average PM_2.5 were 0.84 and 0.97, respectively, with a very low root mean square error of 7.2. µg/m³.[21] The data were available at a 1 × 1 km spatial resolution in raster format, with X and Y cell sizes of 0.1 km each. Monthly PM_2.5 data were derived from satellite-captured AOD, a count of 240 raster images (12 months *20 years = 240 images). Cell-to-cell raster analysis was performed to create a single map showing the average of 20 years of PM2.5 data.

To make the data comparable with diabetes mellitus and hypertension data, district-wise zones were created by performing zonal statistics in ArcGIS Pro (Version 2.9.0, Environmental Systems Research Institute, Inc., India).

Non-communicable diseases

Data on risk factors for NCDs and the prevalence of diabetes and hypertension were obtained from the National NCD Monitoring Survey (NNMS, 2017–2018). The survey was conducted to generate national-level estimates of key indicators related to NCDs in India. A detailed methodology for this data collection is given elsewhere.[22] The NNMS defined diabetes as a fasting blood sugar level of 126 mg/100 ml or more. Hypertension was defined as systolic and diastolic blood pressure greater than 140 and/or 90 mm of Hg, respectively.

Data points

The cut-offs for fasting blood sugar and blood pressure were used to identify cases of diabetes and hypertension from the NNMS data. The prevalence was calculated using the following formula:

Number of cases in each primary sampling unit (PSU)/(sample size of PSU) ×1000. The sample size in each PSU was 20; thus, it was too small to calculate a prevalence rate.

However, since the sample size was 20 in each PSU, all PSUs in a given district were aggregated, and subsequently, district-level prevalence was calculated. Each district was represented as a polygon in spatial analysis.

Data analysis

The concentration of PM_2.5 constituted the exposure variable, while persons with diabetes and/or hypertension were included as the outcome variable.

Using geospatial techniques and GIS tools, person-level data were analyzed to examine the distribution of diabetes and hypertension across geographic space. ArcGIS Pro Version 2.9.0 (Environmental System Research Institute Inc., USA) was used to create maps and perform spatial statistical analysis. NNMS (2017–18) data included individuals’ fasting blood sugar, systolic blood pressure, and diastolic blood pressure.

Spatial classification and hotspot analysis methods were used to see the distribution and variation in diabetes and hypertension prevalence across India. Five classes were created using the equal interval grouping method, based on the prevalence value. Each class has a range of prevalences comparable to other classes, which helps us understand the variation of the disease in our study area. Hot spots and cold spots were identified using the spatial hotspot analysis method, which involves spatial clusters of high values (hot spots) and low values (cold spots). The statistical significance of these clusters was measured using the “Getis-Ord Gi” statistic, as calculated through Gi_Bin fields, to determine statistically significant hot and cold spots. The hot spot analysis method comes with a P value and a z-score, where the P value reflects the significance of the analysis, and the z-score interprets the intensity of clustering. A P value of < 0.05 was considered statistically significant. The higher the z-score with a positive value is, the more intense the clustering of high values. However, the lower the z-score with a negative value is, the more intense the clustering of low values. A single layer was created using monthly average PM_2.5 levels. A zonal statistic was used to generate region-wise PM_2.5 concentrations, where district boundaries were used as the zonal input. A total of 240 raster images (12 months × 20 years) were combined to create a single map showing a 20-year monthly average of PM_2.5.

The linear relationship between PM_2.5 and diabetes and hypertension prevalence was assessed by importing the data into IBM SPSS software (Version: 28.0.0.0 (190), SPSS India). Linear regression was performed between district-wise PM_2.5 average concentration (independent variable), and the prevalence of diabetes and hypertension (dependent variables), using the adjusted R-squared value to determine the association. A scatter plot was also generated to represent the values of two variables used to observe the relationship between PM_2.5 and diabetes and hypertension. Geographically weighted regression (GWR) was performed in ArcGIS Pro to identify the spatial relationship across geographic spaces on the same map. GWR determines relationships, especially when independent variables are non-stationary, and identifies statistically significant relationships, if any, using local entropy. The study area is divided into multiple classes, each showing different relationships over space. In our analysis, GWR used PM_2.5 concentration as a nonstationary as well as independent variable to identify associations with diabetes and hypertension, revealing different local R-squared values across geographic regions. A model diagnosis is generated during GWR, which interprets the analysis’s output. The model diagnosis includes r², Adjusted r², Corrected Akaike Information Criterion (AICc), Sigma-Squared, Sigma-Squared maximum likelihood estimate (MLE), and Effective Degrees of Freedom. R-squared (r²) is a measure of goodness of fit, and its value ranges from 0.0 to 1.0. Adjusted r² is an adjusted value of crude r², normalizing the numerator and denominator by their degrees of freedom. AICc measures model performance and can be used to compare regression models. Sigma-squared is the estimate of variance for the residuals. Sigma-squared MLE represents the MLE of the variance of the residuals. The value of Effective Degrees of Freedom reflects a tradeoff between the variance of the fitted values and the bias.

