University of Southern California Dissertations and Theses

Spatiotemporal studies of out-of-hospital cardiac arrests and bystander cardiopulmonary resuscitation in Los Angeles

(USC Thesis Other)

Spatiotemporal studies of out-of-hospital cardiac arrests and bystander cardiopulmonary resuscitation in Los Angeles

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Spatiotemporal Studies of Out-of-Hospital Cardiac Arrests and Bystander Cardiopulmonary
Resuscitation in Los Angeles

by

Douglas Owen Fleming

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(Population, Health and Place)

December 2021

Copyright © 2021 Douglas Owen Fleming
ii

Dedication
To my family and friends for their support during my Ph.D. journey.

iii
Acknowledgements
I am grateful to my mentors and co-advisors, Dr. Ann Owens and Dr. Stephen Sanko, for their
support, guidance, and patience. I am also grateful to my other dissertation committee members,
Dr. Karen Kemp, Dr. Sarah Axeen, and Dr. Jennifer Ailshire, for their commitment and
thoughtful contributions to the work of this dissertation. To the faculty of the Spatial Sciences
Institute, I am thankful for the support and assistance they gave me when requested. I am grateful
for the data that these studies are based on provided to me by the Los Angeles City Fire
Department. I would like to thank my Geospatial Research, Analysis, and Services Program
leadership and teammates for their flexibility and patience while I completed some of this work.
Lastly, I would like to thank my second family at Sweetens Cove Golf Club for welcoming me
and providing me a place to relax and decompress during this process.

iv
Table of Contents
Acknowledgements ........................................................................................................................ iii
List of Tables ................................................................................................................................ vii
List of Figures .............................................................................................................................. viii
Abbreviations .................................................................................................................................. x
Abstract .......................................................................................................................................... xi
Chapter 1 Analysis of Out-of-Hospital Cardiac Arrest (OHCA) and Bystander CPR Literature
Pre- and Post-2015 Institute of Medicine Report .................................................................... 1
1.1. Introduction .........................................................................................................................1
1.2. Methods...............................................................................................................................3
1.2.1. Eligibility Criteria ......................................................................................................3
1.2.2. Data Sources ..............................................................................................................3
1.2.3. Search Strategy ..........................................................................................................3
1.2.4. Study Records ............................................................................................................4
1.2.5. Bibliometric Analysis ................................................................................................4
1.3. Results .................................................................................................................................6
1.3.1. Descriptive Statistics of the Literature .......................................................................6
1.3.2. Authorship Characteristics .........................................................................................8
1.3.3. Journal and Citation Characteristics ..........................................................................9
1.3.4. Scoping Analysis of Significant Publications ..........................................................10
1.4. Discussion .........................................................................................................................16
1.5. Conclusions .......................................................................................................................20
Chapter 2 Spatiotemporal Analysis of Out-of-Hospital Cardiac Arrest in the City of Los Angeles,
2011–2019 ............................................................................................................................. 22
2.1. Introduction .......................................................................................................................22
2.2. Study Design .....................................................................................................................23
2.3. Data Sources .....................................................................................................................23
2.4. Unit of Analysis ................................................................................................................24
2.5. Measures and Methods .....................................................................................................24
2.5.1. Incidence Rate Calculations .....................................................................................24
2.5.2. Non-Performance of Bystander CPR (NBCPR) Event Rate Calculation ................25
2.5.3. Excess Risk Calculation ...........................................................................................25
2.5.4. Mapping of Rates .....................................................................................................25
2.5.5. Spatial Variation in Temporal Trends (SVTT) Clustering ......................................25
2.5.6. Gi* Statistic by Year—Hotspot Mapping ................................................................26
2.5.7. Identifying and Categorizing Neighborhood Criteria ..............................................27
2.5.8. Analytical Software .................................................................................................28
2.6. Results ...............................................................................................................................28
2.6.1. Study Area Description ............................................................................................28

v
2.6.2. OHCA Maps, Tables, and Figures ...........................................................................29
2.7. Discussion .........................................................................................................................37
2.8. Conclusion ........................................................................................................................39
Chapter 3 Mapping Spatiotemporal Relative Risk of Non-Performance of Bystander CPR Out-
of-Hospital Cardiac Arrest and Community Public Health Resource Capacity in Los
Angeles .................................................................................................................................. 40
3.1. Introduction .......................................................................................................................40
3.2. Data ...................................................................................................................................41
3.2.1. Data ..........................................................................................................................41
3.2.2. Study Population and Study Area ............................................................................43
3.3. Measures and Methods .....................................................................................................43
3.3.1. NBCPR Risk ............................................................................................................43
3.3.2. Population Density ...................................................................................................44
3.3.3. Community Public Health Resources ......................................................................44
3.3.4. Spatiotemporal Analytical Workflow ......................................................................45
3.3.5. Zonal Analysis .........................................................................................................47
3.3.6. Bivariate Mapping ...................................................................................................47
3.3.7. Analytical Software .................................................................................................48
3.4. Results ...............................................................................................................................48
3.4.1. NBCPR Relative Risk ..............................................................................................48
3.4.2. Community Public Health Resource Capacity ........................................................51
3.4.3. Zonal Analysis .........................................................................................................52
3.4.4. Bivariate Mapping ...................................................................................................53
3.5. Discussion .........................................................................................................................55
3.6. Conclusion ........................................................................................................................57
Chapter 4 Impact of COVID-19 on Out-of-Hospital Cardiac Arrest and Bystander CPR in Los
Angeles .................................................................................................................................. 59
4.1. Introduction .......................................................................................................................59
4.2. Study Area, Periods, and Population ................................................................................60
4.2.1. Study Area and Periods ............................................................................................60
4.2.2. Study Population ......................................................................................................60
4.3. Data and Measures ............................................................................................................61
4.3.1. Out-of-Hospital Cardiac Arrest (OHCA) Events .....................................................61
4.3.2. OHCA Incidence and NBCPR Event Rate ..............................................................62
4.3.3. OHCA Incidence and NBCPR Event Rate Severity Tiers.......................................62
4.4. Methods.............................................................................................................................62
4.4.1. Analytical Software .................................................................................................64
4.5. Results ...............................................................................................................................64
4.5.1. OHCA Incidence and NBCPR Event Rates.............................................................64
4.5.2. Spatial Analysis of OHCA Incidence and NBCPR Event Rate ...............................67
4.5.3. Fixed Effects Regression Analysis ..........................................................................70
4.6. Discussion .........................................................................................................................72
4.7. Conclusion ........................................................................................................................74

vi
Chapter 5 A Tutorial for Replicating Spatial and Spatiotemporal Analysis of OHCA and
Bystander CPR in the City of Los Angeles and Beyond ....................................................... 76
5.1. Introduction .......................................................................................................................76
5.2. Tutorial Structure ..............................................................................................................77
5.3. Setting Up the RStudio Session and Rmarkdown file ......................................................77
5.4. R Packages ........................................................................................................................78
5.5. Areal Unit Analysis...........................................................................................................80
5.5.1. Analytical Workflow: ..............................................................................................80
5.5.2. R Packages Used for this Analysis ..........................................................................81
5.5.3. Load OHCA and Census Tract Data, Conduct Spatial Join, and Aggregate OHCA
Data.............................................................................................................................82
5.5.4. Preparing for OHCA Incidence and Non-performance of Bystander CPR Event
Rate Calculations ........................................................................................................86
5.5.5. OHCA Incidence Calculations .................................................................................88
5.5.6. NBCPR-Event Rate Calculations ............................................................................89
5.5.7. Excess Risk Calculations .........................................................................................90
5.5.8. Spatiotemporal Hotspot Analysis ............................................................................91
5.5.9. Categorizing Census Tracts into Severity Tiers.......................................................96
5.5.10. Mapping the Results of the Areal Unit Analysis ...................................................97
5.6. Non-Areal Unit Analysis ..................................................................................................97
5.6.1. Analytical Workflow ...............................................................................................97
5.6.2. R Packages Used for This Analysis .........................................................................99
5.6.3. Load Boundary, OHCA, and LandScan Data ..........................................................99
5.6.4. Spatial Relative Risk Surfaces of OHCA and NBCPR .........................................104
5.6.5. Mapping the Results ..............................................................................................109
5.7. Conclusion ......................................................................................................................109
References ................................................................................................................................... 110

vii
List of Tables
Table 1.1 Descriptive statistics of selected literature from 1995 to 2019 ..................................... 19
Table 1.2 Descriptive statistics of selected literature pre- and post the IOM report .................... 21
Table 1.3 Descriptive Information on most citied articles Pre the 2015 IOM report ................... 25
Table 1.4 Descriptive Information on most citied articles pre the 2015 IOM report ................... 26
Table 2.1 Study area and patient characteristics ........................................................................... 40
Table 2.2 Characteristics of OHCA and NBCPR severity tier groupings .................................... 48
Table 3.1 NBCPR relative risk and community organization density for LAFD
battalions................................................................................................................................ 64
Table 3.2 Battalion categorization by bivariate mapping category and statistical surface
characteristics ........................................................................................................................ 65
Table 4.1 Descriptive characteristics of OHCA incidence and NBCPR event rate for
2017-2019 and 2020 .............................................................................................................. 78
Table 4.2 OHCA incidence fixed effect modelling results ........................................................... 82
Table 4.3 NBCPR event rate fixed effect modelling results ......................................................... 82

viii
List of Figures
Figure 1.1 Annual scientific production for the OHCA, bystander CPR, spatial analysis, and
neighborhood effect literature from 1995-2019 .................................................................... 20
Figure 2.1 Out-of-hospital cardiac arrest (OHCA) spatial empirical bayes smoothed (SEBS)
OHCA incidence rate (per 100,000) Map (A) and excess risk map of Non-performance of
Bystander CPR (NBCPR) OHCA map (B) for the City of Los Angeles, 2011–2019 .......... 42
Figure 2.2 Spatiotemporal analysis of OHCA incidence (SVTT) map (A) and OHCA incidence
trends per cluster compared to the city for the City of Los Angeles, 2011–2019 (B) ........... 44
Figure 2.3 SEBS Non-performance of Bystander CPR (NBCPR) event rate hotspot map (A) and
NBCPR event
rate trends per hotspot scores compared to the city for the City of Los Angeles, 2011–2019
(B) .......................................................................................................................................... 45
Figure 2.4 Out-of-hospital (OHCA) and Non-performance of Bystander CPR (NBCPR) severity
tier typology map for the city of Los Angeles, 2011–2019 (A) with trend graphs illustrating
OHCA Incidence (B) and NBCPR event rates (C) ............................................................... 47
Figure 3.1 Spatiotemporal analytical workflow ........................................................................... 57
Figure 3.2 Map series depicting the annual NBCPR relative risk from 2011-2019 on a
normalized scale of 0-100 for the City of Los Angeles......................................................... 61
Figure 3.3 Final combined spatiotemporal representation of NBCPR relative risk on a
normalized scale of 0-100 for the City of Los Angeles with LAFD Battalions .................... 62
Figure 3.4 Community organization density on a normalized scale of 0-100 for the City of Los
Angeles based on data from the LA City Geohub and County Based GIS system ............... 63
Figure 3.5 Map series showing the NBCPR Surface (A), community organization
density (B), and bivariate mapping based on zonal analysis for the LAFD Battalions for
the City of Los Angeles (C)................................................................................................... 66
Figure 4.1 OHCA incidence rate comparison between the average event rate of 2017-2019 and
the incidence rate of 2020 ...................................................................................................... 76
Figure 4.2 NBCPR event rate comparison between the average event rate of 2017-2019 and
the incidence rate of 2020 ...................................................................................................... 77
Figure 4.3 Map series depicting OHCA Incidence by census tract for 2017-2019 (A), 2020 (B),
percent change between the two periods (C) ......................................................................... 80

ix
Figure 4.4 Map series depicting NBCPR event rate by census tract for 2017 to 2019 (A), 2020
(B), percent change between the two periods (C) ................................................................. 81

Figure 5.1 Areal unit analysis workflow ...................................................................................... 93
Figure 5.2 Non-areal unit analysis workflow ............................................................................ 108

x

Abbreviations
LAFD City of Los Angeles Fire Department
OHCA Out-of-Hospital Cardiac Arrest
CPR Cardiopulmonary Resuscitation
BCPR Bystander CPR
NBCPR Non-Performance of Bystander CPR
SVTT Spatial Variation in Temporal Trends

xi
Abstract
Out-of-hospital cardiac arrests (OHCA) continue to be a persistent public health issue.
Decreasing the non-performance of bystander cardiopulmonary resuscitation (NBCPR) is one
key element in improving survivorship with a good neurological outcome when an OHCA
occurs. One way to reduce NBCPR is by understanding where and how persistent this issue is in
certain communities. The primary objective of this dissertation is to provide an analytical
framework for creating actionable intelligence on OHCA and NBCPR. This dissertation seeks to
accomplish that objective through two avenues – (1) using spatiotemporal methods to analyze
OHCA and NBCPR and (2) increasing understanding of OHCA and NBCPR through analyzing
OHCA and NBCPR through the lens of the COVID-19 pandemic. Spatiotemporal methods,
currently, are virtually absent from the OHCA literature that uses spatially enabled data. This
dissertation shows that spatiotemporal methods are viable and can provide more relevant and
useful information for identifying high-risk areas than spatial analysis alone. This dissertation
also shows that the COVID-19 pandemic had differential impacts on communities in Los
Angeles, which also helps inform public health practitioners about resource utilization for public
health and medical emergencies. The final portion of this dissertation furthers the goal of
creating an analytical framework by providing a high-level tutorial on two of the analyses in this
dissertation.

1
Chapter 1 Analysis of Out-of-Hospital Cardiac Arrest (OHCA) and
Bystander CPR Literature Pre- and Post-2015 Institute of Medicine Report
The OHCA and bystander CPR literature is relatively new and has expanded significantly with
the advent of data registries. In this chapter, an analysis of the literature is completed to set up
the more analytical chapters of this dissertation.
1.1. Introduction
Cardiac arrest is a leading cause of mortality in North America, especially in the United States.
Of all sudden cardiac arrests, 360,000 (60%) are out-of-hospital cardiac arrests (OHCA). The
overall survival rate for OHCA in the United States has not exceeded 12% since the
establishment of event-specific registries, such as the Cardiac Arrest Registry to Enhance
Survival (CARES) and the Resuscitation Outcomes Consortium (ROC) registry. Interventions
such as bystander cardiopulmonary resuscitation (CPR) or the use of publicly available
automated external defibrillators (AEDs) are major factors associated with increased survival.
Despite the importance of bystander CPR and AED usage with respect to surviving an OHCA,
usage rates have remained low over the past decade, rarely exceeding 40% for bystander CPR
and even lower, below 15%, for AED usage, according to recent reports (Virani 2020).
Recently, OHCA research that uses spatial analysis and neighborhood effects has blossomed,
bolstered by methodological advances. OHCA-specific registries that include geo-referenced
data have enabled investigations seeking to answer questions such as whether bystander CPR
rates vary by neighborhood (i.e., census tracts) or determining the optimal locations for publicly
accessible AEDs. As part of this growing body of literature, studies examining neighborhood-

2
level variation in OHCA and bystander CPR have found that neighborhood-level measures of
socioeconomic status (SES), race and ethnicity, and language spoken in part explain the
observed geographic variation (Mitchell 2009; Sasson 2011; Raymond 2014; Sasaki 2011; Folke
2009).

In 2015, the Institute of Medicine (IOM) published a summary report on OHCA research that
highlighted existing research and set future research directions (Graham 2015). There are two
aims to this paper. The primary aim of the paper is to bibliometrically analyze OHCA and
bystander CPR research that specifically includes spatial methods and neighborhood effects from
1995 to 2019, bifurcating the literature before and after the IOM report. The secondary aim of
this review is to scope out and compare the primary research objectives, outcomes, covariates,
and methods used in the literature, pre- and post the IOM report. There has yet to be a formal
comparison of the literature after the publication of the IOM report. Developing a sense of how
research changes pre and post the IOM report can help inform research directions that remain to
be explored, or research domains that should be refined and improved if they were addressed in
the literature post the 2015 IOM report.

The results in this review aim to inform future analyses that seek to promote neighborhood
focused OHCA research. The hybrid bibliometric analysis and review presented here will thus
serve as an anchor point on which to base OHCA and bystander intervention research moving
forward (Aria 2020).

3
1.2. Methods
1.2.1. Eligibility Criteria
Articles included in this analysis met two pre-defined criteria: articles must (1) have been written
in English as the primary language; and (2) have been published between 1995 and 2019 in
academic journals, conference proceedings, or in professional reports. These criteria were set up
based on previous reviews of OHCA and bystander CPR research (van Nieuwenhuizen 2019).
1.2.2. Data Sources
The data sources used in this bibliometric analysis were ‘SCOPUS’ and ‘Web of Science’
(SCOPUS 2020; Web of Science 2020). Both databases offer extensive records in which to
search. Furthermore, SCOPUS and Web of Science index important journals for OHCA and
bystander CPR research, including Resuscitation and Annals of Emergency Medicine.
1.2.3. Search Strategy
To identify articles that focus on research on OHCA, bystander CPR, and spatial analysis
together, five keywords or terms were employed—OHCA, bystander, CPR, neighborhood, and
spatial—while searching SCOPUS and Web of Science. Variations in each term were also
included, such as cardiopulmonary resuscitation instead of CPR or geospatial instead of just
spatial, to capture as much of the literature present within both indexing services as possible. The
search was carried out using the Boolean operators ‘AND’ or ‘OR’. The operator AND was used
when combining terms such as OHCA AND bystander CPR AND spatial, while the OR operator
was used to include variations in terms when executing a search of SCOPUS or Web of Science.

4
1.2.4. Study Records
Data management is of the utmost importance when conducting a bibliometric analysis. As such,
for this analysis, records were downloaded as ‘.bib’ files from SCOPUS and Web of Science,
and imported directly into the R package ‘bibliometrix’ for analysis (Aria 2017). The files were
then merged, and duplicates were removed prior to analysis.
1.2.5. Bibliometric Analysis
Bibliometric analysis helps codify a research field by enabling views of both the historical and
current state of a specific topic. The analysis is well regarded and is viewed as a reliable and
objective way to investigate the written literature on a multitude of scales (Aria 2020). There are
two critical procedures performed during bibliometric analysis: performance analysis and science
mapping. Performance analysis evaluates which authors and articles are the most productive and
impactful based on available bibliographic data, whereas science mapping involves visualizing
topics of interest, or historical citation graphs within a given field. This review paper employed
only performance analysis.

The performance analysis started by compiling a set of basic descriptive statistics from the
relevant literature from 1995 to 2019. A separate set of basic descriptive statistics was compiled
for 2010–2014 and 2015–2019 for comparison; the two periods were of equal length and thus
comparable. This basic comparison of the literature helped quantify any changes in the volume
of research before and after the IOM report.

5
After generating initial descriptive statistics for the literature that met the search criteria, the next
step in the performance analysis focused on describing authorship patterns, the most common
journals that publish OHCA studies that include spatial analysis and neighborhood effects, and
the academic reach of these journals. The primary methods in this part of the analysis included
examining authors per document, coauthors per document, and calculating a collaboration index
(ratio between the number of authors of multi-authored documents and the number of multi-
authored documents). These three measures were used in concert to account for the complexities
of how authorship may be measured. For instance, if an author wrote more than one article, that
author would only be counted once in the authors per document calculation, but they would
appear multiple times (e.g., five times for five publications). The collaboration index helps to
confirm which of the two previous ratios are more representative of authorship behavior. Other
measures included looking at citation characteristics and which journals were most popular to
publish in during the time periods considered.

The last portion of this review did not use bibliographic analysis and was instead more
qualitative in nature, centered on determining the most popular themes, concepts, and methods
present within the literature compiled from SCOPUS and Web of Science. To do this, the top 10
most cited original research articles, pre and post the IOM report (meta-analyses and review
articles were excluded), were carefully read, with critical elements collated and tabulated for
easier review.

6
1.3. Results
1.3.1. Descriptive Statistics of the Literature
Table 1.1 provides basic descriptive statistics for the results of the Scopus and Web of Science
queries from 1995 to 2019. There were a total of 161 documents from 81 different sources.
These documents were evaluated on a document-by-document basis to ensure that they met the
criteria for inclusion. The annual percentage growth rate for this type of literature was 10.5%,
with an average of 11.5 documents published per year over the entire query period. The
preponderance of documents was published post-2010, with most being published after the 2015
IOM report.
Table 1.1. Descriptive statistics of selected literature from 1995 to 2019
General statistics
Years 24
Sources 81
Documents 161
Average documents per year 11.5
Author statistics
Authors 884
Single-authored documents 5
Multi-authored documents 879
Annual growth rate in documents 10.5
Documents per author 0.18
Authors per document 5.49
Collaboration index 5.63
Citation statistics
Citations 3723
Average citations per document 21.9
Average citation per year per document 2.6

Figure 1.1 provides a direct, visual comparison of annual scientific production of the literature
pre and post the IOM report. This is highlighted by the line denoting the year in which the IOM
report was published. There is also a clear increase of the scientific production starting in 2010.

