GIS PUBLIC HEALTH PROCEEDINGS 2011

URISAs 2011 GIS in Public Health Conference PROCEEDINGS
The following papers/presentations were presented during the conference and were submitted by the presenters for
inclusion in this PROCEEDINGS publication. Click on the title you wish to view to go to the presentation

Satellite Remote Sensing for Forecasting and Mapping Mosquitoborne Disease Risk
Michael C. Wimberly, Ph.D, South Dakota State University, Brookings, SD

Environmental Modeling of Mosquito Population Dynamics: A Comparison of Groundbased Measurements and Satellite Remote Sensing
TingWu Chuang, South Dakota State University, Brookings, SD

Remotely Sensed Environmental Correlates of Malaria Risk in a Highland Region of Ethiopia
Alemayehu Midekisa, South Dakota State University, Brookings, SD

Alcohol Outlet and Assaultive Violence in the City of Philadelphia
Tony Grubesic, Ph.D, Drexel University, Philadelphia, PA

Geocoding Uniform Crime Report Data and Modeling Neighborhood Violent Crime
Russell Kirby, Ph.D, MS, FACE, University of South Florida College of Public Health, Tampa, FL

Creating a GIS Application for Health Care Facilities Planning at Jeddah City
Abdulkader Murad, King Abdulaziz University, Jeddah, Saudi Arabia

Identifying Gaps in Available Healthcare Resources in Three Southern States
Suman Chatterjee, The University of Mississippi Medical Center, Jackson, MS
Fazlay S. Faruque, PhD, RPG, GISP, The University of Mississippi Medical Center, Jackson, MS

Accessible Online Tools for Health Policy and Planning
Jennifer Rankin, Ph.D, MPH, MHA, Robert Graham Center, Washington, DC

Online Geographic Information Systems for Improving Health Planning Practice: Lessons Learned From the Case Study of LoganBeaudesert,
Australia
Ori Gudes, Queensland University of Technology and Griffith University Brisbane, Queensland, Australia

Evaluating Geospatial Visualization Methods for West Nile Virus Risk Mapping
Aashis Lamsal, South Dakota State University, Brookings, SD
Michael Wimberly, Ph.D, South Dakota State University, Brookings, SD

A PerimeterBased Mapping Framework for Visualization of Disease State
Peichung Shih, Siemens Healthcare, Malvern, PA
David Scholl, Siemens Healthcare, Malvern, PA

Estimated Small Area Prevalence of Mental Disorders in Cuyahoga County, Ohio
Terese Lenahan, MBA, GISP, Center for Community Solutions, Cleveland, OH

The Effect of Changing Spatial Extents when Deriving Environmental RiskFactors of Canine Leptospirosis
Ram Raghavan, Ph.D., GISP, Department of Diagnostic Medicine/Pathobiology, College of Veterinary Medicine, Kansas State University, Manhattan,
KS

Groundwater Contamination Buffer Mapping and Spatial Analysis Using GIS/GPS
Mark Hansell, RS, MS, Oakland County Health Division, Pontiac, MI

SmartMaps and Built Environment Exposure: Extending Measurement Paradigms
Philip Hurvitz, Ph.D, University of Washington Urban Form Lab, Seattle, WA

Using SmartMaps to Capture Environmental Characteristics of Daily Travel Patterns
Lin Lin, Ph.D, University of Washington Urban Form Lab, Seattle, WA

Using SmartMaps to Characterize Home and Non Home Built Environment Exposure
Philip Hurvitz, Ph.D, University of Washington Urban Form Lab, Seattle, WA

Built Environment and Travel Behavior to Grocery Stores
Philip Hurvitz, Ph.D, and Junfeng Jiao, Ph.D., University of Washington Urban Form Lab, Seattle, WA

Working People: PlaceBased Effects on Hypertension, Type 2 Diabetes and Obesity
Alberto Colombi, MD, MPH, PPG Industries, Inc, Pittsburgh, PA

Racial Residential Segregation and Stroke Mortality in Atlanta
Sophia Greer, MPH, Centers for Disease Control and Prevention, Atlanta, GA

Availability to Tobacco Outlets and Other GIS at the Swedish National Institute of Public Health
AnnaKarin Johansson, Swedish National Institute of Public Health, stersund, Sweden

Developing a Community Health Information Web Portal
Paul Juarez, Ph.D, Meharry Medical College, Nashville, TN
Wansoo Im, Ph. D, Meharry Medical College, Nashville, TN

Health Outcomes and Consumer Health & Education Spending in Central Appalachia
Timothy Hare, Ph.D, Morehead State University Institute for Regional Analysis & Public Policy, Morehead, KY