RESULTS

Distribution of PM_2.5

Figure 1 shows a 20-year average PM_2.5 distribution map of India. The values ranged from 3.14 μg/m³ to 85.92 μg/m³, with a mean of 42.28 μg/m³ and a standard deviation of 15.89 μg/m³. The red area represents high PM2.5 values_, and the purple shade indicates low values. The western part of India (Rajasthan and Gujarat) had a high PM2.5 due to the sandy area of the Thar Desert and the Rann of Kutch. However, high-altitude regions and environmentally cooler regions showed low PM_2.5 in the Himalayan part of India (Kashmir, Himachal Pradesh, Uttarakhand, Sikkim, and a few parts of Arunachal Pradesh). A portion of Jharkhand (Jamshedpur, Bokaro, Rourkela) has a high concentration of PM_2.5 due to industrial clusters. The concentration of PM_2.5 in the coastal regions (Malabar and Coromandel coasts) is not high, even though a few coastal zones have a high concentration of industrial units. It is likely due to seasonal and diurnal variations in wind velocity and changes in wind direction between the ocean and coastal areas that pollutants are swept away, unlike in landlocked cities, where there is no scope for winds to wipe away pollutant particles for most of the year. This is evident from the observed moderate to high PM_2.5 concentrations in the Indo-Gangetic plains. The cities located in the Indo-Gangetic plains are densely populated and have a high concentration of industrial units, as well as an equally high density of vehicles. As India is a highly diversified country in terms of geography and physiography, there is a significant variation in ambient PM2.5 concentrations from region to region.

Figure 1.

Distribution of mean PM_2.5 in India over 2000–2019

Distribution of diabetes and hypertension

Diabetes and hypertension distribution maps were generated using their prevalence value per 1000 population. Maps (a) and (b) in Figure 2 show the distribution of hypertension and diabetes, respectively. The prevalence of diabetes was higher in the districts of Kerala, Karnataka, Tamil Nadu, and Telangana. One district in Gujarat also had a high prevalence rate of 441 per 1000 population. Districts in Rajasthan, Maharashtra, Uttar Pradesh, Punjab, Haryana, Delhi, Uttarakhand, Himachal Pradesh, Jammu, Bihar, West Bengal, and the northeast part of India have lower prevalence rates, that is, below 220 per 1000 population. Regarding hypertension, most districts show prevalence rates ranging between 251 to 380 per 1,000 population. The Jaipur district in Rajasthan and the Jhajjar district in Haryana showed higher prevalence rates (>500). The distribution of hypertension across the states was more homogeneous, unlike that of diabetes mellitus.

Figure 2.

(a) Distribution map of hypertension prevalence (b) Distribution map of diabetes prevalence

Hot spots and cold spots

Figure 3a shows diabetes hot spot and cold spot zones with varying confidence levels. The map shows that the southern part of India has hot spot zones with a magnitude of ±3 Gi Bin, which implies a 99% confidence level, indicating a high number of hot spots compared to other districts. The southern part of India, particularly Kerala and Tamil Nadu, is in the hot spot zones. Unlike hotspots, there were few cold spots on the map, with confidence levels of 90% and above visible in the eastern part of India, in the northern part of West Bengal, and the eastern part of Bihar. The rest of the areas in the country did not show any evidence of significant clustering.

Figure 3.