7

Figure 1.1. Annual scientific production for OHCA, bystander CPR, spatial analysis, and
neighborhood effect literature from 1995–2019

Most of the scientific publication activity took place after the publication of the 2015 Institute of
Medicine (IOM) report, with 66.5%, or 107 of the 161 documents being published in this period,
compared to 54 of the 161 documents published during the 2010–2014 time period. There was an
annual percentage growth rate of this literature of 34.3% during the 2010–2014 period, with an
average of eight documents published per year, whereas after the IOM report in 2015, there was
a growth rate of 18.9%, with an average of 20.2 documents published. Additional information is
presented in Table 1.2.

Year IOM report is
published

8
Table 1.2. Descriptive statistics of selected literature pre and post the 2015 Institute of Medicine
(IOM) report
Pre the IOM
report
Post the IOM
report
General statistics
Years 5 5
Sources 28 58
Documents 44 107
Average documents per year 8.8 21.4
Author statistics
Authors 241 506
Single-authored documents 0 4
Multi-authored documents 241 502
Annual growth rate in documents 34.3 18.9
Authors per document 5.48 5.01
Collaboration index 5.48 5.18
Citation statistics
Citations 1168 611
Average citations per document 26.55 6
Average citations per year per document 3.1 1.5

1.3.2. Authorship Characteristics
There were 884 authors listed in the 161 documents published during the 1995–2019 timeframe.
Of those 161 documents, only 5 were single authored documents for the entire time frame as
seen in table 1.1. 4 of those documents were published in the period after the IOM report as table
1.2 indicates. The other single authored document occurred outside of the pre-IOM report period
(i.e., 2010-2015). There were, on average, 5.5 authors per document. The collaboration index
during this time frame was 5.63.

After the IOM report, there were 107 documents written by 506 authors. The authors per
document ratios were less similar to each other when analyzing the time after the IOM report
than prior to the IOM report. Specifically, the authors per document ratio was 5.01 and the

9
collaboration index was 5.18 over the period after the IOM report. These data can be found in
Tables 1.1 and 1.2, respectively.
1.3.3. Journal and Citation Characteristics
For both the entire period and post-IOM report only, the most popular journal for publication of
research meeting the search criteria was Resuscitation. There were a total of 41 documents
published in Resuscitation, with 10 articles published during the 2010–2014 time period. A total
of 26 of the articles were published after the IOM report in 2015. Notably, prior to the IOM
report, Academic Emergency Medicine was tied as the second most popular journal with Pre-
Hospital Emergency Care in which to publish, with seven documents in total and four published
between 2010 and 2014. However, after the IOM report, PLoS One was listed as the second most
popular destination to publish in, while Pre-Hospital Emergency Care was third, with five and
three documents, respectively. Academic Emergency Medicine did not make the list of the top 10
journals after the IOM report.

Citations per document is one way to measure the impact of a body of literature. During the
entire study period, documents averaged 2.65 citations per year per document, and an average of
23.11 citations per document. Documents had 26.55 citations per document and 3.082 citations
per year per document before the IOM report. After the IOM report, documents averaged 1.5
citations per year per document, and 6.1 citations per document. Leading journals, such as
Resuscitation, as of 2019, had an impact factor of 4.6, while journals such as Pre-Hospital
Emergency Care and PLoS One had impact factors of 2.4 and 2.7, respectively.

10
1.3.4. Scoping Analysis of Significant Publications
Table 1.3 provides summary information from the top 10 most cited articles prior to the 2015
IOM report, while Table 1.4 provides summary information after the 2015 IOM report (Uray
2015; Blewer 2018; Chan 2016; Patterson 2017; Lee 2015; Ro 2016; Starks 2018; Sasson 2014;
Sun 2016; Ro 2017; Yasunaga 2011; Sasson 2011; Yokoyama 2011; Sasson 2013; Sasson 2010;
Reinier 2011; Hasegawa 2013; Sasson 2012). There were some common research objectives pre
and post the 2015 IOM report.

The most prevalent research objective was to examine the impact that the neighborhood had on
the provision of bystander CPR, and outcomes from an OHCA, or some variation on that
objective. In total, 7 of the 20 articles that underwent analysis had this as the major, or one of the
key research objectives of the investigation. Prior to the IOM report, a common objective of the
literature was to assess regional or other scales of variation in the survivorship outcomes of
OHCA. Post the IOM report, however, a common research objective was developing or
assessing interventions to improve community-based OHCA care, such as AED access,
dispatcher-assisted CPR instructions, or assessing barriers to using 911.

There were some commonalities between the study types and primary methodologies. The most
common study type across the literature prior to and after the publishing of the IOM report was
observational/cross-sectional analysis. 60% of the studies examined were observational/cross-
sectional. Further, much of the literature used similar covariates. Most often, two levels of
covariates were included in the analyses; those levels were patient- and neighborhood-level. At
the patient level, characteristics like age, sex, race, or ethnicity, location of OHCA event, initial

11
heart rhythm, and bystander intervention (AED or CPR) were the most frequently included.
Socioeconomic status variables such as median household income, education levels,
race/ethnicity composition, and population density were the most common neighborhood-level
characteristics. Lastly, the most frequently used methodology was regression analysis,
specifically some form of logistic regression (e.g., multilevel or multivariable logistic
regression).

From a spatial methods perspective, geocoding event locations was the most common
methodology employed both pre and post the IOM report. Geocoding was used to place the
events within a specific areal unit, most often census tracts, and to link the events to a set of
socioeconomic and demographic data. In some instances, the events were linked and then
aggregated to a census tract or other areal unit to calculate incidence rates.

More advanced spatial methods were used more frequently after, when compared to before, the
IOM report. Specifically, publications post the IOM report saw an increase in analyses that used
spatial optimization models that sought to address the maximum coverage location problem. The
most spatially advanced analysis pre-IOM report came from a paper that utilized local clustering
to perform a ‘hot- and cold-spot’ analysis of bystander CPR education rates.

12
Title Year Citations Study type Objective
Primary outcome
variable Primary methods
Spatial
methods
Association of
Neighborhood
Characteristics with
BCPR
2012 120 Observational/Cohort
Determine what impacts
performance of BCPR
Performance of
BCPR
Hierarchical
regression
Geocoding and
spatial
aggregations
Post-resuscitation Care
with Mild Therapeutic
Hypothermia and
Coronary Intervention
After OHCA
2011 86
Observational/Cross
Sectional
Determine impact of
therapeutic hypothermia
Survivorship and
good neurological
outcome
Binary logistic
regression
Geocoding
Rates of CPR Training in
the United States
2014 64
Observational/Cross
Sectional
Determine regional
variation of BCPR training
in the United States
Rate of CPR training
at the county level
Multivariable
Logistic regression
Aggregation
and spatial
clustering of
CPR training
rates
Regional Variability in
Survival Outcomes of
Out-of-Hospital Cardiac
Arrest: The All-Japan
Utstein Registry
2013 63
Observational/Cross
Sectional
Determine if OHCA
survivorship varied across
Japan
Good neurological
outcome and 1-
month survival
Multivariable
Poisson regression
Geocoding and
spatial
aggregations
Socioeconomic Status
and Incidence of Sudden
Cardiac Arrest
2011 61
Observational/Cross-
sectional
Determine if disparities
exist for sudden cardiac
arrest by socioeconomic
status in Canada
IRR of sudden
cardiac Arrest
Poisson regression
Geocoding and
spatial
aggregations
Small Area Variations in
OHCA: Does the
Neighborhood Matter?
2010 52
Observational/Cross-
sectional
Determine the extent to
which OHCA and BCPR
vary by neighborhood (i.e.,
census tract)
Incidence of OHCA
Multilevel Poisson
regression
Geocoding and
spatial
aggregations;
empirical
bayes
smoothing
Barriers and Facilitators
to Learning and
Performing Bystander
Cardiopulmonary
Resuscitation in
Neighborhoods with
Low Bystander CPR
Prevalence and High
Rates of Cardiac Arrest
in Columbus, Ohio
2013 45
Community based
participatory research
Identify barrier and
facilitators to learning
BCPR
Knowledge barriers
and facilitators to
learning and
performing bystander
CPR
Qualitative analysis
of focus groups
Geocoding and
spatial
aggregations
Table 1.3. Descriptive information on most citied articles Pre-2015 IOM report

13
Impact of Therapeutic
Hypothermia in the
Treatment of Patients
with Out-of-Hospital
Cardiac Arrest from the
J-PULSE-HYPO study
registry
2011 45 Clinical trial
Determine the
effectiveness of
therapeutic hypothermia
Favorable
Neurological
Outcome, 1-month
survival
Two-sided t-tests
Geocoding and
spatial
aggregations
Examining the
Contextual Effects of
Neighborhood on Out-
of-Hospital Cardiac
Arrests and the Provision
of Bystander CPR
2011 44
Observational/Cross-
sectional
Determine associations
between individuals,
neighborhoods, and the
performance of BCPR
Performance of
Bystander Initiated
CPR; Survival to
Discharge
Hierarchical non-
linear regression
Geocoding and
Spatial
Aggregations;
Empirical
Bayes
estimates of
incidence rates
Population Density Call
Response Interval and
Survival of Out-of-
Hospital
2011 39
Observational/Cross-
sectional
Determine how population
density relates to 911
response time
1-month survival, 1-
month survival with
good neurological
outcome
Hierarchical logistic
regression
Geocoding and
Spatial
Aggregations;
Population
Density
Calculations

14

Title Year Citations Study Type Objective Primary Outcome
Variable
Primary
Methods
Spatial
Methods
Effect of Dispatcher-
Assisted
Cardiopulmonary
Resuscitation and
Location of Out-of-
Hospital Cardiac Arrest
of Survival and
Neurologic Outcome
2016 51
Observational/Cross-
sectional
Determine the impact of
dispatcher assisted CPR
efforts on OHCA
survivorship
Good neurological
outcome, survival to
discharge, OHCA
ROSC
Multivariable
logistic regression
Geocoding
Overcoming Spatial and
Temporal Barriers to
Public Access
Defibrillators Via
Optimization
2016 31 Observational/Cohort
Develop an AED
optimization model that
accounts for spatial and
temporal accessibility
AED accessibility
Spatiotemporal
optimization
Spatiotemporal
optimization
Barriers to Calling
911and Learning and
Performing
Cardiopulmonary
Resuscitation for
Residents of Primarily
Latino, High-Risk
Neighborhoods in
Denver, Colorado
2015 30 Mixed methods
Identify barriers and
facilitators to calling 911
Barriers to calling 911
and performance of
BCPR
Qualitative
interviews
Geocoding and
spatial
aggregations
Association of
Neighborhood
Demographics with Out-
Of-Hospital Cardiac
Arrest Treatment and
Outcomes…
2017 24 Observational/Cohort
Evaluate associations
between bystander and
EMS treatment actions on
survivorship of an OHCA
Survival to discharge,
ROSC on arrival to
ED, good neurological
status
Multilevel mixed
effect logistic
regression
Geocoding and
spatial
aggregations
Public Awareness and
Self-Efficacy of
Cardiopulmonary
Resuscitation in
Communities and
Outcomes of Out-of-
Hospital Cardiac
Arrest…
2016 22
Observational/Cross-
sectional
Determine the impact of
CPR awareness at a
community level has on
OHCA survivorship
BCPR
Multilevel logistic
regression
Geocoding and
spatial
aggregations
Interaction Effects
between Highly Educated
Neighborhoods and
2016 21
Observational/Cross-
sectional
Determine the interactions
effects of neighborhood
BCPR
Multilevel logistic
regression
Geocoding and
spatial
aggregations
Table 1.4. Descriptive information on most citied articles Post-2015 IOM report

15
Dispatcher-Provided
Instructions on Provision
of Bystander CPR
education level and
dispatcher assisted CPR
…The ARREST pilot
randomized trial
2017 19 Randomized pilot study
Asses the feasibility of
conducting large scale
clinical trials on OHCA
transfers
30-day all-cause
mortality, good
neurological outcome
at discharge
Odds ratios,
Kaplan–Meier
survival curves, t-
tests
Geocoding
Optimizing the
Deployment of Public
Access Defibrillators
2016 18 Spatial optimization
Develop the first data-
driven framework for
AED deployment
AED coverage Spatial optimization
Spatial
optimization
Gender Disparities
Among Adult Recipients
of Bystander
Cardiopulmonary
Resuscitation in the
Public
2018 18 Observational/Cohort
Determine if differences
by gender exist in the
provision of BCPR by
location
BCPR
Multivariable
logistic regression
Geocoding
Socioeconomic Factors
Associated with Outcome
After Cardiac Arrest in
Patients Under the Age of
65
2015 15
Observational/Cross-
sectional
Describe the associations
between socioeconomic
status and OHCA
survivorship
Survival to discharge,
good neurological
condition at discharge
Multivariable
backwards stepwise
logistic regression
Geocoding

16

1.4. Discussion
The literature relating to OHCA, bystander CPR, spatial analysis and neighborhood effects has
undergone noteworthy changes over the course of 1995–2019. Prior to the 2015 IOM report,
there was a steady increase in the annual scientific production of articles and other documents
investigating this persistent public health issue.

The creation of various OHCA-specific registries that use the Utstein guidelines enabled the
growth of this field of research (McNally 2009; Morrison 2008; Perkins 2015). The initial surge
in research around 2007 after the creation of groups such as the ROC and CARES centered
around detecting and describing statistical variation, not necessarily spatial variation in OHCA,
related outcomes, and performance of bystander CPR (Sasson 2012; Root 2013). Papers focusing
on spatial variation began to show up in the literature post 2010. This increase in research likely
enabled the writing and publication of the IOM report in 2015, as a critical amount of literature
had been amassed. Post the IOM report, the body of literature saw a rapid increase in quantity, as
evidenced on Figure 1.1. This rapid increase in research is mostly indicative of how publishing
works in medicine and public health, where major publications by key institutions often spur
major research initiatives and publication booms.

Examining descriptive statistics for authorship characteristics provided interesting insights into
the OHCA, bystander CPR, spatial analysis, and neighborhood effects literature. For the entire
study period, the collaboration index was 5.63. In the period prior to (2010–2014) and after
(2015–2019) the IOM report, the collaboration index was 5.48 and 5.18, respectively. There

17
were 241 (pre-IOM report) and 506 (post-IOM report) authors who published a document during
the time frames. Taking into account the number of authors in each period, it appears there was a
smaller, more collaborative, and comparatively productive, group of authors during the 2010–
2014 period as compared to the 2015–2019 period, based on the number of publications.

The descriptive statistics for journal characteristics remained consistent before and after the IOM
report in 2015. Resuscitation was the journal in which the majority of the documents were
published, followed closely by Prehospital Emergency Care. However, a contrasting pattern
emerged when evaluating the journals of the top-cited articles before and after the IOM report.
Prior to the IOM report, many of the highest cited journal articles were published in non-
resuscitation or emergency medicine journals. About 50% of the most cited articles prior to the
IOM report were published in journals with a broader focus, such as The New England Journal
of Medicine or The Journal of the American Medical Association. Conversely, after the IOM
report, a majority (90%) of the articles were published in either Resuscitation or another
emergency medicine-focused journal, such as Annals of Emergency Medicine.

Moving deeper into the OHCA, the bystander CPR, spatial methods, and neighborhood effects
literature required a comparative scoping analysis of articles published pre and post the 2015
IOM report. Research objectives distinctly shifted between publications before and after the IOM
report, from investigations centered on describing variation or explaining the role of
neighborhood-level covariates in the provision of bystander CPR to more solution-focused
investigations. However, the use of spatial and aspatial methodologies remained consistent for
the entire period considered, despite advancements in both types of analyses. Mostly, logistic

18
regressions were used for defining relationships, while geocoding event location was the most
frequently used spatial method. This was a somewhat surprising result, given the intention of this
investigation, as the expectation was that spatial analysis and associated methods would be more
common and more advanced.

Only 3 of the top 20 cited articles pre or post the IOM report employed spatial methods beyond
geocoding or aggregation of variables. Ideally, spatial methods should be more prevalent in
research that describes variation and seeks to explain neighborhood effects, assuming limitations
such as the modifiable areal unit are appropriately addressed or acknowledged. For instance,
given the likelihood that spatial autocorrelation would exist for either the outcome variables, or
explanatory variables and covariates, spatial lag or spatial error terms should be directly
incorporated into models, as in conditional autoregressive analyses, to provide more accurate
results (Duncan 2013).

Based on the results of the analyses conducted in this paper, there are multiple pathways through
which future research may continue to make an impact in this field. One option moving forward
is to focus on incorporating more spatial methods, and even spatiotemporal methods, into the
literature. Other spatial methods could focus on spatial smoothing or kernel risk surface
representations of OHCA and bystander CPR risk. Statistical surfaces have been touted as a
superior method for creating maps of disease and other health outcomes and behavior (Beyer
2012). These techniques offer the distinct advantages of areal unit analysis and are commonly
used in other types of spatial epidemiology (Beyer 2012). Spatiotemporal methods would also be
a logical step forward. The CARES and ROC registry have been in continuous operation since

19
2007 and 2005, respectively. Given the length of time over which both have been in operation,
there is a substantial body of data with which to implement spatiotemporal analyses.
Spatiotemporal analyses would provide the opportunity to develop a more in-depth
understanding of OHCA and bystander CPR risks in communities for which data are available.
Indeed, some analyses have already begun to focus on the spatiotemporal patterns of OHCA and
bystander CPR (Demirtas 2016).

Another pathway to improve OHCA research centers on developing and implementing more
substantial theoretical frameworks relating to the mode of action by which neighborhoods
influence OHCA and bystander CPR risk. The results from the thematic analysis of the top 10
articles pre and post the IOM report demonstrate that they used same covariates at the
neighborhood level. Specifically, they used race and ethnicity, income, age, and education levels.
However, the existing literature does little to move beyond the socioecological interaction of
individuals and their neighborhood. Using theoretical frameworks such as the ecology of social
support popularized by Klinenberg could be an interesting path forward (Klinenberg 2002 and
2012). The exploration of other theoretical frameworks could also provide new insights into
OHCA and bystander CPR public health challenges. These new insights could be published in
journals that enjoy a broader readership, which would, in turn, open up the field to a broader
audience again. As an example of this, walkability is a potentially important covariate, but it has
only just begun to appear in the OHCA and bystander intervention literature (Chen 2019).

The bibliometric analysis and scoping review outlined here is not without its limitations. Firstly,
this is a relatively niche body of literature to examine. This is evident when considering that

20
there have been only 161 documents published on the topic since 1995. However, while this is a
niche of the literature, it is nonetheless an important one, given that OHCA remains a critical
public health issue, with over 360,000 occurring every year, and the survivorship hovering at
around only 10%. Further, bystander CPR and bystander AED usage rates have remained
relatively stagnant, despite the publication of the IOM report and recent intervention efforts.
1

Another potential weakness of this analysis is that alterations to the keywords, or differences in
the execution of the search for academic documents, might paint a different bibliometric or
thematic picture. For example, the removal of “neighborhood” or “neighborhood effects” and
other variations in the search strategy might have resulted in articles that included a higher
variety of spatial methods other than geocoding or areal aggregation. There are multiple studies
that use spatial clustering techniques to identify high-risk neighborhoods from an OHCA and
bystander CPR perspective, but none of these papers made the top 10 most cited lists for before
or after the IOM report (Nassel 2014). Additional limitations are in the use of descriptive
statistics only. This makes it difficult to be fully confident in any inferences made in this review.
1.5. Conclusions
The body of literature relating to OHCA, bystander CPR, spatial analysis, and neighborhood
effects has grown tremendously over the 1995–2019 period, which largely occurred after the
creation of OHCA specific registries. The 2015 IOM report provided a summary of research up
to that point and helped direct future research. As a result, researchers began testing and
evaluating new interventions, such as dispatcher-assisted CPR and public access AED programs.
Moving forward, new spatial and spatiotemporal methods should be employed, given the amount
of data now available. New theoretical frameworks and neighborhood measures should be
included in research to help open up the discipline and bring in new scholarship. OHCA and

21
bystander CPR remain difficult public health challenge. Thus, research must continue to improve
to better reduce the public health burden that comes from OHCAs with no or limited bystander
action.

22
Chapter 2 Spatiotemporal Analysis of Out-of-Hospital Cardiac Arrest in the
City of Los Angeles, 2011 –2019
OHCA and bystander CPR analyses have largely underutilized spatiotemporal analysis to create
risk stratifications for different environments. This chapter focuses on introducing a
spatiotemporal blueprint to categorize census tracts according to OHCA and NBCPR severity.
2.1. Introduction
Over 360,000 OHCAs occur annually in the United States, and they account for a
disproportionate share of sudden cardiac death mortality, including over 2,400 EMS-treated
cases each year in the City of Los Angeles (Rea 2003; Virani 2020). The use of Utstein
guidelines and both local and national registries has enabled broad spatial analysis of OHCA
events, which has led to a better understanding of determinants of survival and the improvement
of OHCA outcomes (Perkins 2015; Morrison 2008; McNally 2009; Sasson 2012; Nassel 2014;
Hasselqvist-Ax 2015).

Spatiotemporal analysis may help confirm the identity of neighborhoods with persistently high
OHCA incident rates and low bystander CPR performance rates and help target research and
public health efforts to reinforce the chain of survival in these discrete communities.

