Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Street connectivity and childhood obesity: a longitudinal, multilevel analysis
(USC Thesis Other)
Street connectivity and childhood obesity: a longitudinal, multilevel analysis
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
1
STREET CONNECTIVITY AND CHILDHOOD OBESITY: A LONGITUDINAL, MULTILEVEL
ANALYSIS
by
Anita Khachikian
A thesis presented to the UNIVERSITY OF SOUTHERN CALIFORNIA
With the conferring degree/major/program of MASTER OF SCIENCE in APPLIED BIOSTATISTICS
AND EPIDEMIOLODY within the DEPARTMENT OF PREVENTATIVE MEDICINE
Degree conferral date of MAY 2016
2
TABLE OF CONTENTS
ABSTRACT ……………………………………………………………………………………………... 3
INTRODUCTION ………………………………………………………………………………………. 5
METHODS …………………………………………………………………………………………….... 9
RESULTS …………………………………………………………………………………………….... 21
DISCUSSION ………………………………………………………………………………………..… 29
CONCLUSION ……………………………………………………………………………………….... 31
SUPPLEMENTARY TABLES AND FIGURES ……………………………………………...………. 32
REFERENCES ………………………………………………………………………………………… 55
3
ABSTRACT
Background: Childhood obesity rates are still of major public concern. Previous studies have
shown associations with various built environment characteristics. One of the characteristics that has not
been thoroughly studied longitudinally is street connectivity, thus this study attempts to clarify this
association.
Methods: The study population included children ages 5-10 years old (N=4,550; 90% between
the ages 5.6-7.9) who were followed for up to four years in Southern California. BMI was measured
annually. Street connectivity was measured using the gamma index and average block size. A sensitivity
analysis was conducted by restricting the dataset to those who never moved from study baseline to end
of follow-up. Multilevel linear models were used to assess the associations between street connectivity
on attained BMI at age 10 and on BMI growth during follow-up. Two models were implemented into
this study. The first employs a modeling approach that outputs the gender-combined effects of street
connectivity. The second employs a modeling approach that outputs the gender-specific effects of street
connectivity. Potential effect modification was explored by social factors, such as population density.
Results: For the combined effect of males and females, after adjusting for gender, race/ethnicity,
community and nodes, a statistically significant effect was found for gamma index on BMI attained at
age 10 (p < 0.0001) and on BMI rate of growth during follow-up (p < 0.0001). In gender-specific
models, with the same adjustments, these effects were also statistically significant in males (p = 0.0003
and p < 0.0001, respectively) and in females (p < 0.0001 and p < 0.0001, respectively). Thus, after
adjustments, every two standard deviation increase of 0.12 in the gamma index was associated with
increases in predicted BMI attained by age 10 of 0.67 kg m
-2
(SE = 0.15 kg m
-2
), 0.62 kg m
-2
(SE = 0.17
kg m
-2
) and 0.74 kg m
-2
(SE = 0.17 kg m
-2
) for combined, male and female populations, respectively.
Furthermore, after adjustments, every two standard deviation increase of 0.12 in the gamma index was
4
associated with increases in predicted BMI growth during follow-up of 0.13 kg m
-2
(SE = 0.02 kg m
-2
),
0.14 kg m
-2
(SE = 0.02 kg m
-2
) and 0.12 kg m
-2
(SE = 0.02 kg m
-2
) for combined, male and female
populations, respectively.
For the combined effect of males and females, after adjustments, a statistically significant effect
was found for average block size on BMI attained at age 10 (p = 0.004) and on BMI rate of growth
during follow-up (p = 0.0006). In gender-specific models, with the same adjustments, these effects were
also statistically significant in males (p = 0.042 and p = 0.018, respectively) and in females (p = 0.027
and p = 0.009, respectively). Thus, after adjustments, every two standard deviation increase of 5.04 in
average block size was associated with decreases in predicted BMI attained by age 10 of 0.40 kg m
-2
(SE = 0.15 kg m
-2
), 0.35 kg m
-2
(SE = 0.20 kg m
-2
) and 0.45 kg m
-2
(SE = 0.20 kg m
-2
) for combined,
male and female populations, respectively. Furthermore, after adjustments, every two standard deviation
increase of 5.04 in average block size was associated with decreases in predicted BMI growth during
follow-up of 0.10 kg m
-2
(SE = 0.05 kg m
-2
), 0.05 kg m
-2
(SE = 0.05 kg m
-2
) and 0.10 kg m
-2
(SE = 0.05
kg m
-2
) for combined, male and female populations, respectively.
A sensitivity analysis revealed no major differences between the complete dataset and the dataset
that restricted to only those who did not move from baseline to follow-up, thus differential
misclassification of the exposure variables are not of concern.
Conclusion: Households with higher gamma indices tend to increase BMI trajectories in
children and households with higher average block size tend to decrease BMI trajectories in children.
This counters the hypothesis that street connectivity is negatively associated with obesity in children.
Population density modified the average block size and BMI relationship. Specifically, a stratified
analysis shows that for both males and females living in low population density communities, average
block size has negative associations with attained BMI and with BMI rate of growth, although the
5
associations are only statistically significant in males Also, for both males and females living in high
population density communities, average block size has positive associations with attained BMI and
with BMI growth, although the only statistically significant association is the rate of growth among
males. Additional research is needed in order to more completely assess differences in effects of street
connectivity measures within males and females across the entire childhood period across different
population densities.
INTRODUCTION
Childhood obesity is of major concern since it is a risk factor for several diseases and has been
linked to various health complications, such as short-term and long-term cancers, as well as
cardiovascular, metabolic, pulmonary, gastrointestinal, skeletal and other diseases that carry through
adulthood [1, 2]. In 2011-2012, about 31.8% of children in the United States were considered
overweight or obese [3]. The current trends in the United States among overweight adolescents are
projected to increase prevalence of obesity of 35-year-olds by the year 2020 to a range of 30-37% in
men and 34-44% in women [4].
Effective obesity prevention is conceivable only when all causal pathways are identified and
understood. The fundamental actuality of weight gain corresponds to overconsumption of energy and
reduction of energy expenditure. The current environment, especially that in the United States, promotes
overconsumption of energy by making it possible to have easy access to inexpensive, good-tasting,
energy dense food, and may promote a reduction of energy expenditure by reducing the need for
physical activity in daily living and increasing behaviors of sedimentary activities, such as driving to
destinations and watching television, respectively. It is crucial to get rid and discourage these behaviors
which are instigated by our built environment in order to abate the current obesity epidemic [5].
6
Past research on obesity predominately focused on genetic research, which expanded knowledge
in regards to physiological characteristics of body mass [6]. However, recent studies have attempted to
infer the associations between obesity and the built environment [7-19]. This shift to studying the built
environment is justified because the rise in obesity prevalence in recent decades cannot be possibly
attributed to a change in our genetic structure in such a short timeframe [7]. Even further, there is
evidence that access to opportunities for walking and physical activity may suppress the genetic risk of
high body mass index (BMI) [20] . Thus understanding the role of the built environment in relation to
obesity is vital in order to provide effective prevention methods that focus on creating environments that
encourage an active lifestyle.
The term ‘‘built environment’’ refers to all facets of a person’s surroundings that are man-made.
In an epidemiologic review, 84% of studies found a statistically significant association between some
aspect of the built environment and obesity, and even more, links the built environment to physical
activity, dietary intake and obesity [8]. Figure 1 shows a conceptual path model showing possible
influences of the built environment in the context of other variables that may affect the attained level of
BMI in children. As we can see Figure 1, there are numerous pathways that effect obesity development.
Diet, physical activity and various morbidities, specifically cardiovascular disease, cancer, diabetes and
asthma, directly influence obesity. The school environment may influence a child’s diet and/or
institutional physical activity and thereby may influence obesity development. Family influence can
directly be related to diet and can also have genetic effects, which may directly influence obesity
formulation or may affect morbidities. The built local environment has numerous pathways. Part of the
local built environment is the type and frequencies of food access, which influences dietary behaviors.
Another part of the built environment relates to frequencies of crime and air pollution, which may
indirectly affect physical activity. Furthermore, crime and air pollution may lead to acute and/or chronic
7
health effects, which also have an effect on physical activity. The last pathway of the local built
environment relates neighborhood characteristics, such as districts, edges, nodes, paths and landmarks,
directly to physical activity. Consequently, there is an interconnected web of factors that affect obesity
development. Thus in order to have a complete and thorough investigation of causes of childhood
obesity, one must incorporate a vast collection of not only individual factors but community factors as
well.
Street connectivity attempts to summarize the degree of intersection of roads in any given area.
Higher street connectivity refers to street networks that are more grid-like compared to low street
connectivity, which may include many cul-de-sacs and long blocks. Basically, it is a proxy for
walkability in an area. Low street connectivity increases the distances between destinations, which are in
turn thought to discourage walking and/or physical activity [21].
In order to obtain a comprehensive knowledge on this topic, Pubmed was used to identify all
possible studies. The search started with all possible pairs of the obesity keywords: obesity, childhood
obesity, BMI; and the environmental keywords: connectivity, street connectivity, walkability, built
environment. Using several reference lists from these identified studies, further relevant studies were
identified, if not yet identified. Table 1 shows identified children’s studies on BMI.
Various components of the built environment have been investigated. For instance in numerous
adult studies, higher presences of fast food restaurants have been found to be significantly positively
associated with obesity [13, 22, 23]. Also, higher accesses to parks[19, 24], land-use mixes [13, 16, 25]
and residential densities [16] have all been found to be significantly negatively associated with obesity.
To date, overall research on street connectivity in urban environments find a positive effect on
physical activity behavior in adult studies [12]. Theoretically, a neighborhood with higher street
connectivity is thought to give more opportunities to residents of that area to walk, in the form of
8
physical exercise and/or transportation, which would in turn reduce the risk of obesity and its related
health troubles. Various studies find a significant positive association with poorer street connectivity and
obesity, majority of which are cross-sectional in nature [14, 19, 26-30]. Further, there is overwhelming
evidence that physical activity is directly related to overweight and obese prevalence [31]. Thus, it is
reasonable to hypothesize that areas of higher street connectivity will have a lower overweight and
obesity prevalence.
However, there are studies that have reported contradictory results [16, 32-34]. Although it is
possible that this may be due to differences in the geographic settings of the studies [35]. Another issue
that may be causing disparities between studies is that street connectivity and walkability, which are
both tentatively measuring the same effect, each have various methods of measurement. For instance,
street connectivity can be measured by the gamma or alpha indices, as well as other methods, while
walkability is measured via different methods, such as the Walk Score. Furthermore, it is important to
note that the vast majority of the large cohort studies conducted have been strictly cross-sectional in
nature, and thus reverse-causality is of concern i.e. those people with higher BMI choose to live in areas
of lower street connectivity perhaps since walkability is of no concern to these individuals.
When compared to adult studies, there are even fewer studies that infer about the relationships
between the built environment and childhood obesity. Identified studies show similar results and
restrictions as with that of adult studies. For instance, various studies have found significant inverse
associations with proximities to parks and recreational resources [36-39] and significant positive
associations with proximities to fast food establishments [37, 38, 40-43]. Also, positive associations
have been identified for property crime [41] and homicide rates [44]. Furthermore, higher densities of
street trees [44] and intersections [46] have been found to be negatively associated with obesity in
children. Moreover, studies have shown that built environment characteristics that increased walkability
9
are associated with lower BMI in children [36, 39, 46-51]. However again, majority of the larger cohort
studies have been cross-sectional. What’s more, there are very few large longitudinal studies that infer
about the childhood obesity and street connectivity association explicitly. Thus this study attempts to
clarify the true association.
The Children’s Health Study (CHS), a cohort planned by the University of Southern California,
is a longitudinal study of Southern Californian children. The CHS is a rich resource for studying
childhood obesity and papers have already published using these data. Negative associations of BMI and
organized physical activity [52] and proximity to urban parks/recreational areas [40]; as well as positive
associations of BMI and traffic-related air pollution [53], asthma [45], all types of food access [54],
parental stress [55], second-hand smoke, maternal smoke during pregnancy and near-roadway pollution
exposure [56] have been established.