The Accessibility and Utilization of Oral Health Services in Belo Horizonte Brazil
Evanilde Martins, Ph.D, Pontifical Catholic University of Minas Gerais, Belo Horitzote, Brazil

Where's the Care? Mapping Health Care Providers with the CMSNPI
Robert Borchers, BA, MAR, MSL Wisconsin Office of Health Informatics, Madison, WI

Interactive Webbased Mapping: Bridging Technology and Data for Health
Linda Highfield, Ph.D, MS, St Luke Episcopal Health Charities, Houston, TX
Jutas Arthasarnprasit, HexaGroup Ltd., Houston, TX

Spatial Patterns of Water Insecurity: Lessons from Accra, Ghana
Justin Stoler, MS, MPH, San Diego State University, San Diego, CA

Exposure Patterns in Air Pollution and Their Population Level Health Impacts
John Pearce, Monash University, Melbourne, Australia

Predicting Mosquito Larval Habitats on Farm Lands Using Remote Sensing and GIS
Thomas Lupher and Lillian Reitz, Texas A&M University, Corpus Christi, TX

PCWASA FOG Program: Protecting the Environment through GIS
Xavier Davis, Peachtree City Water & Sewerage Authority, Peachtree City, GA

More Than a Dot on a Map: The Geography of Brownfield and Health
Marilyn Ruiz, Ph.D, University of Illinois, Urbana, IL

Physical and Social Environmental Impacts on Obesity and Risk Factors in the US
Akihiko Michimi, Ph.D, and Michael Wimberly, Ph.D, South Dakota State University, Brookings, SD

Approaching Childhood Obesity from a Placebased Angle: GIS Prototype Evolution
Jennifer Cannon, MURP, University of Florida, Gainesville, FL

Spatial Model to Evacuate Special Needs Populations to Appropriate Shelters
David Bandi, Ph.D, Sudha Yerramilli, Ph.D, and Dayakar Nittala, National Center for Biodefense Communications, Jackson, MS
Fazlay Faruque, GISP, Ph.D., University of Mississippi Medical Center, Jackson, MS

Geographic Opportunities in Medicine
Amy Blatt, Ph.D, Quest Diagnostics, West Norriton, PA

Web GIS Framework for Surveillance of Malaria: A Case Study of Allahabad District, India
Dr. R.D. Gupta, Motilal Nehru National Institute of Technology, Allahabad, India

GIS and Health: Where Can We Go from Here?
Ellen K. Cromley, PhD, Assistant Clinical Professor, Department of Community Medicine and Health Care, University of Connecticut School of
Medicine

GOLD CONFERENCE SPONSOR

Thank you for the generous support

Did you know there is a Georgia Chapter of URISA?

Find out about chapter activities and get involved at www.gaurisa.org

1
Justin Stoler
Department of Geography
San Diego State University
San Diego, CA 92182
(619) 594-5437
stoler@mail.sdsu.edu

A micro-history of geographic perspectives in public health research from the human-
environment tradition

Abstract
Research at the intersection of public health and geography has a long history. One distinct
feature of the academic literature is that it necessarily aims to be proactive and interventionist;
scientific results are intended to inform public health policy rather than simply fill pages in
academic journals. Within this literature, researchers have theorized in different ways about how
place influences individual health outcomes. This paper offers a micro-history of health and
place that traces the evolution of human-environment theory, while maintaining a focus on
public health outcomes. The result is a representative though not nearly comprehensive
review of how we conceptualize place and frame public health research.

Early interventions: From Hippocrates to germ theory
A quick Google search of father of geography will credit Eratosthenes as the first to
use the word geography. Greeks may tell you it is Anaximander, for his invention of the
gnomon, or Homer for his account of the world in The Odyssey a thousand years earlier. Few
will dispute Hippocrates status as the father of medicine, whose indelible mark on history was
left just a century before Eratosthenes. Hippocrates also laid the groundwork for modern
epidemiology over two thousand years before John Snows famous cholera intervention. But
what is less commonly highlighted is how Hippocrates thinking about illness was also
fundamentally spatial. Among his greatest works, Airs, Waters, and Places provides perhaps the
earliest glimpse into disease as inherently geographic in nature. One needs to look no further for
the origins of the tradition that Snow would reclaim during the infamous cholera epidemic of
1854.
As exemplified by the title, Airs, Waters, and Places, Hippocrates recognized the
interactions between human health, the environment, and location that today are central to the
field of medical geography (Meade & Earickson, 2000). Writing in the 4
th
and 3
rd
centuries BCE,
Hippocrates observed distinctly seasonal categories of disease induced by climate: the
respiratory ailments of winter, and the intestinal and malarial fevers of summer. He recognized
the danger of certain water sources (e.g. springs in rocky soil and wash from snowmelt) and
favored rain water and earthy springs from higher elevations. He describes the symptoms of
malaria and relates them to marshy areas, noting relief for those located at higher elevations
away from the water. In several volumes of his Epidemics work, he clearly describes not only
malaria, but also dysentery, diarrheal diseases, and other febrile illnesses, and warns of greater
prevalence after a rainy spring. This departure from the prevailing theories of demonology and
metaphysical medicine was a Kuhnian paradigm shift in the way disease was perceived. What
Hippocrates was missing was the notion of a contagion, something his successors would
postulate and develop into the miasma theory of disease that would persist for nearly two