There have
been some studies that evaluate the temporal characteristics of OHCA at multiple scales, such as
annually, seasonally, or monthly, and those studies found that OHCA in the United States do
have temporal variation (Buick 2018).

However, few studies have examined OHCA distribution
using space and time together (Demirtas 2019). More in-depth spatiotemporal analysis is a
needed next step in OHCA and bystander CPR research.

23
This paper reports on a descriptive spatiotemporal analysis of 9 years of incident OHCA and
non-performance of bystander CPR (NBCPR) risk in Los Angeles.
2.2. Study Design
This is a descriptive, cross-sectional study of data from the City of Los Angeles, a city of
approximately 4 million people. The objective is to identify areas with high rates of incident
OHCA and also a high risk for NBCPR.
2.3. Data Sources
Data for this study was obtained from the Los Angeles Fire Department cardiac arrest registry
consisting of adult (≥ 18 years of age) with EMS-treated OHCA events from January 1, 2011 –
December 31, 2019. The LAFD cardiac arrest registry conforms to current Utstein guidelines.3
Furthermore, the registry uses electronic health record data acquired by paramedics on scene
containing the location and bystander CPR status of each event. Cases were considered for
inclusion in the registry and this study if any of the following items were selected by on-scene
paramedics: Provider Impression Cardiac Arrest, Provider Impression Respiratory Arrest,
Treatment CPR, Rhythm Pulseless VF/VT/PEA/Asystole, or Protocol Non-traumatic Cardiac
Arrest. Cases were subsequently manually reviewed to ensure that each represented a true
LAFD-treated OHCA case. Reports containing multiple inclusion criteria were included once.

Cases were excluded if they were non-adult, of traumatic etiology, occurred after EMS arrival,
were in a healthcare facility or nursing home, or were pronounced dead on arrival. All other

24
events were included in the analysis. The included cases were geocoded on the response address,
with latitudes and longitudes coming from the dispatch system of the LAFD.
Population data in this analysis were sourced from the 2010 census (United States Census
Bureau 2010). This included population estimates per census tract for characteristics such as sex,
age, race, and ethnicity, and median household income.
This study was approved by the Institutional Review Board of the Keck School of Medicine at
the University of Southern California.
2.4. Unit of Analysis
Census tracts were used as the spatial unit of analysis; this is a well-established precedent in
OHCA and other health research (Krieger 2003). Some census tracts (n = 27, 2.6%) were
removed from the analysis, as their residential population counts were extremely low, while their
OHCA counts and event rates were high (e.g., Los Angeles World Airport, Universal Studios,
Griffith Park, and Studio City). Other atypical census tracts, such as the Port of Los Angeles,
were included due to higher and more traditional residential population dynamics.
2.5. Measures and Methods
2.5.1. Incidence Rate Calculations
Crude OHCA incidence calculations relied on standard incidence formulas. Of note, the total
number of adults in each census tract gathered from the census was multiplied by the number of
years this analysis examined to generate normed incidence rates, following previous research
(Hayward 2007). After crude incidence calculations, a spatial empirical Bayes smoother (SEBS)
in GeoDa was used to calculate the smoothed incidence rates for all census tracts. The process of
smoothing spatial event incidence data results in the statistical stabilization of data to avoid the

25
small number problem, which refers to how smaller event counts can result in less statistically
stable rates per unit of analysis (Anselin 2006).
2.5.2. Non-Performance of Bystander CPR (NBCPR) Event Rate Calculation
NBCPR event rates were calculated for all census tracts included in the study sample to gauge
where NBCPR occurred in the study area. These rates were calculated by dividing the total
number of NBCPR events in a census tract by the total number of OHCA events in that census
tract. Spatial empirical Bayes smoothing was used to calculate event rates by year to maintain
consistency in methods with OHCA incidence.
2.5.3. Excess Risk Calculation
Mapping excess risk is a commonly used spatial epidemiology technique. Often, this refers to the
mapping of measures, such as the standardized mortality ratio (SMR) (Anselin 2006). In this
analysis, mapping excess risk focused on NBCPR events and constituted calculating a ratio of
the observed NBCPR event rate in the census tract to the average event rate for the entire city of
Los Angeles. Higher value of this ratio indicates a greater risk of NBCPR.
2.5.4. Mapping of Rates
Each map was classified according to the distribution of each variable using a 1.5-hinge boxplot
classification method. Excess risk was calculated so that higher values indicate a higher chance
of an event not receiving bystander CPR, and lower values indicate a lower risk of not receiving
bystander CPR.
2.5.5. Spatial Variation in Temporal Trends (SVTT) Clustering
The SaTScan software provides multiple options for spatiotemporal analysis of health data. The
SVTT clustering analysis from SaTScan uses a scanning window (i.e., a circle or ellipse that

26
varies in size for each census tract location in the data set) that is only spatial in nature (Kulldorf
2009). The scanning window size was adjusted for every possible location in the study area
(Kulldorf 2009). The scanning window size can vary from a size of 0 (i.e., a singular feature or
census tracts) to a user-designated percentage of the population at risk through distance or a
specific distance absent of a percentage of the population at risk.

Temporal trends of incidence rates per census tract, either increasing or decreasing, depending
on the observed data for the area, are then calculated for inside and outside each location and for
the size of the scanning window (Kulldorf 2009).
16
This clustering approach relies on discrete
Poisson probability (Kulldorf 2009). Statistical significance (p ≤ .05) is then assessed using a
likelihood based on the null hypothesis that trends are the same inside and outside the window
(Kulldorf 2009).

2.5.6. Gi* Statistic by Year—Hotspot Mapping
The Getis-Ord Gi* statistic calculates spatial hotspots and coldspots for a variable of interest,
and the higher the z-score, the more intense the clustering of high values (hotspots), while the
smaller the z-score, the more intense the clustering of low values (coldspots) (Getis 1992). Z-
score calculations take the local sum of a variable for a census tract and its neighbors (Getis
1992). For this analysis, the Getis-Ord Gi* statistic was calculated in GeoDa and repeated for
each year in the data set using the spatially smoothed NBCPR event rate as the variable of
interest, and census tracts with z-scores greater than 1.96 were chosen as statistically significant
clusters with high amounts of NBCPR. From there, a hotspot score was generated to quantify
how a census tract might oscillate between a hot or coldspot of NBCPR events over the study
period. If the census tract was a hotspot in a given year, it was assigned the value of 1;
conversely, a value of -1 was assigned if the census tract was a coldspot in a given year. If the

27
census tract’s z-score was not significant in a given year, it was assigned a value of 0. The values
were then summed for the entire study period. A high positive score indicates that a census tract
was a hotspot for NBCPR events for more years in the study period but does not necessarily
represent consecutive years.
2.5.7. Identifying and Categorizing Neighborhood Criteria
After implementing the analytical methods described above, the results of each method were
used to categorize census tracts into a three-level typology that factors in OHCA incidence and
the risk of NBCPR using R statistical software via RStudio. Tier 1 census tracts were considered
the most severe, while tier 3 census tracts were considered the least severe. Below is a bulleted
list breakdown of the decision criteria for each severity level:
• Tier 1
Within an SVTT cluster with an increasing trend AND
Hotspot score ≥ 4 AND
Excess risk of NBCPR > 1.0

• Tier 2
Within an SVTT cluster with an increasing trend AND
Hotspot scores ≥1 and ≤ 3 AND
Excess risk of NBCPR > 1.0

• Tier 3
SEBS incidence was greater than the mean AND
Excess risk of NBCPR > 1.0

28
2.5.8. Analytical Software
Analysis was performed using R 3.5.3/RStudio 1.2.1335, SaTScan, and GeoDa, while all map
visualizations utilized ArcGIS Pro 2.3 (R Development Core Team 2018; Esri 2018; Anselin
2006).

2.6. Results
2.6.1. Study Area Description
Table 2.1 provides descriptive data of OHCA incidence and NBCPR event rates, as well as
patient and neighborhood sociodemographic characteristics.
Table 2.1. Study area and patient characteristics
OHCA incidence characteristics (per 100,000) [95% CI]
Mean crude incidence 53.7 [47.5–59.9]
Mean smoothed incidence 52.6 [47.3–57.9]
NBCPR event rate characteristics (%) [95% CI]
Mean crude NBCPR event rate 53.3 [52.2–54.3]
Mean smoothed NBCPR event rate 52.8 [52.3–53.3]
Excess risk of NBCPR [95% CI]
Excess risk of NBCPR 1.01 [.99–1.03]
Patient characteristics (n = 15904)
Mean age [SD] 66 [17.4]
Gender (%)
Male 6044 (38.0)
Female 9861 (62.0)
Witnessed Arrest (%)
No 8913 (56.0)
Yes 6848 (43.1)
Unknown 143 (.9)
Bystander CPR (%)
Yes 7389 (46.5)
No 8439 (53.0)
Unknown 76 (.5)
Witnessed arrest by CPR Status [95% CI]
Non-performance of Bystander CPR 3748 (55.0)
Bystander CPR 3100 (45.0)
Census tract sociodemographic characteristics (n = 985)
Gender (%)
Male 49.9
Female 50.1
Age (%)
Under 18 22.7
18–24 11.2
25–34 17.1
35–44 15.1

29
45–54 13.2
55–64 9.9
65 and over 10.7
Race and ethnicity (%)
White 29.2
Black 8.8
Hispanic 48.0
Asian 11.4
Median household income ($)
Median household income 45540

Trends for OHCA incidence and NBCPR event rates can be seen in Figures 2.1 and 2.3,
respectively. From 2011-2019, there were a total of 15,904 EMS-treated, non-traumatic adult
OHCA events that occurred in 985 census tracts in the study period of 2011 through 2019.
Aggregating those events to census tracts resulted in a mean crude incidence OHCA rate for the
City of Los Angeles of 53.7 per 100,000 people, and the mean NBCPR event rate per census
tract was 53.3% for the entire nine-year study period. Both incidence rates and NBCPR event
rates remained relatively stable over the course of the study and across the city, excluding census
tracts identified as high risk. Furthermore, the mean percentage of witnessed events where
bystander CPR occurred for all census tracts was 45.0%.
2.6.2. OHCA Maps, Tables, and Figures
2.6.2.1. SEBS OHCA Incidence Rates and Excess Risk of NBCPR
The mean SEBS incidence rate was 52.6 OHCA events per 100,000 people. Figure 2.1 maps the
SEBS OHCA incidence rates for the city (map A) along with the excess risk of NBCPR events
(map B). 522 census tracts had an excess risk > 1, and 54 census tracts had an excess risk of ≥
1.5. Many of the areas of higher OHCA incidence are also at risk for no or low bystander CPR

30

Figure 2.1. Out-of-hospital cardiac arrest (OHCA) spatial empirical Bayes smoothed (SEBS) incidence rate (per 100,000) map (A) and excess
risk map of Non-performance of Bystander CPR (NBCPR) OHCA map (B) for the City of Los Angeles, 2011–2019

B A

31
2.6.2.2. Spatial Variation in Temporal Trends (SVTT) Cluster Detection of OHCA Incidence
The SVTT analysis was set to detect statistically significant (p < .05) spatial clusters with mean
annual increasing trends of OHCA incidence rates in the City of Los Angeles. There were three
such clusters identified, as presented in Figure 2.2A. The average annual increase in OHCA
incidence rates for the three identified clusters was 19.6%. Figure 2.2B provides additional
context for the increasing trends in incidence rates observed in these three clusters.
2.6.2.3. Getis-Ord Gi* Statistic—Hotspot and ColdSpot Mapping by Year
Figure 2.3A contains the clustering results of the Getis-Ord Gi* and the hotspot analysis of the
spatially smoothed NBCPR event rates for each year in the study period. With regard to hotspots,
336 (34%, n = 985) census tracts in the study area were identified as having a hotspot score
greater than 0, indicating that those census tracts had higher NBCPR event rates and were
surrounded by areas of higher NBCPR event rates for ≥ 1 year of the 9-year study period.
Furthermore, 66 (7%, n = 985) census tracts in the study area were a hotspot in between 4 and 7
years of the 9 years in the study period. Based on the trend lines in Figure 2.3B, census tracts that
were hotspots for between 1 and 2 years closely mirrored the city average for NBCPR event rates
at the beginning of the study period. Regarding census tracts that were a hotspot for at least 3
years, there were markedly higher increase in mean NBCPR values for each year compared to
the city. The trends for census tracts that were hotspots for at least 3 of the 9 years trend where
relatively stable. Furthermore, many of these areas are contiguous and were located in the
central, south, and eastern areas of the city.

32

Figure 2.2 Spatiotemporal analysis of OHCA incidence (SVTT) map (A) and OHCA incidence trends per cluster compared to the city
for the City of Los Angeles, 2011–2019 (B)
A

B
B

33

Figure 2.3. SEBS Non-performance of Bystander CPR (NBCPR) event rate hotspot map (A) and NBCPR event rate trends per hotspot
scores compared to the city for the City of Los Angeles, 2011–2019 (B)
B
A

34
2.6.2.4. Identifying and Categorizing Neighborhoods
This analysis was designed to identify and spatiotemporally categorize census tracts by OHCA
incidence and NBCPR event status into three different severity levels. In total, 182 census tracts
were identified as severe areas with respect to OHCA incidence and NBCPR event rates,
according to the decision criteria. Figure 2.4 and Table 2.2 provide insights into the various
severity tiers and what those tiers look like from multiple perspectives. The trend line charts in
Figures 2.4B and 2.4C show that the observed OHCA incidence and NBCPR event rates in these
spatiotemporally categorized census tracts were generally higher than the city average across the
nine-year study period. Of note, tier 2 and tier 3 census tracts tended to have higher amounts of
incident OHCA, but tier 1 census tracts had higher amounts of OHCA classified as NBCPR.
These areas were largely contiguous and located in the central, and southern parts of the city of
Los Angeles; there were also some groupings in the northern part of the city. Lastly, and similar
to the rest of the city, the age and gender make-up of individuals who experienced an OHCA
remained stable for each of the different tiers.

35

Figure 2.4. Out-of-hospital (OHCA) and Non-performance of Bystander CPR (NBCPR) severity tier typology map for the city of Los Angeles,
2011–2019 (A) with trend graphs illustrating OHCA incidence (B) and NBCPR event rates (C).

A
B
C

36
Tier 1 Tier 2 Tier 3
Number of census tracts 17 Number of census tracts 36 Number of census tracts 129
OHCA incidence characteristics (per 100,000) OHCA incidence characteristics (per 100,000)

OHCA incidence characteristics (per 100,000)

Mean crude OHCA incidence 86.6 Mean crude OHCA incidence 50.5 Mean crude OHCA incidence 85.5
Mean smoothed OHCA incidence 121.5 Mean smoothed OHCA incidence 72.5 Mean smoothed OHCA incidence 74.0
NBCPR event rate characteristics (%) NBCPR event rate characteristics (%) NBCPR event rate characteristics (%)
Mean crude NBCPR event rate 66.5 Mean crude NBCPR event rate 64.6 Mean crude NBCPR event rate 61.9
Mean smoothed NBCPR event rate 65.7 Mean smoothed NBCPR event rate 58.8 Mean smoothed NBCPR event rate 57.9
Excess risk of NBCPR 1.29 Excess risk of NBCPR 1.20 Excess risk of NBCPR 1.23
Patient characteristics (n = 428) Patient characteristics (n = 896) Patient characteristics (n = 3562)
Age [SD] 58.9 [15.9] Age [SD] 63.6 [18.4] Age [SD] 65.3 [17.1]
Gender (%) Gender (%) Gender (%)
Male 352 (66.0) Male 394 (63.3) Male 2124 (60.5)
Female 181 (34.0) Female 228 (36.7) Female 1386 (39.5)
Witnessed Arrest (Y/N) (%) Witnessed arrest (Y/N) (%) Witnessed arrest (Y/N) (%)
No 296 (55.5) No 371 (59.6) No 1952 (55.6)
Yes 232 (43.6) Yes 250 (40.2) Yes 1499 (42.7)
Unknown 5 (.9) Unknown 1 (.2) Unknown 59 (1.7)
Bystander CPR (%) Bystander CPR (%) Bystander CPR (%)
BCPR 164 (30.7) BCPR 214 (34.4) BCPR 1256 (35.8)
NBCPR 369 (69.3) NBCPR 407 (65.5) NBCPR 2218 (63.2)
Unknown 0 (0.0) Unknown 1 (.1) Unknown 36 (1.0)
Percent bystander CPR per witnessed
event
34.9 Percent bystander CPR per witnessed
event
35.6 Percent bystander CPR per witnessed
event
35.7
Sociodemographic characteristics Sociodemographic characteristics Sociodemographic characteristics
Gender (%) Gender (%) Gender (%)
Male 57.9 Male 51.3 Male 49.5
Female 42.1 Female 49.7 Female 50.5
Age (%) Age (%) Age (%)
Under 18 18.7 Under 18 29.6 Under 18 24.7
18–24 9.6 18–24 12.8 18–24 10.8
25–34 17.4 25–34 17.8 25–34 15.5
35–44 15.6 35–44 14.3 35–44 14.4
44–54 17.1 44–54 11.2 44–54 13.5
55–64 12.0 55–64 7.3 55–64 9.9
65 and over 9.6 65 and over 7.0 65 and over 11.2
Race and ethnicity (%) Race and ethnicity (%) Race and ethnicity (%)
White 11.4 White 8.8 White 19.5
Black 29.0 Black 12.8 Black 24.1
Hispanic 50.6 Hispanic 64.3 Hispanic 46.9
Asian 6.4 Asian 12.4 Asian 7.0
Median household income ($) 26114.38 Median household income ($) 34227.30 Median household income ($) 45013.43
Table 2.2. Characteristics of OHCA and NBCPR severity tier groupings

37
2.7. Discussion
This study used a spatiotemporal approach to classify census tracts based on OHCA incidence
and NBCPR severity and provides a detailed depiction of areas of high risk of cardiac arrest in
one of the largest cities in the United States.
Areas of the highest risk were observed seen by the large grouping of census tracts in the south,
central, and south-east areas of Los Angeles, illustrated in the dark red areas of Figure 2.4A.
Using Table 2.2 to explore the classified census tracts, there are interesting, but not wholly
unexpected, patterns of sociodemographic characteristics based on previous research (Sasson
2010; Root 2013; Wissenberg 2013).

Compared to city-wide averages, the census tracts in Tiers
1–3 had higher non-white populations and lower median incomes. Tier 1 had higher Black and
Hispanic populations, Tier 2 was predominantly Hispanic, and Tier 3 was most
disproportionately Black. The population dynamics of these areas reflect well-known
neighborhoods such as East Side and Boyle Heights in Tier 2, with Tier 3 representing areas in
South and Central Los Angeles such as Downtown, Hyde Park, and Florence, and Tier 1 largely
representing the confluence of these two areas.
Comparing the overall trends for Los Angeles to national rates, there was a similar mean
incidence and NBCPR. For example, at a national level, OHCA incidence for EMS-treated
events was 73.0 per 100,000, while Los Angeles had 53.0 per 100,000 events (Virani 2020).
High risk areas identified in this study had higher rates of OHCA and NBCPR compared to
national statistics (Virani 2020). Sociodemographically, the high-risk areas identified in this
analysis were, again, similar to high-risk areas identified in other studies and found to be stable

38
demographically when evaluating decadal census data in the most recent City of Los Angeles
demographic data (LA City Planning 2019).

This analysis provides a spatiotemporal blueprint upon which the foundation of bystander CPR
public health programming and future research can be improved, as research shows that targeted
intervention can lead to improved outcomes (Daya 2015; Fordyce 2017). Compared to spatial
analysis only, spatiotemporal analysis provides a greater amount of context for public health
professionals and EMS leadership to use toward focusing their efforts on the areas that need help
the most before moving to other areas of the city. OHCA research specialists might consider
using spatiotemporal analysis that combines different OHCA rates and proportions into a
singular spatiotemporal analytical framework for other areas around the United States or other
countries. The blueprint put forth in this paper could be combined with other spatiotemporal
analyses that target different elements of the chain of survival, such as AED placement. Existing
research has already examined the spatiotemporal optimization of AED placement (Sun 2018).
Ultimately, the spatiotemporal analysis in this study and in studies examining AED coverage
show that spatiotemporal analysis should be used to determine how public health practitioners
should intervene upon their communities. Furthermore, spatiotemporal blueprints laid out in this
paper and other studies could focus on other intransigent public health issues attended primarily
by front-line EMS providers, such as stoke and opioid overdose calls.

There are some limitations in this analysis. One of the major limitations stems from not using
hierarchical approaches to adjust the clustering performed in this analysis. There could be
confounders that are directly related to and impact the outcomes of the clustering procedures in
this analysis. However, the ability to implement adjustments is limited by the temporal aspects of

39
census tract-level population data. An additional limitation of this analysis is the small numbers
problem. Statistical smoothers helped with this issue but are not a perfect solution. Hence,
multiple rates, measures, and methods were combined into the severity tiers generated during
this analysis. Further, each analysis method and associated results that made up the tiers were
given equal weights. A more nuanced approach might involve weighting the results to better
capture risk. A fourth limitation comes from how bystander CPR is classified in this data. Since
it is field verified, bystander CPR might not be fully captured. However, this is a persistent data
artifact in many OHCA registries.

