EXPLORING NEIGHBORHOOD PATHWAYS TO HEALTH: 

AN INTEGRATED ANALYSIS ACROSS SCALES 

 

By 

Amanda T. Rzotkiewicz 

 

A THESIS 

Submitted to 

Michigan State University 

in partial fulfillment of the requirements 

for the degree of 

 

Geography – Master of Science 

 

2018 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

ABSTRACT 

EXPLORING NEIGHBORHOOD PATHWAYS TO HEALTH: 

AN INTEGRATED ANALYSIS ACROSS SCALES 

 

 By 

Amanda T. Rzotkiewicz 

This research is a synthesis and discussion of two papers that apply diverse geographic 

techniques to closely examine neighborhoods and health, introduced in Chapter 1. Chapter 2 is 

titled, “Systematic review of the use of Google Street View in health research: Sampling, 

exposure assessment, prevention or monitoring, and health policy compliance” and of 54 studies 

qualifying for the review, one (2%) utilized GSV for sampling, forty-four (82%) for exposure 

assessment, and six (11%) for policy monitoring. Most studies reported considerable benefits of 

GSV, when compared to non-virtual methods, through the reduction of research time and costs, 

making it a promising tool for automated environmental assessment for health research. Chapter 

3 explores a relatively novel pathway to health (the microbiome) and is titled, “Evaluating the 

relationship between neighborhood vegetation and the human microbiome: implications for 

green space-health research”. Neighborhood vegetation scores and impervious surface area were 

compared to the microbial genera and biodiversity of the mouth, ears, eyes, nose and rectum (a 

surrogate of the gut) human microbiomes of postmortem residents of Wayne County, Michigan 

(n = 98). Relationships between neighborhood greenness and microbial composition varied by 

neighborhood size and area of the body.  Results suggest that each body area is a unique 

microbial niche that interacts with the environment in different ways, an important consideration 

for targeted modification of the microbial environment.  Overall, this research illustrates how an   

integrated analysis of neighborhoods and health has the potential to improve both health research 

and public policy across a wide range of geographic contexts and scales. 

 

 

For Felix, Sadie, Mari, Ollie, Sasha, Little Peanut, Bubbles, Cody, Inca (Gadget), Brooke 

(Stevie), Emma, the storm drain kittens and Lady, Ava, Nena & Minks, Apricot, Louie, Sugar 

Plum, Peaches, and most of all Nectarine, who has come the farthest of any of them. 

iii 

ACKNOWLEDGEMENTS 

 

My thanks to: 

Dr. Ashton Shortridge, for allowing me the opportunity to prove myself in the field of 

Geography, infinite wisdom and patience, and lenience with my (in)ability to work in R. 

Dr. Amber Pearson, for the introduction to Geography and consequent years of knowledge, 

guidance, friendship, and opportunities to travel and learn that I could only dream of 

before this. 

Dr. Jennifer Pechal, for providing the data and expertise that allowed me to explore Geography 

and learn Microbiology while also integrating my lifelong passion for data collected at a 

morgue. 

The Human Postmortem Microbiome (HPMM) team, specifically Dr. Jennifer Pechal (MSU 

Department of Entomology), Dr. M. Eric Benbow (MSU Department of Entomology and 

MSU Department of Osteopathic Medical Specialties), Dr. Carl J. Schmidt (Chief 

Medical Examiner, Wayne County Medical Examiner’s Office and U of M Department 

of Pathology), and Dr. Heather Jordan (Mississippi State University Department of 

Biological Sciences). It has truly been an honor to collaborate with the team responsible 

for creating an avenue of research that not only targets many issues important to my 

home community but takes place at the heart of it. 

Benjamin V. Dougherty, my steadfast co-author and co-pilot in academia and in real life.  

But most importantly,     E 

Dr. Nick Wilson, for invaluable contributions and expertise towards Chapter 2 of this thesis and 

its successful publication as a review article. 

iv 

Jonnell Sanciangco, for writing the infamous code that made it possible to run in six seconds the 

data analysis that was originally taking me six hours to complete. 

JEM & TMG, the two people who have always supported me more than I could ever deserve. 

I would also like to thank M.M. Brewer and C.R. Weatherbee for assistance in sample 

processing. This research was funded by a grant from the National Institute of Justice, 

Office of Justice Programs, U.S. Department of Justice awarded (2014-DN-BX-K008) to 

J.L.P, C.J.S., H.R.J., and M.E.B. Points of view in this document are those of the author 

and do not necessarily represent the official position or policies of the U.S. Department of 

Justice. 

 

 

v 

TABLE OF CONTENTS 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

vi 

LIST OF TABLES 
 
LIST OF FIGURES 
 
Chapter 1 Introduction 
 
Chapter 2 Systematic review of the use of Google Street View in health research  3 
 
5 
7 
 
 
15 
 
Chapter 3 Evaluating the relationship between neighborhood vegetation and  

 
2.1 Methods 
2.2 Results 
 
2.3 Discussion  and conclusion 

 
 
 

 
 
 

 
 
 

 
 
 

 
 
 

1 

 
 

 
 

vii 

 

 

     the human microbiome: implications for green space-health research 
3.1 Methods 
 
 
 
3.2 Statistical Analyses 
3.3 Results 
 

22 
 
24 
24 
24 
25 
27 
28 
3.3.1 Neighborhood greenness/impervious surfaces and microbial  29 

 
 
 
3.1.1 Study site 
3.1.2 Microbial data 
 
3.1.3 Neighborhood greenery and impervious surface data   
 
 

 
 
 

 
 
 

 
 
 

 
 
 

 
 
 

 
 

 
 

 
 

 
 

 
 

 

 
 
 
 
 
 

diversity by body area 

 

3.3.2 Correlations between neighborhood greenness/impervious   32 
         surfaces and abundances of genera by body area 

3.4 Discussion  
 

Chapter 4 Conclusion  
 
REFERENCES 
 
 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

36 

42 

46 

vi 

LIST OF TABLES 

 

Summary of studies included in this systematic review 

Sample statistics stratified by sex 

 

 

 

 

 

Hierarchical linear regression model results for neighborhood 
greenness/impervious surfaces and number of observed microbial  
species, for each body area. Significant associations (p ≤ 0.10) in  
bolded text 

 

 

 

Hierarchical linear regression model results for neighborhood  
greenness/impervious surfaces and Shannon diversity index, for each  
body area. Significant associations (p ≤ 0.10) in bolded text  

 

Hierarchical linear regression model results for neighborhood  
 
greenness/impervious surfaces and phylogenetic diversity, for each  
body area. Significant associations (p ≤ 0.10) in bolded text 

9 

29 

30 

31 

32 

Three strongest, significant correlations between proportion of  
 
neighborhood greenness/impervious surface and individual genera  
at each body area for the 200m neighborhood block. Cell color indicates 
significance level; blue (p ≤ 0.10), green (p ≤ 0.05), and yellow (p ≤ 0.01),  
nd indicates None Detected 

33 

Three strongest, significant correlations between proportion of  
 
neighborhood greenness/impervious surface and individual genera  
at each body area for the 400m neighborhood. Cell color indicates  
significance level; blue (p ≤ 0.10), green (p ≤ 0.05), and yellow (p ≤ 0.01),  
nd indicates None Detected 

34 

Three strongest, significant correlations between proportion of  
 
neighborhood greenness/impervious surface and individual genera  
at each body area for the 600m neighborhood. Cell color indicates  
significance level; blue (p ≤ 0.10), green (p ≤ 0.05), and yellow (p ≤ 0.01),  
nd indicates None Detected 

35 

Three strongest, significant correlations between proportion of 
 
neighborhood greenness/impervious surface and individual genera  
at each body area for the 800m neighborhood. Cell color indicates  
significance level; blue (p ≤ 0.10), green (p ≤ 0.05), and yellow (p ≤ 0.01),  
nd indicates None Detected 

36 

Table 1. 
 
Table 2. 
 
Table 3.  

 
Table 4.  

 
Table 5.  

Table 6.  

Table 7.  

Table 8.  

Table 9.  

 

vii 

LIST OF FIGURES 

 

Figure 1. 
 
Figure 2. 

Search process for inclusion of studies in the review  

 

Block (blue) and 400m (green), 600m (yellow), 800m (red)  
neighborhood buffers 

 

 

7 

26 

 

 

 

 

 

 

 
 

viii 

 

Chapter 1 

Introduction 

The condition of a neighborhood and the features contained within it can both threaten 

and promote the physical and mental health of its residents. For example, exposure to chemical 

contaminants, urban blight, or high crime rates can negatively impact health by causing chronic 

illness or stress and anxiety. Similarly, the existence of features such as well-maintained parks, 

bike lanes, and a cohesive social environment can promote healthy behaviors (e.g., physical 

activity). Public health policy and interventions are developed around the understanding of these 

associations and the improvement of health policy relies upon deepening our understanding of 

these complex relationships. To uniquely contribute to this knowledge base, this work extends 

beyond the standard reaches of public health and applies a geographic perspective—a careful 

consideration of the importance of place, space, and time and how these components interact 

within and across multiple scales—to public health issues. Utilizing geospatial techniques when 

researching public health concerns can provide new avenues for achieving the goal of improving 

both research techniques and public policy.  

This thesis illustrates and contributes to our knowledge of this symbiotic relationship 

between geography and public health in two ways. The first chapter reviews the use of Google 

Street View in health research—elucidating the research avenues and conclusions this relatively 

new resource has been able to provide thus far, and the new directions in health research that it 

can provide from a neighborhood to global level. In particular, this review focuses on how the 

intimate panoramic views of neighborhoods provided by GSV have been innovatively used to 

study health-related outcomes without the resources required to have a study team “on the 

ground”. Second, a novel approach is applied to explore the health of individuals and their 

1 

 

neighborhood conditions via the human postmortem microbiome. In this chapter, a measure of 

neighborhood greenness and proportion of impervious surface cover is compared to the 

microbial diversity and composition of residents of Wayne County, Michigan. While certain 

microbes can cause sickness, humans also rely on their microbial profile for healthy function. 

This chapter provides an exploratory look into how the human microbial profile may potentially 

be influenced by the vegetation (or lack thereof) present in the surrounding environment and 

consequently, the microbes it has to share. In contrast to the first chapter, the second chapter uses 

an aerial analysis to objectively quantify ground cover at four iterations of neighborhood size. 

This birds-eye view method allows for a more expansive, complete, and consistent assessment of 

the content of a neighborhood; however, it lacks the nuance provided by street-level analysis 

when targeting additional features hypothesized to play a role in health. By integrating the 

knowledge gleaned from both perspectives, a more holistic understanding of neighborhoods and 

health is achieved. Thus, the techniques applied in both chapters enhance our understanding of 

the potential for neighborhood conditions to affect the health and well-being of the individuals 

that reside within them, an important contribution towards the development of successful public 

health policy and initiatives to the benefit of the populations being studied.  

 

 

2 

 

Chapter 2 

Systematic review of the use of Google Street View in health research 

Since its development and official launch in the United States in 2007, Google Street 

View (GSV, a component of Google Maps) has become a source of ‘big data’ characterized by 

high spatial resolution, freely available images that provide panoramic views of homes, streets, 

businesses and neighborhoods at eye level1. Currently, GSV is available for a number of cities 

globally, including almost complete coverage of the cities in many developed countries. Less-

developed and rural areas or paths are also being added to the GSV database2. On the other hand, 

public backlash driven by privacy norms in countries such as Austria and Germany (Germany, 

for example, has made it easy for residents to opt-out) has restricted the amount and detail of 

GSV coverage available3. The image data for GSV is typically collected from cars equipped with 

a special camera called the Google Trekker, capturing overlapping images that are reconstructed 

into a 360° view linked to GPS data identifying the location of the image. Neighborhood image 

data are then updated by Google at a frequency that is dependent upon population density and 

weather conditions2. The older images, dating back to 2007, are retained and viewable through a 

timeline feature released in 20144. As such, GSV provides a data source for health researchers. 

These uses may include auditing the built environment of neighborhoods via the development of 

outcome- or exposure-specific tools, which rely on GSV data in lieu of the costly, time-

consuming and/or impractical in-person auditing that was previously necessary and is considered 

the ‘gold standard’1,5. Assessment of exposures in the built environment is of particular 

importance in health research for its associations with both mental and physical health via 

various pathways6. For example, maintained and visible urban green spaces have been associated 

with a multitude of positive health outcomes for nearby residents, including the facilitation of 

3 

 

physical activity and the promotion of positive mental well-being7. Conversely, areas 

characterized by disorder (e.g., broken windows, graffiti) have been associated with negative 

social and health outcomes, such as fear of crime8. In addition to exposure assessment, emerging 

research suggests that GSV may be useful for other facets of the health research process, such as 

designing a sampling scheme, identifying study sites and enhancing the quality of available data 

for features such as food outlets, which are theorized to have varying impacts on health-related 

behaviors9,10. Still other work has identified uses of GSV data for smoking-related policy 

compliance11.  

Prior studies have reviewed various geospatial, non-GSV approaches for health research 

including the use of Geographic Information Systems (GIS) to study the food environment12, the 

use of GPS technology to study physical activity13, and methods of neighborhood observation14. 

A systematic review by Charreire, et al. 1, examined the capacity of free geospatial tools to 

assess features of the built environment and their relationship to health outcomes to aid in the 

development of the SPOTLIGHT project, and found GSV and Google Earth to be reliable tools 

in all 13 studies reviewed. Building on this information, there has been no systematic review, to 

date, on the use of GSV for health research, though it has been partly considered in a wider 

overview of emerging geospatial technologies by Schootman, et al. 15, reviewed specifically for 

tobacco-related issues16, and discussed as a source of secondary food source data5. Therefore, the 

existing literature on the use of GSV in health research was reviewed, in order to understand its 

usefulness, limitations and potential for future research. More specifically, the review questions 

were: 1) How has GSV been used as a sampling tool in health research?; 2) How has GSV been 

used to evaluate exposures in health research?; and 3) How has GSV been used to evaluate 

health-related policy compliance, prevention or monitoring? These questions were then 

4 

 

addressed by reviewing identified health studies and noting implications for future health 

research. 