(a) Hotspot map of diabetes showing high and low clustered diabetes regions (b) Hotspot map of hypertension showing high and low clustered hypertension regions

Figure 3b shows the distribution of hotspot and coldspot clusters of hypertensions. The map shows that most of the hotspot zones are in the southern part of India, and there is only one state (Punjab) in the northern part with confidence levels of 90% and above. Only two districts in Kerala have a 99% confidence level for the hot spots. Unlike the diabetes hotspot map, the hypertension hotspot map has a good count of cold spot districts. All of Uttar Pradesh, Bihar, and parts of Gujarat have cold spot zones where the majority have a confidence level between 99% and 95% and the clustering of cold spots is more intense than that of hotspots. The point is noted that diabetes hotspot areas are more intense and have higher confidence levels than hypertension hotspot areas. However, cold spot regions in hypertension are more intense with higher confidence levels than in diabetes. So, a variation in clustering is shown in both diabetes and hypertension across India.

Association between PM_2.5 and hypertension and diabetes

Figure 4 shows the scatter plots and model summary of the linear regression, along with the related R-squared (R²) value, to represent the relationship between PM_2.5 and hypertension and diabetes. The scatter plot indicates a wide scattering of points. The adjusted r² for diabetes and hypertension were 0.034 and 0.023. Such small values indicate a weak predictive relationship in both cases.

Figure 4.

(a) Scatter plot and model summary of linear regression between PM_2.5 and diabetes (b) Scatter plot and model summary of linear regression between PM_2.5 and hypertension

GWR modeling was carried out to examine the relationship across geographic space. GWR was generated by fitting a single regression equation to every object within the dataset. Figure 5a shows the GWR map of hypertension with five classes of local R² values, indicating a positive association across all geographic regions. The r² values lie between 0.24 and 0.35, implying a weak strength of association. A few districts in Madhya Pradesh and Bihar had shown comparatively higher values, that is, 0.28–0.35. Most districts belonged to the first and second classes, where the r² value ranged from 0.024 to 0.15.

Figure 5.

Map of spatial association of PM_2.5 with (a) diabetes and (b) hypertension

Figure 5b shows the GWR map of diabetes, with a legend of five classes of local R² values. The first class contains values ranging from −0.056 to 0.0, indicating a negative correlation between the variables. The rest of the categories showed a positive correlation up to 0.34. A model diagnostic has been generated and attached to the map, giving a statistical description of the relationship. It displays the adjusted R-squared value, which is 0.21, and the remaining descriptions are presented in Table 1. Comparatively, a high positive r² value has been seen in Rajasthan, parts of Gujarat, Uttar Pradesh, Sikkim, and the upper part of West Bengal. A negative correlation was observed in certain areas of West Bengal, Bihar, and Madhya Pradesh.

Table 1.

Model diagnostics of GWR

Diagnostic Components	PM2.5 and Diabetes Mellitus	PM2.5 and Hypertension
R 2	0.3001	0.2167
Adj R2	0.2135	0.1350
Sigma-Squared	18073.6748	19055.1746
Sigma-Squared MLE	16090.1271	17259.8447
AICc	3967.1691	3980.4593
Effective Degrees of Freedom	277.7587	282.6042

AICc=Corrected akaike information criterion, GWR=Geographically weighted regression, MLE=maximum likelihood estimate

DISCUSSION

The present study was conducted on a nationally representative sample to study the spatial variation in PM_2.5, diabetes, and hypertension, and to assess the association between PM_2.5 and these two NCDs. We observed that spatial variation of diabetes was higher in districts in the southern states of Kerala, Telangana, Karnataka, and Tamil Nadu, consistent with the findings from other studies.[17,23,24,25] This was corroborated by hotspot zones, which indicated a spatial clustering of diabetes cases in the south. The spatial variation was less pronounced for hypertension, wherein the prevalence rate in most districts varied between 251 and 380/1000 population. While we could not observe any specific geospatial variation, geospatial analysis of NFHS-4 data observed the highest prevalence of hypertension in Telangana, Delhi, Sikkim, and Nagaland, followed by Himachal Pradesh, Arunachal Pradesh, and Assam.[25]

A national-level study conducted in China concluded that an interquartile range increase in PM_2.5 was significantly associated with increased diabetes prevalence.[8] Another national-level study in the US stated that diabetes prevalence increases with increasing PM_2.5 concentrations.[9] In Japan, Mihye Lee and Sachiko Ohde found that long-term exposure to PM2.5 was associated with an increased risk of diabetes.[10] The correlation of PM_2.5 concentration with diabetes and hypertension was found to be weak in the present study. Curto et al. found no evidence of the association between PM_2.5 and blood glucose levels among 5,056 adults in peri-urban villages of Andhra Pradesh.[15] However, Jacob et al. observed a 77.5% higher prevalence rate among adults exposed to high levels of suspended particulate matter in the residential areas of Chennai compared to those living in areas with lower PM_2.5 exposure.[12] There may be a link between PM2.5 and hypertension, as geographically weighted regression revealed a positive correlation across all geographic areas. Whatever the nature of the relationship between PM_2.5 and diabetes and hypertension, the strength of the relationship is too weak, which does not provide any conclusive evidence of association. Growing evidence suggests that there might be a link between air pollution and NCDs from studies conducted in countries like the US, China, and Japan.[7,8,9,10,11] However, neither the biological plausibility nor the causal relationship has been conclusively established. If we accept the magnitude of effect that pollutants are reported to have on these diseases based on the studies conducted in low pollutant exposure areas, then in a country like India, where the average exposure to PM_2.5 is 3 to 10 times higher, the magnitude of the problem should be extremely high, which is not the case. Thus, there is insufficient evidence to conclude that PM2.5 is a significant risk factor for hypertension and diabetes. In our study, we did not observe spatial patterns or statistical evidence that suggests a strong association between exposure to PM_2.5 and the spatial distribution of diabetes and hypertension. It is also important to understand that findings in arid regions[13] differ from those in coastal regions[15] of India. Hence, further studies are needed to establish any conclusive evidence, coupled with histopathological evidence, of the effect of PM_2.5 on target organs, namely the respiratory and endocrine systems.

We had several strengths in our study. As we understand the insidious onset of both the NCDs discussed here, prolonged exposure to the putative risk factors is expected to contribute to the pathogenesis. First, a large PM_2.5 dataset was utilized, which is continuous across geographic space and spans a 20-year time, lending strength to estimating the relationship between the dependent and independent variables. Second, this kind of study was conducted for the first time using nationally representative data. Third, the relationship was visualized across geographic space using a geostatistical approach; therefore, differentiation has been noted in the nature of the relationship, which is not possible with the classic statistical method.

Limitations of our study include the coarse resolution of PM2.5 data, which is based on a 1 km by 1 km grid. The prevalence of hypertension and diabetes mellitus was determined from a cross-sectional survey, which did not have enough strength in our analysis to establish the temporality of causation. Similarly, data on duration and severity of illness were also not available, thereby restricting our ability to determine a dose-response relationship. The analysis was not adjusted for the prevalence of other established risk factors among the cases, thus making it difficult to determine the attribution ratio.

Nonetheless, the study provides preliminary insight into the potential association between exposure to PM_2.5 and the occurrence of hypertension and diabetes mellitus.

CONCLUSION

Using a nationally representative dataset of a cross-sectional survey conducted for the prevalence of diabetes and hypertension, and satellite imagery based on 20-year average values of PM_2.5 in ambient air, we did not find any significant association between PM_2.5 and common NCDs. The global scientific community is divided over the potential role of PM2.5 as a risk factor, suggesting that we need longitudinal data on hypertension and diabetes mellitus in India, including the time of onset, duration of illness, and severity, to correlate with PM2.5 levels over the past 20 years. This should also be adjusted for other known risk factors in GWR analysis to determine the spatio-temporal evidence of association. Fine-resolution data, such as village or ward level, can yield precise results through higher-level analysis. In the future, if we conduct this kind of research, we can generate more evidence to establish the association, if any.

Ethical approval and/or Institutional Review Board (IRB) approval

The study was approved by the Ethics Committee of ICMR-NIIRNCD

Data availability

Monthly PM_2.5 data were taken from The Centre of Excellence for Research on Clean Air (CERCA), Indian Institute of Technology, Delhi (IITD), and data related to non-communicable diseases data were taken from the Indian Council of Medical Research (ICMR) - National Centre for Disease Informatics and Research (NCDIR). These data are not publicly available.

Conflicts of interest

There are no conflicts of interest.

Funding Statement

Nil.