2.8. Conclusion
Using a novel three-tiered neighborhood risk classification tool, specific neighborhoods with
chronic or accelerating rates of OHCA and low bystander CPR were identified in the second
largest city in the U.S. Further study of bystander response and community-based public health
campaigns to increase bystander CPR are needed in these communities.

40
Chapter 3 Mapping Spatiotemporal Relative Risk of Non-Performance of
Bystander CPR Out-of-Hospital Cardiac Arrest and Community Public
Health Resource Capacity in Los Angeles
Most analysis of OHCA that involves spatial methods beyond geocoding and spatial aggregation
uses areal units like census tracts of counties. Non-areal unit analysis offers some distinct
advantages. The previous chapter used spatiotemporal methods focused on areal unit analysis.
This chapter focuses on spatiotemporal analysis that does not require areal units to calculate risk.
3.1. Introduction
Out-of-hospital cardiac arrests (OHCA) are a tremendous public health issue, with approximately
400,000 individuals suffering one every year (Rea 2003; Virani 2020). Using spatial analysis, a
multitude of studies has investigated how the location of an OHCA event and the
sociodemographic makeup and spatial organization of a community play a role in the provision
of bystander CPR. This led to the creation of many public health initiatives and educational
programs that are targeted at certain communities in various cities around the United States
(Sasson 2010; Root 2013).

Traditionally, studies examining OHCA and bystander CPR use census tracts or some other areal
unit as a representation of local communities, while incorporating OHCA incidence, bystander
CPR rates, and sociodemographic context into efforts to identify areas that are at risk (Beyer
2012; Nassel 2014; Sasson 2011 and 2012).

However, areal unit analysis has its limitations, chief
among them being the small numbers problem, the modifiable areal unit problem, and the limited
variability problem (Cebrecos 2016).

Each of these issues centers on the fact that selecting an
areal unit for choropleth mapping forces the exploration of disease rates by the selected areal

41
unit, which can result in more variability for the rates (i.e., small numbers problem), changing
the observed spatial problem (i.e., modifiable areal unit problem), or misrepresentation of the
actual spatial variation (i.e., limited variability problem) (Beyer 2012).

An avenue of spatial analysis that would better operationalize OHCA and bystander CPR
registry data is to build statistical surfaces that are not tied to one specific areal unit. Statistical
surfaces have been discussed as an ideal method for mapping health and disease risk, because
surfaces allow for applying areal units after rate calculation.

Spatial analysis is a critical methodological area for understanding and creating better OHCA
and bystander CPR outcomes; however, new spatial methods need to be demonstrated to gain
broader acceptance by public health practitioners. This paper seeks to introduce the aggregation
of annual spatial relative risk with temporal weighting, zonal analysis, and bivariate mapping to
better understand public health resource distribution.
3.2. Data
3.2.1. Data
3.2.1.1. Out-of-Hospital Cardiac Arrest (OHCA) Events
OHCA event data came from the City of Los Angeles Fire Department (LAFD) OHCA registry
and conformed to the 2015 Utstein guidelines (Perkins 2015). The registry uses data from LAFD
electronic health records (EHR), which contain the location of each OHCA and bystander CPR
status of the event. Paramedics and dispatchers serve as the primary data collectors for each
OHCA call, as they enter data for each event into the EHR, which then feeds into the LAFD
OHCA registry. Cases were identified from the electronic health record system if they included

42
any of the following criteria: provider impression cardiac arrest, provider impression respiratory
arrest, treatment CPR, rhythm pulseless VF/VT/PEA/asystole, or protocol non-traumatic cardiac
arrest. Cases were then manually reviewed to ensure that they represented OHCA with
resuscitation attempted. Reports meeting multiple criteria were included only once. Cases were
subsequently sorted according to the most recently published Utstein guidelines. Exclusion
criteria included events that were non-adult, occurred in nursing homes because, or were of
traumatic etiology. After the final case list was determined, the cases were geographically
displayed using latitude and longitude.
3.2.1.2. Population Data
Population data needed to be in a raster-based, or a gridded matrix of cells, format for this
analysis. Therefore, LandScan data was the source for population data as it offers high-quality
population estimates in raster format (Rose 2018). LandScan data are produced by the Oak Ridge
National Laboratory (Rose 2018). The modeling relies on multiple data inputs, such as census,
administrative boundary, landcover, other spatial, and satellite imagery data (Rose 2018).
Ultimately, LandScan modeling produces 24-hour ambient average population counts per 1 km
square area. These data were converted into population density according to Oak Ridge National
Laboratory recommendations (Rose 2018).

3.2.1.3. Community Public Health Resource Data
Data for community public health resource capacity included locations for organizations that
could potentially host or help coordinate CPR education. The organizations included were
community organizations and non-profits, churches, schools, LAFD stations, and community
health locations (e.g., federally qualified health centers, hospitals, public health offices).
Theoretically, the idea of including community organization density as a measure of capacity

43
stems from work by Robert Sampson and Pat Sharkey (Sampson 2012; Sharkey 2018). In their
respective works, both argued that the number of organizations per community plays an
important role in affecting positive change for that area. Thus, the community organization data
as of 2017 were sourced from both the City of Los Angeles geoportal and the County of Los
Angeles geoportal (County of Los Angeles 2020). Each location was displayed using latitude and
longitude. All locations were used for community public health resource capacity calculations;
they were not separated by organization type.

3.2.2. Study Population and Study Area
The City of Los Angeles has an estimated population of 4 million residents. This analysis
included adult (  18 years old) OHCA patients described above between January 1, 2011, and
December 31, 2019, who experienced a non-traumatic event within the City of Los Angeles
boundary and were serviced by the LAFD, regardless of bystander CPR status. Bystander CPR
was classified according to what was found in the field by paramedics.
3.3. Measures and Methods
3.3.1. NBCPR Risk
NBCPR risk was measured by calculating spatial relative risk surfaces for each year in the study
period (i.e., 2011-2019). Risk surfaces in this context refer to statistical surfaces created with a
specific spatial relative risk function in the sparr package in R (Davies 2017). The spatial relative
risk function relies on a modified kernel density estimator (KDE) that fits a mathematical surface
comprised of pixels that represent the relative risk of a specific area given the distribution of
cases and controls while adjusting for the underlying population (Davies 2017). It works by

44
calculating a log-normalized ratio of two KDE surfaces, one for cases and one for controls, for a
common study window (Davies 2017). Cases in this study were represented by OHCA where
bystander CPR did not occur, while controls represented OHCA events where CPR did occur.
The adaptive bandwidth option was selected to avoid over smoothing areas that are densely
populated with cases without compromising the statistical stability of areas that are less densely
populated (Davies 2017). The relative risk KDE function was run for each year in the study
period to develop the ability to temporally aggregate relative risk for the City of Los Angeles for
the entire time period.
3.3.2. Population Density
LandScan data came in the form of a gridded surface and represented the estimated number of
people in a given cell (Rose 2018). This does not represent population density. Density needs to
be calculated for the spatial relative risk to be estimated accurately by accounting for the
underlying population density distribution (Sutton 2003). To do this, the area of each cell had to
be calculated using guidance from the Oak Ridge National Laboratory (Rose 2018).

The
resulting population density surface was then incorporated into the OHCA and bystander CPR
relative risk calculations.

3.3.3. Community Public Health Resources
The community public health resource measure represents the aggregated count of locations
critical to community public health resource capacity. To convert community public health
resource capacity into mathematical surfaces, such as NBCPR relative risk surfaces, an adaptive
kernel KDE was used to estimate the density of community public health resource capacity
locations across the City of Los Angeles. The community public health resource capacity surface

45
was then combined with the final NBCPR surface. This process is detailed in the spatiotemporal
analytical workflow section.
3.3.4. Spatiotemporal Analytical Workflow
The purpose of this study was to combine multiple statistical surfaces using zonal analysis and
bivariate mapping. Figure 3.1 provides a graphic of the steps of the workflow.

46
Figure 3.1. Spatiotemporal analytical workflow
It started with creating a point population density surface using LandScan data, followed by
creating NBCPR relative risk surfaces that were adjusted for the underlying population (i.e.,
LandScan data) of the city of Los Angeles. The relative risk surfaces were calculated for each
year of the study using an adaptive kernel width. Next, the community public health resource
capacity was calculated using an adaptive bandwidth KDE. Both surfaces underwent uniform
edge correction.

Map algebra enables the combination of pixel-based, or raster surfaces (Longley 2005). Before
combining the NBCPR surfaces, the data were normalized on a scale of 0–100 to make them
comparable and more easily combinable. For an additional context, higher values in the
normalized scale represent higher amounts of relative risk for an OHCA not having bystander
CPR conducted. The map algebra operation relied on a temporally weighted average of the
various annual risk surfaces. The temporal weighting scheme was accomplished via a decreasing
geometric sequence where the starting weight was 1 and the common ratio between the terms of
the sequence was .5. Therefore, the 2019 NBCPR risk surface had a weight of 1, while 2018 had
a weight of .5 and 2017 had a weight of .25 and so on. The temporal weighting scheme was done
to ensure that more recent NBCPR relative risk surfaces had higher weights to more accurately
capture the current status of NBCPR risk in the City of Los Angeles. After the map algebra step
was performed, a final NBCPR relative risk surface that represented the cumulative
spatiotemporal relative risk was created.

47
3.3.5. Zonal Analysis
To better operationalize both statistical surfaces, a zonal analysis was conducted for multiple
areal units. Zonal analysis works by calculating statistics of raster data for specific areal units or
zones (Lonely 2005). LAFD battalion was the areal unit of choices. In this instance, all available
statistics were calculated for the pre-specified areas.

3.3.6. Bivariate Mapping
Ensuring that both the final spatiotemporal relative risk surfaces and the community capacity
surface are fully operationalized requires bivariate mapping. Bivariate mapping takes two
variables at an area level and maps them simultaneously (Biesecker 2020). In this instance, the
two area-level variables stem from the zonal analysis of the relative NBPCR-OHCA risk surface
and the community capacity surface. Furthermore, the bivariate mapping scheme was set up to
prioritize high–low values of NBCPR relative risk. The four available categories in the bivariate
mapping scheme were (1) high-high; (2) high-low; (3) low-high; (4) low-low. For each category,
NBCPR was categorized first, followed by community organization density. As an example, a
battalion with a high-high designation would indicate high levels of NBCPR risk and high
amounts of community capacity, indicating this area has high NBCPR risk, but also has the
capacity to host more bystander CPR education courses. Conversely, a high-low designation
would mean a battalion has high NBCPR risk but a lower capacity to host bystander CPR
classes. Lastly, the breaks for these bivariate designations were done using natural breaks and a 3
by 3 bivariate grid orientation according to the statistical distributions of each surface.

48
In summary, after creating the various data surfaces, a map algebra operation was used to
combine the various measurements of relative risk into a singular surface. An adaptive kernel
bandwidth operation was then performed to create a community public health resource capacity
surface. After the map algebra operations, the surfaces underwent zonal analysis to generate
descriptive statistics for LAFD battalions of Los Angeles. Lastly, those descriptive statistics,
specifically the average of all pixel values for an area, were displayed bivariately to highlight
areas where the NBCPR risk and community public health resource capacity are high, low, or
some mix of the two.
3.3.7. Analytical Software
Analysis was performed using R 3.5.3/RStudio 1.2.1335 and ArcGIS Pro 2.3. All visualizations
were done in ArcGIS Pro 2.3

(R Core Development Team 2018; Esri 2018).
3.4. Results
3.4.1. NBCPR Relative Risk
Figure 3.2 shows the spatiotemporal progression of the NBCPR relative risk for the City of Los
Angeles. All data values here show relative risk normalized on a 0–100 scale. Of note, the
highest concentrations of relative risk across all years were in the southern area of the city before
the extension down to the Port of Los Angeles. Another area that shows a consistently high
amount of relative risk is in the northeast area of the city. This area of the city was higher than
normal for 5 of the 9 years in the progression, whereas the southern part of the city was higher
than normal for 7 of the 9 years in the analysis.

49

Figure 3.2. Map series depicting the annual NBCPR relative risk from 2011–2019 on a normalized scale
of 0–100 for the City of Los Angeles

Figure 3.3 represents the weighted surface, which is a combination of all the annual surfaces. As
expected, the southern and northeastern areas of the city have higher-than-average amounts of
relative risk (i.e., greater than a score of 50 on the normalized scale). Additionally, there is a
band of moderately elevated relative risk connecting the area north of the Pacific Palisades to the

50
northeast area that has been mentioned previously. Another area of elevated risk is the Port of
Los Angeles, as it has elevated relative risk for 4 of the 9 years in the analysis.

Figure 3.3. Final combined spatiotemporal representation of NBCPR relative risk on a
normalized scale of 0–100 for the City of Los Angeles with LAFD battalions

51
3.4.2. Community Public Health Resource Capacity
Figure 3.4 illustrates the density per square kilometer of community public health resource
organizations ideal for CPR education in the city of Los Angeles. Of note, there is an intense
density of organizations in the central or downtown area of Los Angeles. There are also some
additional areas of higher density near Beverly Hills, Canoga Park, and the Hunting Park areas of
Los Angeles.

52
Figure 3.4. Community public health resource density on a normalized scale of 0–100 for the
City of Los Angeles based on data from the LA City Geohub and county-based GIS system

3.4.3. Zonal Analysis
Table 3.1 includes a summary of the zonal analysis of the relative risk and community
organization density surfaces for the LAFD battalions. The overall mean for NBCPR normalized
risk value was 46.5, whereas the overall mean for community organization density per square
kilometer was 7.9 organizations.
Table 3.1. NBCPR relative risk and community organization density for LAFD battalions

Normalized value
Spatiotemporal NBCPR relative risk surface characteristics
Lowest relative risk value 5.7
Highest relative risk value 100
Mean 46.5
Standard deviation 16.9
Community public health resource density surface characteristics
Lowest density 0
Highest density 96.8
Mean 7.9
Standard deviation 9.3

Table 3.2 categorizes the results of the LAFD battalion zonal analysis while providing
information on the mean values for each statistical surface. To be categorized as high in the
NBCPR relative risk, a battalion had to have a mean relative risk score on the normalized value
scale of 25.6 or greater, while a battalion needed to have a mean community organization density
greater than 5.5 to be considered high. High–High was the most frequent categorization with 6
battalions, while High–Low was the second most frequent categorization as there were 5
battalions in that category.

53

Table 3.2. Battalion categorization by statistical surface characteristics

Battalion
Number
Bivariate
Category
Mean NBCPR Relative
Risk
Mean Community Organization
Density
1 High–High 38.9 25.5
2 High–High 32.0 5.5
4 Low–Low 14.9 2.8
5 High–High 27.5 5.6
6 High–Low 38.5 4.8
9 High–Low 27.0 1.8
10 High–Low 25.9 2.6
11 High–High 30.5 32.4
12 High–Low 29.6 1.4
13 High–High 67.9 7.8
14 High–Low 30.0 4.7
15 Low–Low 18.6 2.3
17 Low–Low 14.6 3.7
18 High–High 25.6 10.1

3.4.4. Bivariate Mapping
Figure 3.5 is a map series that depicts the results of the zonal analysis and bivariate mapping. In
Map C of Figure 3.5, the bivariate classification shows that Battalions 1, 2, 5, 11, 13, and 18
skewed towards both high NBCPR risk and community organization density. Battalion 2 had the
lowest community organization density of the census tracts that skewed toward being high in
both categories. Battalions 1 and 11 had the greatest balance between relative risk and
community organization density. Battalions 6, 9, 10, 12 and 14 all have higher amounts of
NBCPR relative risk and lower amounts of community organizations. Only battalions 4, 15, and
17 were categorized as both low in NBCPR risk and community public health resource density.

54

B
C
A
Figure 3.5. Map series showing the NBCPR surface (A), community public health resource density (B), and bivariate mapping
based on zonal analysis for the LAFD battalions for the City of Los Angeles (C)

55
3.5. Discussion
This paper illustrates an application of spatiotemporal methods to better understand NBCPR
relative risk using statistical surfaces, zonal analysis, and bivariate mapping to better understand
relative risk and public health resource distribution.

Figures 3.2 and 3.3 provide helpful insights about the spatiotemporal relative risk of NBCPR in
Los Angeles. Specifically, the southern and northeastern areas of the city have an insidious
public health challenge, as those are areas of high risk across time and space. Figures 3.3 through
3.5 and Table 3.2 highlight the results of operationalizing the spatiotemporal analysis and public
health resources through zonal analysis and bivariate mapping. These maps and the table
demonstrate an uneven distribution of both NBCPR relative risk and public health resources.
Namely, that community public health resources tend to be located where populations with more
needs are in a study area. In this instance, the community public health resources are
concentrated in the battalions that make up the downtown, south, and central areas of Los
Angeles where areas like skid row are located.

Public health practitioners, specifically OHCA and bystander CPR analysts, should consider
using methods not tied to areal units. Spatiotemporal aggregation of kernel density surfaces and
zonal analysis provide benefits over traditional areal unit analysis. One advantage is that kernel
density mapping, prior to zonal analysis, provides a more accurate representation of the urban
environment because it offers smooth transitions that are more representative of an actual
environment, or in other words, the surfaces avoid the limited variability problem (Carlos 2010;
Charreire 2010). Furthermore, the surfaces are not constrained by the small number problem

56
because the space is factored directly into the rate calculation as disease rates are conceptualized
to occur continuously over space instead of in a specific areal unit. Zonal analysis coupled with
KDE surfaces supplies another benefit over traditional areal unit analysis in that the KDE
surfaces can be attributed to multiple types of areal units efficiently. In this paper, we illustrate
the benefits of KDE mapping and performing zonal analysis to map to meaningful administrative
boundaries, such as the LAFD battalions. Analysts using these techniques can provide more
flexible and accurate information to decision makers.

Bivariate choropleth mapping is another underutilized cartographic technique compared to other
cartographic symbology, and public health practitioners should use this symbology more.
Bivariate mapping offers advantages such as enabling the simultaneous representation of two
variables. There are methods such as combining variables of interest into a single representative
synthetic index; however, this only works when the data has high dimensionality. A prime
example of this would be creating a social deprivation index (Messer 2006). Specifically, for this
study, bivariate mapping was a better choice compared to creating a synthetic index, given that
only two variables were examined. Furthermore, without bivariate mapping, the results could be
misinterpreted. For example, an area with a high amount of spatial relative risk but low
community organization density could score as “medium” or “high” in a synthetic index
representing both variables. This would be misleading, as the area only has high risk and not
high community public health resources.

Lastly, the methods in this paper are not limited to NBCPR and community public health
resource capacity. Bivariate mapping has many uses across multiple disciplines (Strode 2020).

57
Utilizing new methods that provide flexibility with respect to operationalizing the spatial data
provides policy makers, public health, and academic professionals with toolsets that provide
accurate and useful data. These data and studies can then be turned into more highly tailored
interventions, which could lead to a reduction in public health burdens, as evidenced by other
studies (MacQuillan 2017; McLuckie 2019).

There are multiple limitations associated with this analysis. Notably, the measurement of
community public health resource capacity is limited. Amassing organizations and calculating
spatial density do not fully capture such a complex concept. However, calculating density does
provide at least a measurement of the concentration of both physical space and organizations that
have vested interest in seeing communities in Los Angeles thrive. Perhaps more importantly is
using bystander CPR education as an additional measure of interest. There have been some
studies that call into the question the effectiveness of traditional bystander CPR education, while
others have shown modest improvements in both bystander CPR rates and associated
survivorship from an OHCA because of these programs. AED placement could be an additional
measure of interest that would fit with this methodology; it has been used in the past in some
spatial optimization location studies.
3.6. Conclusion
This paper highlights methods that provide the opportunity to measure the spatial relative risk of
NBCPR and community CPR public health resource capacity that is not tied to any specific areal
unit. Through zonal analysis and bivariate mapping, this paper provides a framework of
flexibility that allows the opportunity to operationalize the data to different areal units to better

58
tailor intervention efforts. Combining the methods in this paper provides a path for other urban
health researchers and analysts to tackle any number of challenging public health issues.

59

Chapter 4 Impact of COVID-19 on Out-of-Hospital Cardiac Arrest and
Bystander CPR in Los Angeles
The COVID-19 pandemic had many direct and indirect impacts on health in communities across
the globe. Understanding how COVID-19 impacted OHCA and NBCPR provides an opportunity
to better understand these impacts. This paper examines whether there was a differential change
in OHCA incidence and NBCPR during the COVID-19 pandemic.
4.1. Introduction
The COVID-19 pandemic, caused by the novel coronavirus SARS-CoV-2, can cause direct and
great stress on the respiratory and cardiovascular systems of humans, potentially resulting in
death. While the direct impacts of the virus and pandemic have been severe, indirect impacts are
also important to consider. Since the SARS-CoV-2 can cause respiratory and cardiac stress, it
should be expected that as a result of the pandemic, there would be an increase in out-of-hospital
cardiac arrests, less bystander CPR for fear of contracting the disease, and slower EMS response
as a pandemic strains all aspects of the healthcare system. Current research shows inconsistent
impacts of the COVID-19 pandemic on OHCA incidence, amount of bystander CPR, and EMS
response in both United States and European cities (Scquizzato 2020; Uy-Evanado 2021).