2.1 Methods 

 

This review adheres to the PRISMA guidelines for systematic reviews where 

applicable17. Studies were eligible for review if they utilized GSV as a component of their 

research and the study was either directly or indirectly related to human health outcomes, e.g., 

direct health outcomes, health policy compliance, environmental audits (with the intent of health-

related aesthetics or facilitating physical activity), accident/injury assessment, prediction or 

prevention and disaster recovery. Studies evaluating aesthetics without explicitly tying methods 

or findings to health were excluded. Excluded studies were also those only peripherally related to 

health including those concerned with privacy, current or potential building damage, heat 

efficiency, learning support, land use and reviews. Studies that utilized non-image, non-spatial 

GSV data, specifically methane sensors installed on the GSV car18, were also excluded. Studies 

that discussed only theoretical applications or implications of observing certain groups via GSV 

(for an example, see Wolthers 19), were also excluded. Lastly, studies that utilized GSV to 

compare or enhance sources of data without a specific relationship to health applications were 

not eligible. There were no restrictions on year of publication prior to the search date due to the 

relatively recent launch of GSV in 2007. However, studies were restricted to peer-reviewed 

articles published in English.  

Various search terms were used in PubMed to develop a comprehensive understanding of 

the depth of the literature that utilizes GSV. Due to the relatively small number of studies using 

GSV, it was determined that all studies pertaining to GSV would be collected and then assessed 

manually for relationships to health so as to not unintentionally exclude less conventional 

5 

 

approaches that may have otherwise not been identified by a health-related search term. The 

search term “GSV” was removed from the search query as it returned results for the “great 

saphenous vein” rather than Google Street View-related studies. Then, a search of existing 

literature was performed by the first author, using the terms: [(“Google Street View”) OR 

(“Google Streetview”) OR (virtual “street audit”)] on PubMed and Web of Science.  This search 

was performed on May 23, 2017. After articles were selected for inclusion, the second 

investigator then examined citations for additional qualifying articles not picked up by the 

search. 

The lead author was responsible for the compilation of the database search results (n = 

261). Then, two reviewers independently performed title and abstract screening using the website 

Covidence20. Differences between reviewer evaluations were identified by Covidence and 

subsequently discussed and resolved in person (n = 65). Criteria for the items considered to be 

health-related (see above) were refined in this stage. Then, both reviewers conducted full-text 

screening and documentation of study characteristics (Table 1). All included studies were 

approved by both reviewers (Figure 1). Significant discrepancies were resolved through 

discussion, but these were rare (n = 3). After qualifying studies were selected, the second 

investigator examined the citations for additional qualifying studies (n = 3). Included studies 

were heterogeneous in their aims, scope and methodology. Thus, results were synthesized, 

described and compared to the other works included in the review. Additional details, such as 

study site, unit of analysis and relationship to health, were also extracted. Because this review 

was focused on the methodology (usage of GSV), it was determined that a meta-analysis of the 

reported quantitative results in each included study would not be appropriate. Likewise, a risk of 

bias assessment was not applicable. 

6 

 

 

Figure 1. Search process for inclusion of studies in the review 

2.2 Results 

The first search returned 46 results from PubMed and 215 from Web of Science, with 36 

duplicates as determined by Covidence (and one by the reviewers), yielding a total of 224 studies 

assessed for eligibility. A total of 158 studies (71%) were removed in the abstract screening 

process. Studies were removed due to being not related to health research, not actually using 

GSV in the study design, or being only theoretical in content. This left 51 studies qualified for 

the final full review, with three additional studies identified via searching citations for a total of 

54 studies (Table 1). 

Of these eligible studies, methodological themes emerged. Most study sites were in North 

America (though only the United States or Canada, n = 31, 57%) and/or Europe (n = 20, 37%), 

followed by four studies based in New Zealand (7%) and one (2%) each from Australia, Japan, 

and Brazil. A majority of the studies eligible for review (n = 35; 65%) used GSV to audit 

multiple aspects of the environment, typically via an itemized checklist. All but one of the 

remaining studies (n = 18; 33%) used GSV to locate or assess specific features within the built 

environment, such as smokefree-policy signage. One study used GSV as a virtual reality 

simulation tool to improve knowledge of disaster preparedness21. General neighborhood audits 

7 

 

typically included the use or development of an itemized checklist either designed to quantify 

multiple aspects of the built environment or to target features related to specific outcomes. 

Specific examples of auditing tools developed for use with GSV include the European-designed 

and employed SPOTLIGHT (S-VAT) tool, which is designed to assess obesogenic features in 

the environment and was used by five studies included in the review22-26. Similarly, the 

American CANVAS tool is a computer-assisted checklist of 187 items pertaining to walkability 

and physical disorder of neighborhoods and was used by four studies included in the review8,27-

29.  

Eight studies (15%) related to the development of checklist-based tools or audits (such as 

SPOTLIGHT) or automated algorithms to quantify some aspect of the built environment. Often, 

these studies were followed up by tests or evaluations of the scoring methods, with the general 

finding being that the tool in question was more effective than in-person street auditing. The 

exception to this being Clews, et al. 30; however, this study focused on feature location, discussed 

below. In addition to itemized checklists, assessment of the aesthetics of the environment were 

accomplished by using automated algorithms to quantify green space6, open sky31, or the 

perceived comparative safety of city images32 visible in GSV imagery. 

8 

 

Table 1. Summary of studies included in this systematic review 

 

 

Reason for using GSV 

 

 

Study 

Less et al. 
(2015)  

Evans-
Cowley and 
Akar (2014) 

Study 
location 

Oakland, 
California and 
Minneapolis 
and St. Paul, 
Minnesota, 
USA 
Columbus, 
Ohio, USA 

Hanson et al. 
(2013)  

New Jersey, 
USA 

Johnson and 
Gabler (2015)  

Michigan, 
USA  

Mooney et al. 
(2016) 

Clews et al. 
(2016)  

New York 
City, New 
York, USA  
Wellington, 
New Zealand 

Adu-
Brimpong et 
al. (2017) 

Washington, 
D.C., USA 

Badland et al. 
(2010) 

Auckland, 
New Zealand 

Ben-Joseph et 
al. (2013) 

Bethlehem et 
al. (2014)  

Brookfield 
and Tilley 
(2016) 

Chudyuk et 
al. (2014) 

Compernolle 
et al. (2016) 

Boston, 
Massachu-
setts, USA  

Randstad 
megalopolis, 
Netherlands  

Edinburgh, 
UK  

Vancouver, 
British 
Columbia, 
Canada 
Five 
European 
urban regions 

Sampling 

Exposure 
assessment 

Study site 
comparability 

  

Policy 

compliance, 
prevention or 

monitoring 

  

Health 

outcome/behavi
or of interest 

Alcohol-related 
health 

GSV image data 
used (quantity) 

Liquor stores (20) 
and corresponding 
neighborhoods 

  

  

  

  

  

  

  

  

  

  

  

  

Bikeability 

  

Injury 
prevention for 
bicyclists 

Bike route (59) 

Pedestrian death 

2351 crash sites 

Traffic injury 

Crash sites (1,001) 

Pedestrian 
injury 

Walk Score items 
(532 intersections) 

Alcohol-related 
health 

Alcohol signage (12 
street segments) 

Physical activity  Active 

Neighborhood 
Checklist and Walk 
Score items (12 
street segments for 
each of 82 homes) 

Physical activity  NZ-SPACES items 
(48 street segments) 

Physical activity  Brownson et al. 
2004 items (84 
street segments) 

Physical activity  SPOTLIGHT items 

(128 street 
segments) 

Physical activity  FASTVIEW items 
(19 walking routes) 

Physical activity  Walk Score items 

(48 street segments) 

Physical activity  SPOTLIGHT items 

(4486 street 
segments) 

  

  

  

  

  

  

  

  

  

  

  

Audit of 
crash site 

Audit of 
guard rails 

Audit of 
intersection
s 
Alcohol 
marketing 

Neighborho
od 
walkability 

Neighbor-
hood 
walkability 
Obesogenic 
features  

Obesogenic 
features  

Neighbor-
hood 
walkability 

Neighbor-
hood 
walkability 

Obesogenic 
features  

9 

 

Table 1 (cont’d) 

Feuillet et al. 
(2016)  

Griew et al. 
(2013) 

Gullon et al. 
(2015)  

Kelly et al. 
(2013)  

Kelly et al. 
(2014)  

Five 
European 
urban regions 

Wigan, UK  

  

  

Madrid, Spain  

  

St. Louis, 
Missouri and 
Indianapolis, 
Indiana, USA  

St. Louis, 
Missouri and 
Indianapolis, 
Indiana, USA  

  

  

  

  

  

  

  

  

  

  

  

Kepper et al. 
(2016)  

Louisiana, 
USA 

Mertens et al. 
(2017)  

Five 
European 
urban regions 

Mygind et al. 
(2016) 

Victoria, 
Australia  

Roda et al. 
(2016)  

Silva et al. 
(2015)  

Five 
European 
urban regions 
Sao Paulo, 
Brazil  

Vanwolleghe
m et al. 
(2016)  

Vanwolleghe
m et al. 
(2014)  

Wilson et al. 
(2012)  

Belgium  

Flanders, 
Belgium  

St. Louis, 
Missouri and 
Indianapolis, 
Indiana USA  

Yin and 
Wang (2016)  

Buffalo, New 
York, USA 

 

 

Obesogenic 
features  

Neighbor-
hood 
walkability 
Neighbor-
hood 
walkability 

Obesogenic 
features  

Obesogenic 
features  

Neighbor-
hood 
"incivili-
ties" 
Bikeability 

Quality, 
public open 
spaces 
Obesogenic 
features  

Obesogenic 
features  

Obesogenic 
features  

  

  

  

  

  

  

  

  

  

  

  

Physical activity  SPOTLIGHT items 

(4486 street 
segments) 

Physical activity  FASTVIEW items 

(54 street segments) 

Physical activity  M-SPACES items 

(500 street 
segments) 

Physical activity  Active 

Neighborhood 
Checklist items (288 
street segments) 

Physical activity  Active 

Neighborhood 
Checklist items (400 
street segments) 

Physical activity  Checklist items (84 

homes and street 
segments) 

Physical activity  SPOTLIGHT items 

(4486 street 
segments) 

Physical activity  POSDAT items 

(171 open spaces) 

Physical activity  SPOTLIGHT items 
(60 neighborhoods) 

Physical activity  Objective 

Evaluation of 
Environment items 
(29 street segments) 

Physical activity  MAPS Global items 

(68 routes) 

Bikeability 

  

Physical activity  EGA-Cycling items 
(50 cycling routes) 

Physical activity  Active 

Neighborhood 
Checklist items (369 
street segments) 

Physical activity  Visible sky (3592 

images) 

  

  

Obesogenic 
features  

Walkability 
(visual 
enclosure) 

10 

 

Table 1 (cont’d) 

Bader et al. 
(2017) 

Bader et al. 
(2015) 

New York 
City, New 
York, San 
Jose, 
California, 
Philadelphia, 
Pennsylvania 
Detroit, 
Michigan, 
USA 
Unidentified 
metropolitan 
locations, 
USA 

Clarke et al. 
(2010) 

Chicago, 
Illinois, USA 

Hyam (2017)   Edinburgh, 

Kepper et al. 
(2017)  

Lafontaine et 
al. (2017)  

Li et al. 
(2015)a 

Li et al. 
(2016)  

Li et al. 
(2015)b 

Li et al. 
(2015)c 

Marco et al. 
(2017) 

Mooney et al. 
(2014)  

UK  

Southeast 
Louisiana, 
USA  

Ottawa, 
Ontario, 
Canada  
Boston, 
Massachusetts 
and New 
York City, 
New York, 
USA; Linz, 
Salzburg, 
Austria  
Hartford, 
Connecticut, 
USA 
New York 
City, New 
York, USA 
Hartford, 
Connecticut, 
USA 
Valencia, 
Spain 

New York, 
California, 
Pennsylvania 
Michigan, 
USA 

  

  

  

  

  

  

  

  

  

  

  

  

  

  

  

  

  

  

  

  

  

  

  

  

Neighbor-
hood 
disorder 

Neighbor-
hood 
walkability 
and 
disorder 
Neighbor-
hood 
features 
Perceived 
naturalness 

Neighbor-
hood 
disorder 
(perceived 
safety) 
Environ-
ment 
aesthetics 
Neighbor-
hood 
disorder 
(perceived 
safety) 

Street 
greenery 

Street 
greenery 

Street 
greenery 

Neighbor-
hood 
disorder 

Neighbor-
hood 
disorder 

11 

Mental health 

CANVAS items 
(1826 street 
segments) 

Mental health 

CANVAS items 
(300 census tracts) 

Mental health 

CCAHS items (224 
street segments) 

Mental health 

Random panoramic 
images (768) 

Mental health 

Street segments (54) 

Mental health 

Residential blocks 
(150) 

Mental health 

GSV images (4126) 

Mental health 

Green space pixels 
(18 images) 

Mental health 

Green space pixels 
(300 sites) 

Mental health 

Green space pixels 
(3000 sites) 

Mental health 

Mental health 

Itemized checklist 
of disorder criteria 
(92 census block 
groups) 

CANVAS items 
(1826 block faces) 

  

  

  

  

  

  

Table 1 (cont’d) 

Naik et al. 
(2014)  

Odgers et al. 
(2012)  

Porzi et al. 
(2015)  

Quinn et al. 
(2014)  

Rundle et al. 
(2011)  

Wu et al. 
(2014)  

Ahemtovic et 
al. (2015) 

Ahemtovic et 
al. (2017) 

Mitsuhara et 
al. (2016)  

Seekins et al. 
(2014) 

Wilson 
(2015) 

Wilson et al. 
(2015)  

Boston, 
Massachusetts 
and New 
York City, 
New York, 
USA; Linz, 
Salzburg, 
Austria  
England and 
Wales, United 
Kingdom 

Boston, 
Massachusetts 
and New 
York City, 
New York, 
USA; Linz, 
Salzburg, 
Austria  
New York 
City, New 
York, USA  
New York 
City, New 
York, USA  
Cambridge-
shire, United 
Kingdom  
San 
Francisco, 
California, 
USA 
San 
Francisco, 
California and 
Manhattan, 
New York, 
USA; Milan, 
Italy 
Japan 