This study examines the relationship between the gamma index and average block size,
separately, as measures of street connectivity on attained BMI at age 10 and on BMI trajectory over 4
years in pre-adolescent school children from the 2003 cohort of the CHS. This dataset is unique in that it
provides both longitudinal and cross-sectional data on a variety of individual and community level
factors that may confound associations between childhood obesity and the gamma index or average
block size. It is crucial to have data on a wide-range of factors since there are known to be various
confounders and effect modifiers of childhood obesity and physical environmental associations [57].
METHODS
Study population and design
In January 1992, the University of Southern California was awarded a contract from the
California Environmental Protection Agency’s Air Resources Board (ARB) to begin a 10-year cohort
10
study of the health effects of air pollution in southern California. The main purpose of the study was to
assess the effects of chronic exposure to criteria pollutants such as ozone (O3), particulates (PM10,
PM2.5), nitrogen dioxide (NO2), and nitric (HNO3) and hydrochloric acids (H
+
) [58].
A preliminary power calculation for conducting a successful multivariate analysis of two or more
pollutants indicated that at least 10 communities were necessary. However, the communities could not
be selected in such a way such that the correlations in pollutant levels across groups would be large.
Four main pollutants were considered during the community site selection process: O3, NO2, HNO3 and
PM10. Sites were selected in ways that would maximize the variation in pollution profiles within and
among the pollutants. The California ARB’s monitoring program had extensive data on these pollutants
readily available. For each pollutant considered, average levels were calculated for 1986 through 1990
for 86 monitoring stations scattered across Southern California. The pollutant levels were then converted
to standardized units for each community. However, for some pollutants, mainly acids, the values had to
be interpolated from other stations on an inverse-distance weighted basis. Each community was assigned
a “profile” by recording it as either above (+) or below (-) the mean level for each pollutant. Then within
each profile, one to three communities were selected on the basis that their sum of squared standardized
pollution levels were large. Due to financial limitations, twelve communities were chosen to be included
in the study.
The following was the criterion for enrolling potential schools: (1) schools resided in a location
that was in a preselected community of interest based on air pollution levels and patterns, (2) there
existed a sufficient population of target-aged children, (3) there existed a multitude of children attending
school from the immediate neighborhood, (4) there existed a demographic similarity with other potential
and participating community school sites, (5) schools resided in reasonably stable communities in terms
of residential migration, in order to improve chances for longitudinal follow-up with participants, (6)
11
there existed a absence of localized air pollution sources and (7) there existed a proximal location to a
fixed-site air monitoring station [59].
A minimum of two elementary schools, one junior high school and one senior high school were
enrolled in each participating community so that follow-up from elementary school through senior high
school of subjects would be conceivable. Approvals of participation were obtained from school districts.
On-site meetings with school administrators and teachers were implemented in order to execute efficient
administration of annual questionnaires, lung function testing and exposure monitoring. A requirement
for participation was the cooperation of the student and also written consent of an informed parent (or
legal guardian) [59].
In 1993, three cohorts were established. Cohort A included 938 tenth graders, Cohort B included
1,048 seventh graders and Cohort C included 2,192 fourth graders. In 1996, another cohort (Cohort D)
was recruited and included 2,081 fourth graders. In each case, students who continued to reside in the
twelve communities and attend participating study schools were evaluated annually through high school
graduation [59].
In 2003, a new cohort (Cohort E) of 45 participating schools located in 13 southern California
communities were included for recruitment and all enrolled kindergarten and first grade students during
that time were given a questionnaire and informed consent to take home to be completed by parents or
guardians. There were nine communities included from Cohorts A-D and five new communities. Figure
2 shows a map of the participating communities by cohort inclusion. The study was approved by the
University of Southern California Institutional Review Board. Informed consent and questionnaires were
completed and returned for 5,341 (65%) of 8,193 eligible children [60].
This study is a secondary analysis of Cohort E that includes data from baseline and annual
follow-up visits over the 4-year. This analysis was restricted to 4,550 children who were followed up for
12
at least 1 year (i.e. two or more measurements of height and weight). There were no missing
measurements of BMI for any participants before they were censored. Average follow-up for the
subjects was 2.9 years before being censored [55]. Children were excluded from the analytic analysis if
they were missing data describing the gamma index or average block size. Thus, the study population
for both the analyses of the gamma index and average block size includes 4,117 children, which is 77%
of all respondents.
Assessing variables from the CHS
Annual questionnaires were administered and included information on demographics, medical
history (including an extensive set on respiratory illnesses and conditions i.e. pneumonia, bronchitis,
coughing, wheezing, etc.), a housing survey, and history of exposure to ETS, pests and pets. Time-
activity questionnaires included information on what kind, where and times of physical activity
occurrences and were administered bi-annually (once in early winter with the demographic and histories
questionnaire and once in the spring while administering lung function testing). Teachers instructed the
students to complete the questionnaires at home with their guardians in order for the answers to be as
accurate as possible. Spanish translated questionnaires were also available. Project staff or locally
recruited volunteers logged in questionnaires each day at the schools, then via personal courier or
expedited mailing, were sent to research personnel at USC [59].
Participant characteristics
The CHS Annual Questionnaire gathered data on various individual and household
characteristics. Gender was classified as male or female. Race/ethnicity was classified as African
American, Asian, Hispanic, Non-Hispanic White or other. For this analysis, parental education, which
13
was used as a marker for socio-economic class, was dichotomized as either above high school education
or high school education and lower. Second-hand smoke was dichotomized as either no one smoked in
the home or anyone smoked in the home. Current coverage of the child by a health insurance plan (yes
or no) was also ascertained. Prevalence of asthma, which was dichotomized as yes or no for this analysis,
was defined by assessing whether or not there was a reported use of controller medications for asthma.
Also, participants without a physician’s diagnosis of asthma but who had a severe wheeze in the
previous 12 months were classified as prevalent asthmatics in order to identify undiagnosed asthma
cases due to poor access to medical care. Allergic characteristics were ascertained by prevalence of hay
fever or whether the participant experienced a sneezing/blocked/runny nose when they did not have a
cold. Being a Spanish speaker (yes or no) was ascertained as whether a Spanish translated questionnaire
was completed. Housing characteristics included presence of pets (dog, cat, bird, other furry or hairy
pets, or other pets), cockroaches, rats/mice, carpeting, water damage/mold/mildew, use of air conditioner
and combustion source for nitrogen dioxide (a gas oven/stove/heating unit with pilot light). Parental
stress was determined using a 4-item version of the Perceived Stress Scale (PSS), which measures the
extent that individuals believe their lives are unpredictable, uncontrollable, or overwhelming [61]. Lung
function testing, which together with school absences data were used to determine the gravity of
respiratory illnesses, incorporated spirometers and its computerized interfaces and data logging
capabilities. Project staff tested subjects individually in a consistent and prearranged manner [59].
Data on the characteristics of participating schools (which describe the social environment of the
children) were obtained from the California Department of Education for the 2002-2003 school year.
Information was collected on the percentage of students receiving free meals (as an indicator of
deprivation among students), percentage of students with minority ethnicity, percentage of students who
were English-language learners, and the Ethnic Diversity Index (which measure how students were
14
distributed among the ethnic categories reported to the Department). In addition, the school-level
academic performance measured by the Academic Performance Index. This measurement includes
whether the school met academic ‘adequate yearly progress’ criteria and whether the school received
funding related to Title I under the No Child Left Behind Act of 2001. Note that these represent both
academic performance and deprivation at schools [62].
In order to accurately measure BMI (kg of weight/height in m
2
), trained technicians measured
height and weight without shoes annually at the child’s school by following a standardized procedure
using a calibrated medical scale. These measurements were recorded to the neatest 1 centimeter for
height and 1 pound for weight [53]. Also, the Centers
for Disease Control and Prevention (CDC) 2000
gender-specific
BMI-for-age reference values were used to calculate the BMI percentiles of the children
in this cohort. According to the CDC’s website, a BMI percentile of less than 5% is underweight, a
percentile between 5% and 85% is normal or healthy, a percentile between 85% and 95% is overweight
and greater than or equal to 95% is considered obese.
Air pollution and traffic related metrics
In order to assign the history of pollutant exposure for each subject, the participants’ residential
histories were geocoded and linked to the pollutant monitoring data using either the ZIP code centroid,
city centroid or Zip Code Weighted Centroid Coordinates, depending on what information was available
for each participant. The pollutants O3, PM10, PM2.5, NOx and nitric, formic and hydrochloric acid levels
were determined at the community level. O3 levels were measured inside and outside the schools. Also,
through a separately funded project, O3, PM10, PM2.5, formaldehyde, air exchange rates and airborne
acids were measured at a sample of the residences [59].
15
Traffic related pollution was determined using the CALINE4 dispersion model. This model uses
Gaussian plume dispersion parameters with traffic data, emissions factors, and local meteorology to
estimate exposure to near-roadway pollutants for the children’s residences based on a model for the
incremental increase in NOx (parts per billion) above regional background levels. Traffic exposure
variables were based on the California Department of transportation Functional Class (FC) data for the
year 2000 [53]. Assessments of traffic-related pollutants were carried out by estimating the distances of
each participant’s baseline residence to the nearest major road (including freeways, highways and
arterial roads). Using the TeleAtlas Multinet road network data, participants’ residences were
standardized and their location was geocoded to 13 m perpendicular to the side of the adjacent road.
Also, ArcGIS software was used to measure the distance to the nearest major road from the child’s
residence. Further, each direction of travel was represented as a separate roadway, and the shortest
distance was estimated from the residence to the middle of the nearest side of the freeway or major road.
Distance from the child’s home to a major road was categorized as less than 75 m, 75-150 m, 150-300 m
and greater than 300 m. Using a line source dispersion model, residential exposure to fresh traffic-
modeled pollutants from freeway and non-freeway sources were measured while accounting for traffic
volume, wind speed, and direction in each community [60]. Also, the annual average daily traffic
(AADT) volumes were combined to the TeleAtlas network. Furthermore, traffic data was based on
continuous measurements on freeways, highways, major arterials and intermittent measurements within
the previous thee years on other major roads. Additionally, a kernel density function was estimated to
smooth the influence of traffic around the home, which down-weights the influence of traffic exposure
as a function of Euclidian distance from the child’s home [53].
16
Built environment characteristics
Built environment characteristics were organized with geographic information system (GIS)
through a classic framework identified by Lynch, which allows the classification of how people perceive
and navigate within cities through five features: districts, edges, paths, nodes and landmarks [63]. The
GIS provides tools that allow for processing of spatial data into information [64]. Districts are defined as
distinct neighborhoods or easily recognizable zones that are internally homogeneous. Edges are
boundaries and barriers between these districts. Paths connect within and between districts. Nodes are
destinations and places where people gather and landmarks are used to guide people through their
neighborhoods. Having food stores (yes or no) and the number of total food stores were measured within
500 m road network buffer. The gamma index, average block size, nodes, parks and recreations,
normalized difference vegetation index (NDVI) green cover and recreation programs (in 5 k and 10 k
buffers around homes) were measured on a continuous scale (within 500 m Euclidean buffer around
children’s residences, unless otherwise noted) [63]. Figure 3 is a summary table of built environment
variables organized under the Lynch framework.
In this study, street connectivity was assessed via two measures. The first measure is the gamma
index within a 500 meter buffer around the participant’s home. The gamma index is a ratio of the
number of links in the network to the maximum possible number of links between nodes. The maximum
possible number of links is expressed as 3 * (# of nodes – 2) because the network is abstracted as a
planar graph. Here, links are defined as roadway or pathway segments between two nodes, and nodes
are intersections or the end of a cul-de-sac. [21, 64] The second measure of street connectivity uses the
average block sizes within a 500-meter buffer around the participants home.