Studies of OHCA and bystander CPR have resulted in greater understanding of this public health
issue and improvement of related outcomes, such as survivorship with good neurological
outcome (Morrison 2008; Root 2014; Nichol 2008; McNally 2010; Daya 2015; Mitchell 2009;
Sasson 2010).

This research was enabled by OHCA registries, based on Utstein guidelines,
which collect data on various aspects of the chain of survival at different scales, from city- to

60
regional- and national-based evaluations (Perkins 2015).

However, additional research is needed
and should include broader research questions that further flesh out our understanding of the
OHCA and bystander CPR paradigm while also furthering our understanding of the impact of the
COVID-19 pandemic. Examining the indirect impacts of the COVID-19 pandemic on OHCA,
bystander CPR, and EMS response is one way to do this.

The results of this study provide additional understanding to those who study OHCA and
bystander CPR risk by using the lens of the COVID-19 pandemic.
4.2. Study Area, Periods, and Population
4.2.1. Study Area and Periods
The City of Los Angeles has an estimated population of 4 million residents (California
Department of Finance 2020). This analysis included adult (  18 years old) OHCA events that
were EMS attended and EMS treated for two different time periods. The first period examined
was between January 1, 2017 through December 31, 2019. This represents pre-pandemic. The
second period included OHCA events from January 1, 2020 through December 31, 2020, and
represents the first year of the COVID-19 pandemic.
4.2.2. Study Population
OHCA events were defined as non-traumatic events within the City of Los Angeles boundaries
that were serviced by the LAFD, regardless of bystander CPR status. Bystander CPR was
classified according to what was found in the field by the paramedics providing care to the event.
Specifically, OHCA events were classified according to the following criteria: provider
impression cardiac arrest, provider impression respiratory arrest, treatment CPR, rhythm

61
pulseless VF/VT/PEA/asystole, or protocol non-traumatic cardiac arrest. Cases were then
manually reviewed to ensure that they represented OHCA with resuscitation attempted. Cases
meeting multiple criteria were only included once. Additional exclusion criteria, outside of non-
adult events or events with a traumatic etiology, included events that occurred in nursing homes,
or occurred in non-residential, special use census tracts such as Los Angeles International
Airport.

Census tracts were used as the spatial unit of analysis. Census tracts were removed from the
analysis if their residential population counts were extremely low, even though their OHCA
counts and event rates were high (e.g., LAX). Other abnormal census tracts, such as the Port of
Los Angeles, were included due to higher and more traditional residential population dynamics.
4.3. Data and Measures
4.3.1. Out-of-Hospital Cardiac Arrest (OHCA) Events
OHCA event data came from the City of Los Angeles Fire Department (LAFD) OHCA registry
and conformed to the 2015 Utstein guidelines (Perkins 2015). The registry uses data from LAFD
electronic health records (EHR), which contain the location of each OHCA and bystander CPR
status of the event. Paramedics and dispatchers serve as the primary data collectors for each
OHCA call, as they enter data for each event into the EHR, which then feeds into the LAFD
OHCA registry. Cases were subsequently sorted according to the most recently published Utstein
guidelines. After the final case list was determined, the cases were geographically displayed
using latitude and longitude.

62
4.3.2. OHCA Incidence and NBCPR Event Rate
In this analysis, there were two different outcome variables. OHCA incidence per 100,000
individuals was one of the outcome measures. OHCA incidence was calculated on a per census
tract basis per each time period. NBCPR Event Rate was the second outcome measure. This
value represents the proportion of OHCA events in a census tract where non-performance of
Bystander CPR occurred. Similarly, to OHCA incidence, these values were calculated on a per
census tract basis per time period in this study.
4.3.3. OHCA Incidence and NBCPR Event Rate Severity Tiers
Census tracts included in this analysis were categorized into severity tiers that represent the
spatiotemporal patterns of OHCA incidence and NBCPR event rates (Fleming 2021). Tracts
were classified according to a methodology that uses multiple spatiotemporal analyses (e.g.,
spatial variation in temporal trends and spatiotemporal hotspot analysis). There were three
severity tiers that included 182 census tracts out of the 895 in the analytical sample. Tier 1 was
considered the most severe or at risk grouping, and tier 3 was less severe than tiers 1 and 2 but
still significantly worse than the rest of the city. Other tracts in this analysis that were not
categorized are referred to as non-tier categorized tracts.
4.4. Methods
This study employed spatial and aspatial analysis using LAFD OHCA data with the ultimate goal
of understanding how the COVID-19 pandemic impacted the amount of OHCA and the
provision of bystander CPR for OHCA in Los Angeles.

This analysis started with a comprehensive evaluation of OHCA and bystander CPR patterns
before and during the first year of the COVID-19 pandemic. The pre-pandemic period was

63
defined as an average of OHCA incidence and NBCPR event rates from January 2017 through
December 2019, while the period representing the pandemic was the entire calendar year. This
period was selected because it fully encapsulates the potential period for the first major spike of
the pandemic followed by the lower rates of COVID-19 before higher rates were observed
starting in November 2020.
The initial step of the analysis was to examine the differences between the period representing
the average NBCPR event and OHCA incidence rates. This was done by calculating simple
differences on a per census tract basis. These data were mapped for visual comparison purposes.
Next, fixed effect regression models were implemented where each census tract served as its
own control group (e.g., average of 2017–2019) and treatment group (e.g., 2020) to determine if
the COVID-19 pandemic and associated independent variables predicted OHCA incidence or
NBCPR event rates. The tract-fixed effects model was selected because it negates all time-
invariant differences between the census tracts in the study area. Essentially, this analysis sought
to compare similar populations by assuming that all relevant populations were stable. The
regression equation for this part of the analysis took the form of:
𝑦 𝑡𝑝
= 𝛽 1
𝑃 𝑝 + 𝛾 𝑡
𝑦 𝑡𝑝
represents the dependent variable in tract t during time period p. 𝑃 𝑝 represents a dummy
variable for the two specific time periods. 0 was for the control time period, while 1 represented
the treatment period. The remaining term represents tract fixed effects. Furthermore, the dummy
variable representing the time period can be interacted with other independent variables to better
understand whether changes over time vary according to neighborhood characteristics measured.

64
In this instance, the period variable was subsequently interacted with variables representing
severity tier categorizations that capture census tracts at high risk of OHCA and Non-
performance of Bystander CPR, as described in previous works and in this dissertation (Sasson
2012; Root 2014.
An analysis of the residuals in the fixed effects models was evaluated for spatial autocorrelation.
No spatial autocorrelation was detected in the residuals. Therefore, no further modeling was
required.
4.4.1. Analytical Software
Analysis was performed using R 3.5.3/RStudio 1.2.1335 while all map visualizations utilized
ArcGIS Pro 2.3 (R Core Development Team 2018; Esri 2018).

4.5. Results
4.5.1. OHCA Incidence and NBCPR Event Rates
Figure 4.1 shows the OHCA incidence (per 100,000) by month by tract for 2017 through 2019
(non-COVID period) and the 2020 (COVID period) OHCA incidence results per month.

65

Figure 4.1. OHCA incidence rate comparison between the average event rate of 2017–2019 and
the incidence rate of 2020

The mean incidences of OHCA for 2017–2019 and 2020 were 54.5 per 100,000 and 70.2 per
100,000, respectively. OHCA incidence in 2020 was greater than the 2017–2019 average OHCA
incidence for 11 of the 12 months used in this study. The only month in 2020 that was lower than
the 2017–2019 average was January, which was prior to the advent of the COVID-19 pandemic.
Furthermore, in Figure 4.1, the peaks of the 2020 trend line appear to roughly mirror spikes in
the COVID-19 pandemic in Los Angeles, which were March, July, and December 2020.
Figure 4.2 displays the NBCPR event rate in 2017–2019 and for 2020, as in Figure 4.1.

66

Figure 4.2. NBCPR event rate comparison between the average NBCPR event rate of 2017–2019
and the event rate of 2020

The mean NBCPR event rates for the two study periods were 43.8% for 2017-2019 and 51% for
2020. Ten of the twelve months in 2020 had higher NBCPR event rates compared to the 2017–
2019 average. Similarly, spikes in the NBCPR event rate appeared to mirror spikes in the
COVID-19 pandemic in Los Angeles. Table 4.1 provides additional OHCA and NBCPR event
rate descriptive statistics for the City of Los Angeles.

Table 4.1. Descriptive characteristics of OHCA incidence and NBCPR event rate for 2017–2019
and 2020 by tract
OHCA incidence characteristics (per 100,000)

67
2017 51.7
2018 60.2
2019 65.6
Mean 2017–2019 54.5
2020 70.2
Percent change in incidence between 2020 and 2017–2019 (%) 28.8
NBCPR event rate characteristics (%)
2017 44.4
2018 44.6
2019 42.3
Mean NBCPR Event Rate for 2017–2019 43.8
2020 51.0
Percent Change in NBCPR Event Rate between 2020 and 2017–2019 (%) 16.4
OHCA incidence characteristics (per 100,000) by high-risk tiers
Tier 1 2017–2019 mean 71.4
Tier 1 2020 84.2
Tier 2 2017–2019 mean 61.8
Tier 2 2020 85.9
Tier 3 2017–2019 mean 83.8
Tier 3 2020 88.2
City 2017–2019 mean 54.5
City 2020 Mean 70.2
NBCPR event rate characteristics (%) by high-risk tiers
Tier 1 2017–2019 mean 54.9
Tier 1 2020 62.8
Tier 2 2017–2019 mean 53.2
Tier 2 2020 62.3
Tier 3 2017–2019 mean 50.0
Tier 3 2020 47.9
City 2017–2019 mean 43.8
City 2020 mean 51.0

4.5.2. Spatial Analysis of OHCA Incidence and NBCPR Event Rate
Figures 4.3a and 4.3b illustrate OHCA incidence in 2017-2019 and 2020, respectively, and
Figure 4.3c illustrates change in OHCA incidence. Tracts that had a negative percent change are
blue, while tracts that had a positive percent change are shaded with red.

68

Figure 4.3. Map series depicting OHCA incidence by census tract for 2017 to 2019 (A), 2020 (B), percent change between the two
periods (C)

A

B C

69

Figure 4.4 Map series depicting NBCPR event rate by census tract for 2017 to 2019 (A), 2020 (B), percent change between the two
periods (C)

A

B C

70

Figures 4.4a–4.4c provide the same view but focus on NBCPR event rates. There were 508
census tracts where the difference in OHCA incidence was positive or 52%, while there was a
total of 461 census tracts or 47% that had a positive difference in the NBCPR event rate between
the two periods.
4.5.3. Fixed Effects Regression Analysis
Table 4.2 displays the results of the fixed-effects modeling on OHCA incidence rates, while
Table 4.3 includes the results of the fixed-effects modeling on NBCPR event rates. In total, there
were four fixed effects models were run, two per each dependent variable.
Table 4.2. OHCA Incidence Fixed Effect Modeling Results
Model 1 Model 2
Period 8.7*** 8.2***
Period × Tier 1 tracts

4.7
Period × Tier 2 tracts
15.9
^

Period × Tier 3 tracts -3.8

N = 1970 for both analyses; Two-tailed significance tests *** p≤ 0.001, ** p≤0.01, *p≤ 0.05, ^ p≤ 0.1

Table 4.3. NBCPR-Event Rate Fixed Effect Modelling Results
Model 1 Model 2
Period 7.1*** 8.5***
Period × Tier 1 tracts

-.6
Period × Tier 2 tracts .6
Period × Tier 3 tracts -10.5*

N = 1970 for both analyses; Two-tailed significance tests *** p≤ 0.001, ** p≤0.01, *p≤ 0.05, ^ p≤ 0.1

Table 4.2 highlights the OHCA incidence fixed effects modeling results. On average, there were
8.7 more events per 100,000 people in 2020 compared to the 2017–2019 average OHCA

71
incidence. This supports the visual exploratory analysis provided in Figure 4.2. In model 2, the
main effect of 8.2 indicates that non-tier categorized tracts experienced an increase of 8.2 more
OHCA events per 100,000 people in 2020 compared to 2017-19. There was a significant
interaction between period and tier only for tier 2 tracts; those tracts had 24.1 (8.2 + 15.9) more
OHCA events per 100,000 people (p <.1) in 2020 compared to the average OHCA incidence
from 2017 through 2019. Tier 1 and Tier 3 tracts’ increase was not significantly different than
non-tier categorized tracts

For the NBCPR event rate, it was found that, on average, the event rate was 7.1 percentage
points greater per census tract in 2020 (p < .001). This indicates that more OHCA had no
bystander CPR performed for EMS-attended and treated events during the COVID-19 pandemic.
Breaking out the high-risk census tracts by tier, model 2 shows that only the tier 3 census tracts
had a statistically significant (p <.05) interaction. The interaction was negative, implying that
more events in this tier had bystander CPR performed during the COVID-19 pandemic compared
to non-tier categorized tracts, which is also backed up by the data in Table 4.1. Specifically, the
NBCPR rate declined by 2 percentage points (8.5-10.5) in tier 3 tracts from pre- to post-
pandemic, compared to the 8.5 percentage point increase in NBCPR in non-tier categorized
tracts. Ultimately, there was significant change in the non-performance of bystander CPR for a
majority of Los Angeles, with only a subset of high-risk areas experiencing a decrease in the
non-performance of bystander or, also stated as an increase in the performance of bystander CPR
between the two periods.

72
4.6. Discussion
The analysis sought to understand the impact of the COVID-19 pandemic on OHCA and
NBCPR excess risk in Los Angeles. In summary, in Los Angeles for the periods considered, the
COVID-19 pandemic resulted in more OHCA events, higher number of NBCPR events, and
changes to how the EMS system was activated. From a spatial analysis perspective, there was
significant geographic clustering only in the difference between the average OHCA incidence of
2017–2019 and the 2020 periods.

The fixed-effects models confirm that there were statistically significant differences resulting in
increases in the incidence of OHCA and NBCPR event rates across the two periods. However,
unexpectedly, the high-risk census tracts were not consistently worse during the COVID-19
pandemic. Only tier 2 census tracts had even more OHCA than non-tier categorized tracts, while
tier 3 census tracts had better results from an NBCPR perspective (i.e., fewer NBCPR events
than non-tier categorized tracts). Looking at the average incidence and NBCPR event rates for
2017–2019, the high-risk census tracts, across the board, had more OHCA events and more
events where NBCPR occurred; however, the tier 3 grouping since they were 2 percentage points
fewer as it relates to NBCPR OHCA events compared to non-tier categorized tracts.

There are many possible explanations for the observed relationships in spatial analysis and fixed-
effect models. First, given the nature of the COVID-19 pandemic, analyses were expected to
statistically confirm increases in the OHCA incidence that occurred during the pandemic. From
an OHCA incidence perspective, COVID-19 greatly impairs the cardiorespiratory health of
individuals, making it more likely that OHCA events would occur. This might also help explain

73
the observation that there were more EMS-attended, not-treated OHCA events in 2020 compared
to previous years. An increase in NBCPR event rates could also be expected, as previous
research has shown that the risk of contracting a disease while performing bystander CPR can be
a mitigating factor for some bystanders (Institute of Medicine 2015). However, the results of the
fixed-effect modeling by high-risk tier were somewhat surprising. The COVID-19 pandemic in
Los Angeles impacted both non-tier categorized, and tier categorized tracts relatively equally
given the amount of non-significant interactions. Only tier 2 had more OHCA when compared to
the non-tiered categorized tracts, while tier 3 saw a decline in NBCPR. Another possible
explanation for these findings is that the COVID-19 pandemic created large scale changes in
workplace and residential dynamics. Specifically, more people were relatively isolated and at
home instead of at work or in public where there would be potentially more formal.

The information gleaned from this analysis is useful in multiple ways. First, this analysis
highlights the importance of understanding the indirect impacts of the COVID-19 pandemic on
public health and other issues. Emergency situations, unfortunately, provide an opportunity to
study how our institutions work. Conducting studies of our institution by examining the indirect
effects of the COVID-19 pandemic provides an opportunity to create improvements. For
example, highlighting indirect impacts such as OHCA incidence supports public health
policymakers and professionals in their efforts to increase community resiliency and hire more
community health workers. Community resiliency and community health workers have been
shown to improve health outcomes across a broad range of health issues during emergency
situations (Nicholls 2015). Secondly, the evidence provides an opportunity to change bystander
CPR education modalities. Bystander CPR education could switch largely to virtual instruction

74
and reach a potentially larger audience, which has been shown to be effective (Wong 2018). The
results of this study support this because they highlight the fact that individuals being located at
their residents during an OHCA can still result in high amounts of non-performance of bystander
CPR. Increasing the reach of CPR education through virtual platforms could result in more
people being educated and potentially prepared to perform CPR, regardless of their location.

This study has several limitations. First, this study only examined group-level variables of
OHCA and NBCPR and did not seek to understand how individual- or patient-level factors
interacted with neighborhood-level factors such as age distribution and race or ethnicity
composition. Previous research has shown significant interactions between patient-level
characteristics and neighborhood-level characteristics, specifically as it relates to the
performance of bystander CPR (Sasson 2012). A second limitation is that bystander CPR is field
verified in this dataset, but with the advent of COVID-19, field verification of bystander CPR
was more difficult to accomplish given the strain the pandemic placed on LAFD. Lastly, the
rates used in this analysis covered a limited timeframe, which could result in the small number
problem given that these rates were for entire census tracts. Small numbers have been shown to
result in less reliable statistical estimates.
32
4.7. Conclusion
This analysis used a variety of analytical techniques. Spatially, there was some geographic
patterning of the differences between the OHCA incidence and the NBCPR event rate for the
two different time periods. However, these areas largely did not overlap with areas previously
identified as being high risk. The regression modeling in this analysis used fixed effects to
determine whether there were statistically significant differences. These models also confirmed

75
that high-risk areas either remained at the same level as before or saw decreases, while the rest of
Los Angeles saw increases in OHCA incidence and NBCPR event rates. This study provides
insight into this process through a combination of spatial and non-spatial analytical techniques.

76

Chapter 5 A Tutorial for Replicating Spatial and Spatiotemporal Analysis of
OHCA and Bystander CPR in the City of Los Angeles and Beyond
Replicating research is of paramount importance in academia and in ensuring that the research is
impactful outside of the setting in which it was conducted. Therefore, this chapter focuses on
providing a tutorial for the first of the two analytical chapters of this dissertation.
5.1. Introduction
Spatial analysis of out-of-hospital cardiac arrests (OHCA) and bystander CPR is a well-
established way to identify high-risk areas that are ready for intervention, while spatiotemporal
analysis of OHCA and bystander CPR is less well established in OHCA and bystander CPR
research. A majority of spatial analysis of OHCA and bystander CPR uses spatial aggregation of
geocoded point data to areal units, such as census tracts. However, there are additional methods
that do not require areal units, such as spatial relative risk functions. Each type of spatial analysis
offers advantages and disadvantages that should be carefully considered when examining OHCA
and bystander CPR in a given study area.

The primary goals of this tutorial are twofold. The first is to provide, in a single document,
spatiotemporal analytical workflows for areal and non-areal analysis of OHCA and bystander
CPR. The second is to improve the accessibility of specialized spatial methods in areal and non-
areal analysis that are present in R for all individuals interested in analyzing OHCA and
bystander CPR.

77
5.2. Tutorial Structure
The structure of this tutorial focuses on two larger sections that walk the user through both the
areal unit analysis and non-areal unit analysis presented within this dissertation. Within each of
those sections, there is a discussion of the overall analytical workflow, followed by instructions
and code chunks. The code chunks will only be examples intended to demonstrate how to
implement the analytical workflow in R code. An example code break is shown below.

> # Comment providing context for what the next line of code does
> Line of R code performing the action that the comment describes
5.3. Setting Up the RStudio Session and Rmarkdown file
R is a statistical programming language and free software maintained by the R Foundation.
RStudio is an interactive development environment that comes with a set of tools designed to
make working in and with R more productive. Both R and RStudio are required to complete this
tutorial.

The next step after installing R and RStudio is to set up a Rmarkdown file. Rmarkdown is a
document that enables the creation of dynamic reports from R code. Critical to Rmarkdown files
are code chunks. Below is an example code chunk for an R markdown file:

```{r set WD}
> # Code to set the working directory
> setwd(“workingdirectoylocation “)
```

78

The start of a code chunk is ```{r description of what the code chunk is doing}, while the code
chunk is closed out by ```. This code chunk example is setting up a working directory for the
Rmarkdown file so that data can be imported and exported to the same place easily. It also sets
up the location of the final report document generated. Standard practice for the working
directory is to set up a file folder location somewhere on the local drive on the computer on
which the analysis is being completed. All code chunks within this tutorial should be set up
within a code chunk.
5.4. R Packages
R packages are groups of functions and analytical tools that complete more advanced data
handling and analysis. R packages in this tutorial include:

• sf – simple features is the main spatial data handling package for this tutorial. It provides
the ability to work with both vector (areal unit analysis) and raster (non-areal unit
analysis) data models.