2
thousand years until Louis Pasteur. Between Hippocrates and Pasteura period of markedly
little progress on the theoretical link between health and placephysicians remained aware of
geographic variations in air, water, soil, vegetation, animals, insects, housing type, and even
some socio-cultural factors (Meade & Earickson, 2000). These variations were the basis for the
earliest health interventions, perhaps the most celebrated being Snows removal of the water
pump.
At about the same time in history as Snows intervention, Charles Darwins writings on
evolution were becoming influential within geography, and notably picked up by the American
physical geographer William Morris Davis, and the German human geographer Friedrich Ratzel.
Ratzels ideas about social adaptation to the local physical environment in Anthropogeographie,
though later modified, became the springboard for the paradigm of environmental determinism
that dominated geography at the turn of the 20
th
century. Environmental determinism reflected
the period of Western colonialism that was well underway, and effectively justified the lack of
intervention by colonists on the part of the conquered, who suffered greatly from introduction of
western diseases (not to mention bullets). Americans such as Ellen Churchill Semple and
Ellsworth Huntington wrote starkly racist papers about how the environment influences personal
character (Huntington, 1921; Semple, 1910). By the 1940s, environmental determinism was
largely discredited due to lack of scientific rigor, Germanys use of this framework to justify
Nazism, and thriving human geography elsewhere in Europe.
As environmental determinism died in the early 20
th
century, European geographers such
as Alfred Hettner, Paul Vidal de la Blache, and Halford Mackinder followed the tradition laid
down by Immanuel Kant, Alexander von Humboldt, and Carl Ritter, emphasizing human-
environment linkages in different ways. Carl Sauer helped reintegrate American geography with
this regional tradition with a famous paper on cultural landscape morphology (Sauer, 1925), and
Richard Hartshorne brought regional geography to the disciplines forefront where it would
remain for several decades with his seminal work The Nature of Geography (Hartshorne,
1939). In the latter half of the 20
th
century, ushered on initially by the quantitative revolution
of the 1960s (for inspiration, see Schaefer, 1953), the nature of geographic research transitioned
from being descriptive to analytic, both quantitatively and qualitatively. Geographic inquiry
became more about why and how in addition to what and where, requiring a shift of
emphasis from form to process (Golledge, 2002). This shift is analogous to the shift in
epidemiology a half-century earlier from descriptive inventories of symptoms to a revived
etiological focus with the advent of germ theory. Unfortunately, the emphasis on objectivity
during the quantitative revolution loosened the ties between geography and interventionism.
The germ theory of disease, or recognition that microorganisms can cause disease, was an
earth-shattering shift in medical knowledge that single-handedly offered the greatest extension of
life-expectancy the world has known. Two great bacteriologists at the forefront of this late 19
th
-
century revolution, Pasteur and Robert Koch, identified bacteria responsible for transmission of
many infectious diseases, and laid the groundwork for early vaccines and advances in serology
and immunology. The biggest takeaway of this era was validation of early cultures beliefs that
cleanliness, particularly with respect to water and sanitation, was closely linked to health. This
period was also the beginning of a shift in increases in life expectancy due to preventive
medicine, rather than curative medicine. However, with medical research predominantly focused
on germs for the first half of the 20th century, there was quickly another shift borne out of
massive global population growth and environmental degradation. Chronic diseases brought the