Five 
communities 
in USA 
Lower North 
Island, New 
Zealand  

Lower North 
Island, New 
Zealand  

  

  

  

  

  

  

  

  

  

  

  

  

 

 

 

Neighbor-
hood 
disorder 
(perceived 
safety) 

Neighbor-
hood 
disorder 

Neighbor-
hood 
disorder 
(perceived 
safety) 

Neighbor-
hood 
disorder 
Neighbor-
hood 
disorder 
Environ-
ment 
aesthetics 
  

  

  

  

  

  

 

12 

Mental health 

Streetscore items 
(4109 GSV images) 

Mental health 

Mental health 

SSO i-Tour items 
(2024 children's 
streets) 

Automatic scene 
recognition in GSV 
images (4136) 

Mental health 

CANVAS items 
(532 block faces) 

Mental health 

Block faces (37) 

Mental health 

REAT items (24 
streets) 

Zebra crosswalks 
(137) 

Locations of 
zebra 
crosswalks 

Injury 
prevention for 
the blind 

Locations of 
zebra 
crosswalks 

Injury 
prevention for 
the blind 

Zebra crosswalks 
(1,292) 

Disaster 
preparedness 

Injury 

Handicap 
accessibility 

Injury/recovery 

Virtual 
representation of 
study site (campus) 

Civic centers in 5 
communities 

Smokefree 
signage 

Smoking-related 
health 

Public hospitals (30) 

Smokefree 
signage 

Smoking-related 
health 

Primary and 
secondary schools 
(50) 

 

One (2%) study pertained to the first research question. In only one study was the 

development of a sample an objective of neighborhood auditing via GSV. The study sample was 

created by utilizing the audit to confirm the homogeneity of multiple study sites9. In this study, 

GSV was used to audit the environment of liquor stores in six target areas in Minneapolis 

(Minnesota, USA) and scores assigned to determine ‘matching’ study sites. This method was 

determined to be useful but limited by similar caveats found in other GSV research; particularly, 

items related to neighborhood disorder such as small, temporary features (e.g., litter) were 

difficult to identify and the severity of disorder (e.g., impact of graffiti) was likely to vary 

between auditors. Additionally, spatio-temporal inconsistency affected the ability to effectively 

audit environments, such as the construction of a building not yet captured by GSV imagery. 

Exposure assessment has thus far been the primary use of GSV in health research (n = 47, 

89%). Across all studies, and as expected, the relationship of the GSV data to health was 

indirect. These various indirect relationships were interconnected in their potential impact on 

physical and mental health outcomes. For example, some studies focused on features related to 

safety, such as potential33 and realized pedestrian-vehicle collision sites34,35. Others used GSV to 

measure signs of neighborhood disorder as a proxy for perceived safety8,28,29,36-38. Physical 

disorder was considered a feature that affects human behavior and thus mental and physical 

health outcomes. It is important to note that while only the primary health outcome of interest for 

each study was recorded within Table 1, outcomes such as physical activity and mental health 

are inextricably linked and thus may be concurrently studied (for an example, see Kepper, et al. 

39). Health-related behaviors, such as walking or biking, were also evaluated for their 

relationships to the built environment assessed via GSV22,40. Similarly, relationships between 

built environment features and negative health-related behaviors (e.g., smoking or consuming 

13 

 

alcohol) were examined9,11,30. Urban greenery and the degree of naturalness were also considered 

as indirect influences on human health 6. While GSV was utilized to quantify these potential 

causal mechanisms, only studies that relied on surveys or used post-event data (such as crash 

sites) integrated a measurement of actual health outcomes. 

 

Six studies (11%) were identified for examining features of the built environment that 

were related to the implementation of public health policy using GSV. In effect, GSV was 

utilized to virtually identify the locations (and thus, confirm the compliance) of such features. 

Two studies used GSV to identify the location of “zebra crosswalks”, which have thick, painted 

lines designed to aid blind pedestrians and prevent crashes33,41. The first study (2015) focused on 

the development of an automatic algorithm to identify zebra crosswalks in satellite imagery and 

confirmed in GSV imagery. In total, 137 zebra crosswalks were identified. A follow-up study 

expanded the scope from San Francisco (California, USA) to an additional European site (Milan, 

Italy) and identified 1,292 zebra crosswalks. To improve this algorithm and the resulting 

database, the Pedestrian Crossing Human Validation web service was created to utilize 

crowdsourcing to identify errors in the recognition data. Other feature-location studies focused 

on negative health behaviors (e.g., features related to smoking and alcohol consumption). Clews, 

et al. 30 used GSV to detect the presence of alcohol-related signage on 12 segments of six streets 

determined to have varying socioeconomic status in Wellington, New Zealand. In contrast to 

most studies included in the review, this study did not find GSV to be more effective than in-

person auditing, mainly due to of poor image resolution. Similarly, GSV was used to identify the 

presence of smoke-free signage at the grounds of hospitals and schools in New Zealand11,42. At 

schools, GSV was found to have a 44% success rate for identifying smokefree signage at school 

entrances, when compared to in-field assessments11. In some cases, the signage was legible in the 

14 

 

GSV image (which were an average of 1.9 years old) but not in the field observation, due to 

fading.  

Probably the most innovative application of GSV to human well-being was the 

integration of GSV imagery into a disaster evacuation simulation21. This study, from Japan, 

addressed the increase in prevalence of “natural” disasters but lack of general disaster 

preparedness among individuals by creating a virtual reality “game” that presented participants 

with choices to effectively teach them how to respond to emergencies. GSV imagery was added 

to the simulation so that participants could experience the evacuation in a realistic setting, with 

the ability to change routes and repeat the exercise. While other studies utilizing GSV as a 

learning tool were excluded (e.g., becoming familiar with a route or other forms of navigation), 

this study was unique in that it provided information with the intent of influencing human health 

outcomes. 

2.3 Discussion and conclusion 

 

Overall, one (2%) study pertained to the first review question (“How has GSV been used 

as a sampling tool in health research?”), 44 (82%) related to the second (“How has GSV been 

used to evaluate exposures in health research?”), and six (11%) were relevant to the third (“How 

has GSV been used to evaluate health-related policy compliance, prevention or monitoring?”). 

Concerning the first research question, only one study was determined to relate to sampling, 

however, multiple studies presented similar challenges with image quality as those discussed by 

Less et al. using GSV as a sampling tool9. Burgoine and Harrison 43 used GSV to supplement the 

identification of food outlets, which are associated with varying obesogenic capacity, in 

comparison to solely relying on official data. Clews, et al. 30 assessed the capacity of GSV to 

collect alcohol-related data in the city of Wellington, New Zealand; for example, the 

15 

 

identification of alcohol-related marketing (both promotional and discouraging) or observations 

of active alcohol consumption. In this study, GSV was proven to be an inadequate tool in 

comparison to on-street observations, as only 50% of alcohol outlets and 52% of associated 

marketing data collected on-street were identified using GSV. The authors note that this method 

may be more reliable as more current image data with increased quality/image resolution become 

available. In September 2017, it was reported in online media that Google would be updating its 

cameras to be higher definition (of interest, stating a goal of being able to read fine print on 

buildings) but Google does not mention the resolution of its cameras or this improvement on 

their official site2,44, however, the Google API services state a maximum resolution of 2048 x 

2048 available to their Premium Plan customers2. The new camera system uses fewer but more 

specialized cameras and machine learning to include features such as business hours in search 

results45. Thus, it is difficult to assess the extent of this limitation, as high definition imagery 

may already be available in some areas and vastly affecting the capacity of GSV in this context. 

Also related to data availability, Curtis, et al. 46 assessed the effectiveness of evaluating change 

over time of street segments in the available data. Both of these image quality may affect the 

researchers’ ability to interpret and compare sites. In sum, the power of GSV as a sampling tool 

is largely dependent on the visibility (size) and permanence of the targeted features as well as 

uniform data availability and resolution. 

Concerning the second research question, a majority of studies that utilized GSV for 

exposure and outcome assessment were focused on auditing (via itemized checklist) features of 

the built environment that impact human health outcomes. Most often, these included obesogenic 

features pertaining to physical activity22-26,40,47-60. The identification of street features that 

contributed to the “walkability” of the environment motivated most checklists; however, studies 

16 

 

also targeted specific physical activities, such as bikeability22,40,59 or specific groups such as 

elderly persons10. Second and most prominently, itemized checklists addressed mental health 

outcomes28,29,37-39,61-64 via features that pertained to neighborhood disorder and potentially fear of 

crime (stress). Of note, mental health outcomes are not mutually exclusive from physical health 

outcomes; for example, neighborhood disorder can discourage walkability8. Physical injury was 

addressed both proactively, via identifying features that could cause it28,40, and retroactively, by 

assessing the sites where a car crash had already occurred34. This variety of application across 

health outcomes suggests that GSV is a useful source for original data collection, particularly in 

relation to the construction and implementation of auditing instruments designed to investigate 

the relationship between the built environment and human health. The particular strength of GSV 

for exposure and outcome assessment studies continues to be the ability to remotely (and thus, 

time- and cost-effectively) assess neighborhoods in a wide range of global contexts, as has been 

identified in previous systematic reviews1,12,13. However, this method is limited by image 

availability, which may result in the exclusion of specific populations, such as those living in 

deprived or rural areas. Similarly, inconsistency in temporal availability and image quality may 

hinder data collection efforts46. Inconsistencies in image date and resolution quality may result in 

the misclassification or erroneous inclusion/exclusion of exposures, negatively impacting the 

validity of the collected health data. 

Beyond checklists, GSV was also used for the identification or quantification of features 

that relate to health using automated methods. Hyam 65 used image classification to assess 

perceived naturalness, which is linked to mental health. Using a foundation of images rated for 

perceived naturalness by humans and the Google Vision API (which uses GSV images), images 

in the city of Edinburgh, Scotland, were assigned a Calculated Semantic Naturalness (CSN) 

17 

 

metric that was found to significantly correlate with human-assessed perceived naturalness. 

Automated methods were also used to successfully assess street greenery by modifying a green 

view index6,66-68. These studies, which quantify green pixels, expand the application of this 

method to assess differences in the socioeconomic, structural, and perceived safety of urban 

neighborhoods. Similar to green pixels, Yin and Wang 31 assessed quantity of visible sky in 

3,592 images of Buffalo, New York. Visible sky, or lack thereof, is associated with “visual 

enclosure” which is theorized to affect street walkability via human perception of the 

environment. These novel automated methods, which are largely unexplored in comparison to 

audit-based methods, require more intricate technical development but are advantageous in their 

ability to quickly and effectively assess large quantities of image data. 

Lastly, relevant to the final research question, existing research pertaining to the use of 

GSV to assess policy compliance prevention or monitoring is currently sparse in comparison to 

exposure auditing; however, it may become more prevalent as the GSV database and image 

resolution improves30. Though limited in number, these studies use multiple approaches to health 

policy; specifically, GSV has been used to monitor the installment of zebra crosswalks to aid 

blind persons33,41, including the development of a volunteer-generated database to expand the 

monitored area. It has also been used for compliance assessment, particularly to evaluate 

handicap accessibility in public buildings69 and identify smoke-free signage at the grounds of 

schools and hospitals11,42. In Japan, where earthquakes can result in serious public disasters, 

Mitsuhara, et al. 21 utilized GSV to develop a disaster simulation exercise to aid preparedness 

skills in the environment familiar to the user. 

A major strength of this review is that it is the first, full systematic review of the use of 

GSV in health research: it identifies 54 studies in comparison to the 12 studies discussed in the 

18 

 

overview of emerging technologies by Schootman, et al. 15 and with a much larger scope than the 

review of GSV and tobacco issues by Wilson, et al. 16. However, this review is limited by the 

lack of coverage of non-peer reviewed grey literature, such as reports published on official 

websites, that may have utilized GSV in their design, and may also benefit from an expanded 

search of available databases. It also does not include other freely available geospatial software, 

such as Bing Maps. However, as discussed in the review by Charreire, et al. 1, GSV and Google 

Earth were consistently used and only one study by Ben-Joseph, et al. 48, also included in this 

review, used Bing Maps in addition to Google products. While this review encompasses a wide 

range of approaches to health, the scope may still be limited by interpretation of scope, such as 

the potential for building heat efficiency affecting the health of its residents, which was excluded 

in the current review. This review also purposely excludes direct observation of humans via 

GSV, an important component of the social environment that is ultimately linked to health that 

would benefit from a separate, more targeted review. 

The content of the studies included in this review suggests that there are a number of gaps 

in the research potential for GSV. For example, at the time of this review, no studies were 

identified that utilized GSV to study park and playground design (as it may relate to physical 

activity, safety, provision of shade, etc.), availability of and features of public drinking fountains, 

or the potential health benefits of different urban trees (for shade and reduction of air and noise 

pollution, etc.). Similarly, the occurrence of detrimental health-related behaviors in public spaces 

(especially those regulated or potentially regulated by policy) such as smoking or features that 

promote smoking (such as ash trays) or consuming alcohol (visibility of bar patios). The health 

outcomes being studied could also be targeted towards regional risks – such as studying the 

prevalence of wearing a hat in the parks of countries or regions with high risk of skin cancer. In 

19 

 

countries that have “footpath view” (in lieu of “street view”) in retail outlets, there could be 

research pertaining to public promotion of adverse health behaviors (e.g., tobacco or alcohol 

displays, prevalence of highly processed foods). In-depth research could also be performed to 

determine the relative cost-effectiveness of using GSV relative to in-person assessment or field 

observations; building on the comparison by Pliakas, et al. 10 of researcher hours required for 

three different neighborhood observation methods. Additionally, the history feature of GSV, 

which provides previous image data, is a component of GSV that has been recognized for its 

research potential15 but is otherwise unexplored in health contexts. For example, there is 

potential for repeating exposure assessment on images of the same location at different time 

points and then comparing these results to longitudinal health data. As the Google Maps database 

expands, other regions of the globe such as those nations in Africa, South America and Southeast 

Asia may also benefit from GSV-integrated research. As it stands, developing countries outside 

of Brazil are unrepresented in this review though this finding may have been exacerbated by the 

exclusion of non-English-language studies. Likewise, the current GSV literature examines only 

urban and suburban communities while rural areas are unrepresented. While GSV is frequently 

used to compare multiple cities within a single study, these study sites have been relatively 

homogenous in that they are Western urban centers, even when the country differs. However, an 

expanding GSV database will provide the opportunity to systematically collect data on many 

diverse study sites, facilitating larger, more comprehensive studies without the additional costs 

typically incurred in global research. 