• spdep – spatial dependence: weighting schemes and statistics provide the ability to
calculate spatial weights matrices and other formula critical for calculating the different
rates in the areal-unit analysis portion of this tutorial.

• tidyverse – tidyverse is a set of packages that provide the ability to work with data using
a consistent set of rules and language.

79
• sparr – spatial and spatiotemporal relative risk provides the ability to calculate the spatial
relative risk surfaces in the non-areal unit analysis part of this tutorial. It also enables the
selection of different bandwidth methods, for example adaptive bandwidth vs. fixed
bandwidth.

• rgdal – R bindings for the geospatial data abstraction library, which provides the ability
to project spatial data if necessary. This package is important for mapping data and
results correctly.

• raster – geographic data analysis and modeling with raster data. The raster package is
critical for non-areal unit analysis. Specifically, it helps convert the relative risk surfaces
created in sparr to a more usable and format. Package sp is required for this package to
work.

• tmap – thematic maps package works similarly, from a syntax perspective, to ggplot2.
This package makes it easy to create dynamic, thematic maps that are layer based. It can
also handle both vector- and raster-based data.

Installing and loading R packages requires two steps. To do this, the user must use the prompts
below:

> # Code to install a specific package
> install.packages(“sf”)

80
> # Calling/loading the package into the R Session
> library(“sf”)

This code block should be repeated for each package listed in this section. Remember, the code
break above this section should be in a code chunk designated to install and load packages into
the R session within RStudio.
5.5. Areal Unit Analysis
The areal methods in this tutorial center on calculating rates of OHCA and bystander CPR across
time and using the local Gi* statistic, or hotspot statistics, by year to determine areas that have
high or low OHCA and bystander CPR risk across time. To accomplish both of these tasks,
OHCA, and population data must be attributed to the areal unit of choice to fully calculate
population adjusted rates. Examples in this tutorial rely on census tracts for the areal unit, as they
are the most prevalent areal unit used in OHCA and bystander CPR research**.
5.5.1. Analytical Workflow:
Figure 5.1 provides a visual of the analytical workflow for the areal unit analysis. The first
portion of this workflow requires spatially joining OHCA event data that has been bifurcated by
bystander CPR status to census tracts within the boundaries of the Los Angeles. After joining
OHCA data to census tracts, the non-performance of bystander CPR and bystander CPR events
must be aggregated into a single count column before the various rates can be calculated. The
next major portion of the workflow involves calculating spatially smoothed OHCA incidence
and non-performance of bystander CPR (NBCPR) event rates. The final parts of the workflow
require implementing the GI* clustering algorithm on an entire study period and annual basis to

81
understand the potential spatial and spatiotemporal patterning of OHCA incidence and the non-
performance of bystander CPR.

Figure 5.1. Areal Unit Analysis Workflow
5.5.2. R Packages Used for this Analysis
• sf
• spdep
• tidyverse
• rgdal

82
• tmap
5.5.3. Load OHCA and Census Tract Data, Conduct Spatial Join, and Aggregate OHCA Data
The first step in the areal unit analysis requires loading the areal unit data into R via a simple line
of code:

> analysistracts <- st_read( “analysistracts.shp ”)

This reads a census tracts shapefile of a study area into R and stores it as a simple features or sf
dataframe.

Next, working on the raw .csv OHCA data, an analyst needs to transform those data into an sf
dataframe.

(1) Example of loading the .csv file into R and creating a simple features dataframe
> # Code to create a dataframe from a .csv file. The name of
the dataframe needs to be specific to the year and whether
cases or controls are being uploaded
> ohca2011nbcpr <- read.csv( “ohca2011nbcpr.csv ”)

> # Code to create a crs object that defines the coordinate
reference system of the census tract layer into a single
object. This will help create the simple feature dataframe
for the OHCA data

83
> crsobject <- st_crs(analysistracts)

> # Code to create a simple features dataframe. Note - the
“coords = c( “long ”, “lat ”) ” portion of the code below
requires an exact name match for the columns that include
the longitude and latitude data for the OHCA .csv file
> ohca2011nbcprshp <- st_as_sf(ohca2011nbcpr, coords =
c( “long ”, “lat ”), crs = crsobject)

After loading both the census tract data and the OHCA data, the OHCA data need to be spatially
joined and a count of OHCA by bystander CPR status created. St_join is the sf function that
forms the core portion of data setup prior to conducting analysis, such as calculating incidence of
bystander CPR event rates. The code for conducting the spatial join and counting NBCPR events
is below. This code will need to be repeated for the bcpr-ohca events.

> # This is where 2011 NBCPR events sf dataframe are joined
to the analysistracts dataframe
> nbcpr11intract <- st_join(ohca2011nbcpr, analysistracts,
join = st_within)
> # This code uses tidyverse language to create a new sf
dataframe that only includes the census tracts where there
were NBPCR-OHCA events. This results in a count of events
per census tract.

84
> nbcpr11tractcount <- count(as_tibble(nbcpr11intract),
tract) %>%
> print()
> # This code joins the new event count per census tract
dataframe to the original analysistracts dataframe
> analysistracts <- left_join(analysistracts,
nbcpr11tractcount)
> # This line of code renames the “n ” column from the
nbcpr11tractcount dataframe to nbcpr2011count
> analysistracts <- analysistracts %>% rename(nbcpr2011count
= n)
> # This line of code takes any “na ” values, or tracts that
had no NBCPR events occurring within them and replaces the
“na ” value with 0.
> analysistracts$nbcpr2011count[is.na(analysistracts$nbcpr201
1count)]<-0

The code chunks above that focus on OHCA data will need to be repeated for events where
bystander CPR occurred. That code was left out of the code chunks to ensure that the tutorial was
concise.

85
After importing both the non-performance of bystander CPR and bystander CPR events, those
two new columns need to be totaled for the entire year. See the code below on how to
accomplish this.

> # Creating a new variable in analysistracts called total1l
that equals the total number of events for 2011 per census
tract.
> analysistracts$total11 <- analysistracts$bcpr2011count +
analysistracts$nbcpr2011count

Again, like all other codes in this tutorial, this will need to be repeated for every year.
Ultimately, all the new columns (e.g., nbcprYYYYcount, bcprYYYYcount, and totalYY) will
need to be combined to create three new columns that aggregate the data. Those new columns
are totalnbcpr, totalbcpr, and totalevents. An example of this code is below.

> ## This line of code creates a column that contains a count
of all NBCPR events per census tracts for the entire study
period. This needs to be repeated for BCPR events.
> analysistracts$totalnbcpr <- analysistracts$nbcpr2011count
+ analysistracts$nbcpr2012count +
analysistracts$nbcpr2013count +
analysistracts$nbcpr2014count +
analysistracts$nbcpr2015count +

86
analysistracts$nbcpr2016count +
analysistracts$nbcpr2017count +
analysistracts$nbcpr2011count +
analysistracts$nbcpr2012count

Similar code should be used to create a totalbcpr variable within the analysistracts dataframe.
From there, the totalnbcpr and totalncpr variables can be combined to create a totalevents
variable.
5.5.4. Preparing for OHCA Incidence and Non-performance of Bystander CPR Event Rate
Calculations
The last two tasks to accomplish before moving onto calculating incidence and non-performance
of bystander CPR event rate calculations. Those tasks are creating an incidence population count
for each census tract and creating the spatial weights matrix required to generate the smoothed
incidence and NBCPR event rate calculations.

Calculating the incidence population for each census tract is quite simple. To do this, the 2010
census population total, or the most recent decadal census count, needs to be multiplied by the
number of years in the study period—in this instance that number is nine. This is accomplished
by using the following code:

> # Code to create the “incpop ” variable. This is only used
when a rate for the entire study period is being calculated
> analysistracts$incpop <- analysistracts$pop2010*9

87

The final step before calculating the spatially smoothed rates requires creating a spatial weights
matrix. In this analysis, contiguity-based spatial weights are used. Contiguity means that two
areal units share a common border. In this analysis, we used a first-order queen criterion to create
the spatial weights matrix. This indicates that neighbors were categorized as such if they shared a
common vertex, not just a common edge. The first-order queen criterion was selected because it
provides more accurate estimates when applying spatial smoothing to various rates based on
small numbers, since it results in more neighbors being a part of the smoothing process. Below is
the code required to create the spatial weight matrix.

> # Create queen ’s contiguity spatial weights matix
> st_queen <- function(a, b = a) st_relate(a, b, pattern =
“F***T**** ”)
> sf.sgbp.queen <- st_queen(analysistracts)

> # Creating function to assign stored attributes of the
queen ’s spatial weight matrix
> as.nb.sgbp <- function(x, ...) {
> attrs <- attributes(x)
> x <- lapply(x, function(i) { if(length(i) == 0L) 0L else i
} )
> attributes(x) <- attrs
> class(x) <- “nb ”

88
> x
> }

> # Use the as.nb.sgbp function to create an nb object that
can be used in the smoothed rate calculations.
> sf.nb.queen <- as.nb.sgbp(sf.sgbp.queen)

> # Create an nb2listw object to use in the spatial smoothing
of NBCPR event rates
> queenweights <- nb2listw(sf.nb.queen)
5.5.5. OHCA Incidence Calculations
With the data prepared, aggregated, and a spatial weights matrix created, rate calculations can
begin. The first rates to be calculated are incidence rates—both crude and smoothed. Crude
incidence was only calculated for the entire study period. The formula and code for that
calculation can be seen below.

Crude OHCA Incidence per 100,000 = (Total Count of OHCA per Census Tract/Incidence
Population)*100,000

> analysistracts$ohcacrudeincidence <-
(analysistracts$totalevents/analysistracts$incpop)*100000

89
Next, the user must calculate smoothed OHCA incidence rates. Smoothing of disease rates is
done because rates have intrinsic variance instability, especially at small numbers, as is the case
with OHCA data. Smoothing corrects for the instability created when mapping rates by
borrowing strength from other observations to improve the precision of the crude rate estimate.
Incidence rates in this analysis were smoothed using a local empirical bayes smoother via the
EBlocal function in the spdep package. The EBlocal function operates in the same way as a
traditional spatial Empirical Bayes smoother. However, when using an EBlocal smoother, the
prior, or reference rate, is calculated using a spatial window based on the neighbors and the
observation itself. The EBlocal function takes an x (or a vector containing counts of something
of interest), a y (or vector containing the populations at risk of x), and a nb object. The code
demonstrating the use of the EBlocal smoother is below.

> #Code for calculating spatial empirical bayes smoothed
incidence in 2011
> # This code should be replicated for each year in the
analysis and for all years in the analysis using the
appropriate variables in each case.
> sebsinc11 <- EBlocal(analysistracts$total11,
analysistracts$pop2010, sf.nb.queen, geoda = TRUE)
5.5.6. NBCPR-Event Rate Calculations
The next portion of the areal-unit analysis is to calculate spatially smoothed non-performance of
bystander CPR (NBCPR) event rates for the census tracts included in the analysis. Bystander

90
CPR event rate is defined as the amount of non-performance of bystander CPR OHCAs (i.e., the
numerator) divided by the total amount of events in a given census tract (i.e., the denominator).

For the NBCPR event rates, the same EBlocal function was used. The example code chunk is
shown below.

> ##NBCPR Event rate
> ## Calculate a crude event rate for the entire study period
> #analysistracts$crudeevr <- analysistracts$totalnbcpr /
analysistracts$totalevents
> sebsevr11 <-EBlocal(analysistracts$nbcpr2011count,
analysistracts$pop2010, sf.nb.queen, geoda = TRUE)
> analysistracts$sebsevr11 <- sebsevr11$est*100
5.5.7. Excess Risk Calculations
The final rate calculation for the areal unit analysis portion of this tutorial is the excess risk rate
of both OHCA and NBCPR. Excess risk in this context represents a relative risk, which is based
on the idea of comparing the observed rate to the expected rate of an event based on standardized
reference risk. In this analysis, mapping excess risk focused on NBCPR events and constituted
calculating a ratio of the observed NBCPR event rate in the census tract to the average event rate
for the entire City of Los Angeles. Higher value of this ratio indicates a greater risk of NBCPR.
The code for calculating the excess risk of OHCA and NBCPR is below.

> ## Excess Risk Analysis

91
> # Excess Risk of an OHCA in a census tract
> sumohcaobserved <- sum(analysistracts$totalevents)
> sumpopulation <- sum(analysistracts$incpop)
> p_i <- sumohcaobserved / sumpopulation
> E_i <- p_i * analysistracts$incpop
> analysistracts$ohcaexcessrisk <- analysistracts$totalevents
/ E_i

> #Excess Risk of an NBCPR in a census tract
> sumnobcprobserved <- sum(analysistracts$totalnbcpr)
> sumtotalevents <- sum(analysistracts$totalevents)
> p_i <- sumnobcprobserved / sumtotalevents

> E_i <- p_i * analysistracts$totalevents
> analysistracts$nbcprexcessrisk <- analysistracts$totalnbcpr
/ E_i
5.5.8. Spatiotemporal Hotspot Analysis
After calculating the various rates per census tract, the next analytical process in the areal unit
analysis workflow requires a hotspot analysis using the Gi* statistic for both OHCA incidence
and NBCPR rates. The Getis-Ord Gi* statistic calculates spatial hotspots and coldspots for a
variable of interest, and the higher the z-score, the more intense the clustering of high values
(hotspots), while the smaller the z-score, the more intense the clustering of low values
(coldspots). Z-score calculations take the local sum of a variable for a census tract and its

92
neighbors. Using a Z-score also helps determine whether that hotspot or coldspot is statistically
significant. A Z-score greater than 1.96 results in the census tract being categorized as a
statistically significant hotspot, whereas a Z-score less than -1.96 results in a census tract being
categorized as a statistically significant coldspot.

To incorporate a spatiotemporal view on the hotspot analysis, the Gi* statistic was run on the
spatially smoothed annual rates, and a hotspot score was calculated. Census tracts that were
identified as a hotpot or coldspot were assigned a value of 1 for a hotspot and -1 for a coldspot. If
a census tract was not statistically significant in a given year, it was assigned a value of 0. From
there, the values of the Gi* analysis for both incidence and NBCPR event rates were then
summed for the entire study period. A hotspot score maximum of 18 or -18 was allowable. This
would have indicated that a census tract was a hotspot or coldspot for both OHCA incidence and
NBCPR for all 9 years in the study period. The code for executing this analysis is below.

> ## Create a new queen ’s contiguity for the hotspot analysis
> nb_q <- poly2nb(analysistracts)
> nb_q
> nbweights.lw <- nb2listw(include.self(nb_q), style= “W ”,
zero.policy=T)

> ## Run the localG function to perform the hotspot analysis
on the spatially smoothed incidence

93
> ## Note that this code will need to be replicated for each
year in the study analysis
> analysistracts$localGvaluesinc11 <- localG(x =
as.numeric(analysistracts$sebsinc11p100k), listw =
nbweights.lw, zero.policy = TRUE)

> ## Categorize the results of the hotspot analysis to
generate values that can be summed for a hotspotscore for
incidence
> ## Note that this code will need to be replicated for each
year in the study analysis
> analysistracts$localGcatinc11[analysistracts$localGvaluesin
c11 >= -1.96 | analysistracts$localGvaluesinc11 <= 1.96] <-
0
> analysistracts$localGcatinc11[analysistracts$localGvaluesin
c11 < -1.96 ] <- -1
> analysistracts$localGcatinc11[analysistracts$localGvaluesin
c11 > 1.96 ] <- 1

> ## Run the localG function to conduct the hotspot analysis
on the NBCPR event rate variables
> ## Note that this code will need to be replicated for each
year in the study analysis

94
> analysistracts$localGevr11 <- localG(x =
as.numeric(analysistracts$ssernbcpr11), listw =
nbweights.lw, zero.policy = TRUE)

> ## Categorize the results of the hotspot analysis to
generate values that can be summed for a hotspotscore for
the event rate
> ## Note that this code will need to be replicated for each
year in the study analysis
> analysistracts$localGcatevr11[analysistracts$localGevr11 >=
-1.96 | analysistracts$localGevr11 <= 1.96] <- 0
> analysistracts$localGcatevr11[analysistracts$localGevr11 <
-1.96 ] <- -1
> analysistracts$localGcatevr11[analysistracts$localGevr11 >
1.96 ] <- 1

> ## Add all the localG cat variables created previously to
create a hotspot score on OHCA incidence per census tract
> analysistracts$hotspotscore <-
analysistracts$localGcatinc11 +
analysistracts$localGcatinc12 +
analysistracts$localGcatinc13 +
analysistracts$localGcatinc14 +
analysistracts$localGcatinc15 +

95
analysistracts$localGcatinc16 +
analysistracts$localGcatinc17 +
analysistracts$localGcatinc18 +
analysistracts$localGcatinc19

> ## Add all the localG cat variables created previously to
create a hotspot score on the NBCPR event rate per census
tract
> analysistracts$hotspotscoreevr <-
analysistracts$localGcatevr11 +
analysistracts$localGcatevr12 +
analysistracts$localGcatevr13 +
analysistracts$localGcatevr14 +
analysistracts$localGcatevr15 +
analysistracts$localGcatevr16 +
analysistracts$localGcatevr17 +
analysistracts$localGcatevr18 +
analysistracts$localGcatevr19

> ## Create a totalhotspotscore variable that can be
incorporated into the typology categorization
> analysistracts$totalhotspotscore <-
analysistracts$hotspotscore +
analysistracts$hotspotscoreevr

96
5.5.9. Categorizing Census Tracts into Severity Tiers
The final part of the areal unit analysis is categorizing the census tracts into a typology that
factors in the results of the spatiotemporal hotspot analysis. Tier 1 census tracts were considered
the most severe, while tier 3 census tracts were considered the least severe. To be in tier 1, a
census tract needed to score 5 or above in the total hotspot score variable. Tier 2 required a score
between 2 and 4, while tier 3 required a total hotspot score of 1. All other census tracts were not
classified. Below is the code required to categorize census tracts according to this rule set.

> ## Code to categorize census tracts based on the
totalhotspotscore variable
> analysistracts$type[analysistracts$totalhotspotscore == 5 |
analysistracts$totalhotspotscore == 6 |
analysistracts$totalhotspotscore == 7 |
analysistracts$totalhotspotscore == 8 |
analysistracts$totalhotspotscore == 9] <- “Tier 1 ”
> analysistracts$type[analysistracts$totalhotspotscore == 2 |
analysistracts$totalhotspotscore == 3 |
analysistracts$totalhotspotscore == 4] <- “Tier 2 ”
> analysistracts$type[analysistracts$totalhotspotscore == 1]
<- “Tier 3 ”
> analysistracts$type[is.na(analysistracts$type)] <- “Not
Classified ”

97
5.5.10. Mapping the Results of the Areal Unit Analysis
The final part of the analytical workflow for the areal unit analysis portion of this tutorial is to
map out the results. There are two maps useful for visualizing the results in this portion of
tutorial. Both use the R package tmap. Tmap makes it easy to create quick thematic maps. Other
options include exporting the data from R so that it can be imported into another GIS for
visualization, such as ArcGIS, QGIS, or Tableau.
5.6. Non-Areal Unit Analysis
Kernel density estimation is a non-parametric approach for creating continuous density functions
of point data. From a public health perspective, kernel density estimation is often regarded as a
strong alternative to areal-unit analysis when point-level data is available. One form of kernel
density estimation specific to health data is spatial relative risk. Spatial relative risk represents
the ratio of two kernel density surfaces; one surface represents the cases, while the other
represents the controls. In this analysis, spatial relative risk was calculated for an NBCPR
occurring somewhere in Los Angeles.
5.6.1. Analytical Workflow
Figure 5.2 provides a graphic of the steps of the workflow. It starts with creating a point
population density surface using LandScan data and then moves on to creating NBCPR relative
risk surfaces that were adjusted for the underlying population (i.e., landscan data) of the city of
Los Angeles.

The next step involved map algebra to combine the two surfaces. The map algebra operation was
a weighted average of the various surfaces that computes the average value of pixels at a given
location. The average was weighted with the more recent NBCPR relative risk surfaces being

98
weighted more heavily to more accurately capture the current status of NBCPR in the city of Los
Angeles. After the map algebra step was performed, a final NBCPR relative risk surface that
represented the cumulative spatiotemporal relative risk was created.

Figure 5.2. Non-areal unit analysis workflow

99
For this portion of the tutorial, code exampled needed to execute the different steps of the
analytical workflow in two years 2011 and 2012.

5.6.2. R Packages Used for This Analysis
• sf
• tidyverse
• rgdal
• sparr
• tmap
5.6.3. Load Boundary, OHCA, and LandScan Data
Similar to the areal unit analysis portion of this tutorial, the first part of the analytical work flow
is loaded in the data required for the analysis. In this analysis, there are three subsets of data
critical to completing this portion successfully. Those subsets are (1) boundary data, (2) OHCA
data, and (3) LandScan data. The boundary data and the OHCA data were wholly imported and
manipulated in R, while the landscan raster data were loaded in ArcGIS and converted to a point
data shapefile. The LandScan data can also be manipulated in R. If desired, the rasterToPoints
function of the raster package should be used; otherwise, the raster-to-point conversion in
ArcGIS Pro should be used. Sparr is the main R package used during this portion of the tutorial.
Sparr requires .ppp and .owin objects for all analytical techniques contained within the package.
Therefore, much of the loading code will us sf to read in the required data, and then most of the

100
data will be converted into the required formats of .ppp for point data and .owin for boundary
data.