3
sudden need to prevent and maintain public health beyond the workings of available vaccines.
As Meade (1977) writes in the opening of her landmark paper:
Such diverse deteriorations of health as the rise of cancer in the American population, the spread of
schistosomiasis in Africa with the extension of irrigated agriculture, the carnage being wrought by the
automobile around the world, and, most poignantly, the existing magnitude of malnutrition and its
consequences have shattered the entrenched prepossession of health professionals with the germ theory of
disease.
Meade culled ideas from throughout the social sciences and constructed an integrated model for
medical geography that was based upon interactions between population, environment, and
culture. The population component stems from demography and encompasses factors such as
genetics and food access. The environmental dimension includes physical and chemical health
risks that are usually studied by physicians, toxicologists, and ecologists. The cultural component
borrows from anthropology and sociology, and accounts for human behaviors interacting with
the environment and other humans, such as diet, housing, and the cultural frameworks for
perception of health and illness. From Meades work, geographers began to pursue the spatial
workings of these factors. This influence blossomed into a modern era of medical geography, but
the cultural component described above represented an important cross-pollination with other
social sciences, particularly the field known as social epidemiology. In the last few decades, in
light of unprecedented urbanization and failure of any nation to boast of universal good health
and welfare for all citizens, one important branch of the human-environment tradition of
geography has taken the form of the study of neighborhood effects.

Medical geography and neighborhood effects
A significant development in modern medical geographywhich overlaps with social
and spatial epidemiological research and has firm roots in the human-environment tradition
(Meade & Earickson, 2000:7)is increased understanding of the complexity between
compositional vs. contextual effects of health in populations at different scales.
A fundamental question of medical geography is essentially how does where you live
affect your health? As practitioners have evolved from a strictly biomedical approach to disease
to a multi-factorial approach incorporating socio-cultural elements, geographers and spatial
epidemiologists face the difficulty of discerning the separate effects of the physical and social
environment. The rising incidence of chronic diseases such as cancers and cardio-pulmonary
disease, relative to infectious diseases, has further increased the emphasis on social factors.
There is a substantial literature on health inequalities that looks at both geographic and social
factors at a variety of scales. Medical geographers straddling both the human-environment and
regionalization traditions often want to know whether regional health differences are accounted
for by the spatial distribution of socio-economic status (SES) or if where you live (i.e.
neighborhood) has additional effects. Researchers often aim to control for SES, or
compositional, factors first, and then see if there are additional spatial, or contextual, effects
(Graham et al., 2004), and multi-level modeling approaches have become a popular technique
allowing the investigation of both effects simultaneously (see for example Gelman & Hill, 2007;
Goldstein, 2003). In the demography literature, Bongaarts and colleagues (1984) reiterate a
distinctionmade previously by Davis and Blake (1956)between proximate determinants and
socioeconomic or background variables with regard to fertility, an analogy that is portable to
other social and health phenomena. Background variables refer to the social, economic, cultural,
political, psychological, and environmental context, while proximate determinants refer to

4
personal biological and behavioral composition through which background variables manifest.
Proximate variables are characterized by having direct influence on some social phenomenon,
while socioeconomic context has indirect influence by serving as a modifier of compositional
factors (Bongaarts et al., 1984).
Graham and colleagues (2004) cite a number of studies attempting to disentangle the
social and geographic effects of mortality; different designs have drawn different conclusions
about individual vs. ecological effects, as well as the degree that context matters in a region, e.g.
absolute or relative to surrounding regions. The authors summarize that (1) contextual health
effects may exist at a variety of scales, leaving many conclusions as tentative; (2) studies that
focus on the areas of lowest SES may underestimate health disadvantages due to lack of
heterogeneity, as the greatest health differentials are often in more affluent areas; and (3) relative
inequality within regions may significantly affect morbidity even after accounting for both
individual- and area-level characteristics, which reinforces the notion that context must continue
to be studied in light of compositional effects (Graham et al., 2004).
Woods (2004), writing about the history of socioeconomic mortality inequality, observes
that prior to the 20
th
century, the environment and rudimentary levels of medical knowledge were
better predictors of mortality. A socio-economic model of health is indeed a 20
th
-century
phenomenon in the developed world, but has implications for the complexity of study in the
developing world. In poorer nations, contextual environment factors still play an important role,
as lack of medical resources rather than knowledge per se often drives mortality. However,
the presence of some (i.e. insufficient) modern medical resources does exacerbate socio-
economic (or compositional) differentials, whether through market conditions or government
policy, and can also create additional contextual differences that may have little-to-nothing to do
with class status. The rapid, unplanned urbanization of the developing world has resulted in cities
with significant heterogeneity of settlement patterns; however, developed nations also exhibit
significant variation in health outcomes (Graham et al., 2004), but with substantially longer
histories of population and medical records available to researchers. Modern research designs
aimed at intervening in the developing world are largely drawn from those that have been fruitful
in the developed world.
When a non-governmental organization (NGO) implements a local health promotion
effort in a developing area there are obvious compositional effects for the receiving population,
but also contextual effects relative to similar areas excluded from the program. When modeling
extremely poor areas, it can be hard to discern one from the other, and this notion of
heterogeneity has been a focus of much contemporary public health research. Blakely et al.
(2002) suggest that can be difficult to disentangle effects of low overall income in an area from
effects of income inequality due to confounding between levels of measurement. This is
consistent with observations seen in recent publications from my own project work in Ghana,
where heterogeneity seems to be obscuring contextual effects for child mortality and malaria
(Weeks et al., 2006; Stoler et al., 2009), thus spurring a multi-level approach for subsequent
neighborhood-level investigations of health outcomes (Weeks et al., 2010; Stoler et al., in
review).
While modern neighborhood effects research has been historically motivated by its
ability to inform policy and intervene in poor health outcomes, its place within the human-
environment tradition has occasionally veered off into emphasis on modeling and theoretical
differences between methodologies. The interpretation and applicability of many neighborhood
analysis techniques have occasionally been contested (see Oakes 2004 for a spirited example).