To conclude, this review suggests that Google Street View has thus far proven to be a 

generally useful tool, albeit a still developing one, for observing features of a wide range of 

health-related environments. Additionally, the current review demonstrates how advances in 

20 

 

technology, public availability of imagery and our general knowledge of human health are 

continually becoming integrated in ways that expand our ability to approach new areas of 

geospatial health research. These works that integrate GSV into their methodology are beneficial 

to public health in two critical ways. First, they provide a means of in-depth exposure 

assessment, which allows for the quick and efficient identification and location of both risks and 

beneficial features within a targeted community. Second, it provides a tool for remotely 

assessing health policy implementation or compliance. The identification of the institutions 

which fail to comply with public health policies could reveal other systemic issues, such as lack 

of funding or poor management, which can then be remedied to the benefit of the communities 

being served. Put broadly, GSV is a powerful tool for identifying areas and features in need of 

improvement, a primary concern for health researchers. 

 

 

 

21 

 

Chapter 3 

Evaluating the relationship between neighborhood vegetation and the human microbiome: 

implications for green space-health research 

There are multiple pathways through which the surrounding environment can impact 

human health70. The condition of neighborhood built and natural environments (e.g., abandoned 

buildings, maintained green spaces) have repeatedly been associated with human health 

outcomes, such as obesity and stress/anxiety53,71-75. Causal mechanisms for these trends such as 

the promotion of physical activity have been investigated in-depth, however, emerging research 

suggests that the human microbiome may also play a role. For example, like physical activity, 

increased human microbial diversity in select body areas such as the gut have been associated 

with improved mental and physical health76,77. This is important because humans are constantly 

sharing and obtaining microbes from their environment78,79. For example, the human microbiome 

has been linked to the microbial composition of the surrounding built environment, such as the 

surfaces (e.g., ATM, subway) present in urban areas like New York City79,80. In addition to 

dispersing microbes, humans obtain microbes from their environment and other occupants such 

as family members or housemates including the family pet81. Pathogens found on a kitchen 

counter have been matched with those found on the hands of household residents, demonstrating 

the human acquisition of microbes found in the environment82. However, urban areas typically 

contain less vegetation than non-urban areas (and instead large areas covered by impervious 

surfaces such as parking lots or roads), which limits microbial biodiversity83. Recent research 

suggests that the decreased exposure to microbes in urban areas (compared to rural) may play a 

mechanistic role in the prevalence of immunological and stress-related disorders84,85. For 

example, urban populations have been found to be at increased risk for health issues such as 

22 

 

allergies due to decreased microbial biodiversity in the surrounding environment compared to 

rural residents86. However, it remains unclear how the presence of the vegetation in 

neighborhoods may affect the microbial biodiversity and composition of the microbes on human 

residents and ultimately, mental and physical health. 

The human body is host to multiple microbial niches, of which the pathways of microbial 

exposure, composition, and the relationship to health is not uniform. Currently, the gut 

microbiome is the most studied and well-understood. A highly diverse system hosted within the 

body, the gut microbiome has been linked to the regulation of multiple mental and physical 

health outcomes, including but not limited to anxiety and depression, obesity, allergies, 

Alzheimer’s, and irritable bowel syndrome, often linked to diet87-93. In contrast to the gut 

microbiome, the oral microbiome is only moderately affected by diet and is also typically highly 

diverse94,95. The areas of the body along the digestive tract, despite their physical connection, 

have been found to host distinct microbiomes and so it remains unclear whether exposure to 

vegetation-associated microbes at external body areas such as the mouth or nose (e.g., through 

breathing) may act as a pathway to the internal gut microbiome96. 

To address the following gaps: i) the potential effect of neighborhood vegetation and 

impervious surfaces on the biodiversity of the human postmortem microbiome; and ii) the 

potential effect of neighborhood vegetation and impervious surfaces on the composition of the 

human postmortem microbiome, this study used high-resolution, remotely sensed imagery to 

examine the relationships between neighborhood greenery, impervious surfaces area, and the 

biodiversity and composition of the human postmortem microbiome at five body areas: rectum (a 

sample site for the gut microbiome), mouth, nose, ear and eyes, sampled from residents of 

Wayne County, Michigan.  

23 

 

3.1 Methods 

3.1.1 Study site 

            With nearly two million residents, Wayne County is the most populated county in 

Michigan and the 13th most populated in the United States. It is comprised of 43 communities, 

including the post-industrial city of Detroit97. The total area of Wayne County is 1,083km2, 

however, 98km2 of the eastern border of the county is the waters of the Detroit River and Lake 

St. Clair98 . As of 2010, 52.3% of residents were white, 40.5% were black or African American 

and 7.2% were another race or ethnicity. The per capita income (2010) for the county was only 

$20,948, with approximately 23.7% of the population below the poverty line [American 

Factfinder, US Census]. Detroit’s role in the heyday of the automobile industry has made Wayne 

County a major industrial center. The variation in vegetation between the suburban 

neighborhoods and the city of Detroit makes Wayne County an ideal place to study the potential 

influence of neighborhood vegetation on the human microbiome. 

3.1.2 Microbial data 

           Postmortem microbial samples were collected during routine death investigation at the 

Wayne County Medical Examiner’s Office (Detroit, MI) between 2014-2015. A total of 98 cases 

(61 male, 37 female) were swabbed at five different body areas (mouth, ears, eyes, nose and 

rectum) less than 48 hours after death. This threshold for swabbing is important, as a microbial 

sample taken in this window is a suitable  proxy for the living microbiome99. For a description of 

sample collection, genomic DNA extraction, 16s rRNA gene targeted amplicon high-throughput 

sequencing, and sequencing analysis, see Pechal, et al. 100. Three measures of biodiversity 

(observed species, Shannon diversity index, and phylogenic diversity whole tree), in addition to 

counts of identified genera (richness), were calculated. Sequence data were archived through the 

24 

 

European Bioinformatics Institute European Nucleotide Archive (www.ebi.ac.uk/ena) under 

accession number: PRJEB22642. The following data were recorded for each case: sex, 

race/ethnicity, age (years), date of autopsy, location of death event (coordinates and 

indoor/outdoor), home address, manner of death (how an injury or disease was determined at 

autopsy to have led to death: natural, accident, homicide, suicide) and body mass index (BMI: 

kg/m2).  For this study, only cases in which the home address or location of death was known 

were included. Hospital deaths without home addresses were excluded (n = 6).  

3.1.3 Neighborhood greenery and impervious surface data 

            Coordinates were assigned to each case based on home address. If the home address was 

unknown, the location of death was used for geocoding (n = 8 or 8%). Then, each case was 

assigned  to the relevant census block-group, to assign median household income for the area 

(US Census American Community Survey 5-year 2010 estimate101). 

            Since the geographic extent to which ‘neighborhood’ vegetation could  affect the human 

microbiome is unknown, varying-sized buffers around each home location were created, 

including i) 200m polygons representing a city block (based on the standard neighborhood block 

size in Detroit); and ii) 400m, 600m and 800m Euclidean buffers, based on prior research on 

neighborhood extent102,103. The 200m polygons representing ‘neighborhood blocks’ were 

individually constructed around each case to represent the most accurate immediate 

neighborhood. This allowed for block neighborhoods to accurately represent cul-de-sacs and 

other non-linear neighborhoods (Figure 2). 

 

 

25 

 

Figure 2. Block (blue) and 400m (green), 600m (yellow), 800m (red) neighborhood buffers 

 

            Next, aerial imagery was compiled for these neighborhoods. Sixty-seven aerial images 

covering Wayne County, provided by the National Agriculture Imagery Program (NAIP) at 1m 

resolution were obtained from USGS Earth Explorer104. The images used in this study were four-

band (RGB, infrared) GeoTIFFs captured either June 28, 2014 or July 5, 2014. All images were 

mosaicked to create a single raster image of Wayne County and bordering areas. The Normalized 

Difference Vegetation Index (NDVI) was then calculated using the red (band 1) and near 

infrared (band 4) bands. To calculate the proportion of neighborhood with high levels of 

greenery, an NDVI of greater than 0.3 was used (hereafter neighborhood greenness; an NDVI 

value of greater than 0.3 represents healthy vegetation and excludes dry grasses and shrubs. This 

threshold was spot-checked in multiple locations throughout the study area.) To calculate the 

proportion of impervious surfaces, the proportion of neighborhood with an NDVI less than or 

equal to zero was calculated. Next, the proportion of neighborhood occupied by water (as NVDI 

is also less than zero) was subtracted. All spatial processes were conducted using ArcGIS 10.5 

26 

 

(ESRI, Redlands, CA) and pgAdmin 4, a graphical front end for a Postgres/PostGIS database 

(Dave Page, open source)105.  

3.2 Statistical Analyses 

Specifically, this study aimed to answer the following: 

1) For each of the anatomical microbiome areas tested, is there a relationship between proportion 

of neighborhood greenness and proportion of impervious surface area and the diversity of the 

human microbiome when controlling for individual and neighborhood covariates? 

2) For each of the anatomical microbiome areas tested, what are the relationships between 

proportion of neighborhood greenness and proportion of impervious surface area and the 

abundances of genera found in the human microbiome? 

To answer the first research question, separate, hierarchical linear regression models for 

each body area were fit using the three diversity metrics (observed species, Shannon diversity 

index, and phylogenic diversity whole tree) as the dependent variable and proportion of 

neighborhood greenness or impervious surface area (at 200m, 400m, 600m and 800m) as the 

independent variable of interest. Hierarchical models were used as both individual-level and 

geographic area-level covariates were included, and such models allow for the “nesting” of 

individual-level within area-level covariates (in this case, block-level median household income) 

to avoid repeated measures across observations, a requirement of regression models. 

Specifically, each model was adjusted for age, sex, race/ethnicity, season of death, BMI, manner 

of death, and census block-level median household income. In total, 120 models were fitted. The 

results for the independent variables of interest are presented here. 

To answer the second research question, Kendall’s tau-b, a conservative, rank-based non-

parametric correlation test was calculated. This test was selected to account for multiple 

27 

 

comparisons and non-normally distributed data. Separate correlation tests between genera counts 

(n = 220) and proportion of neighborhood greenness and impervious surface area at each buffer 

size for each body area were performed. For each neighborhood buffer, the three correlations 

with the largest, significant tau-b values observed at each body area are presented here (full 

results available as a Supplementary File). All statistical analyses were conducted using Stata 

v15 (StataCorp, College Station, TX). 

3.3 Results 

All cases were adults and ranged in age from 18 to 88 years (mean 43; sd 14) [Sample 

characteristics are shown in Table 2]. The sample was evenly split between black and white 

(50%), but a majority were male (62%). Most cases died indoors (87%), in the spring (53%), and 

through a manner of death deemed an ‘accident’ (44%). The median household income of census 

blocks in our sample was similar to those in Wayne County (~ $40,000)101. Males and females 

were very similar in age (mean 43; sd 14), ethnicity (50% black/white) and BMI (mean 29; sd 8). 

They were also similar in season of death; however, no males and only three females died in the 

Autumn for this dataset. Though males and females were relatively equally like to have their 

manner of death classed as natural or an accident, males were more frequently a victim of a 

homicide while women more frequently died of suicide for this subset of the data. 

In terms of the descriptive statistics for the exposures of interest, the 200m neighborhood 

blocks and the 400m buffers had less than one percent up to 44% neighborhood greenness (mean 

18%; sd 10%). Similarly, 600m and 800m buffers had 2 – 41% neighborhood greenness (mean 

19%; sd 9%). For impervious surfaces, the 200m neighborhood blocks had 21 – 94% (mean 

56%; sd 13%). Similarly, the 400m neighborhood buffer had 20 – 89% (mean 54%; sd12%), the 

28 

 

600m neighborhood buffer had 16 – 86% (mean 54%; sd 11%), and the 800m had 18 – 85% 

(mean 54%; sd 11%). 

Table 2. Sample statistics stratified by sex 

Variable 

  

Female (n = 37) 

Male (n = 61) 

Total (n = 98) 

 
black 
white 

 
2014 
2015 

Age, mean (sd) 
Ethnicity, n (%) 
  
BMI, mean (sd) 
Year of death, n (%) 
  
Homeless, n (%) 
Site of death, n (%) 
  
Manner of death, n (%) 
  
  
  
Season of death, n (%) 
  
  
  
Median income, mean (sd) 

 
outdoors 
indoors 
natural 
accident 
homicide 
suicide 
spring 
summer 
autumn 
winter 

44 (12) 
18 (49) 
19 (51) 
28 (9) 
18 (49) 
19 (51) 

3 (8) 
4 (11) 
33 (89) 
10 (27) 
16 (43) 
5 (14) 
6 (16) 
19 (52) 
9 (24) 
3 (8) 
6 (16) 

42 (15) 
31 (51) 
30 (49) 
30 (7) 
27 (44) 
34 (56) 

5 (8) 
9 (15) 
52 (85) 
14 (23) 
27 (44) 
17 (28) 

3 (5) 

33 (54) 
18 (30) 

0 (0) 

10 (16) 

43 (14) 
49 (50) 
49 (50) 
29 (8) 
45 (46) 
53 (54) 

8 (8) 

13 (13) 
85 (87) 
24 (25) 
43 (44) 
22 (22) 

9 (9) 

52 (53) 
28 (28) 

3 (3) 

16 (17) 

40,152 (17,918) 

38,248 (17,328) 

38,967 (17,485) 

 

3.3.1 Neighborhood greenness/impervious surfaces and microbial diversity by body area 

In Tables 3-5, results are provided by each dependent variable. For observed species and 

the independent variables of interest, only two relationships were significant at p ≤ 0.10 (Table 

3). These were negative associations between the nose microbiome and neighborhood greenness 

at the 600m (β = -0.61, p = 0.08) and 800m (β = -0.66, p = 0.06) buffers, respectively. In other 

words, and surprisingly, for every 1% increase in neighborhood greenness, one would expect the 

observed species in the nose microbiome to decrease by 0.61 operational taxonomic units 

(OTUs). There were no significant relationships between neighborhood impervious surfaces and 

observed species for any of the body areas. The directions of the observed relationships were 

29 

 

mostly consistent across neighborhood buffer sizes, except for impervious surface area and the 

microbiomes of the nose and eyes. Interestingly, an increase in the proportion of both 

neighborhood greenness and impervious surface area was associated with an expected decrease 

in detected OTUs in the gut microbiome (sampled at the rectum), though not significantly in any 

model. 