17
Population-level characteristics
Community characteristics, which also describe the social environment of the children, used the
same measurements as that for school characteristics, only difference being that the measurements are
based on data describing census block groups that were aggregated to the community level. Also, two
measures of the dissimilarity index were used to describe residential segregation of African Americans
and Hispanics, respectively, compared to all other races. Furthermore, data from the Federal Bureau of
Investigation Uniform Crime Reports for the California Department of Justice for the year 2008 were
obtained to gather information on crime for each participating community and stress relating to exposure
of violence at the individual level. Specifically, these data report various types of violent (including
murder, rape, robbery and aggravated assault) and property (including burglary, larceny, and motor-
vehicle theft) crimes and were measured as crimes per 100,000 people [62].
Neighborhood characteristics were defined by matching the census tract of residence to data
from the U.S. 2000 Census. Some items included were measures of median household income,
percentage with no high school diploma, percentage unemployed, percentage living in poverty,
population density and racial compositions. All factors were measured within a 500 meter Euclidean
network buffer around the children’s homes. The Gini coefficient was used to describe income
inequality within neighborhoods [62].
Statistical Methods
Flexible multilevel linear models have been previously proposed to analyze this type of data [65].
This modeling approach properly adjusts for age- and gender- specific effects on BMI growth in
children, provides an effective mechanism for assessing effects of risk factors on BMI level and growth,
and also implicitly adjusts for baseline levels of BMI [53]. For this study, two modeling approaches
18
were considered. The first employed a combined model that outputs the effect of the gamma index (and
again for average block size) on attained BMI at age 10 and on the rate of growth during the follow-up
period, for the total study population. The second employed a modeling approach that outputs the
gender-specific effects of the gamma index (and again for average block size) on attained BMI at age 10
and on the rate of growth during the follow-up period separately for males and females based on a
jointly fitted model. Executing these multilevel linear models allows for the examination of the effects
within individuals, between individuals and between communities.
For the model with gender-combined effects, let Ycij denote the outcome measure (i.e. BMI) for
subject i in community c at attained age ŧij (centered at 10 years of age) for the j
th
year. Consider the
following three-level linear model:
Level 1: Ycij=Aci+Bciŧij+ϒ1zcij+ecij
where Aci denotes the subject-specific BMI level at age 10, Bci denotes the subject-specific BMI rate of
growth, adjusted for the time-dependent covariates denoted by zcij (e.g., asthma status).
Level 2a: Aci=Ac+β1G ci+ϒ2zci+eci
Level 2b: Bci=Bc+δ1Gci+δ2zci+fci
where Ac and Bc denote community specific attained level of BMI at age 10 and rate of BMI growth,
respectively, adjusted for time-independent covariates zci (e.g., race-ethnicity). Here, eci and fci denote
subject-specific random intercept and slope, assumed to follow normal distributions with means zero
and variances Ve(ci) and Vf(ci), respectively, while allowing random effects to be correlated. . Here, β1
represents the regression parameter associated with the effect of the exposure variable of interest (i.e.
gamma or average block size) on the attained level of BMI at age 10, and δ1 represents the regression
19
parameter associated with the effect of the exposure variable of interest (i.e. gamma or average block
size) on the linear rate of growth of BMI (essentially forming an interaction between Gci and age (ŧij)).
Level 3a: Ac=β0+ϒ3zc+ec
Level 3b: Bc=δ0+ϒ4zc+fc
where zc denotes community level covariates. Combining levels (1)-(3) gives the following mixed
effects model:
Ycij= β0+β1Gci +δ0ŧij+δ1Gciŧij+ϒ2zci+ϒ3zc+δ2zciŧij +ϒ1zcij +ϒ4zcŧij+ecij+ec +eci+ +fcŧij+ +fciŧij
The terms we are interested in correspond to β1 and δ1, which simultaneously estimate effects of
the main exposure variable on BMI level attained at age 10 and on the change in BMI during the follow-
up period, respectively.
For the second model with gender-specific effects, let Ycij denote the outcome measure (i.e. BMI)
for subject i in community c at attained age ŧij (centered at 10 years of age) for the j
th
year. Consider the
following three-level linear model:
Level 1: Ycij=Aci+Bciŧij+ϒ1zcij+ecij
where Aci denotes the subject-specific BMI level at a given age, Bci denotes the subject-specific BMI rate
of growth, adjusted for the time-dependent covariates denoted by zcij (e.g. asthma status).
Level 2a: Aci=Ac+βMGci+βFGci +ϒ2zci+eci
Level 2b: Bci=Bc+δMGci+δFGci +δ2zci+f ci
20
where Ac and Bc denote community specific attained level of BMI at age 10 and rate of BMI growth,
respectively, adjusted for time-independent covariates zci (e.g. race-ethnicity). Here eci and fci denote
subject-specific random intercept and slope, assumed to follow normal distributions with means zero
and variances Ve(ci) and Vf(ci), respectively, while allowing random effects to be correlated. Here, βM and
βF represent the regression parameters associated with the effect of the exposure variable of interest (i.e.
gamma or average block size) on the attained level of BMI at age 10 for males and females, respectively.
Further, δM and δF represent the regression parameters associated with the effect of the exposure
variable of interest on the linear rate of growth of BMI (essentially forming an interaction between Gci
and age (ŧij)) for males and females, respectively. We note that this modeling is a reparametrized version
of a model that has appropriate interaction terms between the main effects of main risk factors (i.e.,
gamma index or average block size) and gender.
Level 3a: Ac=β0+ϒ3zc+ec
Level 3b: Bc=δ0+ϒ4zc+fc
where zc denotes community level covariates. Combining levels (1)-(3) gives our cohesive mixed effects
model:
Ycij= β0+ βMG ci+βFG ci +δ0ŧij+ δMGci ŧij +δFGci +ϒ2zci+ϒ3zc+δ2zciŧij +ϒ1zcij +ϒ4zcŧij+ecij+ec +eci+ +fcŧij+
+fciŧij
The terms we are interested in correspond to βM and δM for males and βF and δF for females.
These denote the regression parameters, separately for males and females, of the main exposure variable
on BMI level attained at age 10 and on the change in BMI during the follow-up period, respectively.
For both the combined and gender-specific models, the same model building techniques were
implemented. Also, these processes were carried out twice. Once for assessing the relationship between
21
the gamma index and BMI, and then again for assessing the average block size and BMI relationship.
All statistical analyses were conducted using the SAS 9.4 package.
Indicator variables were created for all categorical variables, including a missing indicator
variable to avoid unnecessary loss of data due to missingness in adjustment variables. We assumed that
data were missing at random [66]. Any missing observations for continuous variables were deleted from
analyses. For individual-level covariates, community was treated as a fixed effect. For population-level
covariates, community was treated as a random effect. First, a univariate analysis was conducted in
order to test the associations of various covariates on BMI. The covariates that were considered for this
study are listed in Table 3. Any covariate that was related to BMI (p ≤ 0.20) was then checked to see if it
was a confounder in the BMI and the main exposure variable (i.e. the gamma index, average block size)
relationship. A covariate was considered to be a confounder if the effect of the main exposure variable
on BMI level attained at age 10 and/or the effect of the main exposure variable on the rate of growth of
BMI during follow-up, changed by more than 10%. The base model for both the gamma index and
average block size models adjusts for gender, race/ethnicity and community. Any significant
confounders identified were added to the base models. Effect modification was assessed for four
variables using the gender-combined model: population density (within 500 m), total freeway NOx (parts
per billion at the child’s residence), property crime rate (per 100,000) and gender. Statistically
significant confounders and interactions (except for the gender interactions) were then implemented into
the gender-specific model.
RESULTS
Participant characteristics were analyzed using baseline values and are listed in Table 2. A total
sample of 4550 participants was analyzed. All participants had at least two BMI measurements. The
22
mean (SD) age of participants was 6.67 (0.71) years, with missing observations for 3 participants. The
gender composition consisted of males (50.48%) and females (49.47%), with missing data for two
participants. The mean (SD) height and weight of participants was 120.15 (6.55) centimeters and 53.84
(13.03) pounds, respectively, with no missing observations. The mean (SD) BMI at baseline was 16.78
(2.81) kg/m
2
. Within males, the mean (SD) BMI at baseline was 16.86 (2.81) kg/m
2
. Within females, the
mean (SD) BMI at baseline was 16.70 (2.80) kg/m
2
. The mean (SD) BMI at the end of follow-up was
19.35 (4.21) kg/m
2
. Within males, the mean (SD) BMI at the end of follow-up was 19.50 (4.26) kg/m
2
.
Within females, the mean (SD) BMI at the end of follow-up was 19.19 (4.15) kg/m
2
. Most participants
(70.24%) were considered to be at or below normal BMI, according to the CDC’s 2000 gender-specific
BMI-for-age reference values. Overweight participants consisted of (14.64%) and the rest were
considered obese (15.12%). The racial composition consisted of African Americans (2.68%), Asians
(3.19%), Hispanics (54.11%), Non-Hispanic whites (32.18%) and others (7.85%), with no missing
observations for these categories. The categories of asthma presence comprised of present (12.40%) and
absent (76.95%), with missing observations for 485 participants. Participants were categorized as
Spanish speaking (75.10%) or not (24.90%), with no missing observations. Parental education consisted
of above high school (54.40%) and high school or less (37.05%), with 389 missing observations.
Presence of second-hand smoke consisted of present (6.79%) and absent (87.08%), with missing
observations for 279 participants. The categories of dog ownership comprised of present (27.32%) and
absent (65.52%), with missing observations for 326 participants. The categories of owning any sort of
pet comprised of present (71.03%) and absent (22.24%), with missing observations for 306 participants.
Presence of medical insurance comprised of present (81.08%) and absent (10.70%), with missing
observations for 374 participants. The mean (SD) of population density was 1,703.90 (1,426.65) within
a 500 m buffer around the participants’ home, with missing observations for 582 participants.
23
Various individual-level community characteristics have also been described in Table 2. The
gamma index had a mean (SD) of 0.40 (0.06) within a 500 m buffer around the participants’ home, with
missing data for 433 participants. Average block size had a mean (SD) of 0.79 (2.52) within a 500 m
buffer around the participants’ home, with missing data for 433 participants. Node presence had a mean
(SD) of 48.39 (20.47) within a 500 m buffer around the participants’ home, with missing data for 433
participants. Presence of parks and recreational areas, which was measured in acres, had a mean (SD) of
4.95 (10.60) within a 500 m buffer around the participants’ home, with missing data for 582 participants.
Recreational programs measured within 5 kilometers had a mean (SD) of 29.75 (34.20), with missing
observations for 433 participants. Recreational programs measured within 10 kilometers had a mean
(SD) of 48.03 (44.34), with missing observations for 433 participants. NDVI green cover had a mean
(SD) of 0.90 (0.10) within a 500 m buffer around the participants’ home, with missing data for 433
participants. Traffic density had a mean (SD) of 49.24 (104.93) within a 150 m buffer around the
participants’ home, with missing data for 86 participants. Total NOx emissions, measured as parts per
billion, had a mean (SD) of 19.49 (18.92) at participants’ home, with missing data for 86 participants.
Presence of food stores within a 500 m road network buffer consisted of present (46.97%) and absent
(43.52%), with missing observations for 433 participants. Total number of food store within a 500 m
road network buffer had a mean (SD) of 2.50 (4.57), with missing data for 433 participants.
Lastly in Table 2, all population-level community characteristics had no missing observations.
Violent crime rate had a mean (SD) of 511.73 (268.04) per 100,000 persons, while the forcible rape
crime rate had a mean (SD) of 26.73 (9.06) per 100,000 persons. Murder crime rate had a mean (SD) of
6.14 (5.55) per 100,000 persons, while the robbery crime rate had a mean (SD) of 139.15 (107.89) per
100,000 persons. The aggravated assault crime rate had a mean (SD) of 339.71 (166.66) per 100,000
persons, while the property crime rate had a mean (SD) of 3,125.15 (1,025.93) per 100,000 persons.
24
Motor-vehicle crime rate had a mean (SD) of 574.62 (318.10) per 100,000 persons. Furthermore for
these participants, the community levels of the percentage living in poverty, percentage of unemployed
and the population density (per square mile) had a mean (SD) of 0.15 (0.06), 0.08 (0.02) and 0.002
(0.001), respectively.