Below are examples of the code for loading the data into R.

> ## Code for loading in the boundary data – LA City boundary
and LAFD battalion shapefiles
> ## Loading LA city boundary
> lacityboundary <-
st_read( “lacityboundary/lacityboundary.shp ”)
> class(lacityboundary)
>
> ## Loading LAFD battalion data
> lafirebattalions <- st_read( “LAFD_Battalions-
shp/LAFD_Battalions.shp ”)
>
> ## Projecting the LA city boundary and LAFD data
> lacityprojected <- st_transform(lacityboundary, 3310)
> lafdbattalionproject <- st_transform(lafirebattalions,
3310)
> plot(lafdbattalionproject)
>

101
> ## Creating an OWIN object for the spatial relative risk
analysis out of the LA City boundary
> boundary <- as.owin((as_Spatial(lacityprojected)))
> plot(boundary)
>
> ## Loading the LandScan data
> ## Note that this will need to be repeated for each year of
LandScan data worked with. LandScan provides accurate
population estimates on an annual basis.
> landscan2011 <-
st_read( “landscanpoint/LandScan2011point.shp ”)
> landscan2011projected <- st_transform(landscan2011, crs =
3310)
> landscan2011ppp <-
as.ppp((as_Spatial(landscan2011projected)))
> landscan2011ppp <- landscan2011ppp[boundary]
>
> landscan2012 <-
st_read( “landscanpoint/LandScan2012point.shp ”)
> landscan2012projected <- st_transform(landscan2012, crs =
3310)
> landscan2012ppp <-
as.ppp((as_Spatial(landscan2012projected)))

102
> landscan2012ppp <- landscan2012ppp[boundary]
>
> ## Loading the OHCA data for SRR Incidence
> ohca2011 <- st_read( “conclusion shapefiles/ohca2011.shp ”)
> ohca2011projected <- st_transform(ohca2011, 3310)
> ohca2011ppp <- as.ppp((as_Spatial(ohca2011projected)))
> ohca2011ppp <- ohca2011ppp[boundary]
> ohca2012 <- st_read( “conclusion shapefiles/ohca2012.shp ”)
> ohca2012projected <- st_transform(ohca2012, 3310)
> ohca2012ppp <- as.ppp((as_Spatial(ohca2012projected)))
> ohca2012ppp <- ohca2012ppp[boundary]
>
> ## Loading the OHCA data by case (NBCPR) and control (BCPR)
for NBCPR Relative Risk
> #2011
> ohca2011cases <- st_read( “conclusion
shapefiles/2011case.shp ”)
> ohca2011projectedcases <- st_transform(ohca2011cases, 3310)
> ohca2011caseppp <-
as.ppp((as_Spatial(ohca2011projectedcases)))
> ohca2011caseppp <- ohca2011caseppp[boundary]
>

103
> ohca2011control <- st_read( “conclusion
shapefiles/2011control.shp ”)
> ohca2011projectedcontrol <- st_transform(ohca2011control,
3310)
> ohca2011controlppp <-
as.ppp((as_Spatial(ohca2011projectedcontrol)))
> ohca2011controlppp <- ohca2011controlppp[boundary]
>
> #2012
> ohca2012cases <- st_read( “conclusion
shapefiles/2012case.shp ”)
> ohca2012projectedcases <- st_transform(ohca2012cases, 3310)
> ohca2012caseppp <-
as.ppp((as_Spatial(ohca2012projectedcases)))
> ohca2012caseppp <- ohca2012caseppp[boundary]
>
> ohca2012control <- st_read( “conclusion
shapefiles/2012control.shp ”)
> ohca2012projectedcontrol <- st_transform(ohca2012control,
3310)
> ohca2012controlppp <-
as.ppp((as_Spatial(ohca2012projectedcontrol)))
> ohca2012controlppp <- ohca2012controlppp[boundary]

104

5.6.4. Spatial Relative Risk Surfaces of OHCA and NBCPR

After loading in all the data and getting those data into .ppp and .owin formats, calculating
spatial relative risk using the sparr package can be completed. There are many decisions to make
when calculating spatial relative risks. For instance, the analyst must decide whether to use a
fixed-distance bandwidth or an adaptive bandwidth. Both have profound implications for the
amount of smoothing and the results of the spatial relative risk surface. Further, questions about
how to ensure the density estimate is accurate arise. Methods such as Least-Square Cross-
Validation or Bootstrapping are commonly used. In this analysis, the spatial relative risk surfaces
were calculated using an adaptive bandwidth with bootstrapping. Both methods avoid the pitfalls
of other commonly used methods for calculating spatial relative risk surfaces. For example,
adaptive bandwidths offer an advantage of fixed bandwidths because they avoid over smoothing
because they provide the ability to reduce the amount of smoothing in densely populated areas of
a study window. The code for calculating both OHCA and NBCPR is as follows:

> ## This the OHCA incidence relative risk for 2012. This
code needs to be done/repeated for each year. This is an
example of the code, since this process takes longer to run
computationally.
> inchpilot2011.f <- BOOT.density(ohca2011ppp, type =
“adaptive ”, edge = “uniform ”)

105
> inchpilot2011.g <- BOOT.density(landscan2011ppp, type =
“adaptive ”, edge = “uniform ”)
> h02011 <- OS(ohca2011ppp,nstar= “geometric ”)
> incrho.hat2011 <-
risk(f=ohca2011ppp,g=landscan2011ppp,h0=h02011,adapt=TRUE,h
p=c(inchpilot2011.f,inchpilot2011.g),tolerate=TRUE)
> inchpilot2012.f <- BOOT.density(ohca2012ppp, type =
“adaptive ”, edge = “uniform ”)
> inchpilot2012.g <- BOOT.density(landscan2011ppp, type =
“adaptive ”, edge = “uniform ”)
> h02012 <- OS(ohca2012ppp,nstar= “geometric ”)
> incrho.hat2012 <-
risk(f=ohca2012ppp,g=landscan2011ppp,h0=h02012,adapt=TRUE,h
p=c(inchpilot2012.f,inchpilot2012.g),tolerate=TRUE)

> ## This the OHCA-NBCPR relative risk for 2012. This code
needs to be done/repeated for each year. This is an example
of the code, since this process takes longer to run
computationally.
> hpilot2011.f <- BOOT.density(ohca2011caseppp, type =
“adaptive ”, edge = “uniform ”)
> hpilot2011.g <- BOOT.density(ohca2011controlppp, type =
“adaptive ”, edge = “uniform ”)
> h02011 <- OS(landscan2011ppp,nstar= “geometric ”)

106
> rho.hat2011 <-
risk(f=ohca2011caseppp,g=ohca2011controlppp,h0=h02011,adapt
=TRUE,hp=c(hpilot2011.f,hpilot2011.g),tolerate=TRUE)
> hpilot2012.f <- BOOT.density(ohca2012caseppp, type =
“adaptive ”, edge = “uniform ”)
> hpilot2012.g <- BOOT.density(ohca2012controlppp, type =
“adaptive ”, edge = “uniform ”)
> h02012 <- OS(landscan2011ppp,nstar= “geometric ”)
> rho.hat2012 <-
risk(f=ohca2012caseppp,g=ohca2012controlppp,h0=h02012,adapt
=TRUE,hp=c(hpilot2012.f,hpilot2012.g),tolerate=TRUE)

The next step after creating different relative risk surfaces is to extract those surfaces and
reformat them into a raster data type that can be further manipulated and combined using map
algebra. Raster math enables the combination on a cell-by-cell basis of raster data using common
mathematical operations like arithmetic, algebra, statistics, and trigonometry. There are different
levels of operations available in raster math. These operations consist of local, focal, global, and
zonal operations. In this analysis, local and global operations are used to combine OHCA and
NBCPR raster surfaces into a single, aggregated spatiotemporal relative risk surface. Below is
the code that demonstrates this process, from extracting the various rasters to using raster math to
combing them.

> # Extract Relative Risk and Format as a Raster

107
> rinc2011 <- incrho.hat2012$rr
> rasterinc2011 <- raster(rinc2011)
> rinc2012 <- incrho.hat2012$rr
> rasterinc2012 <- raster(rinc2012)
>
> ## Assign projected coordinate system
> crs(rasterinc2011) <- “+proj=aea +lat_1=34 +lat_2=40.5
+lat_0=0 +lon_0=-120 +x_0=0 +y_0=-4000000 +ellps=GRS80
+datum=NAD83 +units=m +no_defs ”
> crs(rasterinc2012) <- “+proj=aea +lat_1=34 +lat_2=40.5
+lat_0=0 +lon_0=-120 +x_0=0 +y_0=-4000000 +ellps=GRS80
+datum=NAD83 +units=m +no_defs ”

> ## Extract Relative Risk and Format as a Raster
> nbcprrr2011 <- rho.hat2011$rr
> rasternbcpr2011 <- raster(nbcprrr2011)
> nbcprrr2012 <- rho.hat2012$rr
> rasternbcpr2012 <- raster(nbcprrr2012)

> ## Assign projected coordinate system
> crs(rasternbcpr2011) <- “+proj=aea +lat_1=34 +lat_2=40.5
+lat_0=0 +lon_0=-120 +x_0=0 +y_0=-4000000 +ellps=GRS80
+datum=NAD83 +units=m +no_defs ”

108
> crs(rasternbcpr2012) <- “+proj=aea +lat_1=34 +lat_2=40.5
+lat_0=0 +lon_0=-120 +x_0=0 +y_0=-4000000 +ellps=GRS80
+datum=NAD83 +units=m +no_defs ”

After completing the export process, the raster surfaces need to be combined with temporal
weighting. To obtain a more accurate representation of the temporality of the spatial relative risk,
the more recent surface, 2012 in this case, was given more weight, while 2011 was given less
weight. Below is the code that demonstrates this process:

> ## Combining the two incidence surfaces. 2011 is weighted
less than the 2012 surface
> ## Again this is an example. The final combined surface
will include all years in the study period, not just 2011
and 2012.
> combinedincraster <- mean((rasterinc2011*.5) +
(rasterinc2012))
> combinednbcprraster <- mean((rasternbcpr2011*.5) +
(rasternbcpr2012))
> combinedincandnbcprrelativerisksurface <-
mean((combinedincraster) + (combinednbcprraster))
> ## Exponentiate the combined relative risk surface to get
it out of the log-relative scale

109
> expcombinedrelativerisksurface <-
exp(combinedincandnbcprrelativerisksurface)
5.6.5. Mapping the Results
The final step of the non-areal unity analysis workflow is to map the results. Similar to the areal
unit analysis, the tmap package from R can be used to create a map of the combined relative risk
surface. Other GIS software can also be used to map the combined raster surfaces generated from
this analysis.
5.7. Conclusion
The first goal of this tutorial was to provide spatial and spatiotemporal analytical workflows and
the subsequent R code in a single document. Placing these workflows in a single document
accomplishes multiple things. First, it makes this research more replicable. The replication crisis
continues to grow in multiple fields. By providing this walkthrough, the opportunity to replicate
the research in this dissertation is increased.

The second goal of this tutorial was to improve accessibility to specialized spatial and
spatiotemporal methods. The underlying drive for this goal was to introduce these methods so
that they can be used on other health topics, such as cerebrovascular events and opioid overdose
calls. Another pathway in which this type of research can be used is in pairing up the OHCA
analysis with other investigations of things like air or noise pollution. The spatiotemporal
aggregation of OHCA relative risk would be particularly useful given that most air pollution and
noise pollution are often interpolated from point data, which results in a raster dataset.

110
In this tutorial, OHCA and bystander CPR in Los Angeles were analyzed using spatial analysis
and spatiotemporal aggregation in R. For the areal unit analysis section of the tutorial, the major
methods used were spatial smoothing to create more stable rates and hotspot analysis using the
Gi* statistic on an annual basis, which was then categorized to create a severity typology. Spatial
relative risk was measured using kernel density estimation and raster math, which were the major
methods explored in the second half of this tutorial.

111
References
Anderson, Monique L, Margueritte Cox, Sana M Al-Khatib, Graham Nichol, Kevin L Thomas,
Paul S Chan, Paramita Saha-Chaudhuri, et al. 2014. “Cardiopulmonary Resuscitation Training
Rates in the United States.” JAMA Internal Medicine 174 (2): 194–201.
Anselin, Luc, Ibnu Syabri, and Youngihn Kho. 2006. “GeoDa : An Introduction to Spatial Data
Analysis.” Geographical Analysis 38 (1): 5–22.
Anselin, Luc, Nancy Lozano-Gracia, and Julia Koschinky. 2006. “Rate Transformations and
Smoothing.” Technical Report. Urbana, IL: Spatial Analysis Laboratory, Department of
Geography, University of Illinois.
Aria, Massimo, and Corrado Cuccurullo. 2017. “Bibliometrix: An R-Tool for Comprehensive
Science Mapping Analysis.” Journal of Informetrics 11 (4): 959–975.
Aria, Massimo, Michelangelo Misuraca, and Maria Spano. 2020. “Mapping the Evolution of
Social Research and Data Science on 30 Years of Social Indicators Research.” Social
Indicators Research 149 (3): 803–831.
Benjamin, Emelia J, Salim S Virani, Clifton W Callaway, Alanna M Chamberlain, Alexander R
Chang, Susan Cheng, Stephanie E Chiuve, et al. 2018. “Heart Disease and Stroke Statistics—
2018 Update: A Report From the American Heart Association.” Circulation (New York,
N.Y.) 137 (1)2: e67–e492.
Beyer, Kirsten M. M, Chetan Tiwari, and Gerard Rushton. 2012. “Five Essential Properties of
Disease Maps.” Annals of the Association of American Geographers 102 (5): 1067–1075.
Biesecker, Claire, Whitney E Zahnd, Heather M Brandt, Swann Arp Adams, and Jan M Eberth.
2020.“A Bivariate Mapping Tutorial for Cancer Control Resource Allocation Decisions and
Interventions.” Preventing Chronic Disease 17: E01–E01.

112
Blewer, Audrey L, Shaun K McGovern, Robert H Schmicker, Susanne May, Laurie J Morrison,
Tom P Aufderheide, Mohamud Daya, et al. 2018. “Gender Disparities Among Adult
Recipients of Bystander Cardiopulmonary Resuscitation in the Public.” Circulation
Cardiovascular Quality and Ooutcomes 11 (8): e004710–e004710.
Buick, Jason E, Ian R Drennan, Damon C Scales, Steven C Brooks, Adams Byers, Sheldon
Cheskes, Katie N Dainty, et al. 2018. “Improving Temporal Trends in Survival and
Neurological Outcomes after Out-of-Hospital Cardiac Arrest.” Circulation Cardiovascular
Quality and Outcomes 11 (1): e003561–e003561.
California Department of Finance. 2020. Population
projections. http://www.dof.ca.gov/forecasting/demographics/projections/. Updated 2020.
Accessed July 10, 2020.
Carlos, Heather A, Xun Shi, James Sargent, Susanne Tanski, and Ethan M Berke. 2010 “Density
Estimation and Adaptive Bandwidths: A Primer for Public Health Practitioners.” International
Journal of Health Geographics 9 (1): 39–39.
Cebrecos, Alba, Julia Díez, Pedro Gullón, Usama Bilal, Manuel Franco, and Francisco Escobar.
2016. “Characterizing Physical Activity and Food Urban Environments: A GIS-Based
Multicomponent Proposal.” International Journal of Health Geographics 15 (1): 35–35.
Chan, Timothy C.Y, Derya Demirtas, and Roy H Kwon. 2016. “Optimizing the Deployment of
Public Access Defibrillators.” Management Science 62 (12): 3617–3635.
Charreire, Hélène, Romain Casey, Paul Salze, Chantal Simon, Basile Chaix, Arnaud Banos,
Dominique Badariotti, Christiane Weber, and Jean-Michel Oppert. 2010. “Measuring the Food
Environment Using Geographical Information Systems: a Methodological Review.” Public
Health nutrition 13 (11): 1773–1785.

113
Chen, John, Karthik Seetharam, Shawn Reginauld, and Stamatios Lerakis. 2019. “Higher Walk
Score Is Associated with Higher Rates of Bystander Automated External Defibrillator Use in
Street-Level Cardiac Arrest from Cardiac Arrest Registry to Enhance Survival
Registry.” Journal of Cardiovascular Medicine (Hagerstown, Md.) 20 (12): 859–860.
County of Los Angeles Enterprise GIS. 2020. County Of Los Angeles Enterprise GIS Web
site. https://egis-lacounty.hub.arcgis.com.
Davies, Tilman M, Jonathan C Marshall, and Martin L Hazelton. 2018. “Tutorial on Kernel
Estimation of Continuous Spatial and Spatiotemporal Relative Risk.” Statistics in Medicine 37
(7): 1191–1221.
Daya, Mohamud R, Robert H Schmicker, Dana M Zive, Thomas D Rea, Graham Nichol, Jason E
Buick, Steven Brooks, et al. 2015. “Out-of-Hospital Cardiac Arrest Survival Improving over
Time: Results from the Resuscitation Outcomes Consortium (ROC).” Resuscitation 91: 108–
115.
Demirtas, Derya, Steven C Brooks, Laurie J Morrison, and Timothy C Chan. 2015.
“Spatiotemporal Stability of Public Cardiac Arrests.” Circulation (New York, N.Y.) 132 (19).
Esri. 2018. “ArcGIS Pro: Release 2.3.” Redlands, CA: Environmental Systems Research
Institute.
Fleming, Douglas, Ann Owens, Marc Eckstein, and Stephen Sanko. “Spatiotemporal Analysis of
Out-of-Hospital Cardiac Arrest in the City of Los Angeles, 2011–2019.” Resuscitation (2021).
Folke, Fredrik, Freddy Knudsen Lippert, Søren Loumann Nielsen, Gunnar Hilmar Gislason,
Morten Lock Hansen, Tina Ken Schramm, Rikke Sørensen, et al. 2009. “Location of Cardiac
Arrest in a City Center: Strategic Placement of Automated External Defibrillators in Public
Locations.” Circulation (New York, N.Y.) 120 (6): 510–517.

114

Fordyce, Christopher B., Carolina M. Hansen, Kristian Kragholm, Matthew E. Dupre, James G.
Jollis, Mayme L. Roettig, Lance B. Becker, et al. 2017. “Association of Public Health
Initiatives with Outcomes for Out-of-Hospital Cardiac Arrest at Home and in Public
Locations.” JAMA Cardiology 2 (11): 1226–1235.
Getis, Arthur, and J. K Ord. 1992. “The Analysis of Spatial Association by Use of Distance
Statistics.” Geographical Analysis 24 (3): 189–206.
Gräsner, Jan T, Patrick Meybohm, Amke Caliebe, Bernd W Böttiger, Jan Wnent, Martin
Messelken, Tanja Jantzen, et al. 2011. “Postresuscitation Care with Mild Therapeutic
Hypothermia and Coronary Intervention after Out-of-Hospital Cardiopulmonary
Resuscitation: A Prospective Registry Analysis.” Critical Care (London, England) 15 (1):
R61–R61.
Hasegawa, Kohei, Yusuke Tsugawa, Carlos A Camargo, Atsushi Hiraide, and David F.M Brown.
2013. “Regional Variability in Survival Outcomes of Out-of-Hospital Cardiac Arrest: The All-
Japan Utstein Registry.” Resuscitation 84 (8): 1099–1107.
Hasselqvist-Ax, Ingela, Gabriel Riva, Johan Herlitz, Mårten Rosenqvist, Jacob Hollenberg, Per
Nordberg, Mattias Ringh, et al. 2015. “Early Cardiopulmonary Resuscitation in Out-of-
Hospital Cardiac Arrest.” The New England journal of medicine 372 (24): 2307–2315.
Hayward, Rodney A, Michele Heisler, John Adams, R. Adams Dudley, and Timothy P Hofer.
2007. “Overestimating Outcome Rates: Statistical Estimation When Reliability Is
Suboptimal.” Health services research 42 (4): 1718–1738.

115
Idris, Ahamed H, Lance B Becker, Joseph P Ornato, Jerris R Hedges, Nicholas G Bircher, Nisha
C Chandra, Richard O Cummins, et al. 1996. “Utstein-Style Guidelines for Uniform Reporting
of Laboratory CPR Research.” Resuscitation 33 (1): 69–84.
Klinenberg, Eric. 2016. “Social Isolation, Loneliness, and Living Alone: Identifying the Risks for
Public Health.” American journal of public health (1971) 106 (5): 786–787.
Klinenberg, Eric. 2015. Heat Wave: A Social Autopsy of Disaster in Chicago. Second edition.
Chicago: University of Chicago Press.
Krieger, Nancy. 2003. “Place, Space, and Health: GIS and Epidemiology.” Epidemiology
(Cambridge, Mass.) 14 (4): 384–385.
Kulldorff M, Information Management Services, Inc. 2009. “SaTScan
TM
v8.0: Software For The
Spatial And Space-Time Scan Statistics”.
LA City Planning. 2019. “LA City Planning Demographic Reports.”
Lee, Duncan. 2013. “CARBayes: An R Package for Bayesian Spatial Modeling with Conditional
Autoregressive Priors.” Journal of statistical software 55 (13): 1–24.
Lee, Sun Young, Young Sun Ro, Sang Do Shin, Kyoung Jun Song, Ki Ok Ahn, Min Jung Kim,
Sung Ok Hong, and Young Taek Kim. 2015. “Interaction Effects Between Highly-Educated
Neighborhoods and Dispatcher-Provided Instructions on Provision of Bystander
Cardiopulmonary Resuscitation.” Resuscitation 99: 84–91.
Longley, Paul. 2001. Geographic Information Systems and Science Chichester: Wiley.
MacQuillan, E.L, A.B Curtis, K.M Baker, R Paul, and Y.O Back. 2017. “Using GIS Mapping to
Target Public Health Interventions: Examining Birth Outcomes Across GIS
Techniques.” Journal of community health 42 (4): 633–638.