Table 3. Hierarchical linear regression model results for neighborhood greenness/impervious 
surfaces and number of observed microbial species, for each body area. Significant associations 
(p ≤ 0.10) in bolded text 

  

  

200m block 

400m 

600m 

800m 

Obs. Species 

β 

Greenness 

Impervious 

Greenness 

Impervious 

Greenness 

Impervious 

Greenness 

Impervious 

Greenness 

Impervious 

0.17 

-0.08 

0.54 

0.31 

0.10 

-0.09 

-0.18 

-0.03 

-0.08 

-0.02 

 

h
t
u
o
M

 
s
r
a
E

 
s
e
y
E

 
e
s
o
N

 

m
u
t
c
e
R

 

p 

0.54 

0.69 

0.54 

0.38 

0.80 

0.75 

0.57 

0.90 

0.74 

0.91 

β 

0.45 

-0.28 

-0.18 

0.10 

0.02 

0.43 

-0.52 

0.28 

-0.11 

-0.09 

p 

0.14 

0.26 

0.73 

0.81 

0.96 

0.21 

0.12 

0.31 

0.68 

0.68 

β 

0.39 

-0.21 

-0.42 

0.06 

0.25 

0.08 

-0.61 

0.23 

-0.06 

-0.16 

p 

0.22 

0.42 

0.44 

0.89 

0.56 

0.81 

0.08 

0.43 

0.84 

0.48 

β 

0.36 

-0.19 

-0.48 

0.31 

0.50 

-0.07 

-0.66 

0.31 

-0.02 

-0.13 

p 

0.27 

0.49 

0.38 

0.50 

0.25 

0.85 

0.06 

0.31 

0.94 

0.58 

For the Shannon diversity index, there were three significant relationships with an 

independent variable of interest at p ≤ 0.10 (Table 4). There was a significant, positive 

association between diversity of the mouth and neighborhood greenness at 600m (β = 0.02, p = 

0.10). In other words, a 1% in neighborhood greenness would be expected to increase relative 

abundance as measured by the Shannon diversity index by 0.02 units. Additionally, for the 

microbiome of the eye and at 400m, there was a significant, negative association with 

neighborhood greenness (β = -0.03, p = 0.06) and a positive association with impervious surfaces 

(β = 0.02, p = 0.04). These results are not entirely unexpected, as contrary to the other body 

30 

 

areas, the eye is considered to be healthy when there is homeostasis of the ocular mucosa to 

protect against infection106.  

Table 4. Hierarchical linear regression model results for neighborhood greenness/impervious 
surfaces and Shannon diversity index, for each body area. Significant associations (p ≤ 0.10) in 
bolded text 
  

200m block 

800m 

  

400m 

600m 

Shannon 

Greenness 

Impervious 

Greenness 

Impervious 

Greenness 

Impervious 

Greenness 

Impervious 

Greenness 

Impervious 

 

h
t
u
o
M

 
s
r
a
E

 
s
e
y
E

 
e
s
o
N

 

m
u
t
c
e
R

 

β 

0.01 

0.00 

-0.01 

0.01 

-0.02 

0.01 

0.00 

-0.01 

0.00 

0.00 

p 

0.30 

0.76 

0.57 

0.55 

0.11 

0.20 

0.81 

0.39 

0.86 

0.69 

β 

0.02 

-0.01 

0.01 

-0.01 

-0.03 

0.02 

-0.01 

0.00 

0.00 

0.00 

p 

0.10 

0.42 

0.71 

0.38 

0.06 

0.04 

0.57 

0.74 

0.97 

0.62 

β 

0.02 

-0.01 

0.00 

-0.02 

-0.02 

0.02 

-0.01 

0.00 

0.00 

0.00 

p 

0.10 

0.47 

0.96 

0.33 

0.17 

0.22 

0.41 

0.73 

0.72 

0.83 

β 

0.02 

-0.01 

0.00 

-0.01 

-0.01 

0.01 

-0.01 

0.00 

0.01 

0.00 

p 

0.11 

0.46 

0.87 

0.67 

0.40 

0.47 

0.27 

0.99 

0.60 

0.86 

For phylogenetic diversity, there were multiple significant associations for the mouth and 

microbial communities at p ≤ 0.10 (Table 5). Neighborhood impervious surface was negatively, 

significantly associated with the microbiome of the mouth at the 400m (β = -0.03, p = 0.03) and 

600m (β = -0.03, p = 0.08) buffers and was approaching significance at the 800m buffer (β = -

0.03, p = 0.11). This suggests that individuals living in neighborhoods with lower levels of 

impervious surfaces had higher phylogenetic diversity. A similar trend was observed for the 

microbiome of the nose and neighborhood greenness, whereby lower levels of greenness were 

associated with higher diversity at the 400m (β = -0.04, p = 0.06), 600m (β = -0.04, p = 0.07) and 

800m (β = -0.04, p = 0.07) buffers. 

Taken together, there were few significant associations between proportion of 

neighborhood greenness and impervious surface area and microbial biodiversity when 

31 

 

controlling for individual and neighborhood-level factors. No significant associations were 

detected at the 200m block neighborhood or for the ear or rectum microbiomes. However, 

significant associations were observed at the 600m buffer for neighborhood greenness and both 

observed species and the phylogenetic diversity of the nose, in which diversity decreased as 

greenness increased as was hypothesized. Also as expected, the Shannon diversity of the mouth 

increased with neighborhood greenness and phylogenetic diversity decreased with increased 

impervious surface at 600m. Only the biodiversity of the nose was significantly associated with 

neighborhood vegetation when assessed up to 800m from the location of residence: observed 

species and phylogenetic diversity decreased with increased neighborhood greenness, contrary to 

the expected direction. Overall, impervious surfaces were less likely to be associated with 

microbial biodiversity. 

Table 5. Hierarchical linear regression model results for neighborhood greenness/impervious 
surfaces and phylogenetic diversity, for each body area. Significant associations (p ≤ 0.10) in 
bolded text 

  

  

200m block 

400m 

600m 

800m 

PD Whole Tree 

β 

p 

0.62 

0.42 

0.01 

-0.01 

-0.02 

-0.68 

0.02 

0.02 

-0.01 

-0.02 

0.00 

0.00 

0.00 

0.28 

0.47 

0.41 

0.35 

0.74 

0.98 

0.70 

 

Greenness 

Impervious 

Greenness 

Impervious 

Greenness 

Impervious 

Greenness 

Impervious 

Greenness 

Impervious 

 

h
t
u
o
M

 
s
r
a
E

 
s
e
y
E

 
e
s
o
N

 

m
u
t
c
e
R

 

 

β 

0.03 

-0.03 

-0.01 

0.01 

0.02 

0.00 

-0.04 

0.01 

0.01 

0.00 

p 

0.11 

0.03 

0.76 

0.59 

0.45 

0.83 

0.06 

0.51 

0.53 

0.83 

β 

0.02 

-0.03 

-0.02 

0.01 

0.03 

-0.01 

-0.04 

0.00 

0.02 

0.00 

p 

0.19 

0.08 

0.46 

0.60 

0.23 

0.59 

0.07 

0.96 

0.31 

0.95 

β 

0.02 

-0.03 

-0.03 

0.03 

0.04 

-0.02 

-0.04 

0.00 

0.02 

0.00 

p 

0.24 

0.11 

0.39 

0.32 

0.11 

0.34 

0.07 

0.89 

0.25 

0.91 

32 

 

3.3.2 Correlations between neighborhood greenness/impervious surfaces and abundances 
of genera by body area 

There were multiple, significant correlations detected in these analyses (full results in the 

Supplementary File). Table 6 shows the three strongest, significant correlations between 

neighborhood greenness and individual genera at each body area for the 200m neighborhood 

block. As expected for all significant relationships neighborhood greenness and impervious 

surface area were correlated in opposing directions. No genera were significantly correlated with 

greenness/impervious surfaces for more than one body area. For the mouth microbiome, all three 

genera were positively correlated with neighborhood greenness (Actinomyces, Actinomycetaceae 

Other, and Selenomonas). For the nose microbiome, conversely, all three genera were negatively 

correlated with neighborhood greenness (Clostridiaceae Other, Coriobacteriaceae Other, 

Leuconostoc). The microbiomes of the ears, eyes and rectum have relationships with mixed 

directions between neighborhood greenness and impervious surfaces. 

Table 6. Three strongest, significant correlations between proportion of neighborhood 
greenness/impervious surface and individual genera at each body area for the 200m 
neighborhood block. Cell color indicates significance level; blue (p ≤ 0.10), green (p ≤ 0.05), 
yellow (p ≤ 0.01), nd indicates None Detected 

Table 7 shows the three strongest, significant correlations between neighborhood 

greenness and individual genera at each body area for the 400m neighborhood buffer. Compared 

 

33 

GreennessImperviousGreennessImperviousGreennessImperviousGreennessImperviousGreennessImperviousActinomyces0.198-0.125-0.011-0.0310.0100.0410.021-0.1070.0120.04Actinomycetaceae Other0.219-0.1520.021-0.0980.102-0.1290.061-0.106ndndSelenomonas0.182-0.1970.108-0.1400.0320.0010.052-0.0320.035-0.07Helicobacter-0.022-0.0740.229-0.2020.0010.0180.108-0.113ndndGemellales Other-0.0640.035-0.2030.158-0.1040.1150.107-0.119-0.1050.05Ruminococcus0.082-0.133-0.1810.0960.022-0.117-0.1330.140-0.019-0.03Neisseria0.061-0.1180.107-0.1130.181-0.1190.038-0.064-0.007-0.01Desulfovibrio0.081-0.0630.005-0.0050.189-0.1720.037-0.077-0.0240.00Tissierellaceae 1-680.0430.101-0.1070.101-0.1780.105-0.1670.1280.0040.05Clostridiaceae Other-0.019-0.034-0.1760.180-0.0690.082-0.2530.210-0.034-0.02Coriobacteriaceae Otherndnd-0.090-0.0130.034-0.041-0.2500.1880.050-0.07Leuconostocndnd0.014-0.0070.061-0.061-0.2220.231ndndEggerthellandnd0.0250.016-0.047-0.051-0.0380.0080.244-0.15Oxalobacteraceae Other0.011-0.028-0.0430.070-0.034-0.0550.008-0.149-0.1870.12TG50.113-0.1860.160-0.1320.088-0.098-0.0090.0540.183-0.18EyesRectumNoseMouthEarsMouthRectum NoseEyesEars 

to the 200m block, the most strongly correlated genera remained somewhat consistent for the 

mouth and nose microbiomes (Selenomonas, Actinomyces, Coriobacteriaceae Other, 

Leuconostoc) while indicating no repeat genera for the microbiome of the ears and eyes. All 

three genera of the mouth (Selenomonas, Actinomyces, Aggregatibacter) and rectum 

(Eggerthella, Clostridium, Escherichia) were positively correlated with neighborhood greenness, 

while the genera of the microbiome of the nose remained negatively correlated with 

neighborhood greenness (Clostridiaceae Other, Bifidobacterium, Leuconostoc). Similar to the 

200m neighborhood block, no strongly correlated genera were significant across multiple body 

areas. Many genera significantly associated with an independent variable of interest at one body 

area were not detected at other body areas, particularly in the mouth. 

Table 7. Three strongest, significant correlations between proportion of neighborhood 
greenness/impervious surface and individual genera at each body area for the 400m 
neighborhood. Cell color indicates significance level; blue (p ≤ 0.10), green (p ≤ 0.05), yellow (p 
≤ 0.01) and red (p < 0.001), nd indicates None Detected 

Table 8 shows the three strongest, significant correlations between neighborhood 

greenness and individual genera at each body area for the 600m neighborhood buffer. Each body 

area had at least one genus in common with the 400m neighborhood. The genera of the mouth 

(Actinomyces, Selenomonas, Catonella) remain positively correlated with neighborhood 

 

34 

GreennessImperviousGreennessImperviousGreennessImperviousGreennessImperviousGreennessImperviousSelenomonas0.231-0.2240.083-0.0940.063-0.0490.082-0.0850.007-0.07Actinomyces0.208-0.1800.001-0.0280.0230.011-0.030-0.0380.086-0.05Aggregatibacter0.217-0.2260.091-0.0480.109-0.0470.029-0.054-0.0340.04Shewanellandnd-0.2550.094-0.0220.0510.0520.028ndndStreptococcaceae Other0.037-0.0490.186-0.1780.0110.0210.0110.061-0.0870.09Leptotrichiaceae Other-0.0260.015-0.1670.162-0.0010.017-0.0470.085-0.0520.10Akkermansia0.113-0.116-0.007-0.0260.238-0.1750.113-0.0630.148-0.11Methylobacteriumndnd-0.0760.052-0.2390.258-0.015-0.011ndndMogibacteriaceae Other0.035-0.047-0.005-0.0310.207-0.108-0.040-0.025-0.0090.05Clostridiaceae Other-0.0550.010-0.0750.075-0.0770.170-0.2370.172-0.053-0.01Bifidobacterium0.1000.0050.024-0.009-0.1040.085-0.2200.1050.0260.01Leuconostocndnd-0.1390.145-0.0580.050-0.1960.212ndndEggerthellandnd0.044-0.022-0.037-0.078-0.0340.0230.216-0.10Clostridium-0.0130.0160.100-0.0900.037-0.043-0.0400.0930.208-0.12Escherichiandnd0.137-0.1110.0130.034-0.0480.1150.198-0.23Rectum EarsNoseEyesEyesRectumNoseMouthEarsMouth 

greenness. However, mixed directions of correlations are observed in the genera of the nose 

(SR1 Other positively correlated with neighborhood greenness, while Bifidobacterium and 

Coriobacteriaceae Other were negatively correlated). At this neighborhood size, the most 

strongly correlated genera of the ears (Shewanella, Enterobacter, Leptotrichiaceae Other) were 

negatively correlated with neighborhood greenness. None of the genera highly correlated with 

the microbiome of the rectum (Eggerthella, Escherichia, Actinobaculum) were detected in the 

mouth. 