5
New perspectives on the effect of place on health outcomes have emerged through a blending of
empiricism and theory.
Lobao and colleagues (2007) describe three common trends for how space or place is
incorporated into social science research. First and all too frequently, space is not explicitly
accounted for, but is rather a contextual background detail for a given study. Broad social theory
across large populations may inadvertently minimize the importance of place specificity. Space
may also be treated as noise, or some impediment to quantitative inference that must be
controlled for. The known spatial autocorrelation of many social phenomena at specific
geographic scales are the basis for introducing distance as a control for spatial patterning (for
example, as is done in spatial econometric modeling). Finally, space may itself be the
phenomenon of interest; in the inequality literature space is conceptualized by use of different
scales, stratifying by location, or as power-geometries that cut across multiple scales (Massey
1994, as cited in Lobao et al., 2007). Explicitly recognizing space still leaves room for how it is
defined, which for populations can mean physical distributions, activity spaces, or perception of
space (the spatial experience). This new treatment of place may portend a potential
reconciliation within the human-environment tradition between critical theorism and positivism,
with implications for future research into the role of place in health outcomes.

Critical theorists, spatial scientists, and place
The role of place is certainly not ignored in the critical theorist tradition of geography.
Through a brief foray into the history of science, Livingstone (1992) steers us toward a context-
driven history that embraces Thomas Kuhns sense of paradigm-driven scientific revolution. He
suggests that our emphasis should not be solely geographic theory, but the socio-political context
of theory which has too often been ignored. To be sure, he writes, we will be interested in
developments in geographical theory and thinking, but we will always want to locate theory in
social and intellectual circumstance (Livingstone, 1992). Livingstone is interested in situated
theory, which incorporates the local context of knowledge. The emphasis of socio-political
context is an important component of public health practice. Socio-political context is often
paramount to international health programs, as the social climate or political posture of a
community or nation deeply affects the community structure itself, as well as the types of
programs that can be implemented for a given health concern (e.g. HIV/AIDS norms in Eastern
Europe vs. Africa vs. Latin America). This is a different conceptualization of place than the
discrete place of positivists within the human-environment tradition, but it has a clear link to the
compositional vs. contextual effects literature augmented by geographers over the last decade,
and an important contribution to creating competent health interventions.
Lobao and colleagues (2007) note two traditions for social research in a spatial context.
The place-in-society tradition focuses on the distinct qualities of a specific place and compares it
with other places through some lens (human ecology, political economy, Marxism, etc.). The
society-in-place tradition focuses more on the operation of social processes that are then
grounded in specific places. Because it emphasizes processes, rather than the rich identity of a
place, the society-in-place tradition is more amenable to analyzing more places and using
quantitative analysis. Insights for understanding spatial inequalities can be found in the
intersection of these traditions: specific places featuring a rich amount of attributes are needed to
elucidate detailed social processes, yet the comparison of inequality across different places is
necessary to understand how processes change at different spatial scales. We can draw an
analogy from the field of geography: spatial scientists model processes based on determinants

6
drawn from all scales, while critical realists view social and spatial processes as more
intertwined and dependent on place settings.
Del Casino and Stevens (2007) contrast the discrete vs. relational conceptualization of
space between spatial scientists and critical realists, and note the middle ground where these
theoretical approaches may complement each other for several research questions. Many
researchers firmly rooted in the spatial science tradition inform their hypotheses by
acknowledging the idiosyncrasies of the study area when modeling health outcomes across
multiple contexts, as clearly these contexts will drive the feasibility of intervention. While spatial
scientists view space and scale as discrete entities, they may be socially constructed or at least
have socially-produced qualities. We can envision health studies where socially-constructed
contexts may be less relevant such as research performed on inmates or military personnel
where spatial and social hierarchies are very clearly defined or even institutionalized. But