Upon assessing univariate associations with these covariates on BMI, we see that all covariates
had a statistically significant (p ≤ 0.20) intercept and/or slope, except for parks/recreations access and
the population property crime rate. Thus, these two covariates were excluded from the rest of the
analyses. Table 3 shows the estimate, standard error and p-value for intercepts and slopes for all
covariates.
In order to assess confounding for the gamma index on BMI relationship, the dataset was
restricted to those who had information on the gamma index (N=4,117). The base model’s effect
estimate for the gamma index effect resulted in 1.3853. The only identified confounder was nodes,
which outputted an effect estimate for the gamma index effect as 2.0634. This results in a 48.9% change
in the gamma index effect. The Pearson correlation coefficient between gamma index and nodes was
found to be 0.41 (p < 0.0001). Thus, gamma index and nodes are not considered highly correlated,
thereby making nodes a confounder in the gamma index and BMI relationship. Table 4 summarizes this
confounding check for the gamma index and BMI relationship.
In order to assess confounding for the average block size on BMI relationship, the dataset was
restricted to those who had information on average block size (N=4,117). The base model’s effect
estimate for the average block size effect resulted in -0.02137. The only identified confounder was
nodes, which outputted an effect estimate for average block sizes effect as -0.03332. This results in a
55.92% change in the average block size effect. In order to assess co-linearity, the correlation structure
between average block size and nodes was identified. The Pearson correlation coefficient between
25
average block size and nodes was found to be -0.31 (p < 0.0001). Thus, average block size and nodes are
not considered highly correlated, thereby making nodes a confounder in the average block size and BMI
relationship. Table 5 summarizes this confounding check for the average block size and BMI
relationship.
The main effects of the gamma index and average block size on BMI are summarized in Table 6.
For the combined effect of males and females, after adjusting for gender, race/ethnicity, community and
nodes, a statistically significant effect was found for gamma index on BMI attained at age 10 (p <
0.0001) and on BMI rate of growth during follow-up (p < 0.0001). In gender-specific models, with the
same adjustments, these effects were also statistically significant in males (p = 0.0003 and p < 0.0001,
respectively) and in females (p < 0.0001 and p < 0.0001, respectively). Thus, after adjustments, every
two standard deviation increase of 0.12 in the gamma index was associated with increases in predicted
BMI attained by age 10 of 0.67 kg m
-2
(SE = 0.15 kg m
-2
), 0.62 kg m
-2
(SE = 0.17 kg m
-2
) and 0.74 kg
m
-2
(SE = 0.17 kg m
-2
) for combined, male and female populations, respectively. Furthermore, after
adjustments, every two standard deviation increase of 0.12 in the gamma index was associated with
increases in predicted BMI growth during follow-up of 0.13 kg m
-2
(SE = 0.02 kg m
-2
), 0.14 kg m
-2
(SE
= 0.02 kg m
-2
) and 0.12 kg m
-2
(SE = 0.02 kg m
-2
) for combined, male and female populations,
respectively.
A model with gender interaction on the gamma index, essentially a reparameterized version of
the same model we discussed above with gender-specific effects, was fitted in order to formally test for
interaction. Based on this model, there was no statistically significant interaction between the gamma
index and gender for attained BMI at age 10 (p = 0.356) and for BMI rate of growth during follow-up (p
= 0.824).
26
In assessing average block size, all models were adjusted for gender, race/ethnicity, community
and nodes. For the combined effect of males and females, after adjustments, a statistically significant
effect was found for average block size on BMI attained at age 10 (p = 0.004) and on BMI rate of
growth during follow-up (p = 0.0006). In gender-specific models, with the same adjustments, these
effects were also statistically significant in males (p = 0.042 and p = 0.018, respectively) and in females
(p = 0.027 and p = 0.009, respectively). Thus, after adjustments, every two standard deviation increase
of 5.04 in average block size was associated with decreases in predicted BMI attained by age 10 of 0.40
kg m
-2
(SE = 0.15 kg m
-2
), 0.35 kg m
-2
(SE = 0.20 kg m
-2
) and 0.45 kg m
-2
(SE = 0.20 kg m
-2
) for
combined, male and female populations, respectively. Furthermore, after adjustments, every two
standard deviation increase of 5.04 in average block size was associated with decreases in predicted
BMI growth during follow-up of 0.10 kg m
-2
(SE = 0.05 kg m
-2
), 0.05 kg m
-2
(SE = 0.05 kg m
-2
) and
0.10 kg m
-2
(SE = 0.05 kg m
-2
) for combined, male and female populations, respectively.
A model with a gender interaction on average block size, essentially a reparameterized version of
the same model we discussed above with gender-specific effects, was fitted in order to formally test for
interaction. Based on this model, there was no statistically significant interaction between average block
size and gender for attained BMI at age 10 (p = 0.787) and for BMI rate of growth during follow-up (p =
0.678).
In order to assess effect modification for the gamma index on BMI relationship, three potential
effect modifiers were examined. These covariates were dichotomized by their medians. These were
population density (within 500 m), total NOx emission at child’s residence and the population-level
violent crime rate per 100,000. Participant’s homes with a higher population density compared to those
with a low population density did not modify the effect of the gamma index and attained BMI at age 10
(p = 0.751), and also it did modify the gamma index effect of the growth in BMI over follow-up (p =
27
0.754). Participant’s homes with a high NOx emission compared to a low NOx emission did not modify
the effect of the gamma index and attained BMI at age 10 (p = 0.855), and it did not modify the gamma
index effect of the growth in BMI over follow-up (p = 0.472). Lastly, a high population-level
community violent crime rate compared to a low community violent crime rate did not modify the effect
of the gamma index and attained BMI at age 10 (p = 0.438), and it did not modify the gamma index
effect of the growth in BMI over follow-up (p = 0.927).
In order to assess effect modification for the average block size on BMI relationship, the same
three potential effect modifiers were examined. Participant’s homes with a higher population density
compared to those with a low population density marginally modified the effect of average block size
and attained BMI at age 10 (p = 0.050), and it did modify average block size’s effect of the growth in
BMI over follow-up (p = 0.038). Participant’s homes with a high NOx emission compared to a low NOx
emission did not modify the effect of average block size and attained BMI at age 10 (p = 0.500), and it
did not modify average block size’s effect of the growth in BMI over follow-up (p = 0.650). Further, a
high population-level community violent crime rate compared to a low community violent crime rate did
not modify the effect of average block size and attained BMI at age 10 (p = 0.115), and it did not modify
average block size’s effect of the growth in BMI over follow-up (p = 0.079).
Among males, the effects of average block size on attained BMI at age 10 and on BMI growth
during follow-up were statistically significantly modified by total population density (p = 0.024 and p =
0.007, respectively). No such effect modification was found among females. Furthermore, we note that
no effect modification was found for the gamma index on attained BMI at age 10 and on BMI growth
during follow-up for gender-specific models. A summary of effect modification by total population
density is described in Table 7.
28
An analysis stratified by gender and population density (dichotomized at the median) was also
implemented. Among males whose homes are in a low population density, a two standard deviation
increase of 0.12 in the gamma index is associated with a 0.14 kg m
-2
(SE = 0.05 kg m
-2
) increase in BMI
growth during follow-up (p = 0.009). Among females who homes are in a low population density, a two
standard deviation increase of 0.12 in the gamma index is associated with a 0.54 kg m
-2
(SE = 0.27 kg
m
-2
) increase in attained BMI at age 10 (p = 0.044). Among females who homes are in a high population
density, a two standard deviation increase of 0.12 in the gamma index is associated with a 0.83 kg m
-2
(SE = 0.35 kg m
-2
) increase in attained BMI at age 10 (p = 0.018). Among males whose homes are in a
low population density, a two standard deviation increase of 5.04 in average block size is associated
with a 0.40 kg m
-2
(SE= 0.20 kg m
-2
) decrease in attained BMI at age 10 (p = 0.034) and a 0.10 kg m
-2
(SE = 0.05 kg m
-2
) decrease in BMI growth during follow-up (p = 0.012). Furthermore, among males
whose homes are in a high population density, a two standard deviation increase of 5.04 in average
block size is associated with a 0.81 kg m
-2
(SE= 0.35 kg m
-2
) increase in BMI growth during follow-up
(p = 0.021). No effects for average block size among females were found. Table 8 summarizes effects
for the gamma index and average block size, stratified by gender and total population density.
Among gender-specific models, differences between males and females were also assessed. The
association of attained BMI at age 10 and gamma index is not statistically different between males and
females (p = 0.426). The association of BMI growth and gamma index is not statistically different
between males and females (p = 0.404). The association of attained BMI at age 10 and average block
size is not statistically different between males and females (p = 0.287). The association of BMI growth
and average block size is not statistically different between males and females (p = 0.090).
A sensitivity analysis was conducted by sub setting the data to children who did not move from
baseline through the four years of follow-up, and reanalyzing the same effects. This reduced the sample
29
size from N=4117 to N=2838. Table 9 shows summaries of the main effects of connectivity on BMI for
non-movers only. By comparing to Table 6, we see the only association that changed significance was
the gender and gamma interaction on BMI growth during follow-up (p = 0.0006 for non-movers
compared to p = 0.824 for total dataset). All other associations remained the same, with p-values only
varying slightly. Table 10 shows summaries estimates when assessing effect modification by population
density. By comparing to Table 7, we see the associations for the interaction terms did not change
drastically. Thus even though there were a large number of children who moved throughout this study, it
did not affect the results of this study and therefore differential misclassification of the exposure
variables are not of concern here.
DISCUSSION
This analysis found statistically significant associations for both measures of connectivity on
BMI growth during follow-up. The gamma index was positively associated with attained BMI at age 10
as well as with BMI growth, meaning that the higher the street connectivity in the community, the more
likely that a child living in that area will have higher BMI while growing up. Average block size was
negatively associated with attained BMI at age 10 as well as with BMI growth, meaning that the larger
the average block size of a child’s community, the more likely that the child will have lower BMI while
growing up. These associations indicate that street connectivity might lead to higher BMI and hence
potential overweight status or obesity in children. We note that this finding counters the conceptual
hypothesis that more street connectivity is supposed to lead to more physical activity and hence
lower BMI. One possible explanation for this might be social barriers that might prevent higher
levels of physical activity in neighborhoods with high street connectivity. Our assessment of
potential effect modification by such factors such as population density revealed that the average
30
block size and BMI during follow-up association was modified by population density. Specifically, a
stratified analysis shows that for both males and females living in low population density communities,
average block size has negative associations with attained BMI and with BMI rate of growth, although
the associations are only statistically significant in males Also, for both males and females living in high
population density communities, average block size has positive associations with attained BMI and
with BMI growth, although the only statistically significant association is the rate of growth among
males. Furthermore, male participants whose homes are in low population density areas have a
statistically significant association between the gamma index and BMI growth. No such association
existed for males whose home is in a high population density, and also no such associations existed for
females. However, among females for both a low and high population densities, the gamma index was
associated with attained BMI at age 10. This indicates that street connectivity not only effect obesity in
males and females differently, but also is different among population densities.
There are a few notably exceptional features of this study. First, this study uses both individual-
level and population-level characteristics to control for possible confounding and effect modification.
Also, this study has information on not only cross-sectional data, but also longitudinal data, which
allows for a more thorough investigation. Furthermore, this analysis adds knowledge on the limited
number of studies previously conducted regarding the association between street connectivity and BMI
in children.
This study also has some limitations. First, the outcome in all models is BMI, which is not a
direct measure of fat mass [63]. For instance, athletes are likely to have high BMI but this does not mean
that they are obese or overweight as their body composition might be made of mainly lean muscle.
Recent studies on children and adolescents have shown a high correlation between BMI and body
fatness [63]. Hence, there is evidence that BMI could serve as a reasonable measure of body fatness [67].
31
Another limitation is the fact that children are already naturally growing, thus it is difficult to determine
if their BMI is increasing due to this fact, or if BMI growth is due to our main effect measures.
CONCLUSION
Children living in households with higher gamma indices tend to experience increases in BMI
trajectories and those with higher average block size tend to experience decreases in BMI trajectories.