Table 8. Three strongest, significant correlations between proportion of neighborhood 
greenness/impervious surface and individual genera at each body area for the 600m 
neighborhood. Cell color indicates significance level; blue (p ≤ 0.10), green (p ≤ 0.05), yellow (p 
≤ 0.01), nd indicates None Detected 

 

Table 9 shows the three strongest, significant correlations between neighborhood 

greenness and individual genera at each body area for the 800m neighborhood buffer. At this 

neighborhood size, the genera of the mouth (Actinomyces, Selenomonas, Catonella), eyes 

(Akkermansia, Mogibacteriaceae Other, Bacillaceae Other) and nose (SR1 Other, 

Coriobacteriaceae Other, Bifidobacterium) most highly correlated with neighborhood greenness 

remained consistent with those of the 600m buffer. Overall, in comparison to the 600m 

neighborhood buffer, directionality remained largely consistent across all body areas.  

35 

GreennessImperviousGreennessImperviousGreennessImperviousGreennessImperviousGreennessImperviousActinomyces0.225-0.1700.004-0.0460.0310.001-0.006-0.0760.095-0.04Selenomonas0.201-0.1650.075-0.0660.074-0.0680.078-0.094-0.0300.01Catonella0.209-0.0680.066-0.0510.128-0.1080.052-0.0310.002-0.11Shewanellandnd-0.2210.081-0.0390.0850.0520.022ndndEnterobacterndnd-0.193-0.0060.057-0.0140.028-0.057ndndLeptotrichiaceae Other-0.0250.019-0.1900.194-0.0150.017-0.0600.077-0.0250.06Akkermansia0.099-0.099-0.008-0.0340.246-0.2030.092-0.0790.182-0.14Mogibacteriaceae Other0.067-0.097-0.021-0.0110.213-0.145-0.031-0.028-0.0320.02Methylobacteriumndnd-0.085-0.024-0.2080.128-0.013-0.010ndndBifidobacterium0.111-0.0060.0090.069-0.1210.084-0.2240.127-0.0220.04SR1 Other0.118-0.1590.0070.0020.144-0.0710.211-0.1680.022-0.12Coriobacteriaceae Otherndnd-0.057-0.034-0.0090.011-0.2100.088-0.0300.03Eggerthellandnd0.030-0.006-0.030-0.128-0.029-0.0520.251-0.08Escherichiandnd0.143-0.1190.0140.017-0.0960.1420.202-0.19Actinobaculumndnd-0.1160.1280.111-0.049ndnd0.205-0.17NoseRectum EyesMouthEars 

Table 9. Three strongest, significant correlations between proportion of neighborhood 
greenness/impervious surface and individual genera at each body area for the 800m 
neighborhood. Cell color indicates significance level; blue (p ≤ 0.10), green (p ≤ 0.05), yellow (p 
≤ 0.01), nd indicates None Detected 

Taken together, only three genera were consistently among the three most strongly 

correlated at all neighborhood sizes. These were Actinomyces and Selenomonas in the mouth and 

Eggerthella in the rectum. Of note, despite inconsistency in the genera most strongly correlated 

across neighborhood size, none of the genera were significantly correlated with neighborhood 

greenness and/or impervious surface at multiple body areas. Generally, neighborhood greenness 

 

and impervious surface area were correlated in opposing direction, but not always. This 

unexpected directionality was often when both independent variables of interest were very 

weakly correlated with the dependent variable (or, a tau-b value of close to 0). 

3.4 Discussion  

The mutualism between the microbiome of the natural environment and the complexity 

of the human microbiome across areas of the body is well-understood, however, the pathways in 

which human microbial niches are affected by the surrounding environment and the implications 

for health are still being investigated84,107,108. Thus, this research examined the microbial 

biodiversity and abundance of genera at five areas of the body in relationship to the proportion of 

neighborhood greenness or impervious surface area. Increasing our understanding of the nature 

36 

GreennessImperviousGreennessImperviousGreennessImperviousGreennessImperviousGreennessImperviousActinomyces0.229-0.1770.010-0.0420.0380.0170.018-0.0700.110-0.04Catonella0.215-0.0950.033-0.0240.133-0.1330.046-0.0320.013-0.12Selenomonas0.192-0.1580.066-0.0530.064-0.0750.080-0.091-0.0700.05Leptotrichiaceae Other-0.017-0.014-0.1970.189-0.0160.002-0.0690.052-0.0070.02Shewanellandnd-0.1920.076-0.0510.0930.0250.046ndndEnterobacterndnd-0.1810.0130.041-0.0110.017-0.045ndndAkkermansia0.096-0.090-0.009-0.0400.242-0.2230.095-0.0660.191-0.17Mogibacteriaceae Other0.080-0.094-0.0480.0170.213-0.123-0.019-0.022-0.0560.08Bacillaceae Other-0.0260.0460.038-0.052-0.2020.147-0.0350.086-0.094-0.01SR1 Other0.141-0.1720.023-0.0090.113-0.0250.235-0.1960.068-0.15Coriobacteriaceae Otherndnd-0.046-0.040-0.0210.002-0.2180.095-0.0250.04Bifidobacterium0.130-0.0430.0000.075-0.0850.077-0.1990.127-0.0540.09Eggerthellandnd0.034-0.0400.018-0.1260.015-0.0690.249-0.08Actinobaculumndnd-0.1160.1250.102-0.057ndnd0.217-0.20Akkermansia0.096-0.090-0.009-0.0400.242-0.2230.095-0.0660.191-0.17EyesRectumNoseMouthEarsEyesMouthEarsNoseRectum  

of our relationship to the microbes that live in, on, and around us can help shape neighborhood-

level interventions that promote healthy microbiomes for all residents. 

When controlling for individual and neighborhood-level factors, the significant 

associations detected between microbial biodiversity and neighborhood greenness or impervious 

surfaces were limited. These results are contradictory to other studies of the human microbiome 

and the built environment, which generally find that the microbes of the immediate environment 

are shared79,82,108. Plants also rely on their microbiome to function, and thus have developed 

multiple methods of dispersal (e.g., pollen, contact with wind and water) to share these vital 

microbes109. However, this unexpected finding may suggest that exposure to vegetation, where 

microbes would most often be passively dispersed has limited impacts on the human 

microbiome. Some existing research supports this as indoor houseplants, specifically the spider 

plant Chlorophytum comosum, have been shown to contribute microbes to the surfaces of the 

surrounding built environment (but not the air)110. These results may also indicate that the 

microbial diversity present in the soil and vegetation of highly green versus less green 

neighborhoods does not significantly differ. This is similar to work by Barberán, et al. 111, who 

found the indicator microbes of rural and urban dust to be nearly identical. 

Only the phylogenetic diversity of the mouth microbiome was significantly associated as 

predicted with both proportion of neighborhood greenness and impervious surface area, whereby 

an increase in neighborhood greenness and decrease in impervious surface area would result in 

an expected increase in phylogenetic diversity. In contrast, the remaining observed significant 

associations were in unexpected directions. For example, lower diversity of the nose was 

associated with higher greenness (for both observed species and phylogenetic diversity). The 

Shannon diversity of the eyes was also significantly associated with both neighborhood 

37 

 

greenness and impervious surface area in opposite directions than predicted. One possible 

explanation is that exposure to vegetation may instead influence other microbial mechanisms that 

affect diversity, such as the provision of an energy source that causes only certain microbes to 

flourish, rather than increased diversity. The microbiome of the mouth, which is less exposed to 

the external environment than the nose (the predominant airway) and eyes, may be more subject 

to other known influences on human microbial biodiversity associated with highly urban areas, 

such as a highly processed diet94.  

Additionally, despite differing levels of significance, effect sizes for Shannon and 

phylogenetic diversity remained almost completely stable across the four neighborhood buffers, 

as indicated by the beta values.  For example, the phylogenetic diversity of the nose was 

expected to decrease by 0.04 units for each percentage increase in neighborhood greenness at the 

400m, 600m and 800m buffers (p = 0.06, 0.06, and 0.07, respectively). This suggests that 

neighborhood size does not play a major role in determining the effect that proportion of 

greenness or impervious surface area has on the relative abundance and evenness of the human 

microbiome. Similarly, there was no consistent trend of the effect size of neighborhood 

greenness/impervious surface area on observed species (richness), except for the effect of 

neighborhood greenness on the nose microbiome. For samples collected at the nose, increasing 

neighborhood buffer size consistently resulted in fewer expected OTUs detected, though notably 

less so for each buffer larger than 200m. These results may also be an effect of positive spatial 

autocorrelation, wherein surface cover is more likely to be similar across smaller spatial scales 

(or in this case, nested buffers). 

When examining the relationships between neighborhood greenness/impervious surfaces 

and abundances of genera, the strongest, significant correlations varied for each body area. This 

38 

 

highlights the uniqueness of the microbial habitats of the body and supports how general 

conceptions of a ‘healthy’ microbiome may not be consistent across the body90,112,113. Notably, 

many genera correlated with the rectum microbiome were not detected in the mouth or the nose, 

despite the potential for the mouth to act as a pathway to the gut microbiome. However, this may 

be a reflection of the complex diversity of the oral microbiome which varies by the surface being 

sampled within the mouth95. Additionally, all but three of the most correlated genera at each 

body area were different when examining results across neighborhood buffer size. Of the three 

genera most strongly, significantly correlated with neighborhood greenness at all neighborhood 

sizes, Actinomyces is an anaerobic commensal of the oral microbiome but may cause chronic 

infection if it is able to penetrate the mucosal membranes of the mouth, causing lesions114. 

Similarly, Selenomonas has been found to potentially play a role in the onset of periodontitis in 

the oral cavity115.  At all neighborhood sizes, the abundance of the genus Eggerthella detected in 

the rectum was most strongly correlated with proportion of neighborhood greenness in a positive 

direction. This anaerobic, non-sporulating, Gram-positive bacteria characteristic of the human 

colon can cause bacteremia (a disease in which bacteria enter the blood) in immunocompromised 

persons116. Overall, the results support that passive exposure to vegetation may not uniformly 

impact the diversity of the human microbiome but may do so at specific anatomic microbial 

niches in addition to the promotion or inhibition of specific genera in ways unique to the area of 

the body being sampled. 

This research has several limitations. First, the life history of the cases is unknown. Thus, 

it was not possible to control for antibiotic use, hygiene, diet, and other factors that are known to 

affect the human microbiome. It is also possible that some individuals more directly interacted 

with the physical environment more often than others or that other locations, such as place of 

39 

 

work, is a better measure of exposure to vegetation.  The use of postmortem cases may also have 

affected the results, for example, a person with a chronic illness may have been limited in their 

exposure to their external environment in comparison to a healthy person killed in an accident. 

Conversely, a chronically ill person may have spent more time at their place of residence and 

less time at a workplace or elsewhere. It is also important to note that these data represent only a 

small number of the death investigations conducted in Wayne County each year and these 

findings may not be representative of the larger population. Additionally, despite the use of high-

resolution imagery, NDVI only captures the density of healthy vegetation and not plant species 

diversity per se. It is possible that dense, single-species vegetation has less influence on human 

microbial diversity than more varied plant species, or that certain plant species disperse more 

microbes than others. Future research would benefit from differentiating between types (e.g., 

broadleaf, coniferous, observed plant species) of vegetation.  

This study also has several strengths. It is the first study to compare an objective, high-

resolution measure of neighborhood vegetation with the diversity of the human postmortem 

microbiome across multiple body areas. Future research should use these findings as a guide to 

investigate more specific relationships between actual exposure to vegetation and the 

composition of the human microbiome. For example, studies can evaluate the potential effect of 

direct exposure to vegetation (such as gardening) on the profile of external (skin, nose) and 

internal (gut) microbiomes, at varying times since the exposure. This could aid researchers in 

identifying the effect of exposure to vegetation on the microbiome. Such research would further 

our understanding of neighborhood effects on the microbiome, level of exposure required to 

effectively promote a healthy microbiome, and potentially health implications of microbial 

diversity. Future research should also examine living individuals from selected high- and low-

40 

 

vegetation neighborhoods to better control for life histories and more directly examine the effect 

of neighborhood environment on the microbiome within and across neighborhood and lifestyle 

types.  

Understanding the potential impact of the neighborhood natural environment on human 

health at a microbial level is important for neighborhood planning, management, and ecological 

restoration. This is especially important in post-industrial cities where residents are likely to live 

in neighborhoods with high levels of vacant lots, with the potential to improve the ecological 

diversity of these urban spaces and possibly improve health. Interventions that promote a diverse 

microbiome, such as increasing neighborhood plant diversity through volunteer gardening 

efforts, have the potential to be both cost-effective and less labor-intensive than other methods of 

impacting health outcomes.

41 

 

Chapter 4 

Conclusion 

In sum, multiple approaches were used to further explore the many relationships between 

the components of a neighborhood and human health. From serious urban blight to well-

maintained neighborhoods, and from megacities to rural areas, the associations observed in 

geographic and public health research manifest differently across multiple contexts. Thus, 

enhancing our knowledge of these phenomena can help shape important public health policy to 

be more effective both in cost and impact. Applying a geographic perspective and corresponding 

spatial techniques to these issues can help trace the transmission of important health issues 

between individuals and their neighborhoods. This thesis contributed to the knowledge base by 

first systematically reviewing the use of Google Street View in health research and second, 

assessing the relationship between individuals and their neighborhood proportion of greenery 

and impervious surface by using a potential proxy for health, the human postmortem 

microbiome.  Both chapters are geographic in nature while traversing through multiple scales; 

in particular, they both stress the importance of place when comparing individual health 

outcomes. The geographic extent of the use of Google Street View varied from assessing 

neighborhoods at a street block level to comparing data from cities in different continents, 

without sacrificing the street-level resolution of the data being collected. Similarly, the second 

study examined the influence of neighborhood groundcover on the postmortem microbiome of 

an individual at four different neighborhood extents while also comparing results between the 

individuals included in the study. While all samples in the second chapter were taken from 

residents of Wayne County, it was quantifying the comprehensive, high-resolution data provided 

by aerial imagery that allowed for the precise extrapolation of the diverse living conditions that 

coexist within mere blocks in this urban area.  

42 

 

To assess the use of Google Street View in health research, 54 qualifying studies were 

reviewed. At the time of review, GSV had been available for public use for one decade—a 

relatively short time in comparison to other common sources of spatial data such as the Landsat 

satellites, which have been in operation since the 1970s. Although the consistency of available 

data is still developing, this free source of big data provides a highly detailed look into the 

condition of an increasing number of neighborhoods across the globe. This was confirmed by the 

studies included in the review, which used GSV to assess neighborhood features associated with 

urban decay, physical activity, naturalness of vegetation, policy compliance, and disaster 

preparedness, in addition to other similar avenues. This in-depth look at neighborhood exposures 

is important for discovering the areas that would benefit from new health initiatives as well as 

assessing the success (or failure) of existing health policies. 

To dive deeper into neighborhood conditions and health, the relationship between the 

proportions of neighborhood greenery, impervious surface cover and the composition of the 

human postmortem microbiome was assessed for ninety-eight residents of Wayne County, 

Michigan. The human microbiome has been linked to several physical and mental health 

outcomes; thus, studying the microbial profiles of persons living in varying amounts of 

vegetation could elucidate important links between the biodiversity of the surrounding 

environment and its human inhabitants. The current study was largely exploratory; it examined 

the relationships between four extents of neighborhood surface cover and the biodiversity and 

abundance of detected genera of the human postmortem microbiome at five sites on the body. 

Overall, limited evidence was found supporting relationships in opposing directions between 

neighborhood surface cover and the biodiversity of the human microbiome, and this evidence 

was limited to the more exposed anatomic sites (eyes, nose, and mouth microbiomes). Stronger 

43 

 

evidence was found to support the relationship between neighborhood surface cover and the 

abundance of specific genera detected in anatomic niches.  

As different microbes can have pathogenic, symbiotic, or no effect on human health, the 

exploration of the presence of these microbes in the environment and on humans could lead to 

important discoveries in what makes a neighborhood healthy. For example, a 2010 study has 

shown the pathogenic colonization of Staphylococcus aureus in the microbiome of the nose to 

exist in competition with the colonization of Actinobacteria groups117. Thus, a potential avenue 

for preventing S. aureus colonization would be to promote the existence of key Actinobacteria in 

the environment, rather than treating the infection with antibiotics after it occurs. This proactive 

approach has the potential to not only reduce medical costs but prevent the further development 

of antimicrobial resistance, which has become a considerable threat to human mortality and a 

quickly growing global health concern. 

 

Despite the major differences in the approaches taken in each chapter to study 

neighborhoods and health, they are not mutually exclusive. Both pieces demonstrate the use of 

new technologies to explore neighborhood features in detail, from image data of neighborhoods 

to the microbes present in those environments. In the future, the information gleaned from more 

resource-intensive techniques (analyzing microbial samples) could be used to enhance less 

resource-intensive methods (auditing image data) as the increasingly specific features associated 

with the transmission of health-related microbes to humans are discovered. For example, if a 

certain species or type of tree is shown to be associated with higher abundances of a salutogenic 

microbe on humans, then neighborhood audits can be improved to select for this specific tree. 

Similarly, GSV could be used to assess for the presence of common features or conditions in the 

residential neighborhoods of individuals with similar features to their microbial profile. This 

44 

 

approach may lead to the discovery of new explanatory components to neighborhood health that 

may not have been considered otherwise.  

In sum, while both of the approaches included in this research consist of a highly-focused 

analysis of health and the environment, the conclusions drawn can apply to a metropolitan—and 

even global—extent, providing the potential to shape and improve health policies and 

interventions to best fit each community’s needs. 

 

 

 

45 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

REFERENCES 

 

 

46 

 

1 

 
2 
 
3 
 
4 
 
5 

 
6 

 
7 

 
8 

 
9 

 
10 

REFERENCES 

 

Charreire, H. et al. Using remote sensing to define environmental characteristics related 
to physical activity and dietary behaviours: A systematic review (the SPOTLIGHT 
project). Health & Place 25, 1-9, (2014). 

Google. Google Street View, <https://www.google.com/streetview/> (2018). 

Miller, C. C. & O'Brien, K. J. in New York Times    (2013). 

Shet, V.  Vol. 2018   (Google Official Blog, 2014). 

Fleischhacker, S. E., Evenson, K. R., Sharkey, J., Pitts, S. B. J. & Rodriguez, D. A. 
Validity of Secondary Retail Food Outlet Data: A Systematic Review. American journal 
of preventive medicine 45, 462-473, (2013). 

Li, X. et al. Assessing street-level urban greenery using Google Street View and a 
modified green view index. Urban Forestry & Urban Greening 14, 675-685 (2015). 

Barton, J. & Pretty, J. What is the Best Dose of Nature and Green Exercise for Improving 
Mental Health? A Multi-Study Analysis. Environmental Science & Technology 44, 3947-
3955, (2010). 

Bader, M. D. et al. Development and deployment of the Computer Assisted 
Neighborhood Visual Assessment System (CANVAS) to measure health-related 
neighborhood conditions. Health Place 31, 163-172, (2015). 

Less, E. L. et al. Matching study areas using Google Street View: A new application for 
an emerging technology. Evaluation and program planning 53, 72-79, (2015). 

Pliakas, T. et al. Optimising measurement of health-related characteristics of the built 
environment: Comparing data collected by foot-based street audits, virtual street audits 
and routine secondary data sources. Health & Place 43, 75-84, (2017). 

 
11  Wilson, N., Thomson, G. & Edwards, R. The potential of Google Street View for 

studying smokefree signage. Aust N Z J Public Health 39, 295-296, (2015). 

 
12 

 
13 

Charreire, H. et al. Measuring the food environment using geographical information 
systems: a methodological review. Public Health Nutrition 13, 1773-1785, (2010). 

Krenn, P. J., Titze, S., Oja, P., Jones, A. & Ogilvie, D. Use of Global Positioning Systems 
to Study Physical Activity and the Environment: A Systematic Review. American 
Journal of Preventive Medicine 41, 508-515, (2011). 

47 

 

14 

 
15 

Schaefer-McDaniel, N., O'Brien Caughy, M., O'Campo, P. & Gearey, W. Examining 
methodological details of neighbourhood observations and the relationship to health: A 
literature review. Social Science & Medicine 70, 277-292, (2010). 

Schootman, M. et al. Emerging technologies to measure neighborhood conditions in 
public health: implications for interventions and next steps. International Journal of 
Health Geographics 15, 20, (2016). 

 
16  Wilson, N., Pearson, A. L., Thomson, G. & Edwards, R. Actual and potential use of 

Google Street View for studying tobacco issues: a brief review. Tobacco Control, (2017). 

 
17  Moher, D., Liberati, A., Tetzlaff, J., Altman, D. G. & The, P. G. Preferred Reporting 

Items for Systematic Reviews and Meta-Analyses: The PRISMA Statement. PLOS 
Medicine 6, e1000097, (2009). 

 
18 

von Fischer, J. C. et al. Rapid, Vehicle-Based Identification of Location and Magnitude 
of Urban Natural Gas Pipeline Leaks. Environmental Science & Technology 51, 4091-
4099, (2017). 

 
19  Wolthers, L. Monitoring the Margins: Street Views of Sex Workers. Photography and 

Culture 9, 239-254, (2016). 

Covidence. <https://www.covidence.org/> (2017). 

 
20 
 
21  Mitsuhara, H. et al. in Advances in Human Factors, Business Management, Training and 
Education Vol. 498 Advances in Intelligent Systems and Computing (eds J. I. Kantola, T. 
Barath, S. Nazir, & T. Andre)  655-666 (2017). 

 
22  Mertens, L. et al. Built environmental correlates of cycling for transport across Europe. 

Health Place 44, 35-42, (2017). 

 
23 

 
24 

 
25 

 
26 

 

Bethlehem, J. R. et al. The SPOTLIGHT virtual audit tool: a valid and reliable tool to 
assess obesogenic characteristics of the built environment. International Journal of 
Health Geographics 13, 52, (2014). 

Compernolle, S. et al. Physical Environmental Correlates of Domain-Specific Sedentary 
Behaviours across Five European Regions (the SPOTLIGHT Project). PLOS ONE 11, 
e0164812, (2016). 

Roda, C. et al. Lifestyle correlates of overweight in adults: a hierarchical approach (the 
SPOTLIGHT project). International Journal of Behavioral Nutrition and Physical 
Activity 13, 114, (2016). 

Feuillet, T. et al. Neighbourhood typology based on virtual audit of environmental 
obesogenic characteristics. Obesity Reviews 17, 19-30, (2016). 

48 

 

27  Michael, D. M. B., Stephen, J. M., Blake, B. & Andrew, G. R. The Promise, 

Practicalities, and Perils of Virtually Auditing Neighborhoods Using Google Street View. 
The ANNALS of the American Academy of Political and Social Science 669, 18-40, 
(2016). 

 
28  Mooney, S. J. et al. Validity of an ecometric neighborhood physical disorder measure 

constructed by virtual street audit. Am J Epidemiol 180, 626-635, (2014). 

 
29 

 
30 

 
31 

 
32 

 
33 

 
34 

 
35 

Quinn, J. W. et al. Neighborhood Physical Disorder in New York City. J Maps 12, 53-60, 
(2016). 

Clews, C. et al. Alcohol in urban streetscapes: a comparison of the use of Google Street 
View and on-street observation. BMC Public Health 16, 442, (2016). 

Yin, L. & Wang, Z. Measuring visual enclosure for street walkability: Using machine 
learning algorithms and Google Street View imagery. Applied Geography 76, 147-153, 
(2016). 

Naik, N., Philipoom, J., Raskar, R., Hidalgo, C., 2014. IEEE, 2014. Streetscore - 
Predicting the Perceived Safety of One Million Streetscapes. In: Proceedings of IEEE 
conference on computer vision and pattern recognition workshops, pp. 793–799. 

Ahmetovic, D., Manduchi, R., Coughlan, J. M. & Mascetti, S. Mind Your Crossings: 
Mining GIS Imagery for Crosswalk Localization. Acm Transactions on Accessible 
Computing 9, (2017). 

Johnson, N. S. & Gabler, H. C. Injury Outcome in Crashes with Guardrail End 
Terminals. Traffic Inj Prev 16 Suppl 2, S103-108, (2015). 

Hanson, C. S., Noland, R. B. & Brown, C. The severity of pedestrian crashes: an analysis 
using Google Street View imagery. Journal of Transport Geography 33, 42-53, (2013). 

 
36  Marco, M., Gracia, E., Martin-Fernandez, M. & Lopez-Quilez, A. Validation of a Google 
Street View-Based Neighborhood Disorder Observational Scale. J Urban Health 94, 190-
198, (2017). 

 
37 

 
38 

 
39 

 

Odgers, C. L., Caspi, A., Bates, C. J., Sampson, R. J. & Moffitt, T. E. Systematic social 
observation of children’s neighborhoods using Google Street View: a reliable and cost‐
effective method. Journal of Child Psychology and Psychiatry 53, 1009-1017 (2012). 

Rundle, A., Bader, M. D. M., Richards, C. A., Neckerman, K. M. & Teitler, J. O. Using 
google street view to audit neighborhood environments. Am J Prev Med 40, (2011). 

Kepper, M. M. et al. A Reliable, Feasible Method to Observe Neighborhoods at High 
Spatial Resolution. American Journal of Preventive Medicine 52, S20-S30, (2017). 

49 

 

40 

 
41 

Evans-Cowley, J. S. & Akar, G. StreetSeen Visual Survey Tool for Determining Factors 
That Make a Street Attractive for Bicycling. Transportation Research Record, 19-27, 
(2014). 

Ahmetovic, D., Manduchi, R., Coughlan, J. M. & Mascetti, S. Zebra Crossing Spotter: 
Automatic Population of Spatial Databases for Increased Safety of Blind Travelers. 
Assets 2015, 251-258 (2015). 

 
42  Wilson, N. & Thomson, G. Suboptimal smokefree signage at some hospitals: Field 

observations and the use of Google Street View. The New Zealand medical journal 128, 
56-59 (2015). 

 
43 

Burgoine, T. & Harrison, F. Comparing the accuracy of two secondary food environment 
data sources in the UK across socio-economic and urban/rural divides. International 
Journal of Health Geographics 12, 2-2, (2013). 

 
44  Weston, P. Google Street View images are about to get a lot clearer: Firm is rolling out 

HD cameras to help AI map the real world in incredible detail. (2017). 
<http://www.dailymail.co.uk/sciencetech/article-4857542/Google-Street-View-using-
HD-cameras-AI.html>. 

 
45 

 
46 

 
47 

 
48 

 
49 

 
50 

 

BBC. Google's Street View cameras get quality boost. (2017). 
<http://www.bbc.com/news/technology-41082920>. 

Curtis, A., Duval-Diop, D. & Novak, J. Identifying Spatial Patterns of Recovery and 
Abandonment in the Post-Katrina Holy Cross Neighborhood of New Orleans. 
Cartography and Geographic Information Science 37, 45-56, (2010). 

Badland, H. M., Opit, S., Witten, K., Kearns, R. A. & Mavoa, S. Can Virtual Streetscape 
Audits Reliably Replace Physical Streetscape Audits? Journal of Urban Health 87, 1007-
1016, (2010). 

Ben-Joseph, E., Lee, J. S., Cromley, E. K., Laden, F. & Troped, P. J. Virtual and actual: 
relative accuracy of on-site and web-based instruments in auditing the environment for 
physical activity. Health Place 19, 138-150, (2013). 

Brookfield, K. & Tilley, S. Using Virtual Street Audits to Understand the Walkability of 
Older Adults’ Route Choices by Gender and Age. International Journal of 
Environmental Research and Public Health 13, 1061 (2016). 

Chudyk, A. M., Winters, M., Gorman, E., McKay, H. A. & Ashe, M. C. Agreement 
Between Virtual and In-the-Field Environment Audits of Assisted Living Sites. Journal 
of Aging and Physical Activity 22, 414-420, (2014). 

50 

 

51 

 
52 

 
53 

 
54 

 
55 

Griew, P. et al. Developing and testing a street audit tool using Google Street View to 
measure environmental supportiveness for physical activity. International Journal of 
Behavioral Nutrition and Physical Activity 10, 103, (2013). 

Gullón, P. et al. Assessing Walking and Cycling Environments in the Streets of Madrid: 
Comparing On-Field and Virtual Audits. Journal of Urban Health 92, 923-939, (2015). 

Kelly, C. et al. The Built Environment Predicts Observed Physical Activity. Frontiers in 
Public Health 2, (2014). 

Kelly, C. M., Wilson, J. S., Baker, E. A., Miller, D. K. & Schootman, M. Using Google 
Street View to Audit the Built Environment: Inter-rater Reliability Results. Annals of 
Behavioral Medicine 45, S108-S112, (2013). 

Kepper, M. et al. Parental Perceptions of the Social Environment Are Inversely Related 
to Constraint of Adolescents’ Neighborhood Physical Activity. International Journal of 
Environmental Research and Public Health 13, 1266 (2016). 

 
56  Mygind, L. et al. Public open space desktop auditing tool—Establishing appropriateness 
for use in Australian regional and urban settings. Urban Forestry & Urban Greening 20, 
65-70, (2016). 

 
57 

 
58 

 
59 

Silva, V., Grande, A., Rech, C. & Peccin, M. Geoprocessing via Google Maps for 
assessing obesogenic built environments related to physical activity and chronic 
noncommunicable diseases: validity and reliability. Journal of healthcare engineering 6, 
41-54 (2015). 

Vanwolleghem, G. et al. Using an audit tool (MAPS Global) to assess the characteristics 
of the physical environment related to walking for transport in youth: reliability of 
Belgian data. International Journal of Health Geographics 15, 41, (2016). 

Vanwolleghem, G., Van Dyck, D., Ducheyne, F., De Bourdeaudhuij, I. & Cardon, G. 
Assessing the environmental characteristics of cycling routes to school: a study on the 
reliability and validity of a Google Street View-based audit. International Journal of 
Health Geographics 13, 19, (2014). 

 
60  Wilson, J. S. et al. Assessing the Built Environment Using Omnidirectional Imagery. 

American Journal of Preventive Medicine 42, 193-199, (2012). 

 
61 

 
62 

Clarke, P., Ailshire, J., Melendez, R., Bader, M. & Morenoff, J. Using Google Earth to 
conduct a neighborhood audit: Reliability of a virtual audit instrument. Health & Place 
16, 1224-1229, (2010). 

Lafontaine, S. J. V., Sawada, M. & Kristjansson, E. A direct observation method for 
auditing large urban centers using stratified sampling, mobile GIS technology and virtual 
environments. International Journal of Health Geographics 16, 6, (2017). 

51 

 

63 

Naik, N., Philipoom, J., Raskar, R. & Hidalgo, C. in Proceedings of the IEEE Conference 
on Computer Vision and Pattern Recognition Workshops.  779-785. 

 
64  Wu, Y.-T. et al. Assessing environmental features related to mental health: a reliability 

study of visual streetscape images. BMC Public Health 14, 1094, (2014). 

 
65 

 
66 

 
67 

 
68 

 
69 

 
70 

 
71 

Hyam, R. Automated Image Sampling and Classification Can Be Used to Explore 
Perceived Naturalness of Urban Spaces. PLoS One 12, e0169357, (2017). 

Li, X., Zhang, C., Li, W., Kuzovkina, Y. A. & Weiner, D. Who lives in greener 
neighborhoods? The distribution of street greenery and its association with residents’ 
socioeconomic conditions in Hartford, Connecticut, USA. Urban Forestry & Urban 
Greening 14, 751-759, (2015). 

Li, X., Zhang, C. & Li, W. Does the Visibility of Greenery Increase Perceived Safety in 
Urban Areas? Evidence from the Place Pulse 1.0 Dataset. ISPRS International Journal of 
Geo-Information 4, 1166 (2015). 

Li, X., Zhang, C., Li, W. & Kuzovkina, Y. A. Environmental inequities in terms of 
different types of urban greenery in Hartford, Connecticut. Urban Forestry & Urban 
Greening 18, 163-172, (2016). 

Hammond, J., Seekins, T. & Rennie, B. in Environmental Contexts and Disability Vol. 8 
Research in Social Science and Disability  Ch. 6, 123-139 (Emerald Group Publishing 
Limited, 2014). 

Kuo, M. How might contact with nature promote human health? Promising mechanisms 
and a possible central pathway. Frontiers in Psychology 6, (2015). 

Jiang, B., Li, D., Larsen, L. & Sullivan, W. C. A Dose-Response Curve Describing the 
Relationship Between Urban Tree Cover Density and Self-Reported Stress Recovery. 
Environment and Behavior 48, 607-629, (2016). 

 
72  Mayne, S. L., Auchincloss, A. H. & Michael, Y. L. Impact of policy and built 

environment changes on obesity‐related outcomes: a systematic review of naturally 
occurring experiments. Obesity Reviews 16, 362-375, (2015). 

 
73 

 
74 

 

Evans, G. W. The built environment and mental health. J Urban Health 80, 536-555, 
(2003). 

Galea, S., Ahern, J., Rudenstine, S., Wallace, Z. & Vlahov, D. Urban built environment 
and depression: a multilevel analysis. J Epidemiol Community Health 59, 822-827, 
(2005). 

52 

 

75 

 
76 

 
77 

 
78 

 
79 

 
80 

 
81 

 
82 

 
83 

 
84 

 
85 

 
86 

 
87 

 
88 

 

Gascon, M. et al. Mental Health Benefits of Long-Term Exposure to Residential Green 
and Blue Spaces: A Systematic Review. International Journal of Environmental 
Research and Public Health 12, 4354 (2015). 

Bowler, D. E., Buyung-Ali, L. M., Knight, T. M. & Pullin, A. S. A systematic review of 
evidence for the added benefits to health of exposure to natural environments. BMC 
Public Health 10, 456, (2010). 

Liddicoat, C. et al. Landscape biodiversity correlates with respiratory health in Australia. 
Journal of Environmental Management 206, 113-122, (2018). 

Gupta, V. K., Paul, S. & Dutta, C. Geography, Ethnicity or Subsistence-Specific 
Variations in Human Microbiome Composition and Diversity. Frontiers in Microbiology 
8, (2017). 

Afshinnekoo, E. et al. Geospatial Resolution of Human and Bacterial Diversity with 
City-Scale Metagenomics. Cell Systems 1, 72-87, (2015). 

Bik, H. M. et al. Microbial Community Patterns Associated with Automated Teller 
Machine Keypads in New York City. mSphere 1, (2016). 

Leung, M. H. Y. & Lee, P. K. H. The roles of the outdoors and occupants in contributing 
to a potential pan-microbiome of the built environment: a review. Microbiome 4, 21, 
(2016). 

Lax, S. et al. Longitudinal analysis of microbial interaction between humans and the 
indoor environment. Science 345, 1048-1052, (2014). 

Hahs, A. K. et al. A global synthesis of plant extinction rates in urban areas. Ecology 
Letters 12, 1165-1173, (2009). 

Stamper, C. E. et al. in International Review of Neurobiology Vol. 131  (eds J. F. Cryan 
& G. Clarke)  289-323 (Academic Press, 2016). 

Hoisington, A. J., Brenner, L. A., Kinney, K. A., Postolache, T. T. & Lowry, C. A. The 
microbiome of the built environment and mental health. Microbiome 3, 60, (2015). 

Hanski, I. et al. Environmental biodiversity, human microbiota, and allergy are 
interrelated. Proceedings of the National Academy of Sciences 109, 8334-8339, (2012). 

Halfvarson, J. et al. Dynamics of the human gut microbiome in inflammatory bowel 
disease. Nature Microbiology 2, 17004, (2017). 

Foster, J. A. & McVey Neufeld, K.-A. Gut–brain axis: how the microbiome influences 
anxiety and depression. Trends in Neurosciences 36, 305-312, (2013). 

53 

 

89 

 
90 

Samsel, A. & Seneff, S. Glyphosate’s Suppression of Cytochrome P450 Enzymes and 
Amino Acid Biosynthesis by the Gut Microbiome: Pathways to Modern Diseases. 
Entropy 15, (2013). 

Ipci, K., Altıntoprak, N., Muluk, N. B., Senturk, M. & Cingi, C. The possible 
mechanisms of the human microbiome in allergic diseases. European Archives of Oto-
Rhino-Laryngology 274, 617-626, (2017). 

 
91  Maruvada, P., Leone, V., Kaplan, L. M. & Chang, E. B. The Human Microbiome and 

Obesity: Moving beyond Associations. Cell Host & Microbe 22, 589-599, (2017). 

 
92 

 
93 

 
94 

Aron-Wisnewsky, J. & Clément, K. The gut microbiome, diet, and links to 
cardiometabolic and chronic disorders. Nature Reviews Nephrology 12, 169, (2015). 

Albenberg, L. G. & Wu, G. D. Diet and the Intestinal Microbiome: Associations, 
Functions, and Implications for Health and Disease. Gastroenterology 146, 1564-1572, 
(2014). 

Kato, I. et al. Nutritional Correlates of Human Oral Microbiome. Journal of the 
American College of Nutrition 36, 88-98, (2017). 

 
95  Wade, W. G. The oral microbiome in health and disease. Pharmacological Research 69, 

137-143, (2013). 

 
96 

 
97 

 
98 
 
99 

Segata, N. et al. Composition of the adult digestive tract bacterial microbiome based on 
seven mouth surfaces, tonsils, throat and stool samples. Genome Biology 13, R42, (2012). 

County of, W. About Wayne County, Michigan. 
http://www.waynecounty.com/county/aboutwc.htm (2017). 

Bureau, C. 2010 Census Gazetteer Files.  (2014). 

Pechal, J. L., Schmidt, C. J., Jordan, H. R. & Benbow, M. E. A large-scale survey of the 
postmortem human microbiome, and its potential to provide insight into the living health 
condition. Scientific Reports (2018). 

100 

Pechal, J. L., Schmidt, C. J., Jordan, H. R. & Benbow, M. E. Frozen: thawing and its 
effect on the postmortem microbiome in two pediatric cases. Journal of forensic sciences 
62, 1399-1405 (2017). 

101  Census, U. American Community Survey 2010 5-year estimates for Michigan 

<www.census.gov> (2010). 

 
102  Moudon, A. V. et al. Operational Definitions of Walkable Neighborhood: Theoretical 
and Empirical Insights. Journal of Physical Activity and Health 3, S99-S117, (2006). 

 

54 

 

103  Villanueva, K. et al. The impact of neighborhood walkability on walking: Does it differ 
across adult life stage and does neighborhood buffer size matter? Health & Place 25, 43-
46, (2014). 

 
104  USGS. Earth Explorer, <https://earthexplorer.usgs.gov/> (2018). 
 
105 
 
106 

pgAdmin 4 (2016). 

St. Leger, A. J. et al. Microbiome-dependent modulation of mucosal immunity at the 
ocular surface. The Journal of Immunology 198, (2017). 

 
107  McDaniel, M. D., Tiemann, L. K. & Grandy, A. S. Does agricultural crop diversity 

enhance soil microbial biomass and organic matter dynamics? A meta‐analysis. 
Ecological Applications 24, 560-570, (2014). 

 
108  Dethlefsen, L., McFall-Ngai, M. & Relman, D. A. An ecological and evolutionary 

perspective on human–microbe mutualism and disease. Nature 449, 811, (2007). 

 
109  Berg, G., Grube, M., Schloter, M. & Smalla, K. Unraveling the plant microbiome: 

looking back and future perspectives. Frontiers in Microbiology 5, (2014). 

 
110  Mahnert, A., Moissl-Eichinger, C. & Berg, G. Microbiome interplay: plants alter 

microbial abundance and diversity within the built environment. Frontiers in 
Microbiology 6, (2015). 

 
111  Barberán, A. et al. Continental-scale distributions of dust-associated bacteria and fungi. 

Proceedings of the National Academy of Sciences 112, 5756-5761, (2015). 

 
112  Heintz-Buschart, A. et al. The nasal and gut microbiome in Parkinson's disease and 

idiopathic rapid eye movement sleep behavior disorder. Movement Disorders 33, 88-98, 
(2018). 

 
113  Kong, H. H. & Segre, J. A. The Molecular Revolution in Cutaneous Biology: 

Investigating the Skin Microbiome. Journal of Investigative Dermatology 137, e119-
e122, (2017). 

 
114 

Thukral, R., Shrivastav, K., Mathur, V., Barodiya, A. & Shrivastav, S. Actinomyces: a 
deceptive infection of oral cavity. Journal of the Korean Association of Oral and 
Maxillofacial Surgeons 43, 282-285, (2017). 

 
115  Gonçalves, L. F. H. et al. Levels of Selenomonas species in generalized aggressive 

periodontitis. Journal of periodontal research 47, 711-718, (2012). 

 
116 

Lau, S. K. et al. Anaerobic, non-sporulating, Gram-positive bacilli bacteraemia 
characterized by 16S rRNA gene sequencing. Journal of Medical Microbiology 53, 1247-
1253, (2004). 

55 

 

117 

Frank, D. N. et al. The Human Nasal Microbiota and Staphylococcus aureus Carriage. 
PLOS ONE 5, e10598, (2010). 

 

56