University of Southern California Dissertations and Theses

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

(USC Thesis Other)

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

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content
Assessing the Impact of Air Pollution on Adverse Birth Outcomes in a
Low Resource Setting
by

Temuulen Enebish

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
DOCTOR OF PHILOSOPHY
EPIDEMIOLOGY

August 2020

Copyright 2020 Temuulen Enebish
ii

Dedication

This dissertation is dedicated to the three most important women in my life:
my mother Munkhtsetseg Sanduijav;
my wife Nomuulin Namdag; and
my daughter Enerel Temuulen.

iii

Acknowledgements

I was only able to complete this dissertation because of the sacrifices made by my family and
enormous amount of help and support I received from many, many people.
I would like to start by thanking my doctoral dissertation committee chair and mentor Dr
Meredith Franklin for guiding, supporting and encouraging me through the most difficult periods.
Thank you so much for your kindness and patience!
I am forever grateful to Dr David Warburton, who provided me with an opportunity to
pursue my PhD via his D43 training grant from the NIEHS. His continuous advice and support that
made it all easier. I also would like to thank my committee members Dr Rima Habre, Dr Roberta
McKean-Cowdin, and Dr Carrie Breton for making sure I conducted the best science I could and
always being there for me when I needed help.
Doing research based on a developing country requires support and collaboration of many
people. I am extremely grateful to colleagues from the National Center for Maternal and Child
Health of Mongolia, particularly director-general Dr Enkhtur Shonkhuuz, deputy director in charge
of research and foreign relations Dr Bayalag Munkhuu, head of the surveillance department Dr
Gantuya Tumur, head of the research department Dr Otgonbaatar Jugder, and Dr Nomindelger
Tuvshindorj. Administration and archive personnel of all public maternity hospitals in Ulaanbaatar
as well as my 8 research assistants (Unurjargal Bayasgalan, Khandsuren Ard, Munkhshur Batmunkh,
Nandin-Erdene Bazarragchaa, Baigal Tsolmonchimeg, Munkhmaa Lachintseren, and Tergel
Munkhtur) were hugely helpful in collecting data. Thank you all very much!
iv

My professor and mentor from my medical school years, Dr Erdenekhuu Nansalmaa, was
greatly influential in the pursuit my degrees and completing them successfully. For this, I am
immensely grateful.
Last, but most of all, I want to thank my mother, Munkhtsetseg Sanduijav, who raised me
and my siblings through one of the most difficult periods in modern Mongolian history alone and
my wife, Nomuulin Namdag, who put her career on hold to support me and provided me with
everything I needed during my doctoral journey. I would not have been able to do this without you.

Temuulen Enebish
June 30, 2020

v

Table of Contents
Dedication ........................................................................................................................................................... ii
Acknowledgements ........................................................................................................................................... iii
List of Tables .................................................................................................................................................... vii
List of Figures .................................................................................................................................................. viii
Abstract .............................................................................................................................................................. ix
Chapter 1: Predicting ambient air pollution in Ulaanbaatar, Mongolia with machine learning
approaches .......................................................................................................................................................... 1
Abstract ........................................................................................................................................................... 1
Background .................................................................................................................................................... 2
Materials and Methods ................................................................................................................................. 4
Setting ......................................................................................................................................................... 4
Statistical analysis ...................................................................................................................................... 8
Results ........................................................................................................................................................... 11
Discussion .................................................................................................................................................... 19
Chapter 2: Investigating the acute effects of ambient air pollution on stillbirth risk in Ulaanbaatar,
Mongolia............................................................................................................................................................ 24
Abstract ......................................................................................................................................................... 24
Background .................................................................................................................................................. 25
Methods ........................................................................................................................................................ 27
Study population ..................................................................................................................................... 27
Exposure data .......................................................................................................................................... 29
Study design ............................................................................................................................................. 31
Statistical analysis .................................................................................................................................... 32
Results ........................................................................................................................................................... 33
Discussion .................................................................................................................................................... 39
Chapter 3: Acute effect of ambient air pollution on select congenital anomalies: a case-crossover
study ................................................................................................................................................................... 45
Abstract ......................................................................................................................................................... 45
Background .................................................................................................................................................. 46
Methods ........................................................................................................................................................ 48
Study population ..................................................................................................................................... 49
Exposure data .......................................................................................................................................... 50
Study design ............................................................................................................................................. 51
Statistical analysis .................................................................................................................................... 52
Results ........................................................................................................................................................... 52
Discussion .................................................................................................................................................... 60
References ......................................................................................................................................................... 64
vi

Appendix ........................................................................................................................................................... 74
A ..................................................................................................................................................................... 74
B ..................................................................................................................................................................... 76

vii

List of Tables

Table 1. The final tuned values for each model .......................................................................................... 10
Table 2. Variables used to predict PM 2.5 ....................................................................................................... 11
Table 3. Comparison of model performance metrics for prediction of PM 2.5........................................ 12
Table 4. Characteristics of Mothers and Stillbirths, Ulaanbaatar 2010–2018 ......................................... 35
Table 5. Relative Odds of Stillbirth Associated with IQR Increases in Mean PM 2.5, NO 2, SO 2, and
CO Concentration by Lag day(s) ................................................................................................................... 37
Table 6. Relative Odds of Stillbirth Associated with IQR Increase in PM 2.5, NO 2, SO 2, and CO
Concentration on Lag Day 6 by Level of Maternal Characteristics ......................................................... 39
Table 7. Outcome Groups and ICD-10 Codes Included in the Analysis ............................................... 51
Table 8. Distribution of Covariates Among Pooled Congenital Anomaly Cases, and Select Outcome
Groups, Ulaanbaatar, 2014–2018 .................................................................................................................. 53
Table 9. Relative Odds of Congenital Anomalies Related to Cardiovascular System ........................... 55
Table 10. Relative Odds of Select Congenital Anomalies .......................................................................... 56

viii

List of Figures

Figure 1. Map of study area and data sources of PM 2.5 from three organizations ................................... 6
Figure 2. Model predictions from leave-one-location-out cross-validation plotted against observed
PM 2.5 for the entire study period .................................................................................................................... 13
Figure 3. Model predictions from leave-one-location-out cross-validation plotted against observed
PM 2.5 for the cold and warm seasons ............................................................................................................ 14
Figure 4. Predictions from the Random Forest model applied to the entire study period are shown
here as cold (Oct-Mar) and warm (Apr-Sep) season averages in the context of population density of
UB ...................................................................................................................................................................... 16
Figure 5. Daily predicted PM 2.5 at the khoroo level in Ulaanbaatar, Mongolia shown as annual and
seasonal averages. Predictions were obtained from the Random Forest model, and seasonal maps
were plotted at different scales....................................................................................................................... 17
Figure 6. Variable importance scores for the 10 most important variables in the Random Forest
model (descending order) ............................................................................................................................... 18
Figure 7. Study map with number of stillbirth cases .................................................................................. 29
Figure 8. Seasonal variation of modelled PM 2.5 concentration in UB ...................................................... 33
Figure 9. Flow chart of study population ..................................................................................................... 34
Figure 10. Effect estimates of air pollution on stillbirth risk ..................................................................... 38
Figure 11. Flow chart of the study population ............................................................................................ 50
Figure 12. Effect estimates of air pollution on cardiovascular anomaly risk .......................................... 57
Figure 13. Effect estimates of air pollution on cardiac septal anomaly risk ............................................ 58
Figure 14. Effect estimates of air pollution on ventricular septal defect risk ......................................... 58
Figure 15. Effect estimates of air pollution on musculoskeletal anomaly risk ........................................ 59
Figure 16. Effect estimates of air pollution on cleft lip and cleft palate risk .......................................... 59

ix

Abstract
The aim of this dissertation is to develop air pollution exposure assessment models and
apply that towards investigating the short-term impact of air pollution on stillbirth and congenital
anomaly occurrence in a low resource setting with high air pollution levels. The current literature is
focused on developed countries with low levels of air pollution and the evidence is scarce on the
populations that are disproportionately affected by air pollution. We ask what exposure assessment
methods work best for settings with inadequate monitoring capacity and how air pollution affects
adverse birth outcomes in that setting to address the shortcoming. We utilized state-of-the-art
developments in statistical computing to evaluate machine learning approaches. Our results show
they perform well in capturing the complex relationship of air pollution with meteorological and
other related factors. Using case-crossover design, we demonstrated increased acute risks of stillbirth
immediately before delivery and select congenital anomalies on certain vulnerable periods during
gestation. We also show certain sub-groups of women are more sensitive to the acute effect of air
pollution on stillbirth and the possible dose-response gradient for select congenital anomalies. The
significance of the dissertation is the improvement of assessing air pollution exposure in low-income
countries and the contribution to the understanding of decision-makers on who and when to
intervene regarding adverse birth outcomes associated with air pollution.

1

Chapter 1: Predicting ambient air pollution in Ulaanbaatar, Mongolia
with machine learning approaches
Abstract
Accurately assessing individual ambient air pollution exposure is a crucial part of
epidemiological studies looking at the adverse health effect of poor air quality. This is particularly
challenging in developing countries with high levels of air pollution but having sparse monitoring
networks with a lack of consistent data. We evaluated the performance of 6 different machine
learning algorithms in predicting fine particulate matter (PM 2.5) concentrations in Ulaanbaatar,
Mongolia from 2010 to 2018. We found that the algorithms produce robust results based on
performance metrics. Random forest (RF) and gradient boosting models performed the best with
leave-one-location-out cross-validated R
2
of 0.82 for when using data from the entire study period.
After applying tuned models on the hold-out test set, R
2
increased to 0.96 for the RF and 0.90 for
the gradient boosting model. We also predicted PM 2.5 concentrations for each administrative area
(khoroo) of the city using RF and maps of predictions show spatiotemporal variations that are in
line with the location of the ger district, city center, and population density. Our results provide
evidence of the advantage and feasibility of machine learning approaches in predicting ambient PM 2.5
levels in a setting with limited resources and extreme air pollution levels.

2

Background
Adverse effects of ambient fine particulate matter (PM 2.5, aerodynamic diameter less than 2.5
𝜇 m) on human health have been studied extensively over the past several decades. While we
generally have strong evidence about how long and short-term exposures to PM 2.5 negatively affect
mortality (Di et al., 2017; Franklin et al., 2007; Pope et al., 2019), morbidity (Lippmann et al., 2000),
respiratory, and cardiovascular diseases (Brook et al., 2010), there are still many questions regarding
its association with a multitude of different health endpoints. Furthermore, most of the existing
literature on the health effects of air pollution comes from studies conducted in developed countries
where air pollution levels are comparatively low. The availability of air pollution exposure data from
different sources such as monitoring networks, satellite data, and chemical transport models has
made studying air pollution health effects in low-concentration environments more accessible.
Unfortunately, this has led to an underrepresentation of populations in developing countries where
the adverse health impact of air pollution is the highest (Cohen et al., 2017).
An example of one such country is Mongolia where daily ambient PM 2.5 levels during winter
in its capital Ulaanbaatar (UB) regularly exceed 150 𝜇 g/m
3
(Figure 1, WHO air quality guideline - 25
𝜇 g/m
3
(World Health Organization, 2006)). There are two main reasons for the high levels of
pollution in UB. First, there was an influx of rural herders who have lost all their livestock in the
Mongolian winter disaster called “dzud” into the city in the early 2000s looking for better
employment and education opportunities. They set up their “ger”, yurt-like traditional dwelling, in
the outskirts of the city and they use coal-burning stoves to heat and cook during the harsh winter
months of Mongolia. This was the start of the so-called “ger district” which mostly inhabited by
low-income families and now occupies almost 60% of the city. Second, this spike in emission during
winter (Figure 1) is exacerbated by UB’s location inside a valley surrounded by mountains which
3

facilitates the formation of temperature inversion where cold air near the ground is enveloped by
warmer air above and traps the pollutants within the breathing zone.
Air pollution disproportionately harms the people of Mongolia since almost half of the
Mongolian population live in UB, the regional and provincial population center with sources
including coal-burning power plants and most of the population living in the surrounding ger area
where people burn coal for heating and cooking. There have been several efforts to measure and
identify sources of particulate matter (Batmunkh et al., 2013; Davy et al., 2011; Guttikunda et al.,
2013; Nishikawa et al., 2011) in UB and assess its adverse health effects (Allen et al., 2013; Enkhmaa
et al., 2014; Enkh-Undraa et al., 2019) in recent years. However, health studies have been weakened
by inaccurate and error-prone exposure assignment methods due to a lack of monitoring stations
and their data availability.
Environmental epidemiologists use statistical and deterministic exposure assessment models
to interpolate measured concentrations to residential and/or school locations. Models such as
inverse distance weighting and kriging are frequently used for spatial interpolation and are based on
the premise that measurements taken at nearby monitors are more alike than those with a large
distance between them. These methods tend to perform better in places with a dense network of
monitoring sites. Extensions of these models to incorporate different spatial and temporal covariates
such as meteorological data, satellite measurements, and products from chemical transport models
have formed the basis of land-use regression. The limitation, however, is that they often assume
linear relationships between the covariates and response, which might not be suitable when
modeling complex relationships in a variety of terrains.
Air pollution prediction studies have found promising results with more adaptable models
that do not require strict linearity assumptions (e.g. generalized additive models) and different
4

machine learning algorithms (e.g. random forest, gradient boosting, and support vector machine).
Specifically, random forest has been used to model PM 2.5 in the conterminous US (Hu et al., 2017),
urban area encompassing seven counties in Ohio (Brokamp et al., 2018), Tehran, Iran (Nabavi et al.,
2019), and gradient boosting was used in prediction of daily PM~2.5 concentrations across China
(Zhan et al., 2017). Also, Di et al. used a combination of machine learning algorithms through an
ensemble method to predict PM 2.5 across the contiguous US (Di et al., 2019). Xu et al. evaluated the
performance of 8 different algorithms in British Columbia (Xu et al., 2018), Watson et al. showed
that random forest and gradient boosting performed well in prediction of ozone during wildfire
events in northern California (Watson et al., 2019), and our previous work demonstrated that
support vector machine was superior to other algorithms when incorporating different mixtures of
Multi-angle Imaging SpectroRadiometer (MISR) aerosol optical depth (AOD) measurements over
UB (Franklin et al., 2018).
In this study, we aim to improve the PM 2.5 exposure assessment for UB by synthesizing a
unique set of locally obtained data into different machine learning algorithms. We evaluate the
performance of various algorithms to predict ambient daily PM~2.5 levels in Ulaanbaatar, Mongolia
by incorporating a variety of spatial and temporal variables related to air pollution in the city.
Materials and Methods
Setting
UB is the political and economic hub of Mongolia. The city is divided into 9 districts and
each district consists of a varying number of administrative units called “khoroo”.
A khoroo is a subdivision of the city similar to the census tract in the United States. Currently, there
are 152 khoroos and they vary widely in size depending on their population density (Figure 1). We
used khoroo as our spatial prediction level instead of a regular grid since automatic geocoding is
currently not possible for Mongolian addresses due to a lack of standardized addresses.
5

Furthermore, it will facilitate assigning exposure estimates to study participants in a subsequent
epidemiological study based on hospital health records. For model validity, we excluded 3 outlying
rural districts (14 khoroos) with no regulatory monitoring sites from our prediction. In terms of
meteorology and geography, UB experiences short summers and cold, dry winters and is located
approximately 1,300 meters above sea level, along the Tuul River in a valley at the foot of the Bogd
Khan mountain.
Air pollution monitoring data
Hourly PM 2.5 concentration data were obtained from 9 monitoring stations that cover the
period between 2010 and 2018. While 5 of these stations are operating under the purview of the UB
Air Pollution Reduction Agency (APRA), 3 of them are administered by the National Agency for the
Meteorological and Environmental Monitoring (NAMEM) and the remaining station is located at
the US Embassy in UB. Monitoring sites are mainly located along the main avenue connecting the
east and west side of the city (Figure 1). The agencies use different equipment to measure particle
concentration: APRA uses optical particle detection (EDM180, GRIMM Aerosol, Germany);
NAMEM (MP101M, ENVEA, France) and the US Embassy (BAM-1020, Met One, US) both use
beta ray attenuation technology. The US Embassy data is provided by the US Department of State
and is not fully verified or validated. We constructed daily averages and considered it to be missing if
more than 25% of the hourly measurements were missing when calculating daily PM 2.5
concentration.
6

Figure 1. Map of study area and data sources of PM 2.5 from three organizations: APRA - Air Pollution
Reduction Agency of Ulaanbaatar; NAMEM - National Agency for Meteorology and Environmental
Monitoring; and the US embassy in Mongolia. a: Showing the 138 khoroos of UB, the extent of city zones,
and locations of monitoring stations in the context of both the continent and the nation. b: Measured daily
mean PM 2.5 showing seasonal variations over the study period (2010-2018).
7

Meteorological data
Eight monitoring sites operated by the Government agencies were co-located with weather
stations from which we obtained hourly measurements on surface temperature, atmospheric
pressure, wind speed, wind direction, and relative humidity. The mean and range of each
meteorological variable were calculated over all available sites and included in the models to reduce
missingness (about 40% in the original) of the predictors in the models since most machine learning
algorithms are sensitive to missing data. Although this approach loses the spatial resolution of
weather variables, we have deemed that completeness of the predictors is more important than the
spatial information they provide and retain them only as temporal covariates.
Land use and population data
Road length is measured by the length (m) of roads, classified as primary, secondary, or
tertiary (Urban Development Agency of UB, UDA), contained in each khoroo. City zone variables
were constructed by combining UDA’s official classification of “Central Ger area”, “Middle Ger
area”, and “Peri-urban Ger area” and categorizing them as “Ger area” while leaving the other 2
zones the same (“Urban area”, “Summer house”). For the above covariates, values assigned to
khoroos that contains the monitoring sites were also assigned to the sites. Elevation (m) data were
obtained from the data made available by NASA’s Shuttle Radar Topography Mission in 2000 (Jarvis
et al., 2008). We selected the closest elevation pixel in the case of assigning elevation to regulatory
monitoring sites. On the other hand, the average elevation of all the pixels that fall inside khoroos
was used to assign elevation to each khoroo.
Yearly number of total populations, number of households with an internet connectivity,
and number of households that live in Ger area for each khoroo were obtained from the UB
Department of Statistics. The number of stoves in each khoroo was obtained from the
“Enumeration of Air Pollution Sources in Ulaanbaatar” (Narmandakh et al., 2018) published by the
8

APRA. Total population number and stove numbers were divided by their respective khoroo areas
(km
2
) to determine the density estimate for each, while the number of households with internet
connectivity and number of households that live in Ger area were divided by the total number of
households in each khoroo to get proportion estimates. We have also constructed day of year, Julian
date variables as well as indicator variables for weekend, Mongolian public holidays, month, and
season using our dataset and author’s (T.E.) local knowledge.
Statistical analysis
We evaluated the following 6 predictive algorithms using leave-one-location-out (LOLO)
cross-validation (CV) and hold-out test set: random forest (RF) (Wright & Ziegler, 2017), gradient
boosting (GBM) (Chen et al., 2019), support vector machine with a radial basis kernel (SVM)
(Karatzoglou et al., 2004), multivariate adaptive regression splines (MARS) (Trevor Hastie &
Thomas Lumley’s leaps wrapper., 2019), generalized linear model with elastic net penalties
(GLMNET) (Friedman et al., 2010), and generalized additive model (GAM) (Wood et al., 2016). We
developed three separate models for each machine learning algorithm 1) the entire study period
(2010-2018), 2) cold-season (Oct-Mar), and 3) warm-season (Apr-Sep). Datasets were preprocessed
for optimization of each learning algorithm using R package “recipes” (Kuhn & Wickham, 2019).
For instance, while no preprocessing was done on tree-based models (RF, GBM) since decision
trees are invariant to monotonic transformations, algorithms such as SVM are sensitive to different
ranges of predictors and require normalization. Each model is trained, validated, and tuned on the
85% of the dataset and the remaining 15% of the data were held out as the test set that has not been
involved in any of the above processes at all.
Model validation and tuning
We used root mean square error (RMSE) and R
2
as our performance metrics to optimize for
LOLO CV in validation and tuning of our model hyperparameters. k-fold CV is a type of non-
9

exhaustive CV technique that validate models using k equal sized subsamples. In each iteration of k,
a single subsample is retained as a validation set and each k subsample will serve as a validation set
exactly once. Although this method is computationally efficient, they can be over-optimistic in their
estimation of performance metrics when used on spatially or temporally dependent data (Roberts et
al., 2017). Air pollution data from fixed-site regulatory monitoring stations are good examples of
spatiotemporally dependent data due to their nature. LOLO CV is a type of exhaustive CV
technique that is suitable for modeling air pollution data. We have used LOLO CV to train the
models on all but one location at a time for the number of unique locations we have in the data (9
sites in our case) and prediction errors are averaged across the repeats to give us an error estimate.
This ensures that no observation from the validation location will be involved in training the model,
unlike k-fold CV where observations are uniformly distributed among folds at random. Besides,
having relatively few locations allows us to estimate model RMSE and R
2
more accurately despite
LOLO CV’s high computational cost.
Each learning algorithm has its own set of tuning parameters and parameter values that are
optimized for a given data gives better prediction performance. In general, parameter tuning is a
process of searching through a parameter space composed of different types of grids (regular,
random) to find optimal values for better performance. We have used a space-filling design called
maximum entropy to fill our parameter grid with 30 rows. Model parameters were chosen after
training 270 models that were consisted of 9 resamples from our 9 sites for each of the 30 different
parameter combinations selected by maximum entropy. The best performing model parameters
selected by the lowest RMSE are used for fitting each model on the full training data (Table 1). The
final fitted models were then used to predict PM 2.5 on the hold-out test set to evaluate their
performance on data that were not involved in the model building process.
10

Table 1. The final tuned values for each model
Parameters Values
Random Forest
Number of predictors at each split 14
Number of trees 961
Minimum number of data points in a node 3
Gradient Boosting
Number of predictors at each split 23
Number of trees 584
Minimum number of data points in a node 7
Maximum depth of the tree 14
Learning rate 0.04
Loss reduction 0
Sample size (exposed proportion) 0.771
SVM
Cost 0.293
Radial basis function (sigma) 0.007
Epsilon 0.146
MARS
Number of terms 11
Degree of interaction 2
Type of pruning forward
GLMNET
Total amount of regularization 0.013
Proportion of regularization for the L2 penalty 0.841
GAM was modeled separately using "mgcv" package without grid search of hyperparameters.
Statistical computing
All data analysis and modeling were conducted in R 3.6.1 (R Core Team, 2019). “Tidyverse”
set of packages (Wickham et al., 2019) were extensively used in data cleaning and manipulation.
Geographic calculations in the WGS 84 / UTM Zone 48N (EPSG:32648) were carried out using the
“sf” (Pebesma, 2018) package. Cross-validation (Kuhn et al., 2019), model tuning (Kuhn, 2019), and
model fitting (Kuhn & Vaughan, 2019) were implemented with the help of “tidymodels” ecosystem
of packages.
11

Results
Summary statistics of observed PM 2.5 concentrations ( 𝜇 g/m
3
) and all the covariates used in
predicting PM 2.5 are shown in Table 2. While spatiotemporal covariates are defined at khoroo levels
spatially and yearly temporally, spatial covariates only differ by khoroo boundaries and temporal
covariates differ by an only daily change in measurement.
Table 2. Variables used to predict PM 2.5
Variables Mean SD Data Source
PM 2.5, µg/m
3

Entire Period 70.44 67.31 3 monitors from National Agency for the
Meteorological and Environmental Monitoring
(NAMEM), 5 monitors from Ulaanbaatar Air
Pollution Reduction Agency (APRA), and 1 monitor
from U.S. Embassy in Mongolia
Cold Season (Oct-Mar) 112.50 70.80
Warm Season (Apr-Sep) 26.74 16.82
Spatial
Monitor Longitude 106.88 0.07
NAMEM
Monitor Latitude 47.92 0.02
Stove Density, per km
2
525.36 559.01 APRA
Road Length, m 1371.16 1123.20
Urban Development Agency of Ulaanbaatar City
City Zones NA NA
Elevation, m 1320.11 17.10 Shuttle Radar Topography Mission, NASA
Temporal
Atmospheric Temperature,
o
C 0.13 14.92
NAMEM
Wind Speed, m/s 0.90 0.55
Wind Direction,
o
194.92 112.33
Relative Humidity, % 56.98 13.47
Surface Pressure, mmHg 866.87 5.85
Spatiotemporal
Proportion of Households with an
Internet Connection
0.39 0.31
Statistics Department of Ulaanbaatar City Proportion of Households who live
in the Ger area
0.54 0.48
Population Density, per km
2
16286.41 12422.48
We show performance metrics separately for the three different periods that models were fit,
in addition to LOLO CV and hold-out test set performance. Table 3 shows the model accuracy
(RMSE) and model consistency/correlation (R
2
) measures for each prediction algorithm. The RF
12

(LOLO CV for the entire period: RMSE = 29.52 𝜇 g/m
3
, R
2
= 0.82) and the GBM models (LOLO
CV for the entire period: RMSE = 30.02 𝜇 g/m
3
, R
2
= 0.82) have consistently the best performance
for all three models (entire period, cold and warm season) as well as for both LOLO CV and hold-
out test set.
Table 3. Comparison of model performance metrics for prediction of PM 2.5

Entire Period
(n = 12 590)
Cold Season
(n = 6 416)
Warm Season
(n = 6 175)

Model RMSE R
2
RMSE R
2
RMSE R
2

Leave-One-Location-
Out Cross Validation
Random Forest 29.52 0.82 39.58 0.72 12.26 0.49
Gradient Boosting 30.02 0.82 40.35 0.71 12.29 0.47
SVM 38.92 0.72 52.29 0.56 14.97 0.29
MARS 37.42 0.72 48.78 0.58 15.37 0.25
GLMNET 38.91 0.70 51.96 0.54 15.73 0.19
GAM 69.81 0.68 108.41 0.53 85.34 0.13
Hold-Out Test
Random Forest 12.92 0.96 21.23 0.92 7.44 0.84
Gradient Boosting 21.29 0.90 28.30 0.84 9.47 0.68
SVM 33.31 0.76 43.94 0.65 13.13 0.42
MARS 34.15 0.75 42.60 0.64 13.35 0.37
GLMNET 38.19 0.70 48.93 0.55 15.40 0.23
GAM 33.06 0.76 39.52 0.69 12.95 0.41
Overall, we saw that the models that used observations from the entire study period perform
better than the models using observations from only cold or warm seasons (Figure 2 and Figure 3).
We also observed a persistent trend of higher correlation metrics in cold season models in
comparison to higher accuracy metrics in warm-season models. In general, the performance metrics
generated from LOLO CV were lower than the metrics derived from the 15% hold-out test set. One
exception to this was the GAM, which produced relatively good performance metrics in the hold-
out test set despite yielding much worse performance metrics than the other models in LOLO CV.
13

Figure 2. Model predictions from leave-one-location-out cross-validation plotted against observed PM 2.5 for
the entire study period
14

Figure 3. Model predictions from leave-one-location-out cross-validation plotted against observed PM 2.5 for
the cold and warm seasons
15

A scatter plot between observed and predicted values from the RF model (Figure 2)
demonstrates that the model is somewhat underpredicting at higher PM 2.5 concentrations in LOLO
CV. This tendency, however, is not as noticeable in the hold-out test set for the entire period. Figure
4 displays the average seasonal predictions from the RF model overlaid on the khoroo map in the
context of the population density of UB. The predictions display higher PM 2.5 concentrations in the
north side of the city where most of the ger area is located (Figure 1) for both cold and warm
seasons.
16

Figure 4. Predictions from the Random Forest model applied to the entire study period are shown here as
cold (Oct-Mar) and warm (Apr-Sep) season averages in the context of population density of UB
17

Figure 5. Daily predicted PM 2.5 at the khoroo level in Ulaanbaatar, Mongolia shown as annual and seasonal
averages. Predictions were obtained from the Random Forest model, and seasonal maps were plotted at
different scales.
18

In addition, we calculated variable importance scores for our best performing model RF and
showed 10 variables with the highest scores in Figure 6. Temperature, wind, date variables as well as
densities of stove and primary road were the most predictive of PM 2.5 concentration in UB
according to the RF model.

Figure 6. Variable importance scores for the 10 most important variables in the Random Forest model
(descending order)

19

Discussion
We evaluated the performance of six different machine learning algorithms and used the best
performing model to predict daily PM 2.5 concentrations from 2010 to 2018 at each khoroos of
Ulaanbaatar, Mongolia, a city with a dangerous air pollution levels yet lacking in monitoring capacity.
Our study has demonstrated the feasibility of predicting ground-level PM 2.5 using machine learning
models at irregularly sized locales such as administrative areas with inadequate air pollution
monitoring network. We found that decision tree-based ensemble models such as RF and GBM had
the most predictive power in both LOLO CV and hold-out test sets. Also, we observed that the
predictions from the RF model at UB khoroos produced maps with good spatial and temporal
variations. Further, the most important variables in predicting PM 2.5 concentrations consisted of a
mix of meteorological and land-use variables.
Decision trees can be either classification or regression trees based on the outcome type.
Briefly, they are constructed by applying splitting rules on each consecutively smaller partitions
(nodes) of the tree. These split rules are usually based on variance (heterogeneity) or class diversity
(node impurity) of the nodes. Although decision trees have the benefits of being very interpretable
and including higher-order interactions, they are prone to overfitting and highly sensitive to small
data disturbances (Breiman, 1998). These limitations are significantly mitigated by using ensemble
methods that use bagging (RF) or boosting (GBM) techniques. In RF, models are forced to be
trained on random subsets of variables at each split instead of all variables, which in turn leads to
higher possibilities of different split candidates that would not have been considered otherwise.
On the other hand, GBM uses gradient descent to optimize any differentiable loss function
continuously while also training models on the subsets of the original data (Bi et al., 2019).
20

Our RF model results (LOLO CV R
2
= 0.82, Test R
2
= 0.96) are comparable or better than
similar studies that predict PM 2.5 using machine learning algorithms over different geographical areas
such as the contiguous US (Di et al., 2019; Hu et al., 2017) or China (Zhan et al., 2017), northern
California (Reid et al., 2015), British Columbia (Xu et al., 2018) as well as metropolitan areas like
Cincinnati, OH (Brokamp et al., 2018) and Tehran, Iran (Nabavi et al., 2019; Zamani Joharestani et
al., 2019). For the entire US, Hu et al. estimated an overall CV R
2
of 0.80 using RF with a
convolutional layer as a covariate (Hu et al., 2017) and Di et al. produced an average CV R
2
of 0.86
using an ensemble model that incorporated information from neural network, RF, and GBM models
(Di et al., 2019) while Zhan et al. modeled PM 2.5 across China using a geographically weighted GBM
which resulted in CV R
2
of 0.76 (Zhan et al., 2017). Multiple studies evaluated machine learning
techniques for the prediction of PM 2.5 concentrations. Reid et al. looked at the performance of 11
different algorithms to predict fine PM exposure during the 2008 Northern California Wildfires and
found that GBM performed the best with CV R
2
of 0.80 (Reid et al., 2015). In British Columbia, Xu
et al. compared 8 models and found the Cubist, RF, and GBM to have better performance with the
Cubist having the highest CV R
2
of 0.48 (Xu et al., 2018). For smaller area prediction, Brokamp et
al. used RF in the Cincinnati, OH metro area and produced a very high overall CV R
2
of 0.91
(Brokamp et al., 2018) while Nabavi et al. and Zamani Joharestani et al. found that their best CV R
2

values are 0.68 using RF and 0.81 using GBM, respectively, after evaluating a few different
algorithms in Tehran, Iran (Nabavi et al., 2019; Zamani Joharestani et al., 2019). Above studies have
used a similar set of covariates, including observed ground-level PM 2.5 levels, meteorological
parameters, land-use variables as well as different types of aerosol optical depth (AOD)
measurements. Some of the studies (Brokamp et al., 2018; Di et al., 2019; Hu et al., 2017) used the
convolutional layer as a predictor in their models to account for spatial and temporal
autocorrelations of PM 2.5 level. The main difference of our models in comparison is that we have
21

not included AOD as a predictor in our models, mainly due to preliminary analyses (not shown)
indicating almost no difference in model performance when incorporating 1 km x 1 km MAIAC
implementation of MODIS AOD (Lyapustin et al., 2018). Our previous work (Franklin et al., 2018)
explored the performance of machine learning algorithms for predicting particulate matter in
Mongolia using a high-dimensional Multi-Angle Imaging SpectroRadiometer (MISR) aerosol
measurements. We found moderate predictive performance (CV R
2
of 0.46 for PM 2.5) using SVM
and demonstrated the ability of the MISR AOD mixture set in differentiating particulate types,
including sulfates from sulfur-rich coal, in UB. There was however an issue with satellite retrievals
during the wintertime due to bright surface from snow cover, so the study mostly focused on
summertime AOD-PM 2.5 associations. The current study has more complete data in the winter and
we predict PM 2.5 into irregular administrative areas (khoroos) instead of regular grids. Future work
will incorporate additional satellite products (MODIS, MISR, MERRA2) with more sophisticated
missing observation gap-filling techniques. Furthermore, once low-cost sensors such as the
“PurpleAir” become more widely used in UB we plan to leverage their data in our models.
There have been several studies to determine the sources and compositions of particulate
matter and model air pollution exposure assessment in UB. Davy et al. determined that combustions
from coal, biomass burning, and motor vehicles are the largest contributors to fine PM
concentration and that coal combustion contribution increased significantly during winter in UB
(Davy et al., 2011). Nishikawa et al. and Batmunkh et al. reached similar conclusions that soot and
organic carbons were highly correlated during the heating season and likely a result of coal
combustion (Batmunkh et al., 2013; Nishikawa et al., 2011). Our PM 2.5 predictions to khoroo from
the RF model capture this pattern very well (Figure 1, Figure 4, and Figure 5) by displaying the
spatial difference of ger area and city area and temporal difference between cold and warm seasons
in PM 2.5 exposure within the city. Moreover, according to Ganbat et al., daytime and nighttime
22

differences in local urban winds affect the temperature inversion layer and trap pollutants in the
boundary layer during wintertime (Ganbat & Baik, 2016). This is in line with what we found from
our variable importance scores of the RF model (Figure 6) and show that the temperature and the
wind speed along with the UB-specific covariates such as density of coal stoves are important in
prediction of PM 2.5 in the city.
Further, efforts to model ambient PM 2.5 exposure for epidemiological studies have been almost
nonexistent in UB. We hope to alleviate this issue and use our RF model to predict PM 2.5 at the
khoroo level of UB in conducting epidemiological studies looking at adverse health effects of
particulate matter.
A few strengths and limitations of our study should be mentioned. First, to our knowledge
this is the first attempt to statistically model ambient PM 2.5 exposure at a small spatiotemporal scale
in UB. Second, the source and composition of particulate matter differs between developed and
developing countries and the inclusion of locale-specific covariates such as stove density, proportion
of households with an internet connectivity (a proxy for socioeconomic status) helped to model this
variability better. Third, leave-one-location-out cross-validation is a much more rigorous and
appropriate technique for spatiotemporally dependent air pollution data than the k-fold cross-
validation. We were able to utilize LOLO due to the relatively low number of monitoring stations in
our data. In terms of limitations, we were unable to incorporate AOD measurement into our model
due to large proportion of missing values caused by snow covers and clouds in UB, predicting on
irregular areas resulted in larger khoroos being assigned the same concentration level across the
whole area, and locations of the monitors were not representative of the whole city with only one
monitor measuring solely ger area PM 2.5 levels (Figure 1).
23

In conclusion, we demonstrated the strengths of utilizing machine learning algorithms to
predict PM 2.5 in a location with a sparse monitoring setting and unique pollution sources by
producing model performance metrics better or comparable to similar works. Starting in May 2019,
the UB city administration has banned the usage of raw coal within the contiguous city boundary
(excluding satellite districts). Our model will allow us to examine the impact this ban has on air
quality in the UB region, and we will be able to assess any health benefits with continued collection
of monitoring data and health records.
24

Chapter 2: Investigating the acute effects of ambient air pollution on
stillbirth risk in Ulaanbaatar, Mongolia
Abstract
Ulaanbaatar city (UB), the capital and the home to half of Mongolia’s total population, has
experienced extreme seasonal air pollution in the past two decades with fine particulate matter
(PM 2.5) levels reaching 500 𝜇 g/m
3
during winter. Based on monitoring data, particulate matter with
aerodynamic diameter less than 2.5 micrometers (PM 2.5), sulfur dioxide (SO 2), nitrogen dioxide
(NO 2), and carbon monoxide (CO) exposures were estimated in residential areas across UB using
Random Forest models. We collected individual-level data on 1093 stillbirths from UB hospital
records (2010-2013) and a surveillance database (2014-2018). Using a time-stratified case-crossover
design, we investigated whether short-term increases in daily ambient air pollutants with different
exposure lags (2 to 6 days) before delivery were associated with stillbirth. Effect estimates were
derived from a conditional logistic regression model and individual level characteristics were
analyzed for effect modification. We observed significantly elevated relative odds of stillbirth per
interquartile range increase in mean concentrations of PM 2.5 (odds ratio [OR]=1.35, 95% confidence
interval [CI]=1.07-1.71), SO 2 (OR=1.71, 95% CI=1.06-2.77), NO 2 (OR=1.30, 95% CI=0.99-1.72),
and CO (OR=1.44, 95% CI=1.17-1.77) 6 days before delivery after adjusting for apparent
temperature with a natural cubic spline during cold season (Oct-Mar). We also found increased risk
of stillbirth for women younger than 25 years of age, nulliparous and without comorbidities and
pregnancy complications during stratified analyses. Women living in an area with higher proportion
of ger households and higher stove density were associated with increased risk of stillbirth compared
to women living in apartments. We conclude that acute exposure to ambient air pollution before
delivery may trigger stillbirth and this risk is higher for certain subsets of women.

25

Background
Despite remarkable achievements in women and children’s health in the last 15 years, 2.6
million stillbirths in the third trimester occurred worldwide in 2015 (Lawn et al., 2016). Stillbirth was
not included in the Millennium Development Goals (United Nations, 2015a) and is still missing
from the Sustainable Development goals (United Nations, 2015b). With the issue of stillbirth is
hardly mentioned in policies and programs of international and national organizations, it remains an
under financed public health concern. The burden of stillbirth affects women and families by
causing psychological and emotional distress, and negatively influences communities and society in
terms of reduced earnings and health-care expenses. An estimated 4.2 million women are living with
depression associated with a previous stillbirth (Heazell et al., 2016). Evidence of adverse effects of
ambient air pollution on birth outcomes such as preterm birth and low birth weight has been
summarized in several reviews (Jacobs et al., 2017; Li et al., 2017; Sapkota et al., 2012; Stieb et al.,
2012). In comparison, there have been relatively few studies that investigated the association
between ambient air pollution and stillbirth. A recent review based on 13 studies found suggestive
evidence (increased effect estimate, but not statistically significant at 0.05 level) of increased risk on
stillbirth due to chronic exposure to both gaseous and particulate pollutants (Siddika et al., 2016).
Although the biologic mechanism by which ambient air pollutants, particularly fine
particulate matters (PM 2.5) may influence fetal survival is not yet well understood, there are several
processes that have been suggested to describe associations between PM and adverse pregnancy
outcomes. One possible pathway indicates that free iron ions on particle surfaces can react with
super oxide or hydrogen peroxide to generate highly reactive hydroxyl radicals which in turn causes
excessive oxidative stress that may lead to degradation of lipids, proteins and DNA of placenta. In
addition, PM-mediated oxidative stress can be caused by the activation of the inflammation system.
Release of pro-inflammatory cytokines are promoted by immunotoxic compounds and they in turn
26

give a positive feedback loop to form more reactive oxygen species (ROS) and oxidative stress (Al-
Gubory et al., 2010). Another explanation is that DNA damage induced by pollution may have
devastating impacts on development of a fetus whose cells are dividing at a high rate. Studies
connecting air pollution to congenital malformations may provide some support to this pathway
(Padula, Tager, Carmichael, Hammond, Lurmann, et al., 2013; Vrijheid et al., 2011). Other studies
indicate that air pollution disturbs placental health through injury or inflammation, initiating serious
lack of nutrient transfer between mother and fetus (van den Hooven et al., 2012). To accurately
examine the potential mechanistic explanations, additional analyses are required to verify these
findings and to explore more thoroughly the times during pregnancy when pollutants may impact
the fetus resulting in stillbirth.
Ulaanbaatar (UB), the capital city of Mongolia, is in the Tuul river valley between big
mountains and resultingly prone to atmospheric inversions in the winter, whereby air pollutants are
trapped close to the ground, unable to become diluted in the mixing layer of the troposphere. The
fact that almost 60% of the city residents use raw coal, a coal extracted from the seam and not
processed or washed, to heat their home and cook their meals in cook stoves makes UB one of the
most polluted cities in the world during winter when there is almost constant atmospheric inversion
over the city. Our study area has 4 major public maternity hospitals that deliver more than 98% of
all babies in the city. This situation makes UB a unique location where stillbirth can be studied in
relation to air pollution in a relatively cost-effective and timely manner despite its average level of
stillbirth rate (7.7 per 1000 total births in 2015) (Lawn et al., 2016).
In this paper, we examined the short-term effect of particulate matter on the risk of stillbirth
across lags of 2-6 days and investigated effect modification of these associations by
sociodemographic risk factors of stillbirth.
27

Methods
Ulaanbaatar city registered about 36 000 deliveries and 250 stillbirths on average per year
between 2010 and 2018 across 9 districts, which are consisted of varying number of khoroos (the
smallest administrative units of the city). The current study includes stillbirths occurring between
January 1, 2010 and December 31, 2018 in 6 contiguous districts (138 khoroos) of UB. The study
design and methods were approved by the Institutional Review Boards at the Health Sciences
Campus of USC and the Children’s Hospital Los Angeles as well as the Medical Ethics Committee
of the Ministry of Health of Mongolia.
Study population
We used two different data sources to collect individual-level stillbirth data. Between 2010
and 2013, records were only available at maternity hospital archives in the form of a birth record
since there was no official surveillance of stillbirth during this period. In order to collect these data,
8 research assistants were hired and trained to help with data abstraction from the paper-based birth
records into a computer database. In August 2017, we abstracted the stillbirth data using 4 two-
person teams where one person read paper records and the other person entered it into the
database. Every record entry was double checked by each team member. In 2014, the Mongolian
Surveillance Department that conducts nationwide surveillance on maternal and child health
indicators was established at the National Center for Maternal and Child Health (NCMCH). They
use specific forms for each maternal and child health endpoint to collect data. We obtained
individual level stillbirth data between 2014 and 2018 using the already digitized database of stillbirth
forms from the Surveillance Department.
See Appendix A for a translated version of the form.
The following information was abstracted from birth records and stillbirth form: maternal
age, residential address, occupation, gravidity, parity, prenatal care status, delivery hospital, date of
28

stillbirth, fetal gender, weight, length and gestational age. Gestational age was defined by the best
obstetric estimate variable on the birth record, which combines last menstrual period and ultrasound
parameters, as is commonly accepted in clinical practice for gestational age estimation. According to
the Ministry of Health of Mongolia, stillbirth is defined as the birth of a fetus with ≥500 grams
weight at or beyond the 22nd week of gestation and who had no signs of life such as heartbeat,
umbilical cord pulse, or muscle movement at birth (Ministry of Health, Guideline for the Calculation
and Definition of Main Health Indicators, 2004, page 19). We included stillbirths based on singleton
status, maternal residence in one of the 6 main districts, delivery at 4 major public maternity
hospitals (Figure 7). Based on biological plausibility, we excluded intrapartum stillbirths, defined as
fetal death occurring after the onset of labor and prior to delivery (Tavares Da Silva et al., 2016).
Based on the residential khoroo of each subjects, we also obtained khoroo level variables such as
percentage of households with internet connectivity (proxy for socioeconomic status) and
percentage of ger households from the UB Office of Statistics, and number of stoves per square
kilometer from the UB Office of Air Pollution Reduction (Narmandakh et al., 2018).
29

Figure 7. Study map with number of stillbirth cases
Exposure data
We used daily air pollution estimates derived from a previously developed and validated
spatiotemporal model (Chapter 1) at residential khoroos and relevant dates for each stillbirth case.
Briefly, our model is based on ground level measurements from stationary stations supplemented
with multiple spatiotemporal covariates. They include meteorological (atmospheric temperature,
relative humidity, wind speed, wind direction), land use (length of roads) as well as population (ger
30

households, stove density) and indicators for time (day of year, season, month). We evaluated 6
different machine learning algorithms and found that decision trees, particularly Random Forest
models, performed the best with leave-one-location-out cross-validation R
2
of 0.82 and hold-out test
R
2
of 0.96 for PM 2.5. The Random Forest model was used to predict daily levels of PM 2.5, SO 2, NO 2,
and CO at each khoroo of UB city between 2010 and 2018. The model was trained on data from
only the 6 contiguous districts of UB because the remaining 3 districts are not adjacent to the other
districts and do not have ground monitoring stations. In addition, we retrieved meteorological
variables from two stations (Chinggis Khaan Intl, Songiin) in UB that sends their info to the Global
Hourly Integrated Surface Dataset of the National Oceanic and Atmospheric Administration
(NOAA National Centers for Environmental Information, n.d.). Using these data, we calculated
Apparent Temperature (AT) as either one of wind chill index or heat index depending on the
temperature. Wind chill combines temperature and wind velocity to capture how cold the weather
feels to the average person when the temperate falls below 0°C. We used equations used by
Environment and Climate Change Canada (Nelson et al., 2002) since Mongolian National Agency
for Meteorology and Environmental Monitoring (NAMEM) currently does not have a formula
adopted. Equation 1 is used when the temperature of the air is ≤ 0°C and the reported wind speed is
≥ 5 km/h and equation 2 is used when the temperature of the air is ≤ 0°C and the reported wind
speed is > 0 km/h but < 5 km/h.
Equation 1:
𝑊 = 1 3 . 1 2 + 0 . 6 2 1 5 × 𝑇 − 1 1 . 3 7 × 𝑉 . + 0 . 3 9 6 5 × 𝑇 × 𝑉 .
Equation 2:
𝑊 = 𝑇 + [ ( − 1 . 5 9 + 0 . 1 3 4 5 × 𝑇 ) / 5 ] × 𝑉
31

Where 𝑊 is the wind chill index in degrees Celsius, 𝑇 is the air temperature in degrees
Celsius (°C), and 𝑉 is the wind speed at 10 metres (standard anemometer height), in kilometres
per hour (km/h).
Heat index is a measure of heat exposure and combines temperature and relative humidity to
obtain perceived temperature when the weather is hot. We used US National Weather Service
algorithm implemented by the R package “weathermetrics” (Anderson et al., 2013) to calculate AT
when the atmospheric temperature is more than 5°C.
Study design
A time-stratified case-crossover design was used to estimate the relative odds of stillbirth
associated with each interquartile range (IQR) increase in mean pollutant concentrations. The case-
crossover design was first proposed by Maclure (Maclure, 1991) to study transient effect of an
exposure on rare-acute onset disease. Conceptually, the design resembles a combination of
retrospective nonrandomized crossover design and matched case-control design. The main
difference is that the case-crossover design only uses a sample of the base population-time by
including only cases, so that it allows control of time-independent confounders within subjects.
Effect estimates are derived from the comparison between exposure just before the event and
exposure at other control (referent) times. Choosing referent times in air pollution case-crossover
studies is particularly important due to time-dependent confounders, time trends, and
autocorrelation in air pollution exposure series. The time-stratified referent scheme is the only
referent selection that can avoid both overlap bias due to conditional logistic regression estimating
equations and time-trend bias by selecting random index dates based on stratification (Janes et al.,
2005). We stratified on year, month and day of the week to control for long-term, seasonal, and day
of the week trends using time-stratified referent selection. We selected referent (control) periods to
32

be every seventh day during the same month as the stillbirth (case period). For instance, if the case
period was on Wednesday, every other Wednesday in the same month were considered as control
periods. We looked at lag days between 2 and 6 as possible case periods because it may take an
average of 48-70 hours for the deceased fetus to be expelled from the womb (Gardosi et al., 1998;
Genest et al., 1992).
Statistical analysis
We utilized conditional logistic regression models to estimate the odds ratios (OR) of
delivering stillbirth for every IQR increase in PM 2.5 levels on lag days 2 to 6. Models were estimated
using a stratified Cox model with each subject assigned to its own stratum (Gail et al., 1981). We
adjusted each model with a natural cubic spline (4 degrees of freedom) of the mean apparent
temperature on corresponding lag days to control for time-variant meteorological factors and their
non-linear relationship with stillbirth as is usually done in previous studies (Faiz et al., 2013). We also
conducted a seasonally stratified analysis, restricting the models to cases that occurred only in the
cold season (October-March) considering the disproportional seasonal variation caused by air
pollution emission sources in UB during cold weather (Figure 8). The same models were fit to SO 2,
NO 2, and CO with their ORs and 95% confidence intervals (CIs) scaled to the IQR of each
pollutant. To evaluate differential risks of stillbirth by individual case characteristics, we conducted
stratified analyses based on maternal age, employment, number of previous pregnancies and
deliveries, comorbidity, pregnancy complication as well as residential level covariates such as
percentage of ger households and households with internet connectivity in each subject’s living
khoroo. All statistical modeling and geospatial computing were performed in R version 3.6.2 (R
Core Team, 2019) using packages “tidyverse” (Wickham et al., 2019), “sf” (Pebesma, 2018), and
“survival” (Therneau & Grambsch, 2000).
33

Figure 8. Seasonal variation of modelled PM 2.5 concentration in UB
Results
Our study population is consisted of 1093 stillbirth cases occurring between 2010 and 2018
after applying inclusion and exclusion criteria (Figure 9).
34

Figure 9. Flow chart of study population
Almost a quarter of stillbirths occurred among women aged 35 years or older and the
subjects divided equally among employed and unemployed status. More than half (54%) of the
women had been pregnant more than 2 times, however a third (35%) of the women had never given
a birth before. About 65% of the subjects had one or more types of pregnancy complications and
approximately half of them had various comorbidities as well. In terms of fetal variables, there was
no dominant fetal gender, the majority weighed less than 2500 grams and had gestational weeks less
than 36 weeks (Table 4). Table 4 also shows that majority of the cases came from khoroos with
lower internet connectivity, more ger households, and higher stove density.

35

Table 4. Characteristics of Mothers and Stillbirths, Ulaanbaatar 2010–2018
[ALL] N
N=1093
Maternal covariates
Age (years), N (%): 1093
<25 286 (26.2%)
25-34 539 (49.3%)
>=35 268 (24.5%)
Employment, N (%): 1088
Employed or Student 574 (52.8%)
Unemployed 514 (47.2%)
Number of pregnancies, N (%): 1093
2 or less 500 (45.7%)
More than 2 593 (54.3%)
Number of deliveries, N (%): 1092
1 or more 712 (65.2%)
None 380 (34.8%)
Comorbidities, N (%) 504 (48.8%) 1033
Pregnancy complications, N (%) 649 (65.2%) 995
Fetal covariates
Gender, N (%): 1092
Female 528 (48.4%)
Male 564 (51.6%)
Weight (grams), N (%): 1092
>=4000 50 (4.58%)
2500-3999 435 (39.8%)
500-2499 607 (55.6%)
Gestational week, N (%): 1078
22-26 weeks 60 (5.57%)
27-36 weeks 566 (52.5%)
>=37 weeks 452 (41.9%)
Residential covariates
Percentage of households with internet connectivity, N (%): 1088
45% or less 823 (75.6%)
more than 45% 265 (24.4%)
Percentage of ger households, N (%): 1088
20% or less 266 (24.4%)
more than 20% 822 (75.6%)
Number of stoves (per sq.km.), N (%): 1088
15 stoves or less 271 (24.9%)
more than 15 stoves 817 (75.1%)
36

The models looking at overall period showed consistently increased risk of stillbirth on 3 and
6 days before delivery for all pollutants examined (Table 5 and Figure 10). Odds ratios on lag day 6
were generally higher than the estimates on lag day 3. Only carbon monoxide had statistically
significant elevated risk on lag day 6 at 𝛼 =0.05 significance level. In models restricted to the cold
season, most of the effect estimates except NO 2 increased considerably. Previously non-significant
odds ratios on lag day 6 in overall models reached statistical significance for PM 2.5 (OR:1.35, 95%
CI:1.07-1.71) and SO 2 (OR:1.71, 95% CI:1.06-2.77). We also found a statistically significant
increased risk of stillbirth per IQR increase in mean PM 2.5 3 days before delivery (Figure 10).

37

Table 5. Relative Odds of Stillbirth Associated with IQR Increases in Mean PM 2.5, NO 2, SO 2, and CO
Concentration by Lag day(s)
Pollutants Lag Days IQR, µg/m
3
OR
1
(95% CI)
2
OR (95% CI)

Overall (n = 1 088)
3
Cold season (n = 515)
3

PM 2.5
2 74.67 0.99 (0.8-1.23) 0.96 (0.76-1.2)
3 77.00 1.23 (0.98-1.54) 1.28 (1-1.62)
4 75.24 0.92 (0.74-1.14) 0.95 (0.76-1.19)
5 74.65 1.01 (0.81-1.26) 1.05 (0.83-1.31)
6 75.72 1.24 (0.99-1.56) 1.35 (1.07-1.71)
SO 2
2 36.40 0.94 (0.61-1.44) 0.84 (0.53-1.33)
3 37.12 1.26 (0.81-1.97) 1.33 (0.82-2.14)
4 35.82 0.94 (0.61-1.46) 1.02 (0.64-1.63)
5 35.73 0.87 (0.56-1.36) 1.00 (0.63-1.61)
6 36.32 1.47 (0.95-2.3) 1.71 (1.06-2.77)
NO 2
2 29.89 1.01 (0.79-1.29) 0.89 (0.68-1.16)
3 30.55 1.18 (0.92-1.52) 1.14 (0.86-1.5)
4 29.73 1.02 (0.8-1.29) 0.94 (0.72-1.23)
5 30.12 1.14 (0.89-1.45) 1.12 (0.85-1.47)
6 30.14 1.21 (0.95-1.55) 1.30 (0.99-1.72)
CO
2 561.03 1.03 (0.87-1.24) 1.05 (0.86-1.27)
3 572.20 1.18 (0.98-1.42) 1.21 (0.99-1.47)
4 560.32 1.01 (0.85-1.21) 1.04 (0.86-1.26)
5 556.54 1.00 (0.83-1.19) 1.04 (0.86-1.27)
6 571.54 1.25 (1.04-1.51) 1.44 (1.17-1.77)
1
Odds Ratio
2
95% Confidence Interval
3
Models are adjusted for apparent temperature.
38

Figure 10. Effect estimates of air pollution on stillbirth risk
Possible modifiers of effect associated with stillbirth and pollutants examined in stratified
models (Table 6) show significant disparities in odds ratios between various maternal and residential
characteristics. They reveal stronger associations for younger, unemployed, women with fewer
pregnancies, and women who had never given birth before. Interestingly, we also see higher odds
ratios for women without comorbidities and pregnancy complications. On the other hand, living in a
39

khoroo with larger percentage of ger households and households without internet connectivity as
well as higher stove density is associated with higher relative odds of stillbirth.
Table 6. Relative Odds of Stillbirth Associated with IQR Increase in PM 2.5, NO 2, SO 2, and CO Concentration
on Lag Day 6 by Level of Maternal Characteristics

PM2.5 SO2 NO2 CO
Characteristic OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI)
Maternal age (years)
<25 1.45 (0.94-2.25) 2.38 (1.02-5.57) 1.22 (0.8-1.88) 1.52 (1.04-2.21)
25-34 1.19 (0.86-1.66) 1.25 (0.65-2.4) 1.20 (0.83-1.73) 1.21 (0.91-1.6)
>=35 1.17 (0.77-1.78) 1.20 (0.51-2.83) 1.21 (0.73-2.02) 1.15 (0.81-1.62)
Maternal employment
Unemployed 1.43 (1.01-2.03) 1.89 (0.98-3.62) 1.53 (1.06-2.2) 1.49 (1.11-1.98)
Employed or Student 1.12 (0.83-1.51) 1.19 (0.64-2.2) 0.98 (0.7-1.37) 1.12 (0.87-1.43)
Number of pregnancies
2 or less 1.67 (1.19-2.34) 2.59 (1.33-5.03) 1.40 (1-1.96) 1.54 (1.15-2.04)
More than 2 0.98 (0.73-1.33) 0.92 (0.51-1.68) 1.04 (0.73-1.48) 1.07 (0.84-1.38)
Number of deliveries
None 1.74 (1.2-2.52) 2.76 (1.32-5.77) 1.39 (0.96-2.01) 1.64 (1.2-2.23)
1 or more 1.02 (0.77-1.36) 1.00 (0.57-1.76) 1.08 (0.78-1.49) 1.07 (0.84-1.35)
Maternal comorbidity
Yes 1.13 (0.83-1.55) 1.07 (0.56-2.05) 1.13 (0.81-1.6) 1.25 (0.96-1.62)
No 1.39 (1-1.93) 1.97 (1.05-3.7) 1.23 (0.86-1.77) 1.25 (0.95-1.64)
Pregnancy complication
Yes 1.12 (0.84-1.5) 1.17 (0.67-2.03) 1.07 (0.79-1.46) 1.05 (0.82-1.34)
No 1.46 (0.98-2.19) 2.44 (1.07-5.58) 1.56 (1-2.44) 1.57 (1.11-2.23)
Percentage of households with internet connection
45% or less 1.37 (1.05-1.78) 1.81 (1.1-2.97) 1.29 (0.98-1.69) 1.41 (1.13-1.77)
more than 45% 0.99 (0.65-1.51) 0.53 (0.19-1.5) 0.86 (0.49-1.53) 0.94 (0.65-1.35)
Percentage of ger households
20% or less 1.10 (0.72-1.68) 0.80 (0.28-2.28) 1.04 (0.62-1.74) 0.89 (0.61-1.3)
more than 20% 1.32 (1.01-1.72) 1.63 (1-2.68) 1.25 (0.94-1.65) 1.42 (1.14-1.77)
Number of stoves (per sq.km.)
15 stoves or less 0.93 (0.61-1.44) 0.58 (0.2-1.66) 0.81 (0.49-1.35) 0.83 (0.56-1.23)
more than 15 stoves 1.39 (1.07-1.81) 1.77 (1.08-2.9) 1.37 (1.03-1.82) 1.42 (1.14-1.77)
Discussion
We found substantially increased risk of stillbirth 3 and 6 days before delivery associated
with IQR increases in the mean concentrations of all pollutants we examined. These associations
40

were controlled for time-varying confounders such as atmospheric temperature, relative humidity,
and wind velocity in the form of apparent temperature. By virtue of time-stratified case-crossover
design, time-invariant confounders such as individual level maternal, fetal, and residential attributes
as well as seasonal, monthly and day of the week time-trends could not have influenced our results.
Our stratified analyses showed disproportionally higher risk for women living in low income ger
districts compared to women living in apartments. They also indicate younger women who have not
given birth before and have no comorbidities and pregnancy complications are at increased risk of
stillbirth associated with ambient air pollution.
The biological mechanisms underlying the effect of air pollutants on stillbirth risk have
varying levels of evidence depending on the pollutant. On one hand, there is a well-established
toxicity of carbon monoxide on the fetus (Penney, 1996). Two main pathways are reduction of
oxygen-carrying capacity of maternal hemoglobin by CO, which leads to oxygen shortage in fetal
blood (Salam et al., 2005), and higher affinity of fetal hemoglobin for binding CO than adult
hemoglobin that further compromises the oxygen delivery (Sangalli et al., 2003). On the other, we
have scarce evidence of how exactly NO 2 and SO 2 might affect fetal death. There is evidence for
crossing placental barrier and affecting cell division as well as activation of hypoxic injury or
immune-mediated damage (Al-Gubory et al., 2010; Proietti et al., 2013). For particulate matter, there
have been several hypotheses regarding their effect on fetus. Major ones are compromise of
maternal blood and nutrients delivery, DNA damage and inflammation via contribution to oxidative
stress, and lowering of the transplacental function by increasing concentration of DNA adducts
(Perera et al., 1992). Timing of exposure to particulate matter along the fetus developmental periods
may also lead to varying effects due to differences in physiologic maturity of the fetus (Perera et al.,
1999).
41

While there have been numerous studies examining the association between air pollutants
and adverse pregnancy outcomes, relatively few investigated the effect on stillbirth, and even fewer
are focused on the acute effects of air pollution on stillbirth risk. A time-series study in Sao Paulo,
Brazil examined the association between daily counts of intrauterine mortality (gestational duration
of >28 weeks) and ecological measures of daily ambient air pollutants. They found increased short-
term (less than 5 days) risk of intrauterine mortality for NO 2 ( 𝛽 =0.0013/ 𝜇 g/m
3
; p<0.01), SO 2
( 𝛽 =0.0005/ 𝜇 g/m
3
; p<0.10) and CO ( 𝛽 =0.0223/ppm; p<0.10) using Poisson regression adjusted for
season and weather (Pereira et al., 1998). This study, to our knowledge, is the first to look at short-
term effect of air pollutants on stillbirth risk and remains an important part of evidence base despite
its limitations due to ecological design and imperfect case definition.
A study by Faiz et al. utilized time-stratified referent selection in a case-crossover design to
look at the triggering effect of ambient air pollution on stillbirth in New Jersey, US (Faiz et al.,
2013). They found significantly increased relative odds of stillbirth on lag day 2 per IQR increase in
mean concentrations of CO (OR = 1.20, 95% CI = 1.05-1.37) and SO 2 (OR = 1.11, 95% CI = 1.02-
1.22) and increased odds ratios for IQR increases in NO 2 (OR = 1.11, 95% CI = 0.97–1.26) and
PM 2.5 (OR = 1.07, 95% CI = 0.93–1.22) levels. Similar increases in stillbirth risk were also observed
for cumulative averages on days 2 to 6 and they did not observe effect modifications by maternal
characteristics. The current study tries to emulate above study conceptually in order to investigate
the acute effect of ambient air pollutants on the risk of stillbirth. Our findings agree on increased
short-term risk of stillbirth associated with IQR increase in the pollutants. They, however, differ in
terms of which lag day has the highest risk and the magnitude of the risk. We found increased
relative odds of stillbirth on lag days 3 and 6 for all pollutants and the ORs were amplified when
restricted to cold season only. Unfortunately, there was almost no overlap between individual
maternal risk factors between the studies to compare stratified analyses.
42

Two recent studies on the short-term association between stillbirth and ambient pollutants
have been published. Mendola et al. used retrospective cohort data across 12 clinical centers in the
US and investigated the acute and chronic effects of air pollutants on the risk of stillbirth at the
community level using Poisson regression with generalized estimating equations. They observed that
acute exposure to ozone was associated with a 13-22% increased risk of stillbirth on lag days 2, 3,
and 5-7 (Mendola et al., 2017). A time-series study in Ahvaz, Iran by Dastoorpoor et al. used a
distributed lag non-linear model estimated by quasi-Poisson regression to investigate the acute
effects of air pollution (per 10-unit increase) on stillbirth and other adverse pregnancy outcomes.
They looked at single day lags of 1 and 2 as well as cumulative lag days 0 through 14 and did not
find an increased effect estimate, but rather an inverse association with ozone on lag day 1 and 2 and
with SO 2 on lag day 2 (Dastoorpoor et al., 2018).
Our findings are largely consistent with previous literature on the topic. However, there are
several differences that should be mentioned. The magnitude of the effect that we observed is
substantially larger than the aforementioned studies. We believe this is mostly because of large IQR
stemmed from extremely high air pollution levels during cold months in UB. Unfortunately, this
kind of change in daily air quality over days and weeks during winter in UB is entirely possible and in
line with significant health and mortality burden due to air pollution in UB found in previous
studies. Allen et al. conservatively estimated that 1 in 10 mortality in UB is due to air pollution using
land-use regression models and mobile monitoring to assess exposure (Allen et al., 2013) while
researchers from the National Center for Maternal and Child Health (NCMCH) found a strikingly
high seasonal correlation between miscarriage and air pollution levels in the city (Enkhmaa et al.,
2014).
43

The main improvement of our study over previous literature comes from our individual level
exposure assessment with fine spatial and temporal resolution. The main challenge in conducting air
pollution epidemiology studies in developing countries comes from the difficulty in assessing
exposure accurately due to inadequate and unreliable measurements provided by scarce pollutant
monitoring stations. In our previous work, we demonstrated the feasibility and advantages of
utilizing machine learning algorithms in capturing complex relationship between air pollution and
other meteorological, land use, and population level variables in a low-resource setting with sparse
monitoring capacity. This work allowed us to assign air pollution exposure to every subject in our
study based on their residential area and relevant days. Although the exposure measurement errors
are unavoidable when we try to estimate personal exposure using modeling based on ambient air
pollution levels, we believe this approach has made significant improvements over previous methods
of assigning exposures and the error will result in non-differential misclassification likely leading to
bias towards the null. We should also mention that we were able to evaluate various modifiable
maternal risk factors that previous studies have not been able to look at. This is mainly thanks to the
manual data collection process from birth records as well as stillbirth form information provided by
the Surveillance Department of NCMCH. The limitations of our current study include exposure
assessment at residential area (khoroo) level, the exact date of fetal death is only an estimate based
on the delivery date as well as possible residual confounding due to imperfect measurement of
apparent temperature.
In conclusion, we found considerably increased risk of stillbirth on 3 and 6 days before
delivery for every IQR increase in mean concentrations of PM 2.5, SO 2, NO 2, and CO. This
association was strengthened when we restricted to analysis to only cold season. Future studies with
finer spatial and temporal exposure and time-varying confounder assessment and better case
44

identification are necessary to further elucidate the underlying mechanism of air pollution effect on
stillbirth.

45

Chapter 3: Acute effect of ambient air pollution on select congenital
anomalies: a case-crossover study
Abstract
A symmetric bidirectional case-crossover study examined the association between short-term
ambient air pollution exposure during weeks 3-8 of pregnancy and select congenital anomalies
occurred in Ulaanbaatar, Mongolia between 2014 and 2018. By using predictions from a Random
Forest regression model, authors assigned daily ambient air pollution exposure of particulate matter
<2.5 um aerodynamic diameter, sulfur dioxide, nitrogen dioxide, and carbon monoxide for each
subject based on their residential administrative area. Conditional logistic regression with adjustment
for corresponding apparent temperature was used to estimate relative odds of select congenital
anomalies per IQR increase in mean concentrations and quartiles of air pollutants. The adjusted
relative odds of cardiovascular defects (ICD-10 subchapter: Q20-Q28) was 2.64 (95% confidence
interval: 1.02-6.87) per interquartile range increase in mean concentrations of particulate matter <2.5
um aerodynamic diameter for gestational week 7. This association was further strengthened for
cardiac septal defects (ICD-10 code: Q21, odds ratio: 7.28, 95% confidence interval: 1.6-33.09) and
isolated ventricular septal defects (ICD-10 code: Q21.0, odds ratio: 9.87, 95% confidence interval:
1.6-60.93). We also observed increasing dose-response trend when comparing the lowest quartile of
air pollution exposure with higher quartiles on week 6 and 7 for Q20-Q28 and Q21 and week 4 for
Q21.0. Our findings contribute to the limited body of evidence regarding acute effect of ambient air
pollution exposure on adverse birth outcomes.

46

Background
Congenital anomalies pose significant risk of infant mortality, life-long illness and disability,
and have a heavy burden on resource-scarce countries around the world. Worldwide estimates show
that about 13% of total mortality among children under 5 years of age were due to congenital
anomalies (CA) in 2016 (World Health Organization, 2017). The most common forms of CA
include congenital heart defects, neural tube defects, hemoglobin disorders, thalassemia, and sickle
cell disease. Although CA are a global problem, 94 percent of the births with serious birth defects
and 95 percent of the mortality among these children occur in middle and low-income countries
(Christianson et al., 2005). This significant difference is partly explained by maternal health and
other risk factors such as poverty, a higher proportion of older mothers, and a larger frequency of
consanguineous marriages. In terms of etiology, about half of all major CA are unknown and
hypothesized to have multiple causal factors, including environmental exposures (Christianson et al.,
2005).
Our previous work looking at the short-term effect of air pollution on the occurrence of
stillbirth (Chapter 2) found significantly increased risk of stillbirth on 3 and 6 days before delivery.
In comparison to stillbirth, there have been a larger number of studies investigating the association
between air pollution and CA. Ritz et al. used data from the California Birth Defects Monitoring
Program to look at associations between CA and average monthly ambient air pollution, which to
our knowledge, was the first attempt to explore these associations using case-control design with
individual level data for confounding controls. They found increased risk of cardiac ventricular
septal defects with a dose-response trend as well as increased risks for a number of other CA (Ritz et
al., 2002). Since then significant number of case-control studies have been conducted in the US,
including Texas (Gilboa et al., 2005), New Jersey (Marshall et al., 2010) and California (Padula,
Tager, Carmichael, Hammond, Lurmann, et al., 2013; Padula, Tager, Carmichael, Hammond, Yang,
47

et al., 2013; Padula et al., 2015) and around the world, including Taiwan (Lin et al., 2014), the UK (P.
Dadvand et al., 2011; Dadvand et al., 2011; Dolk et al., 2010; Rankin et al., 2009), Australia (Hansen
et al., 2009), Israel (Agay-Shay et al., 2013), Spain (Schembari et al., 2014), and Italy (Gianicolo et al.,
2014). The results of these studies varied due to design heterogeneities derived from different case
ascertainment procedures and exposure assessment methodologies. A few reviews attempted to
synthesize the findings and found suggestive evidence of increased congenital cardiac anomaly risk
associated with NO 2, SO 2, and PM 2.5 (Vrijheid et al., 2011) as well as a significant association
between NO 2 and coarctation of the aorta (Chen et al., 2014). While these studies were able to
utilize state or country-wide birth defect surveillance systems for outcome ascertainment, the
exposure assessment was usually limited to either spatial gradient based on distance from monitoring
stations or temporal gradient based on gestation periods. This may lead to substantial exposure
misclassification which in turn exacerbates the residual confounding problem with using registry
data with limited covariate information.
Although the biological mechanism of how air pollutants may cause birth defects is not yet
certain, we have several plausible pathological pathways based on laboratory studies. Several studies
showed that products of polycyclic aromatic hydrocarbons (PAH) attach to the DNA activating
DNA repair processes and producing DNA adducts which in turn can change gene function and
alter fetal development (Perera et al., 1999). Oxidative stress and inflammatory cytokines triggered
by air pollution can adversely affect fetus via change in placental functions such as trophoblast
proliferation and differentiation as well as apoptosis (Jonakait, 2007; Kannan et al., 2006; Slama et
al., 2008). The other possible mechanisms relate to phenotypical changes caused by air pollution due
to epigenetic processes such as DNA methylation and acetylation, histone modifications and
microRNA expressions etc. (Baccarelli & Bollati, 2009; Hou et al., 2012). Some studies
48

demonstrated that decreased placental or global methylation are associated with increased level of
PM 2.5 and PAH (Herbstman et al., 2012; Janssen et al., 2013).
Ulaanbaatar (UB), the capital city of Mongolia, has hazardous levels of air pollution during
winter due to most of its residents burning raw coal combined with an atmospheric inversion that
traps emissions within the breathing zone. Although Mongolian infant mortality rates have
substantially declined in the last 20 years (63.4 per 1000 live births in 1990 to 13.4 per 1000 live
births in 2018), the proportion of deaths due to congenital anomalies have increased in the last 5
years from 12.7 percent in 2014 to 16 percent in 2018 (Enkhtur & Bayalag, 2019). We have seen
some alarming results indicating high attributable mortality (Allen et al., 2013) due to air pollution
and strong correlation with miscarriage (Enkhmaa et al., 2014) in the city. In our previous studies,
we have tried to minimize the limitations of a resource-poor country like Mongolia in terms of
exposure (inadequate and unreliable monitoring network) and outcome (overworked and
underfinanced healthcare system) ascertainment by using exposure assessment via machine learning
algorithms (Chapter 1) and utilization of surveillance database.
The objective of the current study is to investigate the short-term effects of ambient air
pollution on the risk of delivering a newborn with select congenital anomalies in Ulaanbaatar,
Mongolia. Specifically, we examined 1) different gestational weeks of exposure periods, and 2) risks
of select congenital anomalies according to ICD-10 for both continuous and categorical (dose-
response) exposure.
Methods
Ulaanbaatar city registered about 200 000 live births in total between 2014 and 2018. During
this period, there were 1487 cases of CA which resulted in rate of 7.3 per 1000 live births. This is
slightly lower than the national average of 8.1 per 1000 live births. The current study includes CA
49

occurring between January 1, 2014 and December 31, 2018 in UB. The study design and methods
were approved by the Institutional Review Boards at the Health Sciences Campus of USC and the
Children’s Hospital Los Angeles as well as the Medical Ethics Committee of the Ministry of Health
of Mongolia.
Study population
We obtained individual level CA data from the Surveillance Department of the National
Center for Maternal and Child Health of Mongolia. The department was established in 2014 by the
Ministry of Health order to set up a nationwide active surveillance system for maternal and child
health indicators. Details of the data collection process are described in Chapter 2 and a translated
version of the data collection form for CA can be found in Appendix B. The following information
was abstracted from the form: maternal age, residential address, education, occupation, parity,
smoking status, folic acid usage, date of birth, fetal gender, weight, and gestational age. Gestational
age was defined by the best obstetric estimate variable in the birth record, which combines last
menstrual period and ultrasound parameters, as is commonly accepted in clinical practice for
gestational age estimation. Congenital anomalies are defined as structural or functional anomalies
that occur during the intrauterine period and can be identified prenatally, at birth, or sometimes may
only be detected later in infancy (World Health Organization, n.d.). For the current study, eligible
cases were singleton births, having a maternal residential address in one of the 6 main districts, and
were non-chromosomal CA (Figure 11) having one of 7 CA categories. However, due to low
number of cases, we were only able to analyze 5 subchapters of ICD-10 chapter Q00-Q99
(Congenital malformations, deformations and chromosomal abnormalities), 1 major diagnosis, and 1
isolated defect (Table 7) in this study (World Health Organization, 2004). We also added khoroo
level variables such as percentage of households with internet connectivity (proxy for socioeconomic
50

status), percentage of ger households, and number of stoves per square kilometer based on the
residential khoroo of each subject.

Figure 11. Flow chart of the study population
Exposure data
We used daily air pollution estimates derived from a previously developed and validated
spatiotemporal model (Chapter 1), linking them to the residential khoroo and averaged over the
week for the relevant dates of each CA case. We also calculated apparent temperature using
atmospheric temperature and wind speed or relative humidity based on the atmospheric temperature
of corresponding weeks in order to control for it. The details of our prediction models (Chapter 1)
and how it was applied to adverse reproductive outcome (Chapter 2) were covered in previous
chapters. In addition to continuous exposure measures, we calculated quartiles of each pollutant
assigned to our control and case periods in order to look at the dose-response gradient.

51

Table 7. Outcome Groups and ICD-10 Codes Included in the Analysis
Outcome group ICD-10 code(s) n
Congenital Malformations of The Circulatory System Q20-Q28 282
Congenital Malformations and Deformations of The Musculoskeletal System Q65-Q79 221
Congenital Malformations of Multiple Organ Systems

175
Cleft Lip and Cleft Palate Q35-Q37 159
Congenital Malformations of Eye, Ear, Face and Neck Q10-Q18 131
Congenital malformations of cardiac septa Q21 122
Ventricular septal defect Q21.0 94
Study design
A symmetric bidirectional case-crossover design was used to estimate the relative odds of
specific CA associated with each interquartile range (IQR) increase in mean pollutant concentrations
as well as 2nd-4th quartile compared to the lowest quartile. We covered the details of this design in
Chapter 2. The main difference in terms of design is the referent selection method. Here, we use
symmetric bidirectional referent selection instead of time-stratified. Unlike time-stratified referent
selection, bidirectional selection creates a non-localizable referent windows which means that the
conditional logistic regression estimating equations are affected by overlap bias. Janes et al. noted
that the bias is usually small, particularly for symmetric bidirectional designs (Janes et al., 2005). This
control selection method still allows us to control for long-term and seasonal time trends by design.
Our referent selection was prompted by the fact that there is no identifiable single event date for
birth defects due to their nature. Instead, we had to consider susceptible gestational weeks as our
case periods and the weekly averages with one-week wash-out period in both directions as control
periods. We looked at gestational weeks 3-8 as susceptible periods since most active fetal organ
development happens during this time, thus making them vulnerable to environmental teratogens.
For instance, we designated week 1 and week 5 as controls when we considered week 3 as our case
period in our analyses. Gestational weeks were determined from conception dates which were
calculated by subtracting gestational weeks at birth from the birth dates of each subject.
52

Statistical analysis
We used conditional logistic regression to estimate the effect of ambient air pollution (PM 2.5,
SO 2, NO 2, and CO) on the occurrence of the selected CA during vulnerable periods of gestation.
Similarly to our previous study on stillbirth, we estimated log odds of each CA per IRQ increase in
continuous measures and quartiles with the lowest as a reference in categorical measures. All the
models were adjusted for weekly mean apparent temperature of corresponding gestational weeks to
control for time-dependent confounding. Confounders such as age, socio-economic status, and
comorbidity were considered to be unchanged in the short-term and therefore were controlled by
study design. Ambient air pollution exposure quartiles were examined to whether there are
identifiable dose-response curves for any of the CA and determine their shape. We evaluated
influential gestational week of exposure based on the magnitude and pattern of the observed risk
estimates and their confidence interval widths. All statistical modeling and geospatial computing
were performed in R version 3.6.2 (R Core Team, 2019) using packages “tidyverse” (Wickham et al.,
2019), “sf” (Pebesma, 2018), and “survival” (Therneau & Grambsch, 2000).
Results
Our study population consisted of 968 congenital anomaly cases occurring between 2014
and 2018 after applying the inclusion and exclusion criteria (Figure 11).
We show the distribution of demographic and residential factors of the study subjects by CA
groups in Table 8. About one in five mothers were 35 years or old and most of them had graduated
college or trade school (61%) and were employed (63%). Only about 6% of the mothers reported
smoking while 43% of them had used folic acid supplement within the first 12 weeks of gestation.
Most mothers had previously given birth (65%) and fetal gender was majority male (58%). In terms
of residential khoroo characteristics, greater number of mothers were from khoroos with less
53

connectivity to internet, more ger households, and higher stove density. There were no substantial
differences between CA groups for any of the demographic or residential characteristics.
Table 8. Distribution of Covariates Among Pooled Congenital Anomaly Cases, and Select Outcome Groups,
Ulaanbaatar, 2014–2018
ICD-10 subchapters
Cardiac
septa
Ventricular
septal defect
Overall

Q20-
Q28
Q65-
Q79
Multiple
organ
systems
Q35-
Q37
Q10-
Q18
Q21 Q21.0 [ALL] N
N=282 N=221 N=175 N=159 N=131 N=122 N=94 N=968
Age (years): 967
<25 23% 24% 16% 30% 21% 20% 18% 23%
25-34 54% 58% 51% 58% 62% 56% 60% 56%
>=35 23% 18% 33% 12% 17% 25% 22% 21%
Education: 955
College or trade school 62% 59% 61% 59% 62% 64% 66% 61%
High school or below 38% 41% 39% 41% 38% 36% 34% 39%
Employment: 945
Employed 62% 65% 60% 61% 67% 69% 68% 63%
Unemployed 38% 35% 40% 39% 33% 31% 32% 37%
Maternal smoking 7% 4% 7% 8% 1% 6% 8% 6% 945
Folic acid 38% 45% 38% 50% 50% 44% 49% 43% 968
Number of deliveries: 968
One 33% 39% 32% 40% 30% 34% 31% 35%
Two or more 67% 61% 68% 60% 70% 66% 69% 65%
Gender: 968
Female 45% 42% 43% 43% 33% 42% 41% 42%
Male 55% 58% 57% 57% 67% 58% 59% 58%
Percentage of households
with internet connection:
963
45% or less 55% 58% 60% 62% 62% 50% 49% 59%
more than 45% 45% 42% 40% 38% 38% 50% 51% 41%
Percentage of ger
households:
963
20% or less 42% 40% 35% 32% 35% 44% 44% 37%
more than 20% 58% 60% 65% 68% 65% 56% 56% 63%
Number of stoves (per
sq.km.):
963
15 stoves or less 39% 34% 32% 28% 34% 39% 39% 34%
more than 15 stoves 61% 66% 68% 72% 66% 61% 61% 66%
54

Odds ratio (OR) estimates and their 95% confidence intervals (95% CI) for each IQR
increase in mean pollutant concentrations for every CA groups examined by different gestational
weeks are shown in Table 9 and 10. We observed significantly increased relative odds (OR: 2.64,
95% CI: 1.02-6.87) of cardiovascular defects (Q20-Q28) per IQR increase in mean PM 2.5
concentration in week 7 (Table 9). This association with PM 2.5 was strengthened in the subgroup of
cardiac septal defects (OR: 7.28, 95% CI: 1.6-33.09) and isolated ventricular septal defects (OR: 9.87,
95% CI: 1.6-60.93). We also obtained similar significantly increased relative odds of cardiac septal
defects (OR: 3.12, 95% CI: 0.99-9.82) and isolated ventricular septal defects (OR: 3.96, 95% CI:
1.05-14.93) per IQR increase in mean concentrations of CO in week 7. The relative odds of the
association between PM 2.5 and ventricular septal defects were increased on week 4 (OR: 6.01, 95%
CI: 1.04-34.7) as well (Table 9). Other notable associations include increased relative odds of cleft lip
and cleft palate subchapter (Q35-Q37) and PM 2.5 (OR: 2.25, 95% CI: 0.62-8.1), SO 2 (OR: 2.6, 95%
CI: 0.61-11.12), and CO (OR: 2.83, 95% CI: 0.92-8.72) in week 4 (Table 10). We did not see
consistent effects on any of the other ICD-10 CA subchapters.

55

Table 9. Relative Odds of Congenital Anomalies Related to Cardiovascular System
Pollutants
Gestationa
l weeks
IQR,
µg/m
3

Circulatory
system
(n = 281)
OR (95% CI)
Cardiac Septa
(n = 122)
OR (95% CI)
VSD
(n = 94)
OR (95% CI)
PM 2.5
3 88.51 1.01 (0.41-2.47) 1.37 (0.38-4.96) 1.54 (0.29-8.13)
4 89.31 1.2 (0.46-3.16) 2.57 (0.61-10.82) 6.01 (1.04-34.7)
5 89.18 0.49 (0.18-1.28) 0.21 (0.05-0.92) 0.14 (0.02-0.84)
6 89.65 1.49 (0.58-3.77) 0.84 (0.22-3.2) 0.47 (0.1-2.26)
7 89.04 2.64 (1.02-6.87) 7.28 (1.6-33.09) 9.87 (1.6-60.93)
8 87.48 0.53 (0.21-1.39) 0.92 (0.23-3.68) 1.71 (0.34-8.56)
SO 2
3 32.69 1.19 (0.45-3.1) 1.24 (0.33-4.65) 1.48 (0.26-8.24)
4 32.72 1.41 (0.51-3.93) 2.05 (0.43-9.7) 5.56 (0.81-38.29)
5 33.45 0.56 (0.21-1.47) 0.4 (0.08-1.89) 0.29 (0.04-1.91)
6 33.03 0.93 (0.38-2.23) 0.96 (0.28-3.25) 0.6 (0.12-2.85)
7 33.45 1.58 (0.64-3.91) 2.17 (0.57-8.18) 2.94 (0.57-15.06)
8 33.11 1.01 (0.43-2.38) 0.86 (0.26-2.77) 1.18 (0.3-4.58)
NO 2
3 26.55 1.25 (0.58-2.69) 1 (0.37-2.73) 1.01 (0.28-3.71)
4 26.14 1.15 (0.54-2.44) 1.35 (0.44-4.21) 2.09 (0.53-8.27)
5 26.07 0.83 (0.42-1.65) 0.71 (0.24-2.1) 0.6 (0.17-2.08)
6 26.14 1.15 (0.61-2.16) 0.89 (0.31-2.55) 0.7 (0.21-2.39)
7 26.17 1.17 (0.6-2.28) 2.14 (0.68-6.68) 2.06 (0.63-6.76)
8 26.30 0.74 (0.39-1.39) 1.12 (0.42-2.97) 1.28 (0.46-3.6)
CO
3 788.01 1.32 (0.6-2.91) 1.1 (0.32-3.78) 1.18 (0.3-4.67)
4 778.43 1.15 (0.52-2.55) 2.17 (0.64-7.39) 3.4 (0.73-15.83)
5 796.44 0.6 (0.29-1.24) 0.52 (0.17-1.58) 0.46 (0.14-1.54)
6 783.77 1.17 (0.55-2.51) 0.9 (0.28-2.92) 0.67 (0.17-2.7)
7 802.66 1.81 (0.84-3.89) 3.12 (0.99-9.82) 3.96 (1.05-14.93)
8 796.83 0.76 (0.35-1.68) 1.13 (0.37-3.49) 1.64 (0.48-5.55)

56

Table 10. Relative Odds of Select Congenital Anomalies
Pollutant
s
Gestatio
nal
weeks
IQR,
µg/m
3

Musculoskeletal
system
(n = 220)
OR (95% CI)
Multiple organ
systems
(n = 173)
OR (95% CI)
Cleft lip and
cleft palate
(n = 159)
OR (95% CI)
Eye, ear, face
and neck
(n = 130)
OR (95% CI)
PM 2.5
3 88.51 1.48 (0.52-4.18) 0.57 (0.19-1.69) 0.75 (0.22-2.57) 1.03 (0.28-3.87)
4 89.31 0.62 (0.23-1.68) 0.49 (0.17-1.38) 2.25 (0.62-8.1) 0.27 (0.06-1.25)
5 89.18 1.06 (0.33-3.37) 0.86 (0.3-2.46) 1.33 (0.42-4.27) 0.99 (0.26-3.81)
6 89.65 1.24 (0.42-3.65) 1.27 (0.4-4.06) 0.63 (0.22-1.82) 2 (0.48-8.3)
7 89.04 0.92 (0.32-2.63) 2.19 (0.71-6.77) 0.78 (0.29-2.11) 0.9 (0.24-3.39)
8 87.48 1.37 (0.42-4.46) 1.78 (0.58-5.46) 1.04 (0.39-2.8) 0.66 (0.19-2.27)
SO 2
3 32.69 1.55 (0.66-3.67) 1.17 (0.45-3.07) 1.37 (0.32-5.9) 1.66 (0.38-7.3)
4 32.72 0.71 (0.29-1.78) 0.76 (0.29-1.99) 2.6 (0.61-11.12) 0.65 (0.16-2.71)
5 33.45 0.53 (0.2-1.41) 0.73 (0.28-1.95) 0.83 (0.23-2.97) 1 (0.27-3.74)
6 33.03 0.95 (0.34-2.67) 0.81 (0.29-2.27) 0.83 (0.24-2.84) 1.07 (0.25-4.51)
7 33.45 1.25 (0.53-2.95) 0.99 (0.36-2.68) 0.86 (0.26-2.77) 0.92 (0.24-3.48)
8 33.11 1.6 (0.57-4.52) 2.03 (0.62-6.57) 0.6 (0.19-1.96) 1.34 (0.35-5.14)
NO 2
3 26.55 1.3 (0.7-2.4) 0.82 (0.36-1.86) 0.91 (0.4-2.09) 0.98 (0.36-2.66)
4 26.14 0.91 (0.48-1.72) 0.75 (0.35-1.58) 1.15 (0.49-2.71) 0.74 (0.27-1.98)
5 26.07 0.89 (0.45-1.77) 0.73 (0.36-1.51) 1.2 (0.53-2.7) 0.79 (0.28-2.19)
6 26.14 0.95 (0.46-1.96) 1.02 (0.45-2.29) 0.84 (0.38-1.82) 0.85 (0.32-2.28)
7 26.17 0.96 (0.48-1.92) 1.25 (0.61-2.59) 0.85 (0.36-2.03) 0.85 (0.28-2.6)
8 26.30 1.49 (0.68-3.23) 1.58 (0.69-3.63) 1.17 (0.54-2.52) 1.42 (0.47-4.35)
CO
3 788.01 1.32 (0.58-3.01) 0.82 (0.36-1.85) 1.32 (0.48-3.66) 0.79 (0.23-2.77)
4 778.43 0.78 (0.35-1.76) 0.56 (0.23-1.35) 2.83 (0.92-8.72) 0.59 (0.17-2.03)
5 796.44 1.13 (0.47-2.7) 1.02 (0.48-2.18) 1.57 (0.5-4.89) 1.19 (0.39-3.63)
6 783.77 1.04 (0.43-2.48) 1.35 (0.59-3.12) 0.58 (0.21-1.62) 1.18 (0.34-4.05)
7 802.66 0.73 (0.33-1.61) 1.11 (0.5-2.49) 0.57 (0.2-1.62) 0.76 (0.26-2.23)
8 796.83 1.28 (0.52-3.18) 1.19 (0.51-2.78) 0.72 (0.31-1.66) 0.95 (0.33-2.73)
Estimates of relative odds and their 95% CI derived from comparing the lowest quartile of
PM 2.5 concentration with 2nd-4th quartiles are presented in Figures 12-16. Here, our main goal was
to visually show general trend of dose-response for different types of CA instead of specific
numbers due to the instability of effect estimates caused by small sample size. For brevity, we show
estimates of PM 2.5 short-term effect on cardiovascular, musculoskeletal, cardiac septal, and
ventricular septal defects (VSD) as examples. We observed dose-response patterns for week 6 and 7
for cardiovascular and cardiac septal defects but for VSD, the increasing dose-response trend
57

occurred in week 4. A less pronounced dose-response trend was observed in weeks 4 and 8 for
musculoskeletal anomalies and for cleft lip and cleft palate. We found similarly suggestive dose-
response patterns in different pollutants, gestational weeks, and congenital anomalies (results not
shown).

Figure 12. Effect estimates of air pollution on cardiovascular anomaly risk
58

Figure 13. Effect estimates of air pollution on cardiac septal anomaly risk

Figure 14. Effect estimates of air pollution on ventricular septal defect risk
59

Figure 15. Effect estimates of air pollution on musculoskeletal anomaly risk

Figure 16. Effect estimates of air pollution on cleft lip and cleft palate risk
60

Discussion
We found considerably increased short-term risk of cardiovascular defects, particularly of
cardiac septal defects, associated with per IQR increase in mean concentrations of PM 2.5 and CO on
gestational week 7. This association was further observed to have dose-response gradients when
using categorical exposure measures. Although there have been multiple studies looking at the
association between ambient air pollution and congenital anomalies in recent years (Girguis et al.,
2016; Huang et al., 2019; Lavigne et al., 2019; Vinceti et al., 2016; Zhang et al., 2016; Zhao et al.,
2018), our results contribute substantially to the evidence base on short-term effects of air pollution
on congenital anomalies by looking at susceptible periods of fetal development at fine temporal
detail. To our knowledge, this is the first study to utilize case-crossover design to control for time-
invariant confounders in examining the association between ambient air pollution and congenital
anomalies.
The findings of our study suggest that there may be differential effects of air pollutants
within the critical period of fetus development (weeks 3-8). This is corroborated by embryological
studies on lab animals showing specific stages of organ development. For instance, neural crest cells
start migrating to develop endocardial tubes at the start of the 4th week and results in formation of
ventricular septation and outflow tracts by week 7 and 8 in cardiac development (Gittenberger-de
Groot et al., 2005). That said there is also evidence of teratogen-induced oxidative stress in earlier
weeks affecting later developments of cardiac structures via apoptosis among migrating cells
(Morgan et al., 2008) or different pathway of neural crest cells (Jain et al., 2011). Therefore, it is
possible that harmful air pollutant exposure at a certain gestational week may not present itself in the
form of CA until later weeks. Additional experimental studies to elucidate this relationship are
needed.
61

In contrast to previous studies, we were unable to look at isolated birth defects except VSD
due to small sample size in each of the ICD-10 subchapters. This is important since it is unlikely that
a specific air pollutant would affect all types of anomalies and the fact that even single, isolated
anomalies could have multifactorial etiology (Ritz, 2010). With that said, our finding of an
association between cardiovascular defects and particulate matter is consistent with several earlier
case-control studies (Agay-Shay et al., 2013; Huang et al., 2019; Padula, Tager, Carmichael,
Hammond, Yang, et al., 2013; Stingone et al., 2014; Zhang et al., 2016). Most of the previous
literature on the topic considered somewhat coarse temporal resolution when it comes to vulnerable
periods of development by using either monthly averages of first 3 months of pregnancy or average
of gestational weeks 3-8. Below we compare our results to a few studies that looked at specific
gestational week exposures. Stingone et al. examined maternal exposure to criteria air pollutants and
congenital heart defects using data from the US National Birth Defects Prevention Study. They used
both 1-week and 7-week averages of pollutants using the closest monitoring station to the maternal
residence and found several individual exposure-weeks associations that were not identified from 7-
week averages (Stingone et al., 2014). Specifically, the association between VSD and particulate
matter were identified to be stronger in week 3, compared to week 4 in our categorical exposure
findings (Figure 16). Two studies from Wuhan, China looked at the association between air
pollution and risk of congenital heart defects (Zhang et al., 2016) as well as oral clefts (Zhao et al.,
2018). Zhang et al. found monotonically increasing relative odds of VSD per 10 𝜇 g/m
3
increase in
mean PM 2.5 concentration during gestational weeks 7-10. This is somewhat comparable to our
finding of significant association between VSD and per IQR increase in mean concentration of
PM 2.5 in week 7. On the other hand, Zhao et al. reported increased risk of oral cleft associated with
per 10 𝜇 g/m
3
change in PM 2.5 in gestational weeks 4-9 whereas we found increased relative odds of
cleft lip and cleft palate subchapter and PM 2.5, SO 2, and CO in week 4 (Table 10).
62

In order to interpret our findings in the context of current literature, we must keep in mind
the emission source of air pollution in UB. As we noted before, the dominant source is a domestic
coal combustion due to need for cooking and heating in ger districts during harsh winters
exacerbated by atmospheric inversion and lower mixing height during cold seasons. This is an
unusual emissions source compared to manufacturing in developing countries and vehicular
emission in developed countries. It is possible that composition of particulate matter are
substantially different from compositions in other countries and this may differentially affect the
susceptible windows of organ development with regards to specific CA. Considering this and the
homogenous population of Mongolia, generalizability of our study results may be limited, but it is
suggestive to see partially consistent results with previous studies despite these differences.
Estimating personal exposure via prediction modeling based on scarce ground level
monitors will inevitably lead to measurement error in exposure regardless of the model performance.
In addition, we were only able to predict air pollution exposure based on residential administrative
unit (khoroo) at the time of delivery. However, recent study on the effect of residential mobility
during pregnancy on air pollution exposure misclassification found minor impact (Warren et al.,
2018). Measurement errors due to above reasons will be non-differential and will likely lead to
underestimation of the risk. The other limitation of our study was inadequate sample size for
investigating the effect of air pollutants on isolated birth defects rather than ICD-10 subchapters
that covered large range of congenital anomalies with possibly differing etiologies. This was
unavoidable due to recent establishment of the National Surveillance Department in Mongolia and
will only strengthen as time goes by.
In this study, we observed increased relative odds of select CA with short-term, higher
exposure to air pollutants. Some of the associations were only on specific weeks and had dose-
63

response trend which suggests that accounting for temporal as well as spatial variability may lead to
better understanding of how and when air pollutants affect CA risk. Future studies in this research
area should focus on improving exposure assessment, incorporating multipollutant approach and
ascertaining more isolated birth defects to facilitate the investigation into different susceptible
windows of various congenital anomalies.

64

References
Agay-Shay, K., Friger, M., Linn, S., Peled, A., Amitai, Y., & Peretz, C. (2013). Air pollution and
congenital heart defects. Environmental Research, 124(2), 28–34.
https://doi.org/10.1016/j.envres.2013.03.005
Al-Gubory, K. H., Fowler, P. A., & Garrel, C. (2010). The roles of cellular reactive oxygen species,
oxidative stress and antioxidants in pregnancy outcomes. International Journal of Biochemistry and Cell
Biology, 42(10), 1634–1650. https://doi.org/10/c2shd7
Allen, R. W., Gombojav, E., Barkhasragchaa, B., Byambaa, T., Lkhasuren, O., Amram, O., Takaro,
T. K., & Janes, C. R. (2013). An assessment of air pollution and its attributable mortality in
Ulaanbaatar, Mongolia. Air Quality, Atmosphere and Health, 6(1), 137–150.
https://doi.org/10.1007/s11869-011-0154-3
Anderson, G. B., Bell, M. L., & Peng, R. D. (2013). Methods to calculate the heat index as an
exposure metric in environmental health research. Environmental Health Perspectives, 121(10), 1111–
1119. https://doi.org/10/gbdvhp
Baccarelli, A., & Bollati, V. (2009). Epigenetics and environmental chemicals. Current Opinion in
Pediatrics, 21(2), 243–251. https://doi.org/10/b3s37g
Batmunkh, T., Kim, Y. J., Jung, J. S., Park, K., & Tumendemberel, B. (2013). Chemical
characteristics of fine particulate matters measured during severe winter haze events in Ulaanbaatar,
Mongolia. Journal of the Air & Waste Management Association, 63(6), 659–670.
https://doi.org/10.1080/10962247.2013.776997
Bi, Q., Goodman, K. E., Kaminsky, J., & Lessler, J. (2019). What is Machine Learning? A Primer for
the Epidemiologist. American Journal of Epidemiology. https://doi.org/10.1093/aje/kwz189
Breiman, L. (Ed.). (1998). Classification and regression trees (Repr). Chapman & Hall [u.a.].
Brokamp, C., Jandarov, R., Hossain, M., & Ryan, P. (2018). Predicting Daily Urban Fine Particulate
Matter Concentrations Using a Random Forest Model. Environmental Science & Technology, 52(7),
4173–4179. https://doi.org/10.1021/acs.est.7b05381
Brook, R. D., Rajagopalan, S., Pope, C. A., Brook, J. R., Bhatnagar, A., Diez-Roux, A. V., Holguin,
F., Hong, Y., Luepker, R. V., Mittleman, M. A., Peters, A., Siscovick, D., Smith, S. C., Whitsel, L., &
Kaufman, J. D. (2010). Particulate Matter Air Pollution and Cardiovascular Disease: An Update to
the Scientific Statement From the American Heart Association. Circulation, 121(21), 2331–2378.
https://doi.org/10.1161/CIR.0b013e3181dbece1
Chen, E., Zmirou-Navier, D., Padilla, C., & Deguen, S. (2014). Effects of Air Pollution on the Risk
of Congenital Anomalies: A Systematic Review and Meta-Analysis. International Journal of
Environmental Research and Public Health, 11(8), 7642–7668. https://doi.org/10/f6f4ps
Chen, T., He, T., Benesty, M., Khotilovich, V., Tang, Y., Cho, H., Chen, K., Mitchell, R., Cano, I.,
Zhou, T., Li, M., Xie, J., Lin, M., Geng, Y., & Li, Y. (2019). Xgboost: Extreme gradient boosting.
https://CRAN.R-project.org/package=xgboost
65

Christianson, A., Howson, C. P., & Modell, B. (2005). March of Dimes: Global report on birth
defects, the hidden toll of dying and disabled children. March of Dimes: Global Report on Birth Defects, the
Hidden Toll of Dying and Disabled Children.
Cohen, A. J., Brauer, M., Burnett, R., Anderson, H. R., Frostad, J., Estep, K., Balakrishnan, K.,
Brunekreef, B., Dandona, L., Dandona, R., Feigin, V., Freedman, G., Hubbell, B., Jobling, A., Kan,
H., Knibbs, L., Liu, Y., Martin, R., Morawska, L., … Forouzanfar, M. H. (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), 1907–1918.
https://doi.org/10.1016/S0140-6736(17)30505-6
Dadvand, P., Rankin, J., Rushton, S., & Pless-Mulloli, T. (2011). Association Between Maternal
Exposure to Ambient Air Pollution and Congenital Heart Disease: A Register-based Spatiotemporal
Analysis. American Journal of Epidemiology, 173(2), 171–182. https://doi.org/10/b2sjbx
Dadvand, P., Rankin, J., Rushton, S., & Pless-Mulloli, T. (2011). Ambient air pollution and
congenital heart disease: A register-based study. Environmental Research, 111(3), 435–441.
https://doi.org/10/dwkgr4
Dastoorpoor, M., Idani, E., Goudarzi, G., & Khanjani, N. (2018). Acute effects of air pollution on
spontaneous abortion, premature delivery, and stillbirth in Ahvaz, Iran: A time-series study.
Environmental Science and Pollution Research, 25(6), 5447–5458. https://doi.org/10/gf4qmd
Davy, P. K., Gunchin, G., Markwitz, A., Trompetter, W. J., Barry, B. J., Shagjjamba, D., &
Lodoysamba, S. (2011). Air particulate matter pollution in Ulaanbaatar, Mongolia: Determination of
composition, source contributions and source locations. Atmospheric Pollution Research, 2(2), 126–137.
https://doi.org/10.5094/APR.2011.017
Di, Q., Amini, H., Shi, L., Kloog, I., Silvern, R., Kelly, J., Sabath, M. B., Choirat, C., Koutrakis, P.,
Lyapustin, A., Wang, Y., Mickley, L. J., & Schwartz, J. (2019). An ensemble-based model of PM2.5
concentration across the contiguous United States with high spatiotemporal resolution. Environment
International, 130, 104909. https://doi.org/10.1016/j.envint.2019.104909
Di, Q., Wang, Y., Zanobetti, A., Wang, Y., Koutrakis, P., Choirat, C., Dominici, F., & Schwartz, J.
D. (2017). Air Pollution and Mortality in the Medicare Population. New England Journal of Medicine,
376(26), 2513–2522. https://doi.org/10.1056/NEJMoa1702747
Dolk, H., Armstrong, B., Lachowycz, K., Vrijheid, M., Rankin, J., Abramsky, L., Boyd, P. A., &
Wellesley, D. (2010). Ambient air pollution and risk of congenital anomalies in England, 1991-1999.
Occupational and Environmental Medicine, 67(4), 223–227. https://doi.org/10.1136/oem.2009.045997
Enkhmaa, D., Warburton, N., Javzandulam, B., Uyanga, J., Khishigsuren, Y., Lodoysamba, S.,
Enkhtur, S., & Warburton, D. (2014). Seasonal ambient air pollution correlates strongly with
spontaneous abortion in Mongolia. BMC Pregnancy and Childbirth, 14(1), 146.
https://doi.org/10.1186/1471-2393-14-146
Enkhtur, S., & Bayalag, M. (Eds.). (2019). Comprehensive Survey of the Status of Maternal and Child
Morbidity and Mortality, and Prevance of Congenital Anomalies in Mongolia - V. National Center for
Maternal and Child Health.
66

Enkh-Undraa, D., Kanda, S., Shima, M., Shimono, T., Miyake, M., Yoda, Y., Nagnii, S., &
Nishiyama, T. (2019). Coal burning-derived SO2 and traffic-derived NO2 are associated with
persistent cough and current wheezing symptoms among schoolchildren in Ulaanbaatar, Mongolia.
Environmental Health and Preventive Medicine, 24(1), 66. https://doi.org/10.1186/s12199-019-0817-5
Faiz, A. S., Rhoads, G. G., Demissie, K., Lin, Y., Kruse, L., & Rich, D. Q. (2013). Does Ambient Air
Pollution Trigger Stillbirth? Epidemiology, 24(4), 538–544. https://doi.org/10/f42tzs
Franklin, M., Chau, K., Kalashnikova, O., Garay, M., Enebish, T., & Sorek-Hamer, M. (2018). Using
Multi-Angle Imaging SpectroRadiometer Aerosol Mixture Properties for Air Quality Assessment in
Mongolia. Remote Sensing, 10(8), 1317. https://doi.org/10.3390/rs10081317
Franklin, M., Zeka, A., & Schwartz, J. (2007). Association between PM2.5 and all-cause and specific-
cause mortality in 27 US communities. Journal of Exposure Science and Environmental Epidemiology, 17(3),
279–287. https://doi.org/10.1038/sj.jes.7500530
Friedman, J., Hastie, T., & Tibshirani, R. (2010). Regularization paths for generalized linear models
via coordinate descent. Journal of Statistical Software, 33(1), 1–22. http://www.jstatsoft.org/v33/i01/
Gail, M. H., Lubin, J. H., & Rubinstein, L. V. (1981). Likelihood calculations for matched case-
control studies and survival studies with tied death times. Biometrika, 68(3), 703–707.
https://doi.org/10/crz3fm
Ganbat, G., & Baik, J. J. (2016). Wintertime winds in and around the Ulaanbaatar metropolitan area
in the presence of a temperature inversion. Asia-Pacific Journal of Atmospheric Sciences, 52(3), 309–325.
https://doi.org/10.1007/s13143-016-0007-y
Gardosi, J., Mul, T., Mongelli, M., & Fagan, D. (1998). Analysis of birthweight and gestational age in
antepartum stillbirths. British Journal of Obstetrics and Gynaecology, 105(5), 524–530.
https://doi.org/10/bvd7h6
Genest, D. R., Williams, M. A., & Greene, M. F. (1992). Estimating the time of death in stillborn
fetuses: I. Histologic evaluation of fetal organs; an autopsy study of 150 stillborns. Obstetrics and
Gynecology, 80(4), 575–584.
Gianicolo, E. A. L., Mangia, C., Cervino, M., Bruni, A., Andreassi, M. G., & Latini, G. (2014).
Congenital anomalies among live births in a high environmental risk area-A case-control study in
Brindisi (southern Italy). Environmental Research, 128(349), 9–14. https://doi.org/10/f5n8xx
Gilboa, S. M., Mendola, P., Olshan, A. F., Langlois, P. H., Savitz, D. A., Loomis, D., Herring, A. H.,
& Fixler, D. E. (2005). Relation between ambient air quality and selected birth defects, seven county
study, Texas, 1997-2000. American Journal of Epidemiology, 162(3), 238–252.
https://doi.org/10.1093/aje/kwi189
Girguis, M. S., Strickland, M. J., Hu, X., Liu, Y., Bartell, S. M., & Vieira, V. M. (2016). Maternal
exposure to traffic-related air pollution and birth defects in Massachusetts. Environmental Research,
146, 1–9. https://doi.org/10/f8ccjx
Gittenberger-de Groot, A. C., Bartelings, M. M., Deruiter, M. C., & Poelmann, R. E. (2005). Basics
of cardiac development for the understanding of congenital heart malformations. Pediatric Research,
57(2), 169–176. https://doi.org/10/ftxzrb
67

Guttikunda, S. K., Lodoysamba, S., Bulgansaikhan, B., & Dashdondog, B. (2013). Particulate
pollution in Ulaanbaatar, Mongolia. Air Quality, Atmosphere and Health, 6(3), 589–601.
https://doi.org/10.1007/s11869-013-0198-7
Hansen, C. A., Barnett, A. G., Jalaludin, B. B., & Morgan, G. G. (2009). Ambient air pollution and
birth defects in Brisbane, Australia. PLoS ONE, 4(4), 6–13. https://doi.org/10/fj9pjr
Heazell, A. E. P., Siassakos, D., Blencowe, H., Burden, C., Bhutta, Z. A., Cacciatore, J., Dang, N.,
Das, J., Flenady, V., Gold, K. J., Mensah, O. K., Millum, J., Nuzum, D., O’Donoghue, K., Redshaw,
M., Rizvi, A., Roberts, T., Saraki, H. E., Storey, C., … Downe, S. (2016). Stillbirths: Economic and
psychosocial consequences. The Lancet, 387(10018), 604–616. https://doi.org/10.1016/S0140-
6736(15)00836-3
Herbstman, J. B., Tang, D., Zhu, D., Qu, L., Sjodin, A., Li, Z., Camann, D., & Perera, F. P. (2012).
Prenatal exposure to polycyclic aromatic hydrocarbons, benzo[a]Pyrene-DNA adducts, and genomic
DNA methylation in cord blood. Environmental Health Perspectives, 120(5), 733–738.
https://doi.org/10/fx8pjv
Hou, L., Zhang, X., Wang, D., & Baccarelli, A. (2012). Environmental chemical exposures and
human epigenetics. International Journal of Epidemiology, 41(1), 79–105. https://doi.org/10/bsvw8b
Hu, X., Belle, J. H., Meng, X., Wildani, A., Waller, L. A., Strickland, M. J., & Liu, Y. (2017).
Estimating PM 2.5 Concentrations in the Conterminous United States Using the Random Forest
Approach. Environmental Science & Technology, 51(12), 6936–6944.
https://doi.org/10.1021/acs.est.7b01210
Huang, C.-c., Chen, B.-y., Pan, S.-c., Ho, Y.-l., & Guo, Y. L. (2019). Prenatal exposure to PM2.5 and
Congenital Heart Diseases in Taiwan. Science of the Total Environment, 655, 880–886.
https://doi.org/10/ggs8qv
Jacobs, M., Zhang, G., Chen, S., Mullins, B., Bell, M., Jin, L., Guo, Y., Huxley, R., & Pereira, G.
(2017). The association between ambient air pollution and selected adverse pregnancy outcomes in
China: A systematic review. Science of the Total Environment, 579, 1179–1192.
https://doi.org/10/f9p5b8
Jain, R., Engleka, K. A., Rentschler, S. L., Manderfield, L. J., Li, L., Yuan, L., & Epstein, J. A. (2011).
Cardiac neural crest orchestrates remodeling and functional maturation of mouse semilunar valves.
The Journal of Clinical Investigation, 121(1), 422–430. https://doi.org/10/fddm5n
Janes, H., Sheppard, L., & Lumley, T. (2005). Overlap bias in the case-crossover design, with
application to air pollution exposures. Statistics in Medicine, 24(2), 285–300.
https://doi.org/10/cj5hqx
Janssen, B. G., Godderis, L., Pieters, N., Poels, K., Kici’nski, M., Cuypers, A., Fierens, F., Penders,
J., Plusquin, M., Gyselaers, W., & Nawro, T. S. (2013). Placental DNA hypomethylation in
association with particulate air pollution in early life. Particle and Fibre Toxicology, 10(1), 22.
https://doi.org/10/gbdj7v
68

Jarvis, A., Reuter, H. I., Nelson, A., & Guevara, E. (2008). Hole-filled seamless srtm data version 4.
International Center for Tropical Agriculture (CIAT), Available at: Http://Srtm. Csi. Cgiar. Org (Last Access:
27 June 2019).
Jonakait, G. M. (2007). The effects of maternal inflammation on neuronal development: Possible
mechanisms. International Journal of Developmental Neuroscience, 25(7), 415–425.
https://doi.org/10/fq8m9h
Kannan, S., Misra, D. P., Dvonch, J. T., & Krishnakumar, A. (2006). Exposures to airbone
particulate matter and adverse perinatal outcomes: A biologically plausible mechanistic framework
for exploring potential effect modification by nutrition. Environmental Health Perspectives, 114(11),
1636–1642. https://doi.org/10/fsz694
Karatzoglou, A., Smola, A., Hornik, K., & Zeileis, A. (2004). Kernlab – an S4 package for kernel
methods in R. Journal of Statistical Software, 11(9), 1–20. http://www.jstatsoft.org/v11/i09/
Kuhn, M. (2019). Tune: Tidy tuning tools. https://github.com/tidymodels/tune
Kuhn, M., Chow, F., & Wickham, H. (2019). Rsample: General resampling infrastructure.
Kuhn, M., & Vaughan, D. (2019). Parsnip: A common api to modeling and analysis functions.
Kuhn, M., & Wickham, H. (2019). Recipes: Preprocessing tools to create design matrices.
https://github.com/tidymodels/recipes
Lavigne, E., Lima, I., Hatzopoulou, M., Van Ryswyk, K., Decou, M. L., Luo, W., van Donkelaar, A.,
Martin, R. V., Chen, H., Stieb, D. M., Crighton, E., Gasparrini, A., Elten, M., Yasseen, A. S.,
Burnett, R. T., Walker, M., & Weichenthal, S. (2019). Spatial variations in ambient ultrafine particle
concentrations and risk of congenital heart defects. Environment International, 130, 104953.
https://doi.org/10/ggs8qz
Lawn, J. E., Blencowe, H., Waiswa, P., Amouzou, A., Mathers, C., Hogan, D., Flenady, V., Frøen, J.
F., Qureshi, Z. U., Calderwood, C., Shiekh, S., Jassir, F. B., You, D., McClure, E. M., Mathai, M.,
Cousens, S., Kinney, M. V., De Bernis, L., Heazell, A., … Draper, E. S. (2016). Stillbirths: Rates, risk
factors, and acceleration towards 2030. The Lancet, 387(10018), 587–603. https://doi.org/10/f3p2cr
Li, X., Huang, S., Jiao, A., Yang, X., Yun, J., Wang, Y., Xue, X., Chu, Y., Liu, F., Liu, Y., Ren, M.,
Chen, X., Li, N., Lu, Y., Mao, Z., Tian, L., & Xiang, H. (2017). Association between ambient fine
particulate matter and preterm birth or term low birth weight: An updated systematic review and
meta-analysis. Environmental Pollution, 227, 596–605. https://doi.org/10/gbknfc
Lin, Y. T., Lee, Y. L., Jung, C. R., Jaakkola, J. J. K., & Hwang, B. F. (2014). Air pollution and limb
defects: A matched-pairs case-control study in Taiwan. Environmental Research, 132, 273–280.
https://doi.org/10.1016/j.envres.2014.04.028
Lippmann, M., Ito, K., N’adas, A., & Burnett, R. T. (2000). Association of particulate matter
components with daily mortality and morbidity in urban populations. Research Report (Health Effects
Institute), 95, 5–72, discussion73–82.
Lyapustin, A., Wang, Y., Korkin, S., & Huang, D. (2018). MODIS Collection 6 MAIAC algorithm.
Atmospheric Measurement Techniques, 11(10), 5741–5765. https://doi.org/10.5194/amt-11-5741-2018
69

Maclure, M. (1991). The Case-Crossover Design : A Method for Studying Transient Effects on the Risk of Acute
Events. 133(2), 144–153.
Marshall, E. G., Harris, G., & Wartenberg, D. (2010). Oral cleft defects and maternal exposure to
ambient air pollutants in New Jersey. Birth Defects Research Part A - Clinical and Molecular Teratology,
88(4), 205–215. https://doi.org/10/fw3nx2
Mendola, P., Ha, S., Pollack, A. Z., Zhu, Y., Seeni, I., Kim, S. S., Sherman, S., & Liu, D. (2017).
Chronic and Acute Ozone Exposure in the Week Prior to Delivery Is Associated with the Risk of
Stillbirth. International Journal of Environmental Research and Public Health, 14(7), 731.
https://doi.org/10/gbs76x
Morgan, S. C., Relaix, F., Sandell, L. L., & Loeken, M. R. (2008). Oxidative stress during diabetic
pregnancy disrupts cardiac neural crest migration and causes outflow tract defects. Birth Defects
Research Part A: Clinical and Molecular Teratology, 82(6), 453–463. https://doi.org/10/fw77gc
Nabavi, S. O., Haimberger, L., & Abbasi, E. (2019). Assessing PM2.5 concentrations in Tehran,
Iran, from space using MAIAC, deep blue, and dark target AOD and machine learning algorithms.
Atmospheric Pollution Research, 10(3), 889–903. https://doi.org/10.1016/j.apr.2018.12.017
Narmandakh, L., Galymbek, K., & Tsatsral, B. (2018). Report on 2018 Enumeration of Air Pollution
Sources in Ulaanbaatar. UB Air Pollution Reduction Agency.
Nelson, C. A., Tew, M., Phetteplace, G., Schwerdt, R., Maarouf, A., Osczevski, R., Bluestein, M.,
Shaykewich, J., Smarsh, D., Derby, J., & others. (2002). 6B. 2 joint development and implementation by the
united states and canada of a new wind chill temperature (wct) index.
Nishikawa, M., Matsui, I., Batdorj, D., Jugder, D., Mori, I., Shimizu, A., Sugimoto, N., & Takahashi,
K. (2011). Chemical composition of urban airborne particulate matter in Ulaanbaatar. Atmospheric
Environment, 45(32), 5710–5715. https://doi.org/10.1016/j.atmosenv.2011.07.029
NOAA National Centers for Environmental Information. (n.d.). Integrated Surface Dataset (Global).
https://data.nodc.noaa.gov/cgi-bin/iso?id=gov.noaa.ncdc:C00532#.
Padula, A. M., Tager, I. B., Carmichael, S. L., Hammond, S. K., Lurmann, F., & Shaw, G. M. (2013).
The association of ambient air pollution and traffic exposures with selected congenital anomalies in
the San Joaquin Valley of California. American Journal of Epidemiology, 177(10), 1074–1085.
https://doi.org/10/f4xsbq
Padula, A. M., Tager, I. B., Carmichael, S. L., Hammond, S. K., Yang, W., Lurmann, F., & Shaw, G.
M. (2013). Ambient air pollution and traffic exposures and congenital heart defects in the san
joaquin Valley of California. Paediatric and Perinatal Epidemiology, 27(4), 329–339.
https://doi.org/10/f42jpx
Padula, A. M., Yang, W., Carmichael, S. L., Tager, I. B., Lurmann, F., Hammond, S. K., & Shaw, G.
M. (2015). Air Pollution, Neighbourhood Socioeconomic Factors, and Neural Tube Defects in the
San Joaquin Valley of California. Paediatric and Perinatal Epidemiology, 29(6), 536–545.
https://doi.org/10.1111/ppe.12244
Pebesma, E. (2018). Simple Features for R: Standardized Support for Spatial Vector Data. The R
Journal, 10(1), 439–446. https://doi.org/10.32614/RJ-2018-009
70

Penney, D. G. (1996). Carbon Monoxide. CRC-Press.
Pereira, L. A. A., Loomis, D., Conceição, G. M. S., Braga, A. L. F., Arcas, R. M., Kishi, H. S., Singer,
J. M., Böhm, G. M., & Saldiva, P. H. N. (1998). Association between air pollution and intrauterine
mortality in Sao Paulo, Brazil. Environmental Health Perspectives, 106(6), 325–329.
https://doi.org/10/bmfrpd
Perera, F. P., Hemminki, K., Gryzbowska, E., Motykiewicz, G., Michalska, J., Santella, R. M.,
Young, T.-L., Dickey, C., Brandt-Rauf, P., & DeVivo, I. (1992). Molecular and genetic damage in
humans from environmental pollution in Poland. Nature, 360(6401), 256. https://doi.org/10/drqgwj
Perera, F. P., Jedrychowski, W., Rauh, V., & Whyatt, R. M. (1999). Molecular epidemiologic research
on the effects of environmental pollutants on the fetus. Environmental Health Perspectives, 107
Suppl(February), 451–460. https://doi.org/sc271_5_1835 [pii]
Pope, C. A., Coleman, N., Pond, Z. A., & Burnett, R. T. (2019). Fine particulate air pollution and
human mortality: 25+ years of cohort studies. Environmental Research, 108924.
https://doi.org/10.1016/j.envres.2019.108924
Proietti, E., Roeoesli, M., Frey, U., & Latzin, P. (2013). Air Pollution During Pregnancy and
Neonatal Outcome: A Review. Journal of Aerosol Medicine and Pulmonary Drug Delivery, 26, 9–23.
https://doi.org/10/f4m4qr
Rankin, J., Chadwick, T., Natarajan, M., Howel, D., Pearce, M. S., & Pless-Mulloli, T. (2009).
Maternal exposure to ambient air pollutants and risk of congenital anomalies. Environmental Research,
109(2), 181–187. https://doi.org/10.1016/j.envres.2008.11.007
R Core Team. (2019). R: A language and environment for statistical computing. R Foundation for Statistical
Computing. https://www.R-project.org/
Reid, C. E., Jerrett, M., Petersen, M. L., Pfister, G. G., Morefield, P. E., Tager, I. B., Raffuse, S. M.,
& Balmes, J. R. (2015). Spatiotemporal Prediction of Fine Particulate Matter During the 2008
Northern California Wildfires Using Machine Learning. Environmental Science & Technology, 49(6),
3887–3896. https://doi.org/10.1021/es505846r
Ritz, B. (2010). Air pollution and congenital anomalies. Occupational and Environmental Medicine, 67(4),
221–222. https://doi.org/10/c8xmsb
Ritz, B. R., Yu, F., Fruin, S., Chapa, G., Shaw, G. M., & Harris, J. A. (2002). Ambient air pollution
and risk of birth defects in southern california. American Journal of Epidemiology, 155(1), 17–25.
Roberts, D. R., Bahn, V., Ciuti, S., Boyce, M. S., Elith, J., Guillera-Arroita, G., Hauenstein, S.,
Lahoz-Monfort, J. J., Schröder, B., Thuiller, W., Warton, D. I., Wintle, B. A., Hartig, F., &
Dormann, C. F. (2017). Cross-validation strategies for data with temporal, spatial, hierarchical, or
phylogenetic structure. Ecography, 40(8), 913–929. https://doi.org/10.1111/ecog.02881
Salam, M. T., Millstein, J., Li, Y.-F., Lurmann, F. W., Margolis, H. G., & Gilliland, F. D. (2005).
Birth outcomes and prenatal exposure to ozone, carbon monoxide, and particulate matter: Results
from the Children’s Health Study. Environmental Health Perspectives, 113(11), 1638–1644.
https://doi.org/10/b5rt9n
71

Sangalli, M., Mclean, A., Peek, M., Rivory, L., & Le Couteur, D. G. (2003). Carbon monoxide
disposition and permeability-surface area product in the foetal circulation of the perfused term
human placenta. Placenta, 24(1), 8–11. https://doi.org/10/cppw9r
Sapkota, A., Chelikowsky, A. P., Nachman, K. E., Cohen, A. J., & Ritz, B. (2012). Exposure to
particulate matter and adverse birth outcomes: A comprehensive review and meta-analysis. Air
Quality, Atmosphere and Health, 5(4), 369–381. https://doi.org/10/bjdd2c
Schembari, A., Nieuwenhuijsen, M. J., Salvador, J., de Nazelle, A., Cirach, M., Dadvand, P., Beelen,
R., Hoek, G., Basagaña, X., & Vrijheid, M. (2014). Traffic-related air pollution and congenital
anomalies in Barcelona. Environmental Health Perspectives, 122(3), 317–323. https://doi.org/10/ggs7w9
Siddika, N., Balogun, H. A., Amegah, A. K., & Jaakkola, J. J. K. (2016). Prenatal ambient air
pollution exposure and the risk of stillbirth : Systematic review and meta-analysis of the empirical
evidence. Occup Environ Med, 73, 573–581. https://doi.org/10/gfvdm4
Slama, R., Darrow, L., Parker, J., Woodruff, T. J., Strickland, M., Nieuwenhuijsen, M., Glinianaia, S.,
Hoggatt, K. J., Kannan, S., Hurley, F., Kalinka, J., Šr’am, R., Brauer, M., Wilhelm, M., Henrich, J., &
Ritz, B. (2008). Meeting report: Atmospheric pollution and human reproduction. Environmental
Health Perspectives, 116(6), 791–798. https://doi.org/10/dgs6xr
Stieb, D. M., Chen, L., Eshoul, M., & Judek, S. (2012). Ambient air pollution, birth weight and
preterm birth: A systematic review and meta-analysis. Environmental Research, 117, 100–111.
https://doi.org/10/f3882q
Stingone, J. A., Luben, T. J., Daniels, J. L., Fuentes, M., Richardson, D. B., Aylsworth, A. S.,
Herring, A. H., Anderka, M., Botto, L., Correa, A., Gilboa, S. M., Langlois, P. H., Mosley, B., Shaw,
G. M., Siffel, C., & Olshan, A. F. (2014). Maternal exposure to criteria air pollutants and congenital
heart defects in offspring: Results from the National Birth Defects Prevention Study. Environmental
Health Perspectives, 122(8), 863–872. https://doi.org/10/f6hc6d
Tavares Da Silva, F., Gonik, B., McMillan, M., Keech, C., Dellicour, S., Bhange, S., Tila, M., Harper,
D. M., Woods, C., Kawai, A. T., Kochhar, S., & Munoz, F. M. (2016). Stillbirth: Case definition and
guidelines for data collection, analysis, and presentation of maternal immunization safety data.
Vaccine, 34(49), 6057–6068. https://doi.org/10/ggp5qk
Therneau, T. M., & Grambsch, P. M. (2000). The Cox Model. In T. M. Therneau & P. M.
Grambsch (Eds.), Modeling Survival Data: Extending the Cox Model (pp. 39–77). Springer.
https://doi.org/10.1007/978-1-4757-3294-8_3
Trevor Hastie, S. M. D. from mda:mars by, & Thomas Lumley’s leaps wrapper., R. T. U. A. M. F.
utilities with. (2019). Earth: Multivariate adaptive regression splines. https://CRAN.R-
project.org/package=earth
United Nations. (2015a). The Millennium Development Goals Report. United Nations, 72.
https://doi.org/978-92-1-101320-7
United Nations. (2015b). Transforming our world: The 2030 Agenda for Sustainable Development.
General Assembley 70 Session, 16301(October), 1–35. https://doi.org/10.1007/s13398-014-0173-7.2
72

van den Hooven, E. H., Pierik, F. H., de Kluizenaar, Y., Hofman, A., van Ratingen, S. W., Zandveld,
P. Y. J., Russcher, H., Lindemans, J., Miedema, H. M. E., Steegers, E. A. P., & Jaddoe, V. W. V.
(2012). Air pollution exposure and markers of placental growth and function: The Generation R
Study. Environmental Health Perspectives, 120(12), 1753–1759. https://doi.org/10/gbb5vr
Vinceti, M., Malagoli, C., Malavolti, M., Cherubini, A., Maffeis, G., Rodolfi, R., Heck, J. E., Astolfi,
G., Calzolari, E., & Nicolini, F. (2016). Does maternal exposure to benzene and PM10 during
pregnancy increase the risk of congenital anomalies? A population-based case-control study. Science of
the Total Environment, 541, 444–450. https://doi.org/10.1016/j.scitotenv.2015.09.051
Vrijheid, M., Martinez, D., Manzanares, S., Dadvand, P., Schembari, A., Rankin, J., &
Nieuwenhuijsen, M. (2011). Ambient air pollution and risk of congenital anomalies: A systematic
review and meta-analysis. Environmental Health Perspectives, 119(5), 598–606.
https://doi.org/10/fb2vcd
Warren, J. L., Son, J.-Y., Pereira, G., Leaderer, B. P., & Bell, M. L. (2018). Investigating the Impact
of Maternal Residential Mobility on Identifying Critical Windows of Susceptibility to Ambient Air
Pollution During Pregnancy. American Journal of Epidemiology, 187(5), 992–1000.
https://doi.org/10/gdkrxx
Watson, G. L., Telesca, D., Reid, C. E., Pfister, G. G., & Jerrett, M. (2019). Machine learning models
accurately predict ozone exposure during wildfire events. Environmental Pollution, 254, 112792.
https://doi.org/10.1016/j.envpol.2019.06.088
Wickham, H., Averick, M., Bryan, J., Chang, W., McGowan, L. D., Francois, R., Grolemund, G.,
Hayes, A., Henry, L., Hester, J., Kuhn, M., Pedersen, T. L., Miller, E., Bache, S. M., Miller, K.,
Ooms, J., Robinson, D., Seidel, D. P., Spinu, V., … Yutani, H. (2019). Welcome to the tidyverse.
Journal of Open Source Software, 4(43), 1686. https://doi.org/10.21105/joss.01686
Wood, S. N., N., Pya, & S"afken, B. (2016). Smoothing parameter and model selection for general
smooth models (with discussion). Journal of the American Statistical Association, 111, 1548–1575.
World Health Organization. (n.d.). Congenital anomalies. https://www.who.int/news-room/fact-
sheets/detail/congenital-anomalies.
World Health Organization. (2004). ICD-10 : International statistical classification of diseases and related
health problems : Tenth revision. Spanish version, 1stedition published by PAHO as Publicación
Científica 544.
World Health Organization. (2006). WHO Air quality guidelines for particulate matter, ozone, nitrogen
dioxide and sulfur dioxide: Global update 2005: Summary of risk assessment. Geneva: World Health
Organization.
World Health Organization. (2017). Causes of child mortality. In Global Health Observatory Data.
http://www.who.int/gho/child_health/mortality/causes/en/.
Wright, M. N., & Ziegler, A. (2017). ranger: A fast implementation of random forests for high
dimensional data in C++ and R. Journal of Statistical Software, 77(1), 1–17.
https://doi.org/10.18637/jss.v077.i01
73

Xu, Y., Ho, H. C., Wong, M. S., Deng, C., Shi, Y., Chan, T.-C., & Knudby, A. (2018). Evaluation of
machine learning techniques with multiple remote sensing datasets in estimating monthly
concentrations of ground-level PM2.5. Environmental Pollution, 242, 1417–1426.
https://doi.org/10.1016/j.envpol.2018.08.029
Zamani Joharestani, M., Cao, C., Ni, X., Bashir, B., & Talebiesfandarani, S. (2019). PM2.5 Prediction
Based on Random Forest, XGBoost, and Deep Learning Using Multisource Remote Sensing Data.
Atmosphere, 10(7), 373. https://doi.org/10.3390/atmos10070373
Zhan, Y., Luo, Y., Deng, X., Chen, H., Grieneisen, M. L., Shen, X., Zhu, L., & Zhang, M. (2017).
Spatiotemporal prediction of continuous daily PM 2.5 concentrations across China using a spatially
explicit machine learning algorithm. Atmospheric Environment, 155, 129–139.
https://doi.org/10.1016/j.atmosenv.2017.02.023
Zhang, B., Liang, S., Zhao, J., Qian, Z., Bassig, B. A., Yang, R., Zhang, Y., Hu, K., Xu, S., Zheng, T.,
& Yang, S. (2016). Maternal exposure to air pollutant PM2.5 and PM10 during pregnancy and risk of
congenital heart defects. Journal of Exposure Science & Environmental Epidemiology, 26(4), 422–427.
https://doi.org/10/f8r4ss
Zhao, J., Zhang, B., Yang, S., Mei, H., Qian, Z., Liang, S., Zhang, Y., Hu, K., Tan, Y., Xian, H.,
Belue, R., Jordan, S. S., Xu, S., Zheng, T., & Du, Y. (2018). Maternal exposure to ambient air
pollutant and risk of oral clefts in Wuhan, China. Environmental Pollution, 238, 624–630.
https://doi.org/10/gdrrp6

74

Appendix
A

Study ID: __ __ __ __ __ __
Date completed (MM‐DD‐YY): __ __ / __ __ / __ __
Name of Abstracter: ________________________________
Maternity Clinic: □ 1st □ 2nd □ 3rd □ NCMCH □ Other (Specify) ____________

STILLBIRTH RECORD ABSTRACTION FORM
A. GENERAL INFORMATION
Date and time reported (MM‐DD‐YY-HH-MM): __ __ / __ __ / __ __/ __ __/ __ __
B. Mother’s information
1. Registration number:
2. Age: __ __
3. Residential address: Province/city _____________ Soum/district ____________________
Bag/horoo ____________
4. Occupation: 1. Unemployed 2. Employed 3. Herder 4. Student 5. Pupil 6. Other
__________
5. Number of pregnancies (including current one):
6. Number of deliveries (not including current outcome):
7. Prenatal care: 1. Yes 2. No 3. No information
8. Which week of pregnancy enrolled in prenatal care? _____ weeks
9. Pregnancy complications:
a. No complications
b. Early toxemia
c. Mild preeclampsia
d. Severe preeclampsia
e. Eclampsia
f. Uterine bleeding
g. Premature rupture of membranes
h. Placenta previa/abruption
i. Amnionitis
j. Comorbidity
k. Fetal distress
l. Preterm labor
m. Other ______________________________________

75

10. Maternal morbidity
a. No comorbidity
b. Communicable disease
c. Cancer
d. Blood disease
e. Metabolic disease
f. Neurologic disease
g. Cardiovascular disease
h. Respiratory disease
i. Gastrointestinal disease
j. Urinal tract and reproductive organ disease
k. Other _____________________________
11. Antenatal medication usage: 1. Dexamethasone 2. Other _________________
12. Delivery type: 1. Vaginal (a. Normal b. Stimulated: oxytocin, misoprostol c. Intensified d.
Vacuum ) 2. Cesarean
13. Reason for Cesarean:
a. Late toxemia
b. Eclampsia
c. Recurrent cesarean
d. Delay in birth way
e. Placenta previa
f. Placenta abruption
g. Fetal distress
h. Maternal comorbidity
i. Other
14. Leading part: 1. Head 2. Bottom 3. Leg 4. Other __________________
15. Delivery complications:
a. Delivery delay
b. 1
st
stage weakening of delivery force
c. 2
nd
state weakening of delivery force
d. Other __________________________
16. Fetal complications:
a. Preterm birth
b. Meconium aspiration syndrome
c. Intrauterine growth retardation
d. Fetal asphyxiation
e. Congenital anomaly: Diagnosis ________________________
f. Other ______________________

76

B

Study ID: __ __ __ __ __ __
Date completed (MM‐DD‐YY): __ __ / __ __ / __ __
Name of Abstracter: ________________________________
Maternity Clinic: □ 1st □ 2nd □ 3rd □ NCMCH □ Other (Specify) ____________

CONGENITAL ANOMALY RECORD ABSTRACTION FORM
A. GENERAL INFORMATION
Date and time reported (MM‐DD‐YY-HH-MM): __ __ / __ __ / __ __/ __ __/ __ __
B. Mother’s information
1. Registration number:
2. Age: __ __
3. Residential address: Province/city _____________ Soum/district ____________________
Bag/horoo ____________
4. Number of deliveries: 1. First 2. 2-4 3. 5 or more 4. No info
5. Outcome of previous pregnancies: (in number)
a. Miscarriage
b. Intentional abortion
c. Pregnancy without growth
d. Extrauterine pregnancy
e. Stillbirth
f. Livebirth
C. Infant’s information
1. Date of birth (MM‐DD‐YY): __ __ / __ __ / __ __
2. Gestational age: ______ weeks_______ days
3. Birth weight: ______ g
4. Gender: 1. Male 2. Female 3. Uncertain 4. No info
5. Twins: 1. Yes 2. No
6. If twin, how many had congenital anomalies: 1. One 2. Two 3. 3 or more 4. No info
7. Whether infant lived or not: 1. Yes 2. No

77

8. If no, age when infant died: ___days _____ hours ______minutes
D. Diagnosis
1. Diagnosis period: 1. Prenatal 2. During delivery 3. During first week 4. 8-28 days
2. Gestational age when diagnosed in prenatal: ____ weeks _____ days
3. Whether diagnosed in pathology: 1. Yes 2. No
4. Detailed diagnoses of every congenital anomalies detected:
a. Neurological system anomalies _______________________________________
b. Eye, ear, face and neck anomalies ____________________________________
c. Cardiovascular system anomalies _____________________________________
d. Respiratory system anomalies ________________________________________
e. Cleft lip and palate _________________________________________________
f. Gastrointestinal system anomalies ____________________________________
g. Reproductive system anomalies ______________________________________
h. Urinary tract anomalies _____________________________________________
i. Musculoskeletal system anomalies ____________________________________
j. Other anomalies ___________________________________________________
k. Chromosomal anomalies ____________________________________________
l. Multiple organ system anomalies ______________________________________

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Air pollution, mitochondrial function, and growth in children

PDF

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

PDF

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

PDF

The environmental and genetic determinants of cleft lip and palate in the global setting

PDF

Ambient air pollution and lung function in children

PDF

Associations between ambient air pollution and hypertensive disorders of pregnancy

PDF

Predicting neonatal outcomes among women diagnosed with severe preeclampsia and HELLP syndrome: a comparison of models

PDF

Genomic risk factors associated with Ewing Sarcoma susceptibility

PDF

Early life air pollution exposure, gestational diabetes mellitus, and autism spectrum disorder

PDF

Age related macular degeneration in Latinos: risk factors and impact on quality of life

PDF

Prenatal air pollution exposure, newborn DNA methylation, and childhood respiratory health

PDF

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

PDF

The role of inflammation in non-Hodgkin lymphoma etiology

PDF

Visual acuity outcomes after cataract extraction in Chinese Americans: the Chinese American Eye Study (CHES)

PDF

Evaluation of new methods for estimating exposure to traffic-related pollution and early health effects for large population epidemiological studies

PDF

The role of heritability and genetic variation in cancer and cancer survival

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

Effect of biomass fuel exposure on infant respiratory health outcomes in Bangladesh

PDF

Antibiotic resistance in a large ICU cohort: 1995-2002

PDF

A cohort study of air-pollution and childhood obesity incidence

Asset Metadata

Creator Enebish, Temuulen (author)

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

School Keck School of Medicine

Degree Doctor of Philosophy

Degree Program Epidemiology

Publication Date 07/21/2020

Defense Date 05/15/2020

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

Tag adverse birth outcomes,Air pollution,birth defect,congenital anomaly,environmental epidemiology,Epidemiology,Mongolia,OAI-PMH Harvest,particulate matter,reproductive epidemiology,stillbirth,Ulaanbaatar

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Franklin, Meredith (committee chair), Breton, Carrie (committee member), Habre, Rima (committee member), McKean-Cowdin, Roberta (committee member), Warburton, David (committee member)

Creator Email enebish@usc.edu,temuulen@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-337778

Unique identifier UC11664022

Identifier etd-EnebishTem-8705.pdf (filename),usctheses-c89-337778 (legacy record id)

Legacy Identifier etd-EnebishTem-8705.pdf

Dmrecord 337778

Document Type Dissertation

Rights Enebish, Temuulen

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