This counters the hypothesis that street connectivity is negatively associated with obesity in children.
Population density modified the average block size and BMI relationship. A stratified analysis shows
statistically significant associations for average block size and the rate of BMI growth in both high
population density and low population density settings around male participant’s homes. No such
associations were observed for females. Additional research using more complete longitudinal data
covering the entire childhood period is needed in order to fully assess differences in the effects of street
connectivity measures on childhood obesity development in males and females across different
population densities.
32
SUPPLEMENTARY TABLES AND FIGURES
Figure 1: Conceptual path model showing possible influences of the built environment in the context of
other variables that may affect the attained BMI in children [63].
33
Table 1: Summary table of children’s studies on BMI identified from Pubmed
Authors/Title/Journal/Year Study Type Age Range (n) Outcome Main risk factor(s)
Shankardass, K., et al. "Parental stress increases
body mass index trajectory in pre ‐ adolescents."
Pediatric obesity 9.6 (2014): 435-442.
Dunton, Genevieve, et al. "Organized physical
activity in young school children and subsequent
4-year change in body mass index." Archives of
pediatrics & adolescent medicine 166.8 (2012):
713-718.
Jerrett, Michael, et al. "Traffic-related air pollution
and obesity formation in children: a longitudinal,
multilevel analysis." Environ Health 13.1 (2014):
49.
Crawford, David, et al. "The longitudinal influence
of home and neighbourhood environments on
children's body mass index and physical activity
over 5 years: the CLAN study." International
journal of obesity 34.7 (2010): 1177-1187.
Hedley, Allison A., et al. "Prevalence of
overweight and obesity among US children,
adolescents, and adults, 1999-2002." Jama 291.23
(2004): 2847-2850.
Ogden, Cynthia L., Margaret D. Carroll, and
Longitudinal
Longitudinal
Longitudinal
Longitudinal and Cross-
sectional
Descriptive
Descriptive
Ages 5-10 at baseline
(4078)
Ages 5-10 at baseline
(4550)
Ages 5-10 at baseline
(4550)
Ages 10-12 at baseline
(301)
Ages 2-18 (N=4018 in
1999-2000 and N=4258 in
2001-2002)
Ages 2-19 (8165)
BMI
BMI
BMI
BMI z-score
BMI
BMI
Parental Stress
Organized physical activity
Traffic-related air pollution
Marital status of parents, moderate-to-
vigorous physical activity of parents and
number of home sedentary items
Sex, age and race/ethnicity
Sex, age and race/ethnicity
34
Katherine M. Flegal. "High body mass index for
age among US children and adolescents, 2003-
2006." Jama 299.20 (2008): 2401-2405.
Ogden, Cynthia L., et al. "Prevalence of high body
mass index in US children and adolescents, 2007-
2008." Jama 303.3 (2010): 242-249.
Popkin, Barry M. "Recent dynamics suggest
selected countries catching up to US obesity." The
American journal of clinical nutrition 91.1 (2010):
284S-288S.
Bibbins-Domingo, Kirsten, et al. "Adolescent
overweight and future adult coronary heart
disease." New England Journal of Medicine 357.23
(2007): 2371-2379.
Daniels, S. R. "Complications of obesity in
children and adolescents." International Journal of
Obesity 33 (2009): S60-S65.
Wolch, Jennifer, et al. "Childhood obesity and
proximity to urban parks and recreational
resources: a longitudinal cohort study." Health &
place 17.1 (2011): 207-214.
Jerrett, Michael, et al. "Automobile traffic around
the home and attained body mass index: a
longitudinal cohort study of children aged 10–18
years." Preventive medicine 50 (2010): S50-S58.
Descriptive
Descriptive
Descriptive
Systematic Review
Longitudinal
Longitudinal
Ages 2-19 (3281) and
ages 0-2 (719)
Ages 6–18
Ages of 12-19
Ages 2-19
Ages 9-10 (3173)
Ages 9-10 (n=3318)
BMI
BMI
BMI
Obesity
BMI
BMI and BMI
percentile
Sex, age and race/ethnicity
Age, age squared, and age cubed
Diastolic blood pressure, LDL and
HDL cholesterol, and diabetes
Cardiovascular, metabolic, pulmonary,
gastrointestinal, skeletal and other disorders
Parks and recreational programs
Traffic density
35
Timperio, Anna, et al. "Perceptions of local
neighbourhood environments and their relationship
to childhood overweight and obesity."
International journal of obesity 29.2 (2005): 170-
175.
Rundle, Andrew, et al. "Association of childhood
obesity with maternal exposure to ambient air
polycyclic aromatic hydrocarbons during
pregnancy." American journal of epidemiology
175.11 (2012): 1163-1172.
Su, Jason G., et al. "Factors influencing whether
children walk to school." Health & place 22
(2013): 153-161.
Nelson, Melissa C., et al. "Built and social
environments: associations with adolescent
overweight and activity." American journal of
preventive medicine 31.2 (2006): 109-117.
Hsieh, Stephanie, et al. "Built environment
associations with adiposity parameters among
overweight and obese Hispanic youth." Preventive
Medicine Reports 2 (2015): 406-412.
Duncan, Dustin T., et al. "Characteristics of
walkable built environments and BMI z-scores in
children: evidence from a large electronic health
record database." Environmental health
Cross-sectional
Longitudinal
Cross-section
Cross-sectional
Cross-sectional
Cross-sectional
Ages 5-6 (291) and ages
10-12 (919)
Ages 5 (422) and 7 (341)
Ages 5-7 at baseline
(4338)
Ages 13-18 (20,745)
Ages 8-18 (576)
Ages 4-19 (49,770)
BMI
BMI z-score
Walking rates
BMI, physical activity
BMI, body fat
percentage, waist
circumference
BMI z-score
Perceptions of heavy traffic and concern
about road safetly
Prenatal polycyclic aromatic hydrocarbons
exposure
Traffic density
Neighborhood categories: rural working
class, exurban, newer suburban, upper-
middle class, older suburban, mixed-race
urban and low-SES inner-city areas
Supermarket access, park access and street
connectivity
Recreational open spaces, residential density,
traffic density, sidewalk completeness, and
intersection density
36
perspectives 122.12 (2014): 1359.
Harrison, Flo, et al. "Environmental correlates of
adiposity in 9–10 year old children: considering
home and school neighbourhoods and routes to
school." Social science & medicine 72.9 (2011):
1411-1419.
Kligerman, Morton, et al. "Association of
neighborhood design and recreation environment
variables with physical activity and body mass
index in adolescents." American Journal of Health
Promotion 21.4 (2007): 274-277.
Hoyt, Lindsay T., et al. "Neighborhood Influences
on Girls’ Obesity Risk Across the Transition to
Adolescence." Pediatrics 134.5 (2014): 942-949
Kolodziejczyk, Julia K., et al. "Influence of
specific individual and environmental variables on
the relationship between body mass index and
health-related quality of life in overweight and
obese adolescents." Quality of Life Research 24.1
(2015): 251-261
Hsieh, Stephanie, et al. "Fast-food restaurants, park
access, and insulin resistance among Hispanic
youth." American journal of preventive medicine
46.4 (2014): 378-387
Cross-sectional
Cross-sectional
Longitudinal
Cross-sectional
Longitudinal
Ages 9-10
(1,995)
Mean age=16 (98)
Ages 8-10 at baseline
(174)
Mean age=12.9 (205)
Ages 8-18 (453)
FMI, calculated as fat
mass (kg)/height (m)
2
BMI and physical
activity
BMI
BMI
Body composition and
fat distribution were
assessed using dual x-
ray absorptiometry and
waist circumference
Presence of major roads
Walkability index and recreational programs
Food and service retail and physical
disorders
Body image, self-esteem, neighborhood
environment and acculturation
Fast food density, park space, insulin
resistance and neighborhood linguistic
isolation
37
Gose, Maria, et al. "Longitudinal influences of
neighbourhood built and social environment on
children’s weight status." International journal of
environmental research and public health 10.10
(2013): 5083-5096
Armstrong, Bridget, Crystal S. Lim, and David M.
Janicke. "Park density impacts weight change in a
behavioral intervention for overweight rural
youth." Behavioral Medicine 41.3 (2015): 123-130
Taylor, Wendell C., et al. "Features of the Built
Environment Related to Physical Activity
Friendliness and Children's Obesity and Other Risk
Factors." Public Health Nursing 31.6 (2014): 545-
555
Wasserman, J. A., et al. "A multi-level analysis
showing associations between school
neighborhood and child body mass index."
International Journal of Obesity 38.7 (2014): 912-
918
Miller, Laura J., et al. "Associations between
childhood obesity and the availability of food
outlets in the local environment: A retrospective
cross-sectional study." American Journal of Health
Promotion 28.6 (2014): e137-e145
Carroll-Scott, Amy, et al. "Disentangling
neighborhood contextual associations with child
Longitudinal
Randomized Control
Trial
Cross-sectional
Cross-sectional
Retrospective cross-
sectional
Cross-sectional
Ages 5-6 (485)
Ages 8-14 (93)
Mean age=8 (911)
Ages 4-12 (12,118)
Ages 5-15 (1850)
Ages 10-12 (1,048)
Children’s BMI
standard deviation score
BMI z-score
BMI
BMI percentile
BMI
BMI
Walkability, street type, socioeconomic
status of the district and perceived frequency
of passing trucks/busses
Park density
Accessibility and comfort features of a
walking environment
Population size, fast food outlets, grocery
stores, parks and fitness centers
Healthy food outlets, physical activity, time
spent sedentary, area disadvantage and fast
food outlets.
Property crimes, distance from grocery stores
and fast food outlets
38
body mass index, diet, and physical activity: the
role of built, socioeconomic, and social
environments." Social Science & Medicine 95
(2013): 106-114
Salois, Matthew J. "The built environment and
obesity among low-income preschool children."
Health & place 18.3 (2012): 520-527
Singh, Gopal K., Mohammad Siahpush, and
Michael D. Kogan. "Neighborhood socioeconomic
conditions, built environments, and childhood
obesity." Health affairs 29.3 (2010): 503-512
Sallis, James F., and Karen Glanz. "The role of
built environments in physical activity, eating, and
obesity in childhood." The future of children 16.1
(2006): 89-108
Dunton, Genevieve Fridlund, et al. "Physical
environmental correlates of childhood obesity: a
systematic review." Obesity reviews 10.4 (2009):
393-402
Oreskovic, Nicolas M., et al. "Obesity and the built
environment among Massachusetts children."
Clinical pediatrics (2009)
Lovasi, Gina S., et al. "Neighborhood safety and
green space as predictors of obesity among
preschool children from low-income families in
Ecologic
Cross-sectional
Systematic Review
Systematic Review
Cross-sectional
Cross-sectional
Ages 2-4 (2,192)
Ages 2-19
Ages 2-19
Ages 2-19
Ages 2-18 (21 008)
Ages 3-5 (11,562)
BMI
BMI
Obesity
Obesity
BMI
BMI
Food environment and physical activity
environment
Poor housing, access to sidewalks, parks and
recreation centers
Walkability, presence of healthy local
markets
Access to equipment and facilities,
neighborhood pattern (e.g. rural, exurban,
suburban) and urban sprawl
Fast food outlets, distance to school, distance
to subway station, amount of open space
Homicide rates, density of trees and other
neighborhood characteristics
39
New York City." Preventive medicine 57.3 (2013):
189-193
Slater, Sandy J., et al. "Walkable communities and
adolescent weight." American journal of
preventive medicine 44.2 (2013): 164-168.
Chiang, Po-Huang, et al. "Fast-food outlets and
walkability in school neighbourhoods predict
fatness in boys and height in girls: a Taiwanese
population study." Public health nutrition 14.09
(2011): 1601-1609
Spence, John C., et al. "Influence of
neighbourhood design and access to facilities on
overweight among preschool children."
International Journal of Pediatric Obesity 3.2
(2008): 109-116
McConnell, R., et al., A longitudinal cohort study
of body mass index and childhood exposure to
secondhand tobacco smoke and air pollution: the
Southern California Children's Health Study.
Environ Health Perspect, 2015. 123(4): p. 360-6
Cross-sectional
Cross-sectional
Cross-sectional
Longitudinal
Ages 13-18 (11,041)
Ages 6-13 (2283)
Ages 4-6 (501)
Ages 8-18 (3,318)
BMI
BMI
BMI
BMI
Community walkability index
Fast food and convenient store densities
Walkability of neighborhood (dwelling
density, land use mix, intersection density,
availability of facilities)
Maternal smoking during pregnancy, second-
hand smoke and near-roadway pollution
exposures
40
Figure 2: CHS participating communities by cohort inclusion (from CHS website:
https://healthstudy.usc.edu/communities.php).
41
Figure 3: Summary table of features used to describe the built environment [63].
Table 2: Participant baseline
a
characteristics and potentially confounding variables
Variable No. % Mean SD
Participant Characteristics
Age 4,547
6.67 0.71
Missing Observations 3
Height 4,550
120.15 6.55
Missing Observations -
Weight 4,550
53.84 13.03
Missing Observations -
BMI at baseline 4,550
16.78 2.81
Males 2,251
16.86 2.81
Females 2,297
16.70 2.80
Missing Observations 2
BMI at end of follow-up 4,550
19.35 4.21
Males 2,251
19.50 4.26
Females 2,297
19.19 4.15
42
Variable No. % Mean SD
Missing Observations 2
BMI CDC percentile at baseline
85 > BMIp 3,196 70.24
85 ≤ BMIp < 95 666 14.64
95 ≤ BMIp 688 15.12
Missing Observations -
Race/Ethnicity
African American 122 2.68
Asian 145 3.19
Hispanic 2,462 54.11
Non-White Hispanic 1,464 32.18
Other 357 7.85
Missing Observations -
Gender
Male 2,297 50.48
Female 2,251 49.47
Missing Observations 2 0.04
Ever Asthma
Yes 564 12.40
No 3,501 76.95
Missing Observations 485 10.66
Spanish Speaker
Yes 1,133 75.10
No 3,417 24.90
Missing Observations -
Parental Education
Above High School Education 2,475 54.40
High School Education or less 1,686 37.05
Missing Observations 389 8.55
Second-hand Smoke
Yes 309 6.79
No 3,962 87.08
Missing Observations 279 6.13
Dog Ownership
Yes 1,243 27.32
No 2,981 65.52
Missing Observations 326 7.16
Any Pet Ownership
Yes 3,232 71.03
No 1,012 22.24
43
Variable No. % Mean SD
Missing Observations 306 6.73
Insurance
Yes 3,689 81.08
No 487 10.70
Missing Observations 374 8.22
Total Population Density
b
3,968
1,703.90 1,426.65
Missing Observations 582
Community Characteristics
Gamma Index
b
4,117
0.40 0.06
Missing Observations 433
Average Block Size
b
4,117
0.79 2.52
Missing Observations 433
Nodes
b
4,117
48.39 20.47
Missing Observations 433
Parks and Recreation
b
(acres) 3,968
4.95 10.60
Missing Observations 582
Recreational Programs (within 5 km) 4,117
29.74 34.20
Missing Observations 433
Recreational Programs (within 10 km) 4,117
48.03 44.34
Missing Observations 433
NDVI
b
Green Cover 4,117
0.09 0.10
Missing Observations 433
Traffic Density
c
4,464
49.24 104.93
Missing Observations 86
Total Nox
g
(parts per billion) 4,464
19.49 18.82
Missing Observations 86
No Food
d
Yes 2,137 46.97
No 1,980 43.52
Missing Observations 433 9.52
Total Food
d
4,117
2.50 4.57
Missing Observations 433
Population Characteristics
Violent Crime Rate
e
4,550
511.73 268.04
Forcible Rape Crime Rate
e
4,550
26.73 9.06
Murder Crime Rate
e
4,550
6.14 5.55
Robbery Crime Rate
e
4,550
139.15 107.89
Aggravated Assault Crime Rate
e
4,550
339.71 166.66
Property Crime Rate
e
4,550
3,125.15 1,025.93
Motor-vehicle Crime Rate
e
4,550
574.62 318.10
44
Variable No. % Mean SD
Percentage in Poverty
f
4,550
0.15 0.06
Percentage of Unemployed
f
4,550
0.08 0.02
Population Density
f
(per sq. mile) 4,550
0.002 0.001
a
N=4550.
b
In 500 m Euclidean buffer of homes
of participants.
c
In 150 m of home.
d
In 500 m road network buffer.
e
Crime per 100,000 at community scale from the California Department of Justice.
f
Information obtained for U.S. 2000 Census.
g
At participants baseline residence.
45
Table 3: Individual Associations of BMI and various covariates
Intercept Slope
Covariate Estimate SE P Estimate SE P
Individual-level Characteristics
Above High School Education
-0.46
0.14
0.001
-0.21
0.02
<.0001
Spanish Questionnaire 0.54 0.17 0.001 0.23 0.02 <.0001
Asthma 0.66 0.18 0.001 0.02 0.03 0.646
Second-hand Smoke
Dog Ownership
0.17
0.04
0.06
0.14
0.003
0.772
0.09
-0.06
0.02
0.02
0.001
0.018
Any Pet Ownership -0.19 0.15 0.221 -0.09 0.03 0.001
Insurance -0.16 0.20 0.432 -0.17 0.04 <.0001
Total Population Density
b
0.01 0.01 0.456 0.01 0.01 <.0001
Gamma Index
b
1.39 1.22 0.257 1.09 0.19 <.0001
Average Block Size
b
-0.02 0.03 0.425 -0.02 0.01 0.001
Nodes
b
-0.01 0.01 0.238 0.01 0.01 0.187
Parks and Recreation
b
0.01 0.006 0.701 0.01 0.01 0.383
Recreation programs (within 5 k) -0.01 0.01 0.524 -0.01 0.01 <.0001
Recreation programs (within 10 k) -0.01 0.01 0.043 -0.01 0.01 <.0001
NDVI Green Coverb -2.96 1.74 0.090 -0.33 0.11 0.003
Traffic Density
c
0.01 0.01 0.071 0.01 0.01 0.008
Total NOx
d
(parts per billion) 0.03 0.01 0.023 0.01 0.01 <.0001
Having no food stores
e
-0.41 0.14 0.003 -0.14 0.02 <.0001
Total food stores
e
0.02 0.01 0.199 0.01 0.01 <.0001
Population-level Characteristics
f
Population Violent Crime Rate 0.01 0.01 0.069 0.01 0.01 0.011
Forcible Rape Crime Rate 0.03 0.02 0.112 0.01 0.01 0.035
Population Murder Crime Rate 0.04 0.03 0.298 0.01 0.01 0.034
Population Robbery Crime Rate 0.01 0.01 0.128 0.01 0.01 0.028
Population Property Crime Rate 0.01 0.01 0.579 0.01 0.01 0.320
Population Aggravated Assault Crime
Rate
0.01 0.01 0.072 0.01 0.01 0.017
Motor-Vehicle Crime Rate 0.01 0.01 0.331 0.01 0.01 0.085
Percentage in Poverty
g
8.20 2.26 0.001 2.00 0.44 <.0001
Percentage of Unemployed
g
Total Population Density
g
21.80
356.31
7.28
135.55
0.003
0.009
5.83
68.15
1.30
32.31
<.0001
0.035
46
a
Unadjusted models include gender, race/ethnicity, community and average block size and uses
N=4,117, with complete observations for average block size.
b
In 500 m Euclidean buffer of homes of participants.
c
In 150 m of home.
d
At participants baseline residence.
e
In 500 m road network buffer.
f
Crime per 100,000 at community scale from the California Department of Justice.
g
Information obtained from U.S. 2000 Census.
Table 4: Assessing confounding for the gamma index
Beta for Gamma Effect Beta for Gamma*Age Effect
Covariate Unadjusted
a
Adjusted
Percent
Difference Unadjusted
a
Adjusted
Percent
Difference
Participant Characteristics
Above High School Education
1.3853
1.1386
0.18
1.0871
1.0885
0.12
Spanish Questionnaire 1.3853 1.1441 0.17 1.0871 1.0875 0.04
Asthma 1.3853 1.3613 0.02 1.0871 1.0888 0.15
Second-hand Smoke 1.3853 1.3810 <0.01 1.0871 1.0870 <0.01
Dog Ownership 1.3853 1.4244 2.82 1.0871 1.0875 0.03
Any Pet Ownership 1.3853 1.4752 6.49 1.0871 1.0879 0.07
Insurance 1.3853 1.3703 0.01 1.0871 1.0870 <0.01
Total Population Density
b
1.3853 1.5047 8.62 1.0871 1.0968 0.89
Nodes
b
1.3853 2.0634 48.95 1.0871 1.0870 <0.01
Recreation Programs (within 5 k) 1.3853 1.4400 3.95 1.0871 1.0871 <0.01
Recreation Programs (within 10 k) 1.3853 1.4819 6.97 1.0871 1.0879 0.07
NDVI Green Cover
b
1.3853 1.0906 0.21 1.0871 1.0882 0.10
Traffic Density
c
1.3853 1.4763 6.56 1.0871 1.0867 <0.01
Total NOx
d
(parts per billion) 1.3853 0.9383 0.32 1.0871 1.0873 0.02
Having No Food Stores
e
1.3853 0.7246 0.48 1.0871 1.0876 0.05
Total Food Stores
e
1.3853 1.1375 0.18 1.0871 1.0874 0.03
47
Beta for Gamma Effect Beta for Gamma*Age Effect
Covariate Unadjusted
a
Adjusted
Percent
Difference Unadjusted
a
Adjusted
Percent
Difference
Population Characteristics
f
Population Violent Crime Rate 1.5519 1.3932 0.10 0.3327 0.3309 0.01
Forcible Rape Crime Rate 1.5519 1.3978 0.10 0.3327 0.3310 <0.01
Population Murder Crime Rate 1.5519 1.4738 0.05 0.3327 0.3316 <0.01
Population Robbery Crime Rate 1.5519 1.3845 0.12 0.3327 0.3308 0.01
Population Aggravated Assault
Crime Rate
1.5519 1.4361 0.07 0.3327 0.3314 <0.01
Motor-vehicle Crime Rate 1.5519 1.4840 0.04 0.3327 0.3318 <0.01
Population Poverty Rate
g
1.5519 1.3324 0.14 0.3327 0.3303 0.01
Total Population Unemployment
Rate
g
Total Population Density
g
1.5519
1.5519
1.3527
1.3832
0.13
0.11
0.3327
0.3327
0.3304
0.3307
0.01
0.01
a
Unadjusted models include gender, race/ethnicity, community and the gamma index and uses
N=4,117, with complete observations for the gamma index.
b
In 500 m Euclidean buffer of homes of participants.
c
In 150 m of home.
d
At participants baseline residence.
e
In 500 m road network buffer.
f
Crime per 100,000 at community scale from the California Department of Justice.
g
Information obtained from U.S. 2000 Census.
Table 5: Assessing confounding for average block size
Beta for AverageBlockSize Effect Beta for AverageBlockSize*Age Effect
Covariate Unadjusted
a
Adjusted
Percent
Difference Unadjusted
a
Adjusted
Percent
Difference
Individual Characteristics
Above High School Education
-0.02137
-0.02018
0.06
-0.01516
-0.01519
0.22
Spanish Questionnaire -0.02137 -0.02042 0.04 -0.01516 -0.01516 0.01
48
Beta for AverageBlockSize Effect Beta for AverageBlockSize*Age Effect
Covariate Unadjusted
a
Adjusted
Percent
Difference Unadjusted
a
Adjusted
Percent
Difference
Asthma -0.02137 -0.02127 <0.01 -0.01516 -0.01518 0.13
Second-hand Smoke -0.02137 -0.02135 <0.01 -0.01516 -0.01516 <0.01
Dog Ownership -0.02137 -0.02169 1.51 -0.01516 -0.01516 0.02
Any Pet Ownership -0.02137 -0.02185 2.22 -0.01516 -0.01515 <0.01
Insurance -0.02137 -0.02106 0.01 -0.01516 -0.01516 <0.01
Total Population Density
b
-0.02137 -0.02274 6.43 -0.01516 -0.01508 0.01
Nodes
b
-0.02137 -0.03332 55.92 -0.01516 -0.01518 0.17
Recreation Programs (within 5 k) -0.02137 -0.02281 6.75 -0.01516 -0.01517 0.05
Recreation Programs (within 10 k) -0.02137 -0.01907 0.11 -0.01516 -0.01518 0.15
NDVI Green Cover
b
-0.02137 -0.01399 0.35 -0.01516 -0.01515 <0.01
Traffic Density
c
-0.02137 -0.01827 0.15 -0.01516 -0.01512 <0.01
Total NOx
d
(parts per billion) -0.02137 -0.01696 0.21 -0.01516 -0.01513 <0.01
Having no food stores
e
-0.02137 -0.01205 0.44 -0.01516 -0.01515 <0.01
Total food stores
e
-0.02137 -0.01930 0.10 -0.01516 -0.01515 <0.01
Population Characteristics
f
Population Violent Crime Rate
-0.02113
-0.01979
0.06
-0.00627
-0.00626
<0.01
Forcible Rape Crime Rate -0.02113 -0.01974 0.07 -0.00627 -0.00626 <0.01
Population Murder Crime Rate -0.02113 -0.02034 0.04 -0.00627 -0.00626 <0.01
Population Robbery Crime Rate -0.02113 -0.01934 0.08 -0.00627 -0.00625 <0.01
Population Aggravated Assault
Crime Rate
-0.02113 -0.02052 0.03 -0.00627 -0.00626 <0.01
Motor-vehicle Crime Rate -0.02113 -0.02023 0.04 -0.00627 -0.00626 <0.01
Population Poverty Rate
g
-0.02113 -0.02084 0.01 -0.00627 -0.00626 <0.01
Total Population Unemployment
Rate
g
Total Population Density
g
-0.02113
-0.02113
-0.01998
-0.01852
0.05
0.12
-0.00627
-0.00627
-0.00625
-0.00624
<0.01
0.01
a
Unadjusted models include gender, race/ethnicity, community and average block size and uses
N=4,117, with complete observations for average block size.
b
In 500 m Euclidean buffer of homes of participants.
49
c
In 150 m of home.
d
At participants baseline residence.
e
In 500 m road network buffer.
f
Crime per 100,000 at community scale from the California Department of Justice.
g
Information obtained from U.S. 2000 Census.
50
Table 6: Main effects of connectivity on BMI
Effect Males Females Combined Effects Gender interaction p-value
Gamma
Attained BMI Rate of
at age 10 BMI growth
0.62 (0.17)
a
0.14 (0.02)
a
Attained BMI Rate of
at age 10 BMI growth
0.74 (0.17)
a
0.12 (0.02)
a
Attained BMI Rate of
at age 10 BMI growth
0.67 (0.15)
a
0.13 (0.02)
a
Attained BMI Rate of
at age 10 BMI growth
0.356 0.824
Average Block
Size
-0.35 (0.20)
c
-0.05 (0.05)
c
-0.45 (0.20)
c
-0.10 (0.05)
b
-0.40 (0.15)
b
-0.10 (0.05)
a
0.787 0.678
*
All effects listed as Estimate (Standard Error) and scaled to a two standard deviation increase of 0.12 and 5.04 for gamma index and
average block size, respectively.
*
All models adjusted for gender, race/ethnicity, community and nodes.
a
p < 0.001.
b
p < 0.01.
c
p < 0.05.
51
Table 7: Effect modification by population density
Effect Males Females Combined
Gamma
Population Density
Gamma ×
Population Density
Attained BMI Rate of
at age 10 BMI growth
0.53 (0.24)
c
0.12 (0.03)
a
0.93 (1.33) 0.20 (0.22)
-2.08 (3.29)
d
-0.37 (0.54)
e
Attained BMI Rate of
at age 10 BMI growth
0.63 (0.23)
b
0.09 (0.03)
c
-0.08 (1.35) 0.07 (0..22)
0.79 (3.29)
d
0.12 (0.54)
e
Attained BMI Rate of
at age 10 BMI growth
0.58 (0.20)
b
0.10 (0.03)
b
0.50 (1.02) 0.14 (0.17)
-0.80 (2.53) -0.13 (0.75)
Average Block Size
Population Density
Average Block Size
× Population
Density
-0.30 (0.20) -0.05 (0.05)
0.23 (0.21)
0.09 (0.03)
b
0.81 (0.38)
c,f
0.17 (0.07)
c, g
-0.35 (0.20)
-0.05 (0.05)
0.15 (0.21)
0.10 (0.03)
b
0.28 (0.42)
f
0.02 (0.07)
g
-0.30 (0.15)
c
-0.05 (0.05)
c
0.19 (0.17) 0.10 (0.03)
a
0.57 (0.29)
c
0.10 (0.05)
c
*
All effects listed as Estimate (Standard Error) and scaled to a two standard deviation increase of 0.12 and 5.04 for gamma index and
average block size, respectively.
*
All models adjusted for gender, race/ethnicity, community and nodes.
a
p < 0.001.
b
p < 0.01.
c
p ≤ 0.05.
d
Association of attained BMI at age 10 and gamma index is not statistically different between males and females (p = 0.426).
e
Association of BMI growth and gamma index is not statistically different between males and females (p = 0.404).
52
f
Association of attained BMI at age 10 and average block size is not statistically different between males and females (p = 0.287).
g
Association of BMI growth and average block size is not statistically different between males and females (p = 0.090).
Table 8: Connectivity Effects Stratified by Gender and Population Density
Males Females
Effect Low Population Density High Population Density Low Population Density High Population Density
Gamma Index
Attained BMI Rate of
at age 10 BMI growth
0.53 (0.31) 0.14 (0.05)
b
Attained BMI Rate of
at age 10 BMI growth
0.36 (0.35) 0.08 (0.06)
Attained BMI Rate of
at age 10 BMI growth
0.54 (0.27)
c
0.07 (0.04)
Attained BMI Rate of
at age 10 BMI growth
0.83 (0.35)
c
0.10 (0.06)
Average Block
Size
Attained BMI Rate of
at age 10 BMI growth
-0.40 (0.20)
c
-0.10 (0.05)
c
Attained BMI Rate of
at age 10 BMI growth
2.32 (2.17) 0.81 (0.35)
c
Attained BMI Rate of
at age 10 BMI growth
-0.15 (0.20) -0.05 (0.05)
Attained BMI Rate of
at age 10 BMI growth
2.42 (2.32) 0.01 (0.35)
*
All effects listed as Estimate (Standard Error) and scaled to a two standard deviation increase of 0.12 and 5.04 for gamma index and
average block size, respectively.
*
All models adjusted for gender, race/ethnicity, community and nodes.
a
p < 0.001.
b
p < 0.01.
c
p < 0.05.
53
Table 9: Main effects of connectivity on BMI among non-movers
Effect Males Females Combined Effects Gender interaction p-value
Gamma
Attained BMI Rate of
at age 10 BMI growth
0.42 (0.21)
c
0.13 (0.03)
a
Attained BMI Rate of
at age 10 BMI growth
0.75 (0.20)
a
0.10 (0.03)
a
Attained BMI Rate of
at age 10 BMI growth
0.59 (0.18)
a
0.11 (0.03)
a
Attained BMI Rate of
at age 10 BMI growth
0.104 0.0006
Average Block
Size
-0.40 (0.20)
c
-0.05 (0.05)
c
-0.45 (0.25)
c
-0.10 (0.05)
c
-0.40 (0.15)
b
-0.05 (0.05)
b
0.766 0.758
*
All effects listed as Estimate (Standard Error) and scaled to a two standard deviation increase of 0.12 and 5.04 for gamma index and
average block size, respectively.
*
All models adjusted for gender, race/ethnicity, community and nodes.
a
p < 0.001.
b
p < 0.01.
c
p < 0.05.
54
Table 10: Effect modification by population density among non-movers
Effect Males Females Combined
Gamma
Population Density
Gamma ×
Population Density
Attained BMI Rate of
at age 10 BMI growth
0.42 (0.29) 0.11 (0.04)
b
0.93 (1.59) 0.16 (0.26)
-1.94 (3.96) -0.21 (0.65)
Attained BMI Rate of
at age 10 BMI growth
0.52 (0.28) 0.08 (0.04)
-0.56 (1.61) 0.22 (0.27)
2.00 (3.97) -0.33 (0.66)
Attained BMI Rate of
at age 10 BMI growth
0.47 (0.24)
c
0.09 (0.04)
c
2.65 (1.22) 0.20 (0.21)
-0.13 (3.04) -0.29 (0.52)
Average Block Size
Population Density
Average Block Size
× Population Density
-0.35 (0.20) -0.05 (0.05)
0.11 (0.24) 0.10 (0.04)
b
1.29 (0.45)
b
0.21 (0.08)
b
-0.40 (0.01) -0.05 (0.05)
-0.03 (0.25) 0.04 (0.04)
0.94 (0.48) 0.10 (0.08)
-0.40 (0.15)
c
-0.05 (0.05)
c
0.04 (0.20)
0.07 (0.03)
c
1.12 (0.34)
a
0.15 (0.06)
b
*
All effects listed as Estimate (Standard Error) and scaled to a two standard deviation increase of 0.12 and 5.04 for gamma index and
average block size, respectively.
*
All models adjusted for gender, race/ethnicity, community and nodes.
a
p < 0.001.
b
p < 0.01.
c
p ≤ 0.05.
55
REFERENCES
1. Daniels, S., Complications of obesity in children and adolescents. International
Journal of Obesity, 2009. 33: p. S60-S65.
2. Dietz, W.H. and T.N. Robinson, Clinical practice. Overweight children and
adolescents. N Engl J Med, 2005. 352(20): p. 2100-9.
3. Ogden, C.L., et al., PRevalence of childhood and adult obesity in the united
states, 2011-2012. JAMA, 2014. 311(8): p. 806-814.
4. Bibbins-Domingo, K., et al., Adolescent Overweight and Future Adult Coronary
Heart Disease. New England Journal of Medicine, 2007. 357(23): p. 2371-
2379.
5. Hill, J.O., et al., Obesity and the environment: where do we go from here?
Science, 2003. 299(5608): p. 853-5.
6. Stunkard, A.J., Genetic contributions to human obesity. Res Publ Assoc Res
Nerv Ment Dis, 1991. 69: p. 205-18.
7. Hill, J.O. and J.C. Peters, Environmental Contributions to the Obesity Epidemic.
Science, 1998. 280(5368): p. 1371-1374.
8. Papas, M.A., et al., The built environment and obesity. Epidemiologic reviews,
2007. 29(1): p. 129-143.
9. Hsieh, S., et al., Built Environment Associations with Adiposity Parameters
among Overweight and Obese Hispanic Youth. Prev Med Rep, 2015. 2: p. 406-
412.
10. Ying, Z., L.D. Ning, and L. Xin, Relationship Between Built Environment,
Physical Activity, Adiposity, and Health in Adults Aged 46-80 in Shanghai, China.
Journal of physical activity & health, 2015. 12(4).
11. Roberts, J.D., et al., Built environment and active play among Washington DC
metropolitan children: A protocol for a cross-sectional study. Archives of
Public Health, 2015. 73(1): p. 22.
12. Pearce, J.R. and R. Maddison, Do enhancements to the urban built environment
improve physical activity levels among socially disadvantaged populations. Int J
Equity Health, 2011. 10(28): p. 1-8.
13. Li, F., et al., Built environment, adiposity, and physical activity in adults aged
50-75. Am J Prev Med, 2008. 35(1): p. 38-46.
14. Frank, L.D., et al., A hierarchy of sociodemographic and environmental
correlates of walking and obesity. Prev Med, 2008. 47(2): p. 172-8.
15. Oliver, M., et al., Neighbourhood built environment associations with body size
in adults: mediating effects of activity and sedentariness in a cross-sectional
study of New Zealand adults. BMC public health, 2015. 15(1): p. 1.
16. Pouliou, T., et al., Building obesity in Canada: understanding the individual-
and neighbourhood-level determinants using a multi-level approach. Geospat
Health, 2014. 9(1): p. 45-55.
17. Booth, K.M., M.M. Pinkston, and W.S. Poston, Obesity and the built
environment. J Am Diet Assoc, 2005. 105(5 Suppl 1): p. S110-7.
18. Michael, Y.L., et al., Built environment and change in body mass index in older
women. Health & place, 2013. 22: p. 7-10.
56
19. Wen, M. and L. Kowaleski-Jones, The built environment and risk of obesity in
the United States: Racial –ethnic disparities. Health & place, 2012. 18(6): p.
1314-1322.
20. Horn, E.E., et al., Behavioral and Environmental Modification of the Genetic
Influence on Body Mass Index: A Twin Study. Behav Genet, 2015. 45(4): p. 409-
26.
21. Dill, J. Measuring network connectivity for bicycling and walking.
22. Xu, Y., M. Wen, and F. Wang, Multilevel built environment features and
individual odds of overweight and obesity in Utah. Appl Geogr, 2015. 60: p.
197-203.
23. Morland, K.B. and K.R. Evenson, Obesity prevalence and the local food
environment. Health Place, 2009. 15(2): p. 491-5.
24. Stark, J.H., et al., The impact of neighborhood park access and quality on body
mass index among adults in New York City. Prev Med, 2014. 64: p. 63-8.
25. Frank, L.D., M.A. Andresen, and T.L. Schmid, Obesity relationships with
community design, physical activity, and time spent in cars. Am J Prev Med,
2004. 27(2): p. 87-96.
26. Xu, Y. and F. Wang, Built environment and obesity by urbanicity in the U.S.
Health Place, 2015. 34: p. 19-29.
27. Chiu, M., et al., Walk Score(R) and the prevalence of utilitarian walking and
obesity among Ontario adults: A cross-sectional study. Health Rep, 2015.
26(7): p. 3-10.
28. Frank, L., et al., Healthy aging and where you live: community design
relationships with physical activity and body weight in older Americans. J Phys
Act Health, 2010. 7 Suppl 1: p. S82-90.
29. Grafova, I.B., et al., Neighborhoods and Obesity in Later Life. American Journal
of Public Health, 2008. 98(11): p. 2065-2071.
30. Lim, S. and T.G. Harris, Neighborhood Contributions to Racial and Ethnic
Disparities in Obesity Among New York City Adults. Am J Public Health, 2015.
105(1): p. 159-165.
31. Janssen, I., et al., Comparison of overweight and obesity prevalence in school-
aged youth from 34 countries and their relationships with physical activity and
dietary patterns. Obesity Reviews, 2005. 6(2): p. 123-132.
32. Hirsch, J.A., et al., Built environment change and change in BMI and waist
circumference: Multi-ethnic Study of Atherosclerosis. Obesity (Silver Spring),
2014. 22(11): p. 2450-7.
33. Michael, Y.L., et al., Does change in the neighborhood environment prevent
obesity in older women? Soc Sci Med, 2014. 102: p. 129-37.
34. Duncan, G.E., et al., Quasi-causal associations of physical activity and
neighborhood walkability with body mass index: a twin study. Prev Med, 2015.
70: p. 90-5.
35. Wang, F., M. Wen, and Y. Xu, Population-adjusted street connectivity,
urbanicity and risk of obesity in the U.S. Applied Geography, 2013. 41: p. 1-14.
36. Hsieh, S., et al., Fast-food restaurants, park access, and insulin resistance
among Hispanic youth. Am J Prev Med, 2014. 46(4): p. 378-87.
57
37. Armstrong, B., C.S. Lim, and D.M. Janicke, Park Density Impacts Weight Change
in a Behavioral Intervention for Overweight Rural Youth. Behav Med, 2015.
41(3): p. 123-30.
38. Wasserman, J.A., et al., A multi-level analysis showing associations between
school neighborhood and child body mass index. Int J Obes (Lond), 2014.
38(7): p. 912-8.
39. Singh, G.K., M. Siahpush, and M.D. Kogan, Neighborhood socioeconomic
conditions, built environments, and childhood obesity. Health Aff (Millwood),
2010. 29(3): p. 503-12.
40. Wolch, J., et al., Childhood obesity and proximity to urban parks and
recreational resources: a longitudinal cohort study. Health Place, 2011. 17(1):
p. 207-14.
41. Carroll-Scott, A., et al., Disentangling neighborhood contextual associations
with child body mass index, diet, and physical activity: the role of built,
socioeconomic, and social environments. Soc Sci Med, 2013. 95: p. 106-14.
42. Oreskovic, N.M., et al., Obesity and the built environment among Massachusetts
children. Clin Pediatr (Phila), 2009. 48(9): p. 904-12.
43. Hoyt, L.T., et al., Neighborhood influences on girls' obesity risk across the
transition to adolescence. Pediatrics, 2014. 134(5): p. 942-9.
44. Lovasi, G.S., et al., Neighborhood safety and green space as predictors of obesity
among preschool children from low-income families in New York City.
Preventive medicine, 2013. 57(3): p. 189-193.
45. Gilliland, F.D., et al., Obesity and the risk of newly diagnosed asthma in school-
age children. Am J Epidemiol, 2003. 158(5): p. 406-15.
46. Spence, J.C., et al., Influence of neighbourhood design and access to facilities on
overweight among preschool children. International Journal of Pediatric
Obesity, 2008. 3(2): p. 109-116.
47. Duncan, D.T., et al., Characteristics of walkable built environments and BMI z-
scores in children: evidence from a large electronic health record database.
Environ Health Perspect, 2014. 122(12): p. 1359-65.
48. Salois, M.J., The built environment and obesity among low-income preschool
children. Health Place, 2012. 18(3): p. 520-7.
49. Gose, M., et al., Longitudinal influences of neighbourhood built and social
environment on children's weight status. Int J Environ Res Public Health, 2013.
10(10): p. 5083-96.
50. Taylor, W.C., et al., Features of the built environment related to physical
activity friendliness and children's obesity and other risk factors. Public Health
Nurs, 2014. 31(6): p. 545-55.
51. Slater, S.J., et al., Walkable communities and adolescent weight. American
journal of preventive medicine, 2013. 44(2): p. 164-168.
52. Dunton, G., et al., Organized physical activity in young school children and
subsequent 4-year change in body mass index. Arch Pediatr Adolesc Med, 2012.
166(8): p. 713-8.
53. Jerrett, M., et al., Traffic-related air pollution and obesity formation in children:
a longitudinal, multilevel analysis. Environmental Health, 2014. 13: p. 49-49.
58
54. Jerrett, M., et al., Food access around the home and growth in the body mass
index of children aged 10 –18 years: a longitudinal cohort study. Epidemiology,
2009. 20(6): p. S223.
55. Shankardass, K., et al., Parental stress increases body mass index trajectory in
pre-adolescents. Pediatr Obes, 2014. 9(6): p. 435-42.
56. McConnell, R., et al., A longitudinal cohort study of body mass index and
childhood exposure to secondhand tobacco smoke and air pollution: the
Southern California Children's Health Study. Environ Health Perspect, 2015.
123(4): p. 360-6.
57. Dunton, G.F., et al., Physical environmental correlates of childhood obesity: a
systematic review. Obes Rev, 2009. 10(4): p. 393-402.
58. Navidi, W., et al., Design and analysis of multilevel analytic studies with
applications to a study of air pollution. Environ Health Perspect, 1994. 102
Suppl 8: p. 25-32.
59. Peters, J.M., Epidemiologic investigatin to identify chronic effects of ambient air
pollutants in Southern California. 2004. Contract No. 94-331.
60. McConnell, R., et al., Traffic, susceptibility, and childhood asthma. Environ
Health Perspect, 2006. 114(5): p. 766-72.
61. Milam, J., et al., Parental stress and childhood wheeze in a prospective cohort
study. J Asthma, 2008. 45(4): p. 319-23.
62. Shankardass, K., et al., Social environment and asthma: associations with crime
and No Child Left Behind programmes. Journal of Epidemiology and
Community Health, 2011. 65(10): p. 859-865.
63. Jerrett, M., et al., Automobile traffic around the home and attained body mass
index: a longitudinal cohort study of children aged 10-18 years. Prev Med,
2010. 50 Suppl 1: p. S50-8.
64. DeMers, M.N., Fundamentals of Geographic Information Systems. 3rd Edition
ed. 2005: John Wiley & Sons, Inc.
65. Berhane, K. and N.T. Molitor, A Bayesian approach to functional-based
multilevel modeling of longitudinal data: applications to environmental
epidemiology. Biostatistics, 2008. 9(4): p. 686-99.
66. Diggle, P., K.Y. Liang, and S.L. Zeger, Analysis of longitudinal data. 1994:
Clarendon Press.
67. Pietrobelli, A., et al., Body mass index as a measure of adiposity among children
and adolescents: A validation study. The Journal of Pediatrics. 132(2): p. 204-
210.
Abstract (if available)
Abstract
Background: Childhood obesity rates are still of major public concern. Previous studies have shown associations with various built environment characteristics. One of the characteristics that has not been thoroughly studied longitudinally is street connectivity, thus this study attempts to clarify this association. ❧ Methods: The study population included children ages 5-10 years old (N=4,550
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Air pollution and childhood obesity
PDF
A cohort study of air-pollution and childhood obesity incidence
PDF
Native American ancestry among Hispanic Whites is associated with higher risk of childhood obesity: a longitudinal analysis of Children’s Health Study data
PDF
Association of maternal and environmental factors with infant feeding behaviors in a birth cohort study
PDF
Disparities in exposure to traffic-related pollution sources by self-identified and ancestral Hispanic descent in participants of the USC Children’s Health Study
PDF
A descriptive analysis of medication use by asthmatics in the Children's Health Study, 1993
PDF
Airway inflammation and respiratory health in the Southern California children's health study
PDF
Factors associated with agreement between childhood cancer survivors (CCS) and their parents on knowledge and attitudes about health-related quality of life, treatment and follow-up care
PDF
Bayesian multilevel quantile regression for longitudinal data
PDF
A randomized controlled trial of hormone therapy on postmenopausal women’s quality of life
PDF
Association of neighborhood characteristics with bystander CPR and out-of-hospital cardiac arrests in the city of Los Angeles
PDF
Evaluating the use of friend or family controls in epidemiologic case-control studies
PDF
Obesity paradox in acute heart failure decompensation
PDF
Air pollution and breast cancer survival in California teachers: using address histories and individual-level data
PDF
Effects of stress and the social environment on childhood asthma in the children' s health study
PDF
Exploring the association of number of cigarettes smoked and confidence to quit smoking in Korean American emerging adults: a multilevel modeling approach
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 α₁-adrenergic antagonist use on clinical outcomes in patients with heart failure
PDF
Surgical aortic arch intervention at the time of extended ascending aortic replacement is associated with increased mortality
PDF
The associations between ultra-processed food consumption and type 2 diabetes and obesity among young adults
Asset Metadata
Creator
Khachikian, Anita
(author)
Core Title
Street connectivity and childhood obesity: a longitudinal, multilevel analysis
School
Keck School of Medicine
Degree
Master of Science
Degree Program
Applied Biostatistics and Epidemiology
Publication Date
04/29/2016
Defense Date
04/21/2016
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
BMI,childhood obesity,multilevel models,OAI-PMH Harvest,street connectivity
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Berhane, Kiros (
committee chair
), Eckel, Sandrah (
committee member
), McConnell, Rob (
committee member
)
Creator Email
akhachik@usc.edu,anitakh3@yahoo.com
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c40-245301
Unique identifier
UC11277136
Identifier
etd-Khachikian-4393.pdf (filename),usctheses-c40-245301 (legacy record id)
Legacy Identifier
etd-Khachikian-4393.pdf
Dmrecord
245301
Document Type
Thesis
Format
application/pdf (imt)
Rights
Khachikian, Anita
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
Tags
BMI
childhood obesity
multilevel models
street connectivity