116
Marshall, Patrick. 2016. “LA GeoHub: a Model for ‘‘Datafying’’ Communities: a Los Angeles
Portal Goes Beyond Offering Location-Based Data by Allowing City Employees and the
Public to Build and Share Apps.” Government computer news 35 (2): 47–.
McLuckie, Colleen, Mai T Pho, Kaitlin Ellis, Livia Navon, Kelly Walblay, Wiley D Jenkins,
Christofer Rodriguez, et al. 2019. “Identifying Areas with Disproportionate Local Health
Department Services Relative to Opioid Overdose, HIV and Hepatitis C Diagnosis Rates: A
Study of Rural Illinois.” International journal of environmental research and public health 16
(6): 989–.
McNally, Bryan, Allen Stokes, Allison Crouch, and Arthur L. Kellermann. 2009. “CARES:
Cardiac Arrest Registry to Enhance Survival.” Annals of emergency medicine 54 (5): 674–
683.e2.
McNally, Bryan, Rachel Robb, Monica Mehta, Kimberly Vellano, Amy L Valderrama, Paula W
Yoon, Comilla Sasson, et al. 2011 “Out-of-Hospital Cardiac Arrest Surveillance — Cardiac
Arrest Registry to Enhance Survival (CARES), United States, October 1, 2005–December 31,
2010.” MMWR. Surveillance summaries 60 (8): 1–19.
Medicine, Institute of, Board on Health Sciences Policy, Committee on the Treatment of Cardiac
Arrest: Current Status and Future Directions, Andrea M Schultz, Margaret A McCoy, and
Robert Graham. 2015. Strategies to Improve Cardiac Arrest Survival: A Time to Act.
Washington, D.C: National Academies Press.
Messer, Lynne C, Barbara A Laraia, Jay S Kaufman, Janet Eyster, Claudia Holzman, Jennifer
Culhane, Irma Elo, Jessica G Burke, and Patricia O’Campo. 2006. “The Development of a
Standardized Neighborhood Deprivation Index.” Journal of urban health 83 (6): 1041–1062.

117
Mitchell, Michael J, Benjamin A Stubbs, and Mickey S Eisenberg. 2009. “Socioeconomic Status
Is Associated with Provision of Bystander Cardiopulmonary Resuscitation.” Prehospital
emergency care 13 (4): 478–486.
Morrison, Laurie J, Graham Nichol, Thomas D Rea, Jim Christenson, Clifton W Callaway,
Shannon Stephens, Ronald G Pirrallo, et al. 2008. “Rationale, Development and
Implementation of the Resuscitation Outcomes Consortium Epistry—Cardiac
Arrest.” Resuscitation 78 (2): 161–169.
Nassel, Ariann F, Elisabeth D Root, Jason S Haukoos, Kevin McVaney, Christopher Colwell,
James Robinson, Brian Eigel, David J Magid, and Comilla Sasson. 2014. “Multiple Cluster
Analysis for the Identification of High-Risk Census Tracts for Out-of-Hospital Cardiac Arrest
(OHCA) in Denver, Colorado.” Resuscitation 85 (12): 1667–1673.
Nichol, Graham, Elizabeth Thomas, Clifton W Callaway, Jerris Hedges, Judy L Powell, Tom P
Aufderheide, Tom Rea, et al. 2008. “Regional Variation in Out-of-Hospital Cardiac Arrest
Incidence and Outcome.” JAMA : the journal of the American Medical Association 300 (12):
1423–1431.
Nicholls, Keith, J. Steven Picou, Joycelyn Curtis, and Janel A Lowman. 2015. “The Utility of
Community Health Workers in Disaster Preparedness, Recovery, and Resiliency.” Journal of
applied social science 9 (2): 191–202.
Patterson, Tiffany, Gavin D Perkins, Jubin Joseph, Karen Wilson, Laura Van Dyck, Steven
Roberston, Hanna Nguyen, et al. 2017. “A Randomised Trial of Expedited Transfer to a
Cardiac Arrest Centre for Non-ST Elevation Ventricular Fibrillation Out-of-Hospital Cardiac
Arrest: The ARREST Pilot Randomised Trial.” Resuscitation 115: 185–191.

118

Perkins, Gavin D, Ian G Jacobs, Vinay M Nadkarni, Robert A Berg, Farhan Bhanji, Dominique
Biarent, Leo L Bossaert, et al. 2015. “Cardiac Arrest and Cardiopulmonary Resuscitation
Outcome Reports: Update of the Utstein Resuscitation Registry Templates for Out-of-Hospital
Cardiac Arrest A Statement for Healthcare Professionals From a Task Force of the
International Liaison Committee on Resuscitation (American Heart Association, European
Resuscitation Council, Australian and New Zealand Council on Resuscitation, Heart and
Stroke Foundation of Canada, InterAmerican Heart Foundation, Resuscitation Council of
Southern Africa, Re.” Circulation (New York, N.Y.) 132 (13): 1286–1300.
R Development Core Team. 2018. “R: A language and environment for statistical computing”.
Vienna, Austria: R Foundation for Statistical Computing.
Raymond, Chase Wesley. 2014. “Negotiating Entitlement to Language: Calling 911 Without
English.” Language in society 43 (1): 33–59.
Rea, Thomas D, Mickey S Eisenberg, Linda J Becker, John A Murray, and Thomas Hearne. 2003
“Temporal Trends in Sudden Cardiac Arrest: A 25-Year Emergency Medical Services
Perspective.” Circulation (New York, N.Y.) 107 (22): 2780–2785.
Reinier, Kyndaron, Elizabeth Thomas, Douglas L Andrusiek, Tom P Aufderheide, Steven C
Brooks, Clifton W Callaway, Paul E Pepe, et al. 2011. “Socioeconomic Status and Incidence
of Sudden Cardiac Arrest.” Canadian Medical Association journal (CMAJ) 183 (15): 1705–
1712.

119
Resuscitation Outcomes Consortium (ROC). 2017. “Association Of Neighborhood
Demographics With Out-of-Hospital Cardiac Arrest Treatment And Outcomes: Where You
Live May Matter.” JAMA Cardiology 2: 1110–1118.

Ro, Young Sun, Sang Do Shin, Kyoung Jun Song, Sung Ok Hong, Young Taek Kim, Dong-Woo
Lee, and Sung-Il Cho. 2016. “Public Awareness and Self-Efficacy of Cardiopulmonary
Resuscitation in Communities and Outcomes of Out-of-Hospital Cardiac Arrest: A Multi-
Level Analysis.” Resuscitation 102: 17–24.
Ro, Young Sun, Sang Do Shin, Yu Jin Lee, Seung Chul Lee, Kyoung Jun Song, Hyun Wook
Ryoo, Marcus Eng Hock Ong, et al. 2016. “Effect of Dispatcher-Assisted Cardiopulmonary
Resuscitation Program and Location of Out-of-Hospital Cardiac Arrest on Survival and
Neurologic Outcome.” Annals of emergency medicine 69 (1): 52–61.e1.
Root, Elisabeth Dowling, Louis Gonzales, David E Persse, Paul R Hinchey, Bryan McNally, and
Comilla Sasson. 2013. “A Tale of Two Cities: The Role of Neighborhood Socioeconomic
Status in Spatial Clustering of Bystander CPR in Austin and Houston.” Resuscitation 84 (6):
752–759.
Rose AN, McKee JJ, Urban ML, Bright EA, Sims KM. 2018. “LandScan 2018”.
Sampson, Robert J. 2011. Great American City: Chicago and the Enduring Neighborhood
Effect Chicago: The University of Chicago Press.
Sasaki, Mie, Taku Iwami, Tetsuhisa Kitamura, Shinichi Nomoto, Chika Nishiyama, Tomohiko
Sakai, Kayo Tanigawa, et al. 2011. “Incidence and Outcome of Out-of-Hospital Cardiac
Arrest With Public-Access Defibrillation: A Descriptive Epidemiological Study in a Large

120
Urban Community.” Circulation journal: Official journal of the Japanese Circulation
Society 75 (12): 2821–2826.
Sasson, Comilla, Carla C Keirns, Dylan M Smith, Michael R Sayre, Michelle L Macy, William J
Meurer, Bryan F McNally, Arthur L Kellermann, and Theodore J Iwashyna. 2011.
“Examining the Contextual Effects of Neighborhood on Out-of-Hospital Cardiac Arrest and
the Provision of Bystander Cardiopulmonary Resuscitation.” Resuscitation 82 (6): 674–679.
Sasson, Comilla, Carla C Keirns, Dylan M Smith, Michael R Sayre, Michelle L Macy, William J
Meurer, Bryan F McNally, Arthur L Kellermann, and Theodore J Iwashyna. 2011.
“Examining the Contextual Effects of Neighborhood on Out-of-Hospital Cardiac Arrest and
the Provision of Bystander Cardiopulmonary Resuscitation.” Resuscitation 82 (6): 674–679.
Sasson, Comilla, Carla C Keirns, Dylan Smith, Michael Sayre, Michelle Macy, William Meurer,
Bryan F McNally, Arthur L Kellermann, and Theodore J Iwashyna. 2010. “Small Area
Variations in Out-of-Hospital Cardiac Arrest: Does the Neighborhood Matter?” Annals of
internal medicine 153 (1): 19–22.
Sasson, Comilla, David J Magid, Paul Chan, Elisabeth D Root, Bryan F McNally, Arthur L
Kellermann, and Jason S Haukoos. 2012. “Association of Neighborhood Characteristics with
Bystander-Initiated CPR.” The New England journal of medicine 367 (17): 1607–1615.
Sasson, Comilla, Jason S Haukoos, Cindy Bond, Marilyn Rabe, Susan H Colbert, Renee King,
Michael Sayre, and Michele Heisler. 2013. “Barriers and Facilitators to Learning and
Performing Cardiopulmonary Resuscitation (CPR) in Neighborhoods with Low Bystander
CPR Prevalence and High Rates of Cardiac Arrest in Columbus, Ohio.” Circulation
Cardiovascular quality and outcomes 6(5).

121
Sasson, Comilla, Jason S. Haukoos, Leila Ben-Youssef, Lorenzo Ramirez, Sheana Bull, Brian
Eigel, David J. Magid, and Ricardo Padilla. 2014. “Barriers to Calling 911 and Learning and
Performing Cardiopulmonary Resuscitation for Residents of Primarily Latino, High-Risk
Neighborhoods in Denver, Colorado.” Annals of emergency medicine 65 (5): 545–552.e2.
Sasson, Comilla, Mary A.M Rogers, Jason Dahl, and Arthur L Kellermann. 2010. “Predictors of
Survival from Out-of-Hospital Cardiac Arrest a Systematic Review and Meta-
Analysis.” Circulation Cardiovascular quality and outcomes 3 (1): 63–81.
Sasson, Comilla, Michael T Cudnik, Ariann Nassel, Hugh Semple, David J Magid, Michael
Sayre, David Keseg, Jason S Haukoos, and Craig R Warden. 2012. “Identifying High-Risk
Geographic Areas for Cardiac Arrest Using Three Methods for Cluster Analysis: Identifying
High-Risk Geographic Areas For Cardiac Arrest.” Academic emergency medicine 19 (2): 139–
146.
SCOPUS. 2020. “OHCA, Bystander CPR, Spatial Analysis, and Neighborhood Effects Query”.
Scopus Accessed: May 2020.
Scquizzato, Tommaso, Giovanni Landoni, Andrea Paoli, Rosalba Lembo, Evgeny Fominskiy,
Artem Kuzovlev, Valery Likhvantsev, and Alberto Zangrillo. 2020. “Effects of COVID-19
Pandemic on Out-of-Hospital Cardiac Arrests: A Systematic Review.” Resuscitation 157:
241–247.
Sharkey, Patrick. 2018. Uneasy Peace: The Great Crime Decline, the Renewal of City Life, and
the Next War on Violence. New York: W.W. Norton & Company.
Strode, Georgianna, John Derek Morgan, Benjamin Thornton, Victor Mesev, Evan Rau, Sean
Shortes, and Nathan Johnson. 2020. “Operationalizing Trumbo’s Principles of Bivariate
Choropleth Map Design.” Cartographic Perspectives.

122
Sun, Christopher L.F., Derya Demirtas, Steven C. Brooks, Laurie J. Morrison, and Timothy C.Y.
Chan. 2016. “Overcoming Spatial and Temporal Barriers to Public Access Defibrillators
Via Optimization.” Journal of the American College of Cardiology 68 (8): 836–845
Sutton, Paul C, Chris Elvidge, and Tom Obremski. 2003. “Building and Evaluating Models to
Estimate Ambient Population Density.” Photogrammetric engineering and remote sensing 69
(5): 545–553.
United States Census Bureau/American FactFinder. 2010. 2010 Census. U.S. Census Bureau.
Web. January 1, 2013.
Uray, Thomas, Florian B Mayr, James Fitzgibbon, Jon C Rittenberger, Clifton W Callaway,
Tomas Drabek, Anthony Fabio, Derek C Angus, Patrick M Kochanek, and Cameron
Dezfulian. 2015. “Socioeconomic Factors Associated with Outcome after Cardiac Arrest in
Patients Under the Age of 65.” Resuscitation 93: 14–19.
Uy-Evanado, Audrey, Harpriya S Chugh, Arayik Sargsyan, Kotoka Nakamura, Ronald Mariani,
Katy Hadduck, Angelo Salvucci, Jonathan Jui, Sumeet S Chugh, and Kyndaron Reinier. 2020.
“Out-of-Hospital Cardiac Arrest Response and Outcomes During the COVID-19
Pandemic.” JACC. Clinical electrophysiology 7 (1): 6–11.
van Nieuwenhuizen, Benjamin P, Iris Oving, Anton E Kunst, Joost Daams, Marieke T Blom,
Hanno L Tan, and Irene G.M van Valkengoed. 2019. “Socio-Economic Differences in
Incidence, Bystander Cardiopulmonary Resuscitation and Survival from Out-of-Hospital
Cardiac Arrest: A Systematic Review.” Resuscitation 141: 44–62.
Virani, Salim S, Alvaro Alonso, Emelia J Benjamin, Marcio S Bittencourt, Clifton W Callaway,
April P Carson, Alanna M Chamberlain, et al. 2020. “Heart Disease and Stroke Statistics—

123
2020 Update: A Report From the American Heart Association.” Circulation (New York,
N.Y.) 141 (9) : e139–e151.
Web of Science. 2020. “OHCA, Bystander CPR, Spatial Analysis, and Neighborhood Effects
Query”. Web of Science Accessed: May 2020
Wissenberg, Mads, Freddy K Lippert, Fredrik Folke, Peter Weeke, Carolina Malta Hansen, Erika
Frischknecht Christensen, Henning Jans, et al. 2013. “Association of National Initiatives to
Improve Cardiac Arrest Management With Rates of Bystander Intervention and Patient
Survival After Out-of-Hospital Cardiac Arrest.” JAMA: The Journal of the American Medical
Association 310 (13): 1377–1384.
Wong, Marie Ann Mae En, Shien Chue, Michelle Jong, Ho Wye Kei Benny, and Nabil Zary.
2018. “Clinical Instructors’’ Perceptions of Virtual Reality in Health Professionals’’
Cardiopulmonary Resuscitation Education.” SAGE Open Medicine 6: 2050312118799602–
2050312118799602.
Yasunaga, Hideo, Hiroaki Miyata, Hiromasa Horiguchi, Seizan Tanabe, Manabu Akahane,
Toshio Ogawa, Soichi Koike, and Tomoaki Imamura. 2011. “Population Density, Call-
Response Interval, and Survival of Out-of-Hospital Cardiac Arrest.” International Journal of
Health Geographics10 (1): 26–26.
Yokoyama, Hiroyuki, Ken Nagao, Mamoru Hase, Yoshio Tahara, Hiroshi Hazui, Hideki
Arimoto, Kazunori Kashiwase, et al. 2011. “Impact of Therapeutic Hypothermia in the
Treatment of Patients With Out-of-Hospital Cardiac Arrest From the J-PULSE-HYPO Study
Registry.” Circulation Journal: Official journal of the Japanese Circulation Society 75 (5):
1063–1070.

Abstract (if available)

Abstract Out-of-hospital cardiac arrests (OHCA) continue to be a persistent public health issue. Decreasing the non-performance of bystander cardiopulmonary resuscitation (NBCPR) is one key element in improving survivorship with a good neurological outcome when an OHCA occurs. One way to reduce NBCPR is by understanding where and how persistent this issue is in certain communities. The primary objective of this dissertation is to provide an analytical framework for creating actionable intelligence on OHCA and NBCPR. This dissertation seeks to accomplish that objective through two avenuesㅡ(1) using spatiotemporal methods to analyze OHCA and NBCPR and (2) increasing understanding of OHCA and NBCPR through analyzing OHCA and NBCPR through the lens of the COVID-19 pandemic. Spatiotemporal methods, currently, are virtually absent from the OHCA literature that uses spatially enabled data. This dissertation shows that spatiotemporal methods are viable and can provide more relevant and useful information for identifying high-risk areas than spatial analysis alone. This dissertation also shows that the COVID-19 pandemic had differential impacts on communities in Los Angeles, which also helps inform public health practitioners about resource utilization for public health and medical emergencies. The final portion of this dissertation furthers the goal of creating an analytical framework by providing a high-level tutorial on two of the analyses in this dissertation.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Association of neighborhood characteristics with bystander CPR and out-of-hospital cardiac arrests in the city of Los Angeles

PDF

Characteristics associated with emergency medical services for cardiac arrest pre- and during COVID-19 in Los Angeles, CA

PDF

Understanding the role of population and place in the dynamics of seasonal influenza outbreaks in the United States

PDF

Spatiotemporal patterns of salt and nutrient contamination in Los Angeles County's groundwater basins

PDF

Incomplete data & insufficient methods: transgender population health research in the US

PDF

Enriching the Demographic Survey sampling for the Los Angeles County Annual Homeless Count with spatial statistics

PDF

Integrating spatial visualization to improve public health understanding and communication

PDF

Lead exposure and health outcomes in older adults

PDF

Characterizing self-reported spatial and provider barriers to maternal health care utilization in Malawi

PDF

A spatiotemporal analysis of racial disparity in the distribution of superfund sites within Santa Clara County, California

PDF

Congestion effects on arterials as a result of incidents on nearby freeway: When should you get off the highway?

PDF

Out-of-school suspensions by home neighborhood: a spatial analysis of student suspensions in the San Bernardino City Unified School District

PDF

Physics-aware graph networks for spatiotemporal physical systems

PDF

Homeless encampments and access to Water, Sanitation, and Hygiene (WaSH) services in Los Angeles, CA

PDF

Spatiotemporal visualization and analysis as a policy support tool: a case study of the economic geography of tobacco farming in the Philippines

PDF

Eye.Earth Pro (Beta v1.0): application development and spatial financial analysis utilizing the PESTELM framework

PDF

Evaluating surface casing depths of oil & gas operations in an effort to protect local groundwater: a GIS enabled process

PDF

Modeling patient access to point-of-care diagnostic resources in a healthcare small-world network in rural Isaan, Thailand

PDF

Children’s environmental and developmental equity within Los Angeles: a spatio-temporal exploration of varying conditions, perspectives, and outcomes

PDF

Soil lead contamination from the Exide battery smelter: the role of spatial scale in cleanup efforts

Asset Metadata

Creator Fleming, Douglas Owen (author)

Core Title Spatiotemporal studies of out-of-hospital cardiac arrests and bystander cardiopulmonary resuscitation in Los Angeles

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Health and Place,Population

Degree Conferral Date 2021-12

Publication Date 10/09/2021

Defense Date 07/19/2021

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

Tag bystander CPR,Cardiac Arrest,OAI-PMH Harvest,OHCA,out-of-hospital cardiac arrest,spatial,spatiotemporal

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Owens, Ann (committee chair), Sanko, Stephen (committee chair), Ailshire, Jennifer (committee member), Axeen, Sarah (committee member), Kemp, Karen (committee member)

Creator Email doflemin@usc.edu,douglas.fleming17@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC16208010

Unique identifier UC16208010

Legacy Identifier etd-FlemingDou-10153

Document Type Dissertation

Format application/pdf (imt)

Rights Fleming, Douglas Owen

Type texts

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

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

Repository Name University of Southern California Digital Library

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

Repository Email cisadmin@lib.usc.edu