University of Southern California Dissertations and Theses

Assessment of the mortality burden associated with ambient air pollution in rural and urban areas of India

(USC Thesis Other)

Assessment of the mortality burden associated with ambient air pollution in rural and urban areas of India

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Assessment of mortality associated with air pollution
1
Assessment of the mortality burden associated with ambient air pollution in rural and urban areas
of India

RASHI ARORA
Thesis Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF SCIENCE
(APPLIED BIOSTATISTICS AND EPIDEMIOLOGY)

May 2018

Assessment of mortality associated with air pollution
2
Acknowledgement
This thesis became a reality with the kind support and help of many individuals. I would
like to extend my sincere thanks to all of them.
Foremost, I want to offer this endeavor to our GOD Almighty for the wisdom he
bestowed upon me, the strength, peace of my mind and good health in order to finish this
research.
I would like to express my deepest gratitude and thanks to my thesis committee chair, Dr.
Meredith Franklin for imparting her knowledge, expertise in this field and guidance she
provided throughout this educational journey with lots of support and help.
I would like to thank the distinguished members of committee Dr. Kiros Berhane and
Dr. Rima Habre for thoughtful insights and review of my work.
I am highly indebted to the department of preventive medicine and Dr. Wendy Mack for
introducing me to the applied biostatistics master’s program at University of Southern California,
and my thanks and appreciation to people who have willingly helped me out with their abilities.
Lastly, I would like to express my gratitude towards my family for the encouragement
which helped me in completion of this paper. My beloved and supportive husband, Rohan
Monga who is always by my side when times I needed him the most and helped me a lot in
making this study.

Assessment of mortality associated with air pollution
3
Table of Contents
Acknowledgement ................................................................................................................. 2
List of Tables .......................................................................................................................... 5
List of Figures ........................................................................................................................ 6
ABSTRACT ............................................................................................................................ 7
1. Introduction........................................................................................................................ 9
2. Methodology and Inputs: ............................................................................................... 11
2.1 Study Domain ........................................................................................................................................................ 11
2.2 Air Quality Data ................................................................................................................................................... 12
2.3 Census Data ........................................................................................................................................................... 14
2.4 Mortality Data....................................................................................................................................................... 15
2.5 Statistical Methods .............................................................................................................................................. 15
2.5.1 Spatial Statistical Analysis of Air Pollution Data ......................................................................... 15
2.6 Health Effects Modeling ................................................................................................................................... 18
3. Results ............................................................................................................................... 20
3.1 Air Pollution Characteristics .......................................................................................................................... 20
3.2 Descriptive statistics of study population .................................................................................................. 32
3.3 Association Between Air Pollution and Mortality .................................................................................. 36
4. Discussion ......................................................................................................................... 42
Bibliography ......................................................................................................................... 46
Assessment of mortality associated with air pollution
4
Appendix 1 ............................................................................................................................ 50
Appendix 2 ............................................................................................................................ 59
Appendix 3 ............................................................................................................................ 60

Assessment of mortality associated with air pollution
5
List of Tables
Table 1 Dimensions of confounders obtained from Census data ................................................ 19
Table 2 Parameter estimates and model statistics for exponential semivariogram fit to annual
PM10 data .................................................................................................................................. 26
Table 3 Summary of year-wise mortality count (in Millions)..................................................... 33
Table 4 Descriptive characteristics of study population (in Millions) ......................................... 33
Table 5 Unadjusted Poisson regression model of PM10 exposure and all cause mortality ........... 38
Table 6 Poisson regression model of PM10 exposure and all cause mortality for the whole district
................................................................................................................................................. 39
Table 7 Poisson regression model of PM2.5 exposure and all-cause mortality for the whole district
................................................................................................................................................. 40
Table 8 Summary statistics of locations where pollution data was available .............................. 50
Table 9 Poisson regression model of PM10 exposure and all-cause mortality in Urban Population
................................................................................................................................................. 60
Table 10 Poisson regression model of PM10 exposure and all-cause mortality in Rural Population
................................................................................................................................................. 61
Table 11 Poisson regression model of PM10 exposure and all-cause mortality in Male Population
................................................................................................................................................. 62
Table 12 Poisson regression model of PM10 exposure and all-cause mortality in Female
Population................................................................................................................................. 63

Assessment of mortality associated with air pollution
6
List of Figures
Figure 1 Locations of ambient PM 10 derived from monitoring stations ...................................... 22
Figure 2 Distribution of daily PM10 concentrations by state ....................................................... 23
Figure 3 Annual semivariograms for PM 10 fitted with exponential, spherical and Gaussian
functions ................................................................................................................................... 25
Figure 4 Predicted PM 10 concentration plots using universal kriging ......................................... 27
Figure 5 District-wise distribution of predicted PM10 (µg/m3) Concentration ............................ 28
Figure 6 Validation of interpolation of annual PM 10 data (2015) ............................................... 30
Figure 7 District-wise distribution of predicted PM2.5 (µg/m3) Concentration ........................... 31
Figure 8 Distribution of state-wise mortality count for rural and urban combined ...................... 36
Figure 9 Distribution of state-wise mortality count for Urban population .................................. 37
Figure 10 Distribution of state-wise mortality count for rural population ................................... 37

Assessment of mortality associated with air pollution
7
ABSTRACT
Background
In India, more than a billion people are at risk from ambient particulate matter (PM)
concentrations exceeding World Health Organization air quality guidelines, posing serious threat
to health. Previous studies, conducted in the United States, have elucidated increases in all-cause
and specific-cause mortality associated with exposure to PM in metropolitan cities. Associations
between mortality and ambient particulate matter exposure are poorly characterized for India,
despite its serious air pollution problem.

Methods
In this study, using 2009-2015 data from the Central Pollution Control board (CPCB) and
Civil Registration System (CRS), we investigated the association between exposure to fine and
coarse PM air pollutants (PM2.5 and PM10, respectively) with all-cause mortality, targeting both
rural and urban centers in India. Spatial statistical methods including variograms and universal
kriging, were applied to annual averages of PM data from 222 districts to generate exposure
estimates at unobserved locations. The estimated exposures were linked to annual counts of all-
cause mortality in the respective districts, and the associations between annual average PM2.5 and
PM10 and mortality were determined using Poisson regression. Sex, occupation, literacy,
geographic location, and year were all considered as potential confounders and were included in
the model.

Assessment of mortality associated with air pollution
8

Results
There was 7.7% (95% confidence interval = 7.4% - 7.8%) increase in all-cause mortality
associated with a 10 µg/m
3
increase in annual averaged PM10 concentration across districts in
India. This association was larger in urban areas (16.7% [16.3% - 16.9%]) than in the rural areas
(4.6% [4.2% - 4.8%]). Similarly, with a 10 µg/m
3
increase in an annual average PM2.5
concentration, there was 23.2% (95% confidence interval = 23.0% - 23.3%) increase in all-cause
mortality. This association was adjusted by combination of covariates such as literacy,
occupational workers, household size, geographical location and year, which were also
aggregated at a district level.

Conclusion
This study shows that after controlling for certain demographic factors, geographic
location and year there is a statistical significant increase in mortality with increase in PM
exposures.

Assessment of mortality associated with air pollution
9
1. Introduction
Air pollution impacts human health, well-being and the environment. It has been widely
recognized as a major contributor to health problems in the past 50 years (Gurjar, et al., 2008).
Among different types of environmental pollutants, air pollution is reported to cause the greatest
damage to health and loss of welfare from environmental causes in Asian countries (Hughes, et
al., 1997). Located in the South-East Asia, India is experiencing rapid urban growth and is
affected by local and regional air pollution. Urbanization along with India’s industrial
development has contributed to high levels of atmospheric pollutants. Lack of services such as
proper management of transport, paved roads and primitive roads and unplanned distribution of
industries are all unable to keep pace with urban growth. All these in turn lead to an increase in
the pollution levels.
World Health Organization declared air pollution as the world’s single most important
environmental health risk factor and attributed around seven million deaths globally to air
pollution in 2012 (Jasarevic, 2014). Airborne pollutants are associated with coughing, burning
eyes, and breathing problems, and over time may even contribute to life threatening diseases
such as cancer. The elderly, young and those with cardiopulmonary diseases, or severe
bronchitis, are most vulnerable to air pollution exposure with children being at greater risk. A
study conducted in the US found that each 10 µg/m
3
elevation in fine particulate air pollution
(PM2.5, defined as particulate matter with aerodynamic diameter less than 2.5µm), was associated
with approximately 4%, 6%, and 8% increased risk of all cause, cardiopulmonary and lung
cancer mortality respectively (Pope III, et al., 2002). In addition, elevated PM2.5 exposure
increases risk for acute cerebrovascular strokes, aggravation of heart and lung disease and pre-
mature mortality (Franklin, et al., 2007, 2008), (Zanobetti & Schwartz, 2009).
Assessment of mortality associated with air pollution
10
Air pollution is a national, pan-India problem that is not limited to urban centers and
metropolitan cities; it affects rural Indians as well. Data from the central pollution control board
(CPCB) reveals that 77% of Indian urban agglomerations exceeded India’s National Ambient Air
Quality Standard (NAAQS) for respirable suspended PM (PM2.5 and PM10) in 2010 (CPCB,
2010). Estimates from WHO suggest that out of 20 cities in the world, 13 cities with the worst
PM2.5 are in India, including Delhi, the worst ranked city. It also estimated that 84,000 deaths are
directly attributable to outdoor air pollution in Indian cities (WHO, 2014). Recently, average
PM2.5 levels on certain days have climbed to more than 900 µg/m
3
in Delhi, which are considered
hazardous to breathe, according to data provided by the Delhi Pollution Control Committee. In
2015, over a million deaths could be attributed to air pollution in India (Cohen, et al., 2017).
For this study, air pollution data were collected from 626 monitoring stations located over
274 distinct districts (counties) that measured PM in 2009 through 2015. Statistical methods
were used to spatially interpolate PM10, creating exposure surfaces over the country. Due to the
limited availability of PM2.5 monitoring data, we used the relationship between PM10 and PM2.5
from 25 co-located monitors in 2016-2017 to generate PM2.5 surfaces from the PM10 surfaces.
The predicted exposures were then combined with district-wise mortality data along with
demographic factors obtained from India census. With these combined data, we aimed to identify
the association between each of the PM pollutants and all-cause mortality with and without
adjustment for district-level population characteristics.
Findings from our analysis are the result of comprehensive, district by district analysis of
types of air pollutants, mainly PM 10 and PM2.5 and what impact they have had on health in India.
It is our hope that information provided through this study would be used by government
Assessment of mortality associated with air pollution
11
agencies to set policies and potential intervention strategies to control pollution, thereby paving a
cleaner and healthier path for India’s future.

2. Methodology and Inputs:
2.1 Study Domain
India is the seventh largest country in the world, covering a total area of 3,287,263 square
kilometers with the second largest population of 1.324 billion. India is situated north of the
equator between 8°44' to 37°6' north latitude and 68°7' to 97°25' east longitude. The country is
divided into 29 states and 7 union territories; the study domain presented in Figure 1 covered 26
states and 3 union territories. In India, district is defined as administrative division of an Indian
state or territory. Each district includes one or two cities (or large towns), a few smaller towns
and dozens of villages. They generally form the tier of local government immediately below that
of India’s subnational states and territories. In India districts are similar to counties in the US. Of
the total 707 districts, this study covered 662 districts for analysis.
This study focused on all-cause mortality data from 2009-2015 and pollution data from
2009 to 2017. Pollution data were not available for districts of some states like Tripura, Sikkim,
and Manipur. Jammu and Kashmir were not considered in the analysis due to non-availability of
both exposure and outcome data.
District-level demographic data were obtained from the India census, which is conducted
every 10 years (the last one being in 2011). Features from the 2011 census data were used as
covariates in the health effects assessments.

Assessment of mortality associated with air pollution
12
2.2 Air Quality Data
The national ambient monitoring program reports 24-h averages of key air pollutants 2-3
times per week at 626 monitoring stations in 225 cities (Figure 1). This program is managed by
the Central Pollution Control Board (CPCB). Access to the monitoring data is limited, and only a
limited number of cities operate continuous monitoring stations, measuring the full array of
criteria pollutants (PM2.5, PM10, Nitrogen Dioxide (NO2), Sulphur Dioxide (SO2)). This data is
not available in an easily consumable manner. Web crawlers and parsers were written to acquire
this data from CPCB websites. These programs were designed to accept a list of URLs
corresponding to monitoring stations of interest and convert the HTML files into data formats
that can be utilized for statistical analysis. PM10 has been more routinely measured than PM2.5.
Summary statistics for the PM10 monitors, by district area (in km
2
) was estimated from district
shapefiles obtained from GADM and population density (per km
2
) for 2009-2015 is presented in
Assessment of mortality associated with air pollution
13
Appendix 1.
Monitoring stations are not evenly distributed over India, they are primarily located in
areas of high population or at certain industrial establishments. Remote areas, such as
mountainous regions or the Thar desert have no monitoring stations.
There are varying factors that contribute to air pollution in different regions of India. The
Indo-Gangetic plain has the largest number of brick kilns with old and inefficient combustion
technology, using a mix of biomass and coal for combustion needs (Maithel, et al., 2012). The
states of Bihar, West Bengal, Jharkhand, Orissa, and Chhattisgarh harbor the largest coal mines
in the country, and a cluster of power plants also exist in the states of Punjab, Haryana, Delhi and
Uttar Pradesh, making the north and north-eastern belt the most polluted part of the country. The
cities in North are landlocked and are affected by meteorological conditions that lead to higher
levels of air pollution. High temperature in summers and shortage of power, leads people to use
diesel generators; cold and humid winters lead to fogs and smogs. This is in stark contrast to
conditions in Southern cities that have land-sea breezes that result in dilution; making pollution
less of a problem (Guttikunda & Jawahar, 2014).
Few researchers have produced air pollution surfaces for India, so kriging was applied to
get predicted PM10 maps (2009-2015) and this was one of our study objectives (methods
described in Section 2.5.1.2). We applied kriging and spatial interpolation to get predicted annual
PM10 maps (2009-2015) and estimate exposures for the epidemiological assessment.
Historically, PM2.5 has not been tracked in India. New monitoring stations are being setup
by CPCB and are starting to report values since 2016. Since PM2.5 measurements were not
available for 2009-2015, we trained a univariate generalized additive model (Wood, 2006) using
co-located monitors in 2016-2017 to generate a means of predicting PM2.5 from PM10. This
Assessment of mortality associated with air pollution
14
prediction model was then applied to the interpolated PM10 concentrations obtained via kriging
on 2009-2015 PM10 to infer PM2.5 concentrations.

2.3 Census Data
Our study design incorporated data on potential confounders such as gender, occupation,
literacy and household size. These data were available from the 2011 census for the districts
included in our analysis (Census, 2011).
The census defined occupation as work, which included participation in any
economically productive activity with or without compensation, wages or profit. Work was
defined not only as actual work, but also effective supervision and direction of work. It included
part time help or unpaid work on a farm, family enterprise, or in any economic activity. All
persons (irrespective of age and sex) who participated in any economically productive activity
for any length of time during the reference period were defined as workers.
These workers were further classified as main workers (who worked for more than 6
months in the reference period), marginal workers (who worked for less than 6 months in the
reference period) and non-workers (who did not work at all, such as, students, homemakers,
pensioners, beggars etc.). The main and marginal workers were further categorized as cultivators,
agricultural laborers, household industry workers (people running small business in their home)
and other workers (government servants, municipal employees, teachers, factory workers etc.).
Literacy was categorized as literate and illiterate and gender as male and female. A
person aged 7 years and above who can both read and write with understanding in any language
was taken as literate. Households are defined as buildings that are a separate unit because they
have a separate main entrance. They can be of 3 types, Normal, Institutional, and Houseless.
Assessment of mortality associated with air pollution
15
Classifications are designed to count groups of people that are linked together in a familial or
dependent relationship. In our study, we used average number of people per household in the
district as covariate.

2.4 Mortality Data
Annual mortality records at the district level, the smallest spatial resolution available,
were managed and acquired from the Civil Registration System (CRS). This data was available
in PDF format; therefore, special parsers and manual extraction was used to transcribe and
convert it into analyzable data frames. These data provided non-confidential and aggregated
information on decedents including gender, state of death, and district of death, which was
further classified into rural and urban areas. CRS is widely used in India as it is mandated by the
government, covering ~84% of all births and 70% of all deaths (CRS, 2011). Information on
cause of death was not available, so the mortality counts include all causes of death.

2.5 Statistical Methods
2.5.1 Spatial Statistical Analysis of Air Pollution Data

2.5.1.1 Semivariogram Analysis
After exploratory analysis of exposure data, we calculated annual average PM10
concentrations at each monitoring station location. There were some stations with missing
concentrations. Using PM10 concentrations from the years where they were available, we used
second order splines to interpolate and filled these values. Interpolation was performed when at
Assessment of mortality associated with air pollution
16
least 4 years’ worth of average PM10 data was available, otherwise the record was left untouched
and subsequently removed during model training.
Empirical semivariograms were applied to the spatially referenced monitoring sites to
characterize the spatial process in PM10 concentrations with respect to distance and direction.
It is an attribute of spatial data based on the fact that observations close together tend to
be more alike than observations farther apart. In the geostatistical field, the shape of the
semivariogram is described by the nugget, which refers to a nonzero semivariogram near the
origin, the range, which refers to the distance after which the spatial process is no longer
correlated, and the sill, which refers to the semivariance at which the range is attained. These
parameters were estimated for all years (2009-2015).

Binned empirical semivariograms were constructed for annual averages of PM10
concentrations (µg/m
3
) measured at the monitoring stations. Semivariance 𝛾(ℎ) is defined by:
𝛾(ℎ)=
1
2|𝑁(ℎ)|
* [𝑍(𝑠
.
)−𝑍(𝑠
0
)]
2
3(4)∈.,0

Where, 𝑍(𝑠
.
)−𝑍(𝑠
0
) is the difference in PM10 concentration Z at locations 𝑠
.
and 𝑠
0

(identified by latitude and longitude), h is the distance lag, and N(h) is the number of pairs for a
particular distance lag. Visual inspection of the empirical semivariogram, plotted as the distance
lag (h) versus semivariance enabled us to examine the spatial variability between all pairwise
locations, and explore which theoretical semivariogram function might best fit data. Similar
analysis was also done using box plots on semivariance at distances (h) across entire distance
range to visualize the distribution of concentrations in the different bins.
Theoretical semivariogram functions (Gaussian, spherical, and exponential) were fitted to
the data using Maximum Likelihood (ML). The estimated spatial parameters, (i.e. nugget, sill,
Assessment of mortality associated with air pollution
17
range), Akaike information criterion (AIC) and Bayesian information criterion (BIC) were
compared. The model with smallest AIC was considered for further analysis.

2.5.1.2 Spatial Interpolation
Various methods of spatial interpolation are available to compute values on a continuous
surface. Kriging aims to produce a best linear unbiased estimator (BLUE) for an unobserved
location (Cressie, 1993). It is linear since the estimated values are weighted linear combinations
of the available data, unbiased because the expected mean of the error is 0, and it aims to
minimize the variance of error. When using universal kriging, the trend is subtracted, and kriging
is then performed with the residuals.

For this study, we assumed that our spatial process of PM10, Z(s), is an intrinsically
stationary Gaussian random field (i.e. has a constant mean and can be defined by a
semivariogram that only depends on the distances separating two locations 𝑠
.
and 𝑠
0
).
The general equation for universal kriging, is:
Z(s)= µ(s)+ ϵ(s)
where, Z(s) is spatial process at observed location s = (x, y), µ(s) the deterministic
function describing the trend component (drift) of Z(s), and ϵ(s) denotes the stochastic locally
varying but spatially auto correlated and intrinsically stationary error, defined by a spatial
variance-covariance matrix parameterized with the nugget, sill, and range described by the
theoretical semivariogram above. The trend component µ(s) assumed in this study is linear in the
coordinates, namely µ(s) = β0+β1x+β2y.
Assessment of mortality associated with air pollution
18
We applied universal kriging with a linear trend to interpolate PM10 from the 626
monitoring sites, generating spatially detailed air pollution surfaces that provide concentrations
at unobserved locations.
To validate the model for prediction evaluation purposes, leave one site out cross
validation (LOSOCV) was investigated. A training set was created on N-1 samples and tested
with the remaining 1 sample. This process was repeated N times, leaving a new sample out for
testing at each iteration. Using the model generated at each iteration, prediction on the test
observation was stored and coefficient of determination was computed between observed and
predicted PM10.

2.5.1.3 Spatial Linkage
Administrative boundaries of India were obtained from Global Administrative Areas
Database (GADM, 2009). From the district polygons, we computed their centroids and using the
spatial models from above, PM10 was predicted for each district at its centroid.
Census and CRS mortality data was linked to PM 10 data using the district and state name.
Some districts that have been created recently are therefore, not available in previous analysis
but this manner of spatial linkage avoids any lack of regional coverage.

2.6 Health Effects Modeling
Poisson regression was used to determine the association between district-specific counts
of all-cause mortality and average annual air pollution concentrations PM10 and PM2.5, estimated
for each district by the kriging methods described above. We controlled for confounding effects
of gender, literacy, occupation and number of children in the Poisson regression model.
Assessment of mortality associated with air pollution
19
The Poisson regression model for the counts of all-cause mortality Yi at each districts i
(also stratified by rural and urban areas) with adjustment for confounders Xi had the form:
Log[E(Yi|Xi)] = 𝛽
8
+ 𝛽𝑋
.

We controlled for confounding effects of gender, literacy, occupation and number of
children aggregated at a district level our model. Each confounder variables manifested in 2
dimensions, gender (male/female) and geographical (rural/urban/total). For example, total
number of literate people was converted into 6 proportional variables as follows (Table 1) and so
on for all other confounding factors.

Table 1 Dimensions of confounders obtained from Census data

“Overall” is the proportion of total district population that exhibit the trait (literate, child,
occupation, etc.) whereas “rural” and “urban” are proportions of rural and urban populations of
the district, respectively.
Separate models were developed to test the association between PM and mortality using
the entire districts, and for their corresponding rural and urban strata. Additional confounders
including latitude, longitude, and year were also evaluated.
Resultant effect estimates were expressed as a percent increase in mortality with a 10-
𝜇𝑔/𝑚
@
increase in PM10 and PM2.5 mass concentrations after adjusting for potential confounders
in the model; 95% confidence intervals(CIs) were also computed.
All analyses were conducted in R and ArcGIS 10.4.1.

Entire District Rural Urban
All
% literate overall % literate in rural areas % literate in urban areas
Female
% literate females overall
% literate females in rural
areas
% literate females in urban
areas
Assessment of mortality associated with air pollution
20

3. Results
3.1 Air Pollution Characteristics
Summary statistics of PM 10 for each of the 225 districts, including population density,
mean PM10 and number of monitoring stations are presented in
Assessment of mortality associated with air pollution
21
Appendix 1. The 10 cities (State) with highest average PM10 concentration were Raipur
(CT), Ghaziabad (UP), Singhbhum (JH), Allahabad (UP), Bareilly (UP), Delhi (DL), Gwalior
(MP), Mathura (UP), and Alwar (RJ). Out of these 10 cities, 7 are neighboring cities of Delhi,
located within a range of 500 km and the states they are in are known for their coal mines and
have some of the largest power plants in the country (Guttikunda & Jawahar, 2014).
The maximum annual mean PM10 concentration was 276.4 µg/m
3
in Raipur, Chattisgarh
and minimum annual mean PM10 concentration was 23.1 µg/m
3
in Pathanamthitta, Kerala.
Figure 1 shows the distribution of air quality monitoring stations over the country from
2009-2015 colored by state. It is noted that the monitoring stations are approximately distributed
evenly among each state, but appear to be collocated in population centers, with larger cities
having more monitoring stations compared to rural areas. Metropolitan areas such as Kolkata,
Ahmedabad, Chennai, Hyderabad, Delhi, Kanpur and Mumbai have more than 9 monitoring
stations each. States with lower population density such as Madhya Pradesh, Bihar, and
Rajasthan have poor coverage of pollution monitoring stations in terms of area.

Assessment of mortality associated with air pollution
22

Figure 1 Locations of ambient PM 10 monitoring stations in India
Further, Figure 2 represents the overall spread of daily PM10 concentration obtained by
monitoring stations over the study period. Box plots of each state are presented with extreme
values labeled by the city in which they occurred. E.g. on March 19, 2011 PM10 reached 1,288
µg/m
3
in the city of Alwar, RJ.
Assessment of mortality associated with air pollution
23

Figure 2 Distribution of daily PM 10 concentrations by state

Empirical semivariograms were explored (Figure 3) and Gaussian, exponential, and
spherical theoretical semivariogram models were fitted by maximum likelihood. Examination of
the plots suggested that PM10 monitoring locations, within 500 km exhibited strong spatial
correlation, but beyond that distance the data were no longer correlated. Although, 500 km is a
large distance when considering pollution concentrations but this makes sense in India’s case as
India is continuously populated, with very few regions lacking human habitation, allowing for
pollutants to be generated and carried over these distances.
After comparing AIC and BIC of Spherical and Exponential models we found that the
best fitting theoretical semivariogram to our data was the exponential as it had the lowest AIC
and BIC values and was therefore used for universal kriging. Table 2 displays the parameter
estimates and performance metrics of the final selected model for each of the study years. The
semivariograms showed a maximum distance of 3,000 km, however, we truncated this distance
Assessment of mortality associated with air pollution
24
to a more reasonable range of 500 km as this is where the semivariance leveled off. Using the
best-fitted semivariogram models for a particular year, trend surfaces were calculated for the
corresponding analysis year.

Assessment of mortality associated with air pollution
25
Figure 3 Annual semivariograms for PM 10 fitted with exponential and spherical functions
2015 2014

2013 2012

2011 2010

2009

Assessment of mortality associated with air pollution
26
Kriging, which resulted in the least average variance between actual and estimated value
at 626 monitoring sites, was employed to estimate air pollution on a 300x300 grid over India.
The summary estimates of PM10 concentrations, their coefficients and variogram parameters are
presented in Table 2. Figure 4 shows the spatial distribution of interpolated PM10 for each study
year using the corresponding model. The assumptions of normality of annual average PM 10 data
were checked and were found to hold true.

Table 2 Parameter estimates and model statistics for exponential semivariogram fit to annual PM 10 data

Coefficients Variogram Parameters Metrics
Year
Intercept B1 B2 Nugget Sill Range AIC BIC RMSE R2
2015
32.61 -0.01 0.03 1126.13 1211.42 158.69 5896.62 5922.75 29.73 0.71
2014
35.17 -0.02 0.04 1299.59 1180.49 199.80 5091.77 5116.96 32.51 0.67
2013
21.40 -0.01 0.04 1132.51 1396.87 148.65 5267.01 5292.43 29.41 0.76
2012
20.68 -0.02 0.05 802.09 2048.99 53.71 4581.52 4606.08 21.76 0.87
2011
32.21 -0.02 0.04 750.34 2166.25 44.93 4999.70 5024.78 20.39 0.88
2010
20.65 -0.02 0.04 988.91 2003.90 64.34 4640.39 4664.97 25.20 0.83
2009
25.88 -0.02 0.04 1332.25 1533.06 37.73 4575.80 4600.22 30.00 0.76

Assessment of mortality associated with air pollution
27
Figure 4 Predicted annual PM 10 concentration (µg/m³) plots using universal kriging
2015 2014

2013 2012

2011 2010

2009

Assessment of mortality associated with air pollution
28

Figure 5 District-wise distribution of predicted annual PM 10 concentrations (µg/m
3
)
Assessment of mortality associated with air pollution
29
Figure 5 depicts the district-level PM10 concentrations over the study domain; as
expected, concentrations are higher in the northern regions and lower around the coasts.
Manufacturing and mining towns also show high concentrations whereas, low population
density, rural and remote regions have low predicted PM10 values.
Finally, we used the 7 spatial models, one for each of the study years (Table 2), in the
universal kriging model to predict PM 10 values at the centroid of the 626 districts of India.
Using leave one site out cross validation on 2015 data, we found that adjusted R
2
for
predicted and observed PM10 was 0.524 (p<0.001). Figure 6 shows that there is some scattering
around the 1:1 line. Most of the values fall within a range 23 µg/m
3
to 300 µg/m
3
with some
outliers in the observed set reaching values of more than 500 µg/m
3
.
Further to compute a relationship between PM10 and PM2.5 concentrations, we applied a
generalized additive (GAM) model to the daily data from co-located PM10 and PM2.5 sites in
2016-2017. The fitted model had an estimated smoothing parameter for PM10 of 0.33 (std. error
0.007, p < 0.001) and adjusted R
2
of 0.32. Using the predicted PM2.5 values we generated an
exposure surface as show in Figure 7.

Assessment of mortality associated with air pollution
30

Figure 6 Validation of interpolation of annual PM 10 data (2015)

Assessment of mortality associated with air pollution
31

Figure 7 District-wise distribution of predicted annual PM 2.5 (µg/m
3
) Concentration
Assessment of mortality associated with air pollution
32
3.2 Descriptive statistics of study population
India has a total population of approximately 1.324 billion as per 2011 census. Table 3
shows year-wise mortality count for India. There were ~23 million rural, ~17 million urban and
~41 million total deaths recorded by Civil Registration System. According to census data, sex
ratio, i.e. proportion of males to females, was very similar over the different demographic
dimensions that we considered.
We converted our confounders into proportions by dividing the total value by the
corresponding total population. For example, the proportion of female children was computed by
taking the total number of female children in a district and dividing the total number of women
in the district. We further subdivided these occupational and demographic confounders into rural
and urban areas. Summary statistics of confounders by district are presented in Table 4. Since
this table contains summary statistics for the entire country, confounders were computed as
proportion of the total corresponding populations of India even though in the model they were
calculated as a proportion of the district or state as the case may be.
Workers were subdivided into agricultural laborers, cultivators, household industries and
other workers as defined in Section 2.3 Census Data. Among these categories, the proportion of
laborers and cultivators was found to be high in rural areas. The data also showed that women
constituted the majority of non-working population both in rural and urban areas.
We did not find any missing data of the potential confounders but there were some
districts that were created after 2011 census. For those cities we did not have any information
and thus were not considered for analysis.
Assessment of mortality associated with air pollution
33
Some interesting insights from the overall statistics are that literacy rate was quite low in
rural female population (49.62%) compare to urban females (70.17%). Moreover, the working
female population was quite low both in rural (30.2%) and urban areas (15.44%).

Table 3 Summary of year-wise mortality count (in Millions)

Rural Urban Total
Year
Male Female Total Male Female Total Male Female Total
2009
1.92 1.54 3.46 1.34 0.89 2.22 3.25 2.42 5.68
2010
1.90 1.45 3.36 1.39 0.92 2.31 3.29 2.37 5.67
2011
1.86 1.43 3.30 1.44 0.98 2.43 3.31 2.42 5.73
2012
1.65 1.25 2.90 1.36 0.92 2.28 3.38 2.46 5.84
2013
1.98 1.45 3.43 1.57 1.08 2.66 3.55 2.53 6.09
2014
1.99 1.45 3.44 1.61 1.07 2.68 3.60 2.51 6.12
2015
1.89 1.37 3.26 1.54 1.03 2.57 3.68 2.56 6.25
Total
13.19 9.94 23.14 10.26 6.90 17.15 24.07 17.28 41.37

Table 4 Descriptive characteristics of study population (in Millions)

Population Type Total Male Female Total % Male % Female %

Total 1,211 623 587 51.47% 48.53%

Urban 377 195 182 51.84% 48.16%

Rural 833 428 406 51.31% 48.69%

Household size 249 4.85

Urban 81 4.66

Rural 169 4.94

Total Children
(0-6y)
164 86 79 13.59% 13.76% 13.40%

Urban 43 23 21 11.45% 11.60% 11.30%

Rural 121 63 58 14.55% 14.75% 14.35%
Assessment of mortality associated with air pollution
34

Literate
763 435 329 63.07% 69.76% 55.97%

Urban 281 153 127 74.47% 78.47% 70.17%

Rural 483 281 201 57.91% 65.78% 49.62%

Total Worker
482 332 150 39.79% 53.26% 25.51%

Urban 133 105 28 35.31% 53.76% 15.44%

Rural 349 227 122 41.83% 53.03% 30.02%
Main
Working
362 273 89 29.94% 43.84% 15.20%
Urban 117 95 22 30.95% 48.65% 11.88%
Rural 246 178 68 29.49% 41.63% 16.69%
Cultivator
96 73 23 7.92% 11.72% 3.89%
Urban 3 3 1 0.82% 1.30% 0.31%
Rural 93 70 22 11.13% 16.48% 5.49%
Agricultural
Laborers
86 55 31 7.12% 8.87% 5.26%
Urban 5 4 2 1.38% 1.85% 0.88%
Rural 81 52 29 9.71% 12.08% 7.22%
Household
Industries
12 8 5 1.02% 1.21% 0.82%
Urban 5 3 2 1.35% 1.72% 0.95%
Rural 7 4 3 0.87% 0.98% 0.76%
Other Workers
168 137 31 13.89% 22.04% 5.24%
Urban 103 86 18 27.39% 43.78% 9.76%
Rural 65 52 13 7.78% 12.10% 3.22%
Marginal
Working
119 59 61 9.85% 9.42% 10.31%
Urban 16 10 6 4.36% 5.11% 3.56%
Rural 103 49 54 12.34% 11.39% 13.34%
Cultivator
23 10 13 1.89% 1.55% 2.24%
Urban 1 0 0 0.16% 0.16% 0.17%
Rural 22 9 13 2.67% 2.19% 3.17%
Agricultural
Laborers
58 27 31 4.80% 4.41% 5.22%
Urban 2 1 1 0.56% 0.61% 0.51%
Rural 56 26 30 6.72% 6.15% 7.33%
Assessment of mortality associated with air pollution
35
Household
Industries
6 2 4 0.50% 0.36% 0.64%
Urban 1 1 1 0.35% 0.28% 0.42%
Rural 5 2 3 0.56% 0.40% 0.74%
Other Workers
32 19 13 2.67% 3.10% 2.21%
Urban 12 8 4 3.29% 4.06% 2.46%
Rural 20 11 9 2.39% 2.66% 2.10%

Non-Working
729 291 438 60.21% 46.74% 74.49%

Urban 244 90 154 64.69% 46.24% 84.56%

Rural 485 201 284 58.17% 46.97% 69.98%

Assessment of mortality associated with air pollution
36
3.3 Association Between Air Pollution and Mortality
The total number of deaths over all districts within states having at least 10,000 deaths
during the 2009-2015 study period are shown in Figure 8. The extreme points observed are the
capital region of states, which tend to have the largest population. Figure 9 and Figure 10 show
the same type of distribution for rural and urban areas respectively.

Figure 8 Distribution of state-wise mortality count for rural and urban combined
Assessment of mortality associated with air pollution
37

Figure 9 Distribution of state-wise mortality count for Urban population

Figure 10 Distribution of state-wise mortality count for rural population

Assessment of mortality associated with air pollution
38
Using these mortality data in Poisson regression, we tested its association with district-
level PM2.5 and PM 10 exposures with the null hypothesis of no association between PM air
pollution and all-cause mortality. We found that all measures of particulate matter and all-cause
mortality were statistically significant. The bivariate association of PM10 and total (rural and
urban) mortality was statistically significant (p<0.001), where the all-cause mortality increased
by 0.3% (95% CI: 0.37% - 0.39%) and 1.26% (95% CI: 1.23% - 1.28%) per 10 µg/m
3
increase in
annual PM10 and PM2.5 respectively. This association was also tested separately on rural and
urban counts of mortality and was found to be statistically significant as depicted in Table 5.
Table 5 Unadjusted Poisson regression model of PM 10 exposure and all cause mortality

Estimate 95% CI z value Pr(>|z|)
Total Mortality
(Intercept)
9.181 9.180- 9.182 18257.19 < 0.001
PM 10
0.00038 0.00037 - 0.00039 89.12 < 0.001
Urban Mortality
(Intercept)
8.21 8.21056 - 8.216 10393.05 < 0.001
PM 10
0.00070 0.00069 - 0.00072 105.83 < 0.001
Rural Mortality
(Intercept)
8.74 8.73 - 8.75 13082.54 < 0.001
PM 10
0.00143 0.00142 - 0.00145 248.88 < 0.001

PM 2.5
(Intercept)
9.219 9.218 - 9.221 11109.40 < 0.001
PM 2.5
0.00126 0.00123 - 0.00129 98.092 < 0.001

Proportion of district with respect to the population of state, year, latitude, longitude,
proportion of female population, household size, proportion of children of age 0-6 years,
proportion of literate people, proportion of all workers categorized as cultivators, agricultural
Assessment of mortality associated with air pollution
39
workers, industrial workers, and other workers were considered as potential confounders in the
study. Pearson correlation among potential confounders showed that there were no two variables
strongly correlated with each other as presented in Appendix 2. These confounders
independently altered the main effect of interest by more than 10% and thus were included in the
final Poisson regression model.
Table 6 represents the positive association between PM10 exposures with the total all-
cause mortality after adjusting for potential covariates considered in the analysis. We rejected the
null hypothesis and concluded that PM 10 exposure was statistically significantly associated with
all-cause mortality after adjusting for covariates (p<0.01). After adjustment, the number of all-
cause mortality increased by 7.7% (95% CI: 7.4%-7.8%) per 10 µg/m
3
increase in annual PM10.
Similarly, Table 7 represents, the number of all-cause mortality increased by 23.21% (95% CI:
23% - 23.3%) per 10 µg/m
3
increase in annual PM2.5 after controlling for covariates (p<0.01).

Table 6 Poisson regression model of PM 10 exposure and all cause mortality for the whole district

Estimate 95% CI z value Pr(>|z|)
(Intercept)
-13.3 -13.5 - -12.9 -83.7 <0.001
PM 10
0.00773 0.00772 - 0.00775 1082.1 <0.001
Prop. district population w.r.t state
-2.203 -2.212 - -2.195 -526.6 <0.001
Year
0.0133 0.0132 - 0.0135 170.5 <0.001
Latitude
-0.0474 -0.0475 - -0.0473 -898.0 <0.001
Longitude
-0.0248 -0.0249 - -0.0247 -647.54 <0.001
Prop. of female w.r.t district
-0.200 -0.227 - -0.173 -14.35 <0.001
Number of people / Household
0.040 0.039 - 0.041 96.46 <0.001
Prop. of children w.r.t district
-10.74 -10.76 - -10.71 -817.3 <0.001
Prop. of literate w.r.t district
-0.829 -0.835- -0.823 -259.6 <0.001
Assessment of mortality associated with air pollution
40
Prop. of main cultivators w.r.t district
-3.77 -3.78 - -3.76 -730.02 <0.001
Prop. of main agricultural workers w.r.t district
-0.887 -0.899 - -0.875 -143.8 <0.001
Prop. of main industrial workers w.r.t district
3.65 3.61 - 3.69 177.38 <0.001
Prop. of main other workers w.r.t district
1.71 1.70 - 1.72 330.52 <0.001
Prop. of marginal cultivators w.r.t district
-5.47 -5.49 - -5.45 -534.01 <0.001
Prop. of marginal agricultural workers w.r.t district
0.081 0.067- 0.095 11.180 <0.001
Prop. of marginal industrial workers w.r.t district
2.73 2.61 - 2.85 45.76 <0.001
Prop. of marginal other workers w.r.t district
-7.95 -7.99 - -7.91 -423.20 <0.001

Table 7 Poisson regression model of PM 2.5 exposure and all-cause mortality for the whole district

Estimate 95% CI z value Pr(>|z|)
(Intercept)
-13.62 -13.93 - -13.31 -86.03 <0.001
PM 2.5
0.0231 0.0232 - 0.0233 1078.72 <0.001
Prop. district population w.r.t state
-2.21 -2.22 - -2.20 -529.71 <0.001
Year
0.0133 0.0132 - 0.0135 169.48 <0.001
Latitude
-0.0476 -0.0477 - -0.0475 -901.45 <0.001
Longitude
-0.0249 -0.0251 - -0.0249 -652.01 <0.001
Prop. of female w.r.t district
-0.314 -0.341- -0.287 -22.50 <0.001
Number of people / Household
0.038 0.0378 - 0.0395 91.63 <0.001
Prop. of children w.r.t district
-10.71 -10.74 - -10.68 -815.2 <0.001
Prop. of literate w.r.t district
-0.832 -0.838 - -0.825 -260.45 <0.001
Prop. of main cultivators w.r.t district
-3.77 -3.78 - -3.76 -728.39 <0.001
Prop. of main agricultural workers w.r.t district
-0.87 -0.88- -0.85 -141.11 <0.001
Prop. of main industrial workers w.r.t district
3.41 3.37 - 3.46 164.52 <0.001
Prop. of main other workers w.r.t district
1.718 1.708 - 1.728 330.75 <0.001
Prop. of marginal cultivators w.r.t district
-5.47 -5.49 - -5.45 -534.19 <0.001
Prop. of marginal agricultural workers w.r.t district
0.029 0.014 - 0.043 4.00 <0.001
Prop. of marginal industrial workers w.r.t district
3.98 3.87 - 4.10 65.85 <0.001
Assessment of mortality associated with air pollution
41
Prop. of marginal other workers w.r.t district
-7.83 -7.87- -7.80 -416.83 <0.001

Each for PM10 and PM2.5 all-cause mortality decreased by 47.4% (95% CI: 47%-48%)
with one degree (111 km) increase in latitude after adjusting for other independent variables
(Table 6 and Table 7) in both the PM 10 and PM2.5 models. This is evident because Northern India
has more population and therefore higher mortality rates. Further, with 10% increase in literacy
rate the mortality rate decreased by 0.56% (p <0.01) after adjusting for other covariates in the
model.
For workers exposed to PM10, expected count of all-cause mortality for industrial
workers (main and marginal) were 38.7 times and 15.43 times higher, respectively, compared to
non-workers. Similarly, for PM2.5, the expected count was 30.26 times and 50.35 times higher
compared to non-workers respectively (p<0.001). Marginal or part-time workers might be
economically disadvantaged and therefore are suffering higher mortality rate from exposure to
PM2.5.
Similarly, rural population and urban population was also statistically significantly
associated with all-cause mortality count after adjusting for all potential confounders as
presented in Table 9 and Table 10. We observed that, in urban population and in rural
population, after adjustment, the count of all-cause mortality increased by 16.7% and 4% per 10
𝜇𝑔/𝑚
@
increase in PM10 respectively.
We observed similar effects of PM10 on mortality count for both males and females in
stratified models as shown in Table 11 and Table 12.

Assessment of mortality associated with air pollution
42
4. Discussion
We examined the association between airborne particulates and mortality in a large
population-based study of 626 districts across India, including over 41 million deaths.
Census based confounders that we have chosen in the study represented more than just
occupation or geographical location of the subjects. Rural and urban areas have vastly different
access to healthcare facilities, both in terms of distance and cost. Similarly, different
occupational categories can be proxies for different economic strata, with “other” being mostly
white-collar workers and “agricultural” and “industrial” being lower wage blue-collar workers.
Mortality estimates obtained from regression analysis mirror these observations. Traditional
gender roles in India are also reflected in the data, with women forming the majority of non-
working population. This, consequently, lead to more time spent indoors, lower particulate
exposure and significantly lower mortality compared to men.
It was interesting to note that rural mortality rate was lower than urban mortality, despite
having more than two times the population (Table 9 and Table 10), lower access to health care
and other socio-economic factors. This may be due to the fact that urban exposure to particulate
matter is different from rural exposure. Previous epidemiological studies that focused on large
metropolitan areas have shown that urban particulate matter originated mostly from vehicle
exhaust, industrial fumes, diesel generator sets, brick kilns and waste burning (Reynolds, et al.,
2011). On the other hand, rural particulate matter largely originated from burning organic matter
for cooking and heat (TERI, 1998-1999). This difference in toxicity, combined with the
difference of sparsity in rural vs. urbans areas explains the difference in the mortality rates.
The major advantage of this study is its large study domain. We have combined 7 years
of particulate matter data with corresponding mortality data. We have covered almost all districts
Assessment of mortality associated with air pollution
43
of India and controlled for population density, sex ratio, occupation, literacy rate among other
confounders. In a previous study, (Pope III, et al., 2002) estimated for every 10 µg/m
3
increase in
PM2.5, there was an associated 6%, increased risk of all-cause mortality which was similar to our
finding that mortality increased by 23% per 10 𝜇𝑔/𝑚
@
increase in PM2.5. The larger effect
estimate could be explained by the fact that the Indian population is exposed to much higher
concentration of population compared to the U.S. population in (Pope III, et al., 2002) study. A
study conducted in Delhi (Rajarathnam, et al., 2011) found an association of 0.15% increase in
all-cause mortality for every 10 µg/m
3
increase in PM10. This is in line with the 0.3% increase
identified by our study. Differences in our results with other studies can be explained by the fact
that we have conducted district-by-district analysis over a large time period.
PM2.5 can penetrate deeper into the lungs and reach the alveolar region compared to
PM10, which potentially leads to higher toxicity. Because of their smaller size and aerodynamic
behavior, PM2.5 particles have a longer residence time in the atmosphere compared to PM10
particles, which settle out of the air faster since they are heavier. Owing to their size, PM2.5 can
penetrate deep into the respiratory region of the lungs and may even enter the circulatory system
(TC, et al., 2005). However, monitoring stations are only recently being set up in India to
measure PM2.5 concentrations. For our study period, we had access to daily PM10 concentrations,
but not PM2.5. Using PM10 and PM2.5 data from 2016-17, meteorological factors and Generalized
Additive Models; we were able to build very good estimators for PM2.5 with R
2
of 0.65. The
relationship of PM2.5 to PM10 probably varies by region, urban vs. rural, season and year.
However, unavailability of meteorological data for 2009-2015 made it impossible to apply these
models to PM10 concentrations from that time and we had to contend with a simplified model.
Assessment of mortality associated with air pollution
44
This simplified model had a poor explained variance and coefficient of determination (R
2
),
which introduced noise in predicted PM2.5 values.
The major limitation of this study was lack of access to cause-specific or daily mortality
data over the study domain. We had access to annual count of all-cause mortality at district level.
If we had daily or monthly mortality data, we could account for seasonal variations in pollution.
Our analysis showed that pollution levels are significantly higher in winter, likely due to stubble
burning practices compounded by higher frequency of stagnant, cold meteorological conditions
and inversions, whereas PM levels are significantly lower during the rainy season.
Previous studies have found a close link between exposure to fine particles and premature
death from heart and lung disease (Cao, et al., 2011, Laden, et al., 2006, Ostro, et al., 2010). Fine
particles are also known to trigger or worsen chronic disease such as asthma, heart attack,
bronchitis and other respiratory problems. Access to cause specific mortality or disease incidence
rates would make our analysis better by focusing on outcomes that are directly affected by air
pollution, such as respiratory diseases or cardiac complications. And if we had individual level
health, demographic and mortality data we could have conducted an individual level model.
The government of India has started taking steps to increase the coverage of air pollution
monitoring stations and collaborating with other organizations such as System of Air Quality and
Weather Forecasting and Research (SAFAR), who have their own monitors but are not part of
CPCB. To increase awareness of air quality, they have also released real time data dissemination
services in the form of mobile apps such as “Sameer” (Sameer, 2015). Media has also played an
important role in making public aware about exposure to extremely high levels of air pollution.
However, people are not cognizant of the health costs, the increase in respiratory and
cardiovascular diseases, decrease in lung function of children and increased cancer risks. CPCB
Assessment of mortality associated with air pollution
45
has estimated the need for 4,000 continuous monitoring stations (2,800 urban and 1,200 rural) to
spatially, temporally, and statistically represent the PM2.5 pollution in the urban and the rural
areas of India (CPCB, 2009). It is our hope that they will acquire the necessary funding and will
help reduce to the heavy cost that Indian people are paying in terms of quality of life and
longevity.
In conclusion, we found that occupational categories as socio-economic composition,
literacy rate, household and population density have important effect on the association of
particulate matter air pollution and all-cause mortality, both in rural and urban areas. We also
found alarmingly elevated health risk estimates for PM2.5 and PM10 all-cause mortality in India in
a nationwide analysis. These results are similar to many previous researches in other locations,
indicating the reliability of our study. These findings suggest that people living in urban areas are
gravely affected compared to those in rural areas. In future, we would like to expand our analysis
to include spatio-temporal models to build more robust estimators that take into account daily
variation in PM data. Merging this data with daily, cause-specific mortality would paint an
accurate picture of air quality and its health impacts in India.

Assessment of mortality associated with air pollution
46
Bibliography
• Brook, R. D. et al., 2004. Air pollution and cardiovascular disease: a statement for
healthcare professionals from the Expert Panel on Population and Prevention Science of
the American Heart Association. Circulation, Volume 109, pp. 2655-2671.
• Cao, J. et al., 2011. Association between long-term exposure to outdoor air pollution and
mortality in China: a cohort study. Journal of hazardous materials. Journal of hazardous
materials, pp. 1594-1600.
• Census, 2011. census. [Online]
Available at: http://www.censusindia.gov.in/
• Cohen, A. J. et al., 2017. Estimates and 25-year trends of the global burden of disease
attributable to ambient air pollution: an analysis of data from the Global Burden of
Diseases Study 2015. The Lancet, 389(10082), pp. 1907-1918.
• CPCB, 2009. National Ambient Air Quality Standards. [Online]
Available at: http://cpcb.nic. in/National_ Ambient_ Air_Quality_Standards.php
• CPCB, 2010. NATIONAL AMBIENT AIR QUALITY STATUS & TRENDS IN INDIA.
[Online]
Available at:
http://cpcb.nic.in/openpdffile.php?id=UHVibGljYXRpb25GaWxlLzYyOF8xNDU3NTA
1MzkxX1B1YmxpY2F0aW9uXzUyMF9OQUFRU1RJLnBkZg==
[Accessed 9 3 2018].
• Cressie, N. A. C., 1993. Statistics for spatial data: Wiley series in probability and
mathematical statistics. Find this article online.
Assessment of mortality associated with air pollution
47
• CRS, 2011. Civil Registration System. [Online]
Available at: http://crsorgi.gov.in/web/index.php/auth/login
• Franklin, M., Koutrakis, P. & Schwartz, J., 2008. The role of particle composition on the
association between PM2. 5 and mortality. Epidemiology (Cambridge, Mass.) 19.5, p.
680.
• Franklin, M., Zeka, A. & Schwartz, J., 2007. Association between PM 2.5 and all-cause
and specific-cause mortality in 27 US communities. Journal of Exposure Science and
Environmental Epidemiology, 17(3), p. 279.
• GADM, 2009. [Online]
Available at: http://gadm.org/
• Gurjar, B. R., Butler, T. M., Lawrence, M. G. & Lelieveld, J., 2008. Evaluation of
emissions and air quality in megacities. Atmospheric Environment, Volume 42, pp. 1593-
1606.
• Guttikunda, S. K. & Jawahar, P., 2014. Atmospheric emissions and pollution from the
coal-fired thermal power plants in India. Atmospheric Environment, Volume 92, pp. 449-
460.
• Hong, Y.-C., Lee, J.-T., Kim, H. & Kwon, H.-J., 2002. Air pollution: a new risk factor in
ischemic stroke mortality. Stroke, Volume 33, pp. 2165-2169.
• Hughes, G. et al., 1997. Can the environment wait: priorities for East Asia.
• Jasarevic, T., 2014. Seven Million Premature Deaths Annually Linked to Air Pollution.
[Online]
Available at: http://www.who.int/mediacentre/news/releases/2014/air-pollution/en/
Assessment of mortality associated with air pollution
48
• Laden, F., Schwartz, J., Speizer, F. E. & Dockery, D. W., 2006. Reduction in fine
particulate air pollution and mortality: extended follow-up of the Harvard Six Cities
study. American journal of respiratory and critical care medicine. pp. 667-672.
• Maithel, S. et al., 2012. Brick kilns performance assessment, emissions measurements, &
a roadmap for cleaner brick production in India. Study report prepared by Green
Knowledge Solutions, New Delhi.
• Mwenda, K. M. & Shi, X., 2012. Spatial Interpolation of Fine Particulate Matter in New
Hampshire using Landuse-based Kriging. Extended Abstract Proceedings of the GI-
Science 2012, Extended abstracts (posters), Colombus, Ohio, Volume 9, p. 2012.
• Ostro, B. et al., 2010. Long-term exposure to constituents of fine particulate air pollution
and mortality: results from the California Teachers Study. Environmental health
perspectives 118, p. 363.
• Pope III, C. A. et al., 2002. Lung cancer, cardiopulmonary mortality, and long-term
exposure to fine particulate air pollution. Jama, Volume 287, pp. 1132-1141.
• Rajarathnam, U. et al., 2011. Part 2. Time-series study on air pollution and mortality in
Delhi. Research Report (Health Effects Institute), Volume 157, pp. 47-74.
• Reynolds, C., Grieshop, A. & Kandlikar, M., 2011. Climate and health relevant emissions
from in-use Indian three-wheelers fueled by natural gas and gasoline. Environmental
Science and Technology, pp. 2406-2412.
• SAMEER, 2015. [Online]
Available at: https://data.gov.in/catalog/sameer-national-air-quality-index
• Samet, J. M. et al., 2000. Fine particulate air pollution and mortality in 20 US cities,
1987--1994. New England journal of medicine, Volume 343, pp. 1742-1749.
Assessment of mortality associated with air pollution
49
• TC, L. et al., 2005. Air pollution-associated changes in lung function among asthmatic
children in Detroit. Environ Health Perspect, pp. 1068-75.
• TERI, 1998-1999. TEDDY: Tata Energy Data Directory Yearbook. New Delhi: Tata
Energy Research institute..
• Tsai, S. S., Chiu, H. F. & Yang, C. Y., 2006. Evidence for an Association Between Air
Pollution and Daily Stroke Admissions. Epidemiology, Volume 17, p. S271.
• WHO, 2014. Ambient (Outdoor) Air Pollution in Cities Database 2014. [Online]
Available at: http://www.who.int/entity/quantifying_ehimpacts/national/countryprofi le/
AAP_PM_database_May2014,xls?ua=1
• Wood, S.N., 2006. Generalized Additive Models: an Introduction with R. Chapman &
Hall/CRC.
• Zanobetti, A. & Schwartz, J., 2009. The effect of fine and coarse particulate air pollution
on mortality: a national analysis. Environmental Health Perspectives, 117(6), p. 898.

Assessment of mortality associated with air pollution
50
Appendix 1
Table 8 Summary statistics of locations where pollution data was available
Location
Area (km
2
) Population Density
(per km
2
)
PM 10 (µg/m
3
) Number of
Monitoring Stations
Anantapur, AP
19130 213 77.97 ± 21.66 1
Chittoor, AP
15152 275 55.43 ± 17.6 2
Eluru, AP
7742 509 103.92 ± 68.68 1
Guntur, AP
11391 429 81.37 ± 12.75 1
Hyderabad, AP
217 18172 91.5 ± 44.09 10
Kakinada, AP
10807 477 59.72 ± 6.62 1
Karimnagar, AP
11823 319 64.44 ± 38.49 1
Khammam, AP
16029 175 61.1 ± 15.22 2
Kothur, AP
7493 707 106.47 ± 31.22 1
Kurnool, AP
17658 230 78.84 ± 18.68 1
Nalgonda, AP
14240 245 79.85 ± 17.02 2
Nellore, AP
13076 227 63.84 ± 7.41 1
Nizamabad, AP
7956 321 55.52 ± 6.93 1
Ongole, AP
17626 193 64.57 ± 9.32 1
Rajahmundry, AP
163 2925 64.52 ± 14.35 1
Ramagundam, AP
94 2443 74.46 ± 43.49 1
Sangareddy, AP
4465 340 88.92 ± 49.59 3
Srikakulam, AP
5837 463 72.85 ± 25.29 1
Tirupati, AP
15359 275 52.46 ± 26.11 1
Vijayawada, AP
8727 520 98.66 ± 19.24 3
Visakhapatnam, AP
11161 384 69.2 ± 39.08 8
Vizianagaram, AP
6539 359 69.44 ± 25.72 1
Warangal, AP
12846 273 53.34 ± 21.55 2
Y.S.R., AP
15379 188 72.93 ± 8.56 1
Itanagar, AR
2875 51 85.36 ± 42.08 1
Assessment of mortality associated with air pollution
51
Naharlagun, AR
2875 51 68.09 ± 27.1 1
Bongaigaon, AS
1093 676 53.99 ± 37.64 2
Daranga, AS
2257 763 69.99 ± 41.17 1
Dibrugarh, AS
3381 392 64.98 ± 46.51 1
Golaghat, AS
3502 305 79.89 ± 46.59 1
Guwahati, AS
328 2900 109.14 ± 66.91 6
Hailakandi, AS
1327 497 88 ± 58.26 1
Lakhimpur, AS
2277 458 83.34 ± 46.56 1
Margherita, AS
3790 350 70.42 ± 43.92 1
Nagaon, AS
3973 711 109.72 ± 64.36 1
Nalbari, AS
1052 733 98.28 ± 53.1 1
Silchar, AS
3786 460 57.01 ± 30.7 2
Sivasagar, AS
2668 431 95.12 ± 49.43 2
Tezpur, AS
40 2600 83.8 ± 55.55 1
Tinsukia, AS
3790 350 78.81 ± 48.91 2
Patna, BR
3202 1823 153.92 ± 93.48 2
Chandigarh, CH
114 9258 94 ± 49.85 5
Bhilai, CT
8535 392 106.55 ± 32.28 3
Bilaspur, CT
8272 322 105.13 ± 25.6 1
Korba, CT
6598 183 92.13 ± 23.46 3
Raipur, CT
12383 328 276.42 ± 70.01 3
Khadoli, DN
491 700 53.32 ± 32.14 1
Silvassa, DN
491 700 71.23 ± 8.05 1
Daman, DD
72 2655 51.21 ± 24.43 1
Delhi, DL
1484 11312 227.92 ± 118.6 9
North Goa, GA
1736 470 64.96 ± 23.34 8
South Goa, GA
1966 330 62.5 ± 35.58 8
Ahmedabad, GJ
464 12000 86.6 ± 24.92 14
Assessment of mortality associated with air pollution
52
Ankleshwar, GJ
6509 238 86.47 ± 13.77 5
Bharuch, GJ
6509 238 83.42 ± 6.21 1
Bhuj, GJ
56 3800 85.92 ± 6.06 1
Jamnagar, GJ
14184 152 98.73 ± 18.37 2
Morbi, GJ
4871 200 93.15 ± 5.8 1
Rajkot, GJ
11198 340 95.45 ± 26.77 4
Valsad, GJ
2947 480 87.79 ± 5.56 2
Surat, GJ
4549 1337 88.57 ± 16.5 6
Vadodara, GJ
7546 552 88.84 ± 32.03 7
Faridabad, HR
741 2442 176.2 ± 47.18 2
Hissar, HR
3983 438 99.95 ± 83.62 2
Yamuna Nagar, HR
1768 687 196.9 ± 125.34 1
Solan, HP
1936 300 100.18 ± 40.99 3
Kangra, HP
5739 263 85.04 ± 51.18 2
Sirmaur, HP
2825 160 146.02 ± 77.84 2
Kulu, HP
5503 69 50.61 ± 34.63 2
Shimla, HP
5131 159 54.62 ± 24.65 2
Mandi, HP
3951 230 76.55 ± 37.33 2
Una, HP
1540 338 76.62 ± 22.1 2
Dhanbad, JH
2040 1316 175.05 ± 66.4 3
Jamshedpur, JH
219 6100 151.83 ± 29.06 2
Ranchi, JH
5097 572 184.34 ± 56.91 1
Saraikela Kharsawan, JH
2657 401 162.86 ± 48.86 1
West Singhbhum, JH
5351 280 246.42 ± 189.56 1
Bangalore, KA
2196 4381 122.4 ± 82.72 7
Belgaum, KA
13433 356 54.43 ± 27.06 1
Bidar, KA
5448 313 63.4 ± 9.94 1
Bijapur, KA
10498 207 99.09 ± 46.28 1
Assessment of mortality associated with air pollution
53
Chitradurga, KA
8436 197 38.49 ± 17.89 1
Davangere, KA
5926 330 107.79 ± 79.87 2
Dharwad, KA
4260 434 75.17 ± 21.91 1
Gulbarga, KA
10954 234 77.69 ± 47.35 1
Hassan, KA
6814 261 33.87 ± 13.21 1
Hubli-Dharwad, KA
4265 434 106.57 ± 45.81 1
Kolar, KA
3979 386 59.48 ± 27.22 1
Mandya, KA
4962 364 42.55 ± 12.47 1
Dakshin Kannada, KA
4559 457 37.86 ± 19.52 1
Mysore, KA
6307 476 52.6 ± 16.65 2
Raichur, KA
8442 228 85.07 ± 32.32 1
Ranebennur, KA
42 2117 58.7 ± 25.56 1
Shimoga, KA
8478 207 42.93 ± 23.96 1
Tumkur, KA
10597 253 121.9 ± 47.56 1
Alappuzha, KL
1415 1504 44.33 ± 17.15 2
Kochi, KL
440 6340 52.88 ± 44.13 8
Kollam, KL
2483 1061 45.59 ± 19.72 2
Kottayam, KL
2206 895 55.96 ± 24.53 2
Kozhikode, KL
2345 1316 45.51 ± 18.91 2
Malappuram, KL
3554 1157 37.12 ± 14.81 1
Palakkad, KL
4482 627 38.44 ± 19.19 1
Pathanamthitta, KL
2652 452 23.13 ± 3.19 1
Thiruvananthapuram, KL
2189 1508 54.5 ± 10.22 4
Thissur, KL
101 3100 45.68 ± 16.11 1
Trivandrum, KL
214 4500 62.69 ± 15.68 1
Wayanad, KL
2130 384 36.98 ± 11.17 1
Bhopal, MP
2772 855 162.59 ± 95.76 5
Chhindwara, MP
11815 177 84.81 ± 7.84 2
Assessment of mortality associated with air pollution
54
Dewas, MP
7020 223 94.09 ± 30.88 3
Gwalior, MP
4560 446 223.98 ± 108.03 2
Indore, MP
3898 841 137.53 ± 67.97 3
Jabalpur, MP
5211 473 88.2 ± 30.77 2
Nagda, MP
120 80 91.06 ± 29.06 3
Dhar, MP
8153 270 119.19 ± 25.08 2
Sagar, MP
10252 232 113.79 ± 56.83 1
Satna, MP
7502 297 163.65 ± 77.89 2
Singrauli, MP
5675 208 72.44 ± 19.33 3
Ujjain, MP
6091 326 88.76 ± 37.35 4
Akola, MH
5676 320 133.36 ± 20.41 3
Amravati, MH
12210 237 101.5 ± 31.44 3
Aurangabad, MH
10107 366 81.51 ± 30.05 3
Chandrapur, MH
11443 193 129.79 ± 80.05 6
Jalana, MH
7718 254 120.25 ± 57.29 2
Jalgaon, MH
11765 360 118.13 ± 25.65 3
Kolhapur, MH
7685 504 95.86 ± 38.19 3
Latur, MH
7157 343 100.81 ± 52.41 3
Ratangiri, MH
8208 197 116.63 ± 54.52 2
Raigad, MH
7152 370 159.3 ± 96.52 3
Mumbai, MH
603 20694 112.1 ± 61.38 9
Nagpur, MH
9892 470 94.84 ± 53.35 7
Nanded, MH
10528 319 89.24 ± 65.25 3
Nashik, MH
15530 393 83.93 ± 36.25 4
Pune, MH
15643 603 93.44 ± 54.33 4
Sangli, MH
8572 329 80.99 ± 44.33 3
Solapur, MH
14895 290 75.74 ± 15.69 2
Thane, MH
9558 1157 85.59 ± 49.1 3
Assessment of mortality associated with air pollution
55
Ri Bhoi, ML
2448 106 146.49 ± 39.94 2
West Jaintia Hills, ML
3819 103 50.14 ± 19.3 1
East Jaintia Hills, ML
2126 58 38.12 ± 15.19 1
West Khasi Hills, ML
5247 56 26.37 ± 6.67 1
Shillong, ML
64 234 68.53 ± 26.43 2
West Garo Hills, ML
3714 140 45.16 ± 13.51 1
Aizawl, MZ
3576 112 44.94 ± 26.94 5
Champhai, MZ
3185 39 45.97 ± 33.05 2
Kolasib, MZ
1382 61 37.65 ± 11.8 2
Lunglei, MZ
4536 36 45.82 ± 15.88 2
Dimapur, NL
927 409 95.28 ± 57.72 2
Kohima, NL
1463 183 82.25 ± 36.76 2
Angul, OR
6375 200 109.74 ± 39.78 2
Baleshwar, OR
3806 610 83.01 ± 13.13 3
Ganjam, OR
8070 429 64.21 ± 27.92 1
Bhubaneshwar, OR
422 2131 78.14 ± 25.76 6
Cuttack, OR
3932 667 83.58 ± 37.79 3
Jajpur, OR
87 658 99.38 ± 15.14 3
Kendujhar, OR
8303 217 82.41 ± 42.42 1
Jagatsinghpur , OR
1759 602 94.89 ± 49.22 3
Puri, OR
3479 488 93.67 ± 48.46 2
Rayagada, OR
7073 137 53.84 ± 15.84 2
Rourkela, OR
200 6696 98.59 ± 29.34 3
Sambalpur, OR
6624 157 54.44 ± 10.85 1
Pondicherry, PY
294 3232 40.37 ± 9.46 3
Karaikal, PY
157 1275 35.04 ± 15.17 3
Amritsar, PB
2683 928 191.81 ± 38.46 3
Bathinda, PB
3353 414 135.42 ± 54.69 1
Assessment of mortality associated with air pollution
56
Dera Baba, PB
3551 647 71.88 ± 15.13 1
Mohali, PB
1094 909 122.31 ± 45.29 2
Faridkot, PB
1458 424 89.75 ± 23.54 1
Gobindgarh, PB
1180 509 176.02 ± 58.92 3
Hoshiarpur, PB
3386 469 73.49 ± 20.61 1
Jalandhar, PB
2624 836 150.9 ± 26.62 4
Ludhiana, PB
3578 978 200.43 ± 73.29 4
Roop Nagar, PB
1356 505 86.21 ± 23.46 2
Patiala, PB
3325 570 108.12 ± 26.5 2
Sangrur, PB
3625 457 100.33 ± 24.55 1
Alwar, RJ
8380 438 203 ± 115.65 3
Jaipur, RJ
11143 595 161.53 ± 101.99 6
Jodhpur, RJ
22850 161 165.59 ± 83.68 6
Kota, RJ
5217 374 128.44 ± 74.45 3
Udaipur, RJ
11724 262 134.56 ± 82.02 3
Chennai, TN
175 26553 63.99 ± 39.27 11
Coimbatore, TN
4732 731 65.18 ± 48.04 3
Cuddalore, TN
3703 704 57.53 ± 23.38 3
Madurai, TN
3710 819 47.28 ± 17.43 3
Salem, TN
5237 665 66.61 ± 35.26 1
Thoothukkudi, TN
4745 369 106.45 ± 71.42 3
Trichy, TN
4404 604 84.9 ± 37.98 5
Agra, UP
4041 1093 175.18 ± 96.43 6
Allahabad, UP
5482 1086 241.05 ± 84.45 5
Sonbhadra, UP
6905 270 132.54 ± 24.31 1
Bareilly, UP
4120 1080 233.04 ± 55.84 2
Firozabad, UP
2407 1038 197.28 ± 97.62 3
Amroha, UP
2321 650 144.74 ± 51.54 2
Assessment of mortality associated with air pollution
57
Ghaziabad, UP
1179 3971 257.1 ± 73.4 2
Gorakhpur, UP
3321 1337 123.22 ± 44.43 3
Jhansi, UP
5024 398 115.82 ± 26.63 2
Kanpur, UP
3155 1452 197.86 ± 78.83 10
Bulandshahr, UP
4512 776 165.95 ± 19.34 2
Lucknow, UP
2528 1816 191.46 ± 111.4 5
Mathura, UP
3340 763 206.26 ± 40.63 2
Meerut, UP
2559 1346 135.18 ± 33.43 2
Moradabad, UP
3718 1283 167.84 ± 61.16 2
Noida, UP
203 2463 139.17 ± 24.66 2
Rai Bareilly, UP
4609 739 158.32 ± 31.97 3
Renusagar, UP
5 4000 135.77 ± 23.85 1
Saharanpur, UP
3689 940 43.48 ± 7.73 2
Unnao, UP
4558 682 106.76 ± 28.26 2
Varanasi, UP
1535 2395 136.43 ± 12.31 2
Dehradun, UT
3088 549 169.75 ± 72.46 3
Haldwani, UT
44 3500 138.48 ± 30.8 1
Hardwar, UT
2360 801 127.94 ± 31.15 1
Kashipur, UT
5 22275 171.06 ± 101.74 1
Rishikesh, UT
11 8851 110.19 ± 26.94 1
Rudrapur, UT
27 5100 139.89 ± 47.18 1
Asansol, WB
326 3500 128.43 ± 85.35 5
Barrackpore, WB
25 6200 110.5 ± 64.23 4
Durgapur, WB
154 3400 140.78 ± 97.96 5
Haldia, WB
102 2036 105.19 ± 54.27 6
Howrah, WB
95 11285 122.83 ± 74.87 5
Kalyani, WB
29 3500 86.55 ± 55.27 1
Kolkata, WB
185 24306 117.24 ± 87.14 19
Assessment of mortality associated with air pollution
58
Malda, WB
3733 1069 81.53 ± 19.77 1
Raniganj, WB
113 1145 120.06 ± 79.13 4
Siliguri, WB
41 12000 89.02 ± 31.2 1

Assessment of mortality associated with air pollution
59
Appendix 2

Assessment of mortality associated with air pollution
60
Appendix 3
Table 9 Poisson regression model of PM 10 exposure and all-cause mortality in Urban Population

Estimate 95% CI z value Pr(>|z|)
(Intercept)
-70.799066 -71.2857 - -70.31244 -285.15402 <0.001
PM 10
0.01672853 0.01671 - 0.01675 1735.22683 <0.001
Prop. district population w.r.t
state
0.97423143 0.96279 - 0.98566 167.006043 <0.001
Year
0.0404298 0.04019 - 0.04067 328.995061 <0.001
Latitude
-0.0874808 -0.08764 - -0.08733 -1104.816 <0.001
Longitude
-0.0359894 -0.0361 - -0.03588 -644.48295 <0.001
Prop. of female w.r.t district
8.09485352 8.05611 - 8.1336 409.489502 <0.001
Number of people / Household
0.05305223 0.05178 - 0.05433 81.584366 <0.001
Prop. of children w.r.t district
-19.204229 -19.25809 - -19.15037 -698.84998 <0.001
Prop. of literate w.r.t district
-2.806435 -2.8206 - -2.79227 -388.22817 <0.001
Prop. of main cultivators w.r.t
district
-45.746974 -45.85613 - -45.63784 -821.50631 <0.001
Prop. of main agricultural
workers w.r.t district
1.83477051 1.78035 - 1.88916 66.0970574 <0.001
Prop. of main industrial workers
w.r.t district
13.0288822 12.96277 - 13.09498 386.289569 <0.001
Prop. of main other workers
w.r.t district
5.49721949 5.47727 - 5.51717 540.085072 <0.001
Prop. of marginal cultivators
w.r.t district
-25.151235 -25.41659 - -24.88629 -185.91878 <0.001
Prop. of marginal agricultural
workers w.r.t district
-40.423534 -40.582 - -40.26512 -500.06653 <0.001
Prop. of marginal industrial
workers w.r.t district
-18.921254 -19.2074 - -18.63523 -129.63037 <0.001
Prop. of marginal other workers
w.r.t district
1.02427595 0.96953 - 1.07902 36.6713308 <0.001

Assessment of mortality associated with air pollution
61
Table 10 Poisson regression model of PM 10 exposure and all-cause mortality in Rural Population

Estimate 95% CI z value Pr(>|z|)
(Intercept)
22.8364091 22.42446 - 23.24836 108.650936 <0.001
PM 10
0.0048716 0.00485 - 0.00489 440.661352 <0.001
Prop. district population w.r.t state
0.91246888 0.88659 - 0.93832 69.1415915 <0.001
Year
-0.0060175 -0.00622 - -0.00581 -57.640349 <0.001
Latitude
-0.0456412 -0.04578 - -0.0455 -631.5211 <0.001
Longitude
-0.0336324 -0.03374 - -0.03353 -631.23543 <0.001
Prop. of female w.r.t district
8.85964798 8.81804 - 8.90125 417.388718 <0.001
Number of people / Household
0.06553863 0.06453 - 0.06655 126.881706 <0.001
Prop. of children w.r.t district
-11.881281 -11.91027 - -11.85229 -803.18036 <0.001
Prop. of literate w.r.t district
-1.1897832 -1.1966 - -1.18297 -342.11464 <0.001
Prop. of main cultivators w.r.t
district
-3.1601016 -3.17068 - -3.14953 -585.58948 <0.001
Prop. of main agricultural workers
w.r.t district
-1.0215795 -1.03383 - -1.00933 -163.40189 <0.001
Prop. of main industrial workers
w.r.t district
2.49971194 2.44508 - 2.5543 89.7111213 <0.001
Prop. of main other workers w.r.t
district
-5.8251147 -5.84172 - -5.80851 -687.57302 <0.001
Prop. of marginal cultivators w.r.t
district
-4.1998885 -4.22051 - -4.17927 -399.18755 <0.001
Prop. of marginal agricultural
workers w.r.t district
-2.5162313 -2.5323 - -2.50016 -306.90525 <0.001
Prop. of marginal industrial workers
w.r.t district
11.6502079 11.52181 - 11.77854 177.883365 <0.001
Prop. of marginal other workers
w.r.t district
-1.5572228 -1.59718 - -1.51728 -76.402351 <0.001

Assessment of mortality associated with air pollution
62
Table 11 Poisson regression model of PM 10 exposure and all-cause mortality in Male Population

Estimate 95% CI z value Pr(>|z|)
(Intercept)
-23.612009 -24.0193 - -23.2048 -113.63459 <0.001
PM 10
0.00810391 0.0081 - 0.0081 870.981165 <0.001
Prop. district population w.r.t state
-2.1263399 -2.1369 - -2.1158 -395.07192 <0.001
Year
0.01824796 0.018 - 0.0184 177.154127 <0.001
Latitude
-0.0466502 -0.0468 - -0.0465 -675.74065 <0.001
Longitude
-0.0283863 -0.0285 - -0.0283 -564.07599 <0.001
Prop. of female w.r.t district
0.00604271 -0.0294 - 0.0415 0.33447135 0.74
Number of people / Household
0.02657068 0.0255 - 0.0277 47.7103286 <0.001
Prop. of children w.r.t district
-10.373481 -10.4073 - -10.3396 -600.92349 <0.001
Prop. of literate w.r.t district
-0.7400747 -0.7483 - -0.7319 -176.59729 <0.001
Prop. of main cultivators w.r.t
district
-3.2481946 -3.2614 - -3.235 -480.71045 <0.001
Prop. of main agricultural workers
w.r.t district
-0.8630022 -0.8789 - -0.8471 -106.45099 <0.001
Prop. of main industrial workers
w.r.t district
3.52527202 3.4723 - 3.5782 130.518569 <0.001
Prop. of main other workers w.r.t
district
2.1002395 2.0869 - 2.1136 309.108958 <0.001
Prop. of marginal cultivators w.r.t
district
-5.4831517 -5.5093 - -5.457 -411.4992 <0.001
Prop. of marginal agricultural
workers w.r.t district
0.07624541 0.0574 - 0.0951 7.93227545 <0.001
Prop. of marginal industrial workers
w.r.t district
3.81929608 3.6647 - 3.9739 48.4190383 <0.001
Prop. of marginal other workers
w.r.t district
-7.71221 -7.7604 - -7.6641 -313.88051 <0.001

Assessment of mortality associated with air pollution
63
Table 12 Poisson regression model of PM 10 exposure and all-cause mortality in Female Population

Estimate 95% CI z value Pr(>|z|)
(Intercept)
1.92E-01 -0.2882 - 0.6713 0.78248871 0.434
PM 10
7.22E-03 0.0072 - 0.0072 645.429069 <0.001
Prop. district population w.r.t state
-2.34E+00 -2.3496 - -2.3235 -350.19682 <0.001
Year
6.31E-03 0.0061 - 0.0066 52.0205338 <0.001
Latitude
-4.86E-02 -0.0488 - -0.0485 -592.107 <0.001
Longitude
-2.00E-02 -0.0201 - -0.0199 -337.86373 <0.001
Prop. of female w.r.t district
-5.11E-01 -0.5537 - -0.4677 -23.257443 <0.001
Number of people / Household
5.98E-02 0.0585 - 0.061 92.7275019 <0.001
Prop. of children w.r.t district
-1.13E+01 -11.3458 - -11.2663 -557.41865 <0.001
Prop. of literate w.r.t district
-9.68E-01 -0.9776 - -0.9582 -195.95897 <0.001
Prop. of main cultivators w.r.t
district
-4.54E+00 -4.5516 - -4.52 -562.79209 <0.001
Prop. of main agricultural workers
w.r.t district
-9.31E-01 -0.9494 - -0.9122 -97.88436 <0.001
Prop. of main industrial workers
w.r.t district
3.85E+00 3.7848 - 3.9098 120.607106 <0.001
Prop. of main other workers w.r.t
district
1.15E+00 1.1326 - 1.1643 142.23487 <0.001
Prop. of marginal cultivators w.r.t
district
-5.49E+00 -5.5195 - -5.4566 -341.89419 <0.001
Prop. of marginal agricultural
workers w.r.t district
1.30E-02 -0.0089 - 0.0349 1.16492903 0.244
Prop. of marginal industrial
workers w.r.t district
1.35E+00 1.1694 - 1.5287 14.7187348 <0.001
Prop. of marginal other workers
w.r.t district
-8.20E+00 -8.2603 - -8.1458 -280.93664 <0.001

Abstract (if available)

Abstract Background: In India, more than a billion people are at risk from ambient particulate matter (PM) concentrations exceeding World Health Organization air quality guidelines, posing serious threat to health. Previous studies, conducted in the United States, have elucidated increases in all-cause and specific-cause mortality associated with exposure to PM in metropolitan cities. Associations between mortality and ambient particulate matter exposure are poorly characterized for India, despite its serious air pollution problem. ❧ Methods: In this study, using 2009-2015 data from the Central Pollution Control board (CPCB) and Civil Registration System (CRS), we investigated the association between exposure to fine and coarse PM air pollutants (PM₂.₅ and PM₁₀, respectively) with all-cause mortality, targeting both rural and urban centers in India. Spatial statistical methods including variograms and universal kriging, were applied to annual averages of PM data from 222 districts to generate exposure estimates at unobserved locations. The estimated exposures were linked to annual counts of all-cause mortality in the respective districts, and the associations between annual average PM₂.₅ and PM₁₀ and mortality were determined using Poisson regression. Sex, occupation, literacy, geographic location, and year were all considered as potential confounders and were included in the model. ❧ Results: There was 7.7% (95% confidence interval = 7.4% - 7.8%) increase in all-cause mortality associated with a 10 μg/m³ increase in annual averaged PM₁₀ concentration across districts in India. This association was larger in urban areas (16.7% [16.3% - 16.9%]) than in the rural areas (4.6% [4.2% - 4.8%]). Similarly, with a 10 μg/m³ increase in an annual average PM₂.₅ concentration, there was 23.2% (95% confidence interval = 23.0% - 23.3%) increase in all-cause mortality. This association was adjusted by combination of covariates such as literacy, occupational workers, household size, geographical location and year, which were also aggregated at a district level. ❧ Conclusion: This study shows that after controlling for certain demographic factors, geographic location and year there is a statistical significant increase in mortality with increase in PM exposures.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Spatial analysis of PM₂.₅ air pollution in association with hospital admissions in California

PDF

Examining exposure to extreme heat and air pollution and its effects on all-cause, cardiovascular, and respiratory mortality in California: effect modification by the social deprivation index

PDF

Spatial modeling of non-tailpipe emissions and its association with children's lung function

PDF

Air pollution and breast cancer survival in California teachers: using address histories and individual-level data

PDF

Associations of ambient air pollution exposures with perceived stress in the MADRES cohort

PDF

Associations between ambient air pollution and hypertensive disorders of pregnancy

PDF

Native American ancestry among Hispanic Whites is associated with higher risk of childhood obesity: a longitudinal analysis of Children’s Health Study data

PDF

A cohort study of air-pollution and childhood obesity incidence

PDF

Personal exposure to particulate matter PM2.5 sources during pregnancy and birthweight

PDF

Disparities in exposure to traffic-related pollution sources by self-identified and ancestral Hispanic descent in participants of the USC Children’s Health Study

PDF

Assessing the impact of air pollution on adverse birth outcomes in a low resource setting

PDF

Surgical aortic arch intervention at the time of extended ascending aortic replacement is associated with increased mortality

PDF

A pilot study of a global approach to assessing air pollution exposure in port communities: passive air monitoring of nitrogen dioxide concentrations

PDF

Air pollution, mitochondrial function, and growth in children

PDF

Ambient air pollution and lung function in children

PDF

Association of traffic-related air pollution and lens opacities in the Los Angeles Latino Eye Study

PDF

Associations between physical activities with bone mineral density in postmenopausal women

PDF

The effects of heat and air pollution on mental-health related mortality

PDF

Comparison of models for predicting PM2.5 concentration in Wuhan, China

PDF

Effectiveness of individual- and household-level protective actions in reducing symptoms associated with hydrogen sulfide chronic low-level exposure

Asset Metadata

Creator Arora, Rashi (author)

Core Title Assessment of the mortality burden associated with ambient air pollution in rural and urban areas of India

School Keck School of Medicine

Degree Master of Science

Degree Program Applied Biostatistics and Epidemiology

Publication Date 04/26/2018

Defense Date 03/23/2018

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag all-cause mortality,health effects,OAI-PMH Harvest,PM₁₀,PM2.5,Poisson regression,statistical analysis,universal kriging,variograms

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Franklin, Meredith (committee chair), Berhane, Kiros (committee member), Habre, Rima (committee member)

Creator Email rashi.19@gmail.com,rashiaro@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-495653

Unique identifier UC11265839

Identifier etd-AroraRashi-6277.pdf (filename),usctheses-c40-495653 (legacy record id)

Legacy Identifier etd-AroraRashi-6277.pdf

Dmrecord 495653

Document Type Thesis

Format application/pdf (imt)

Rights Arora, Rashi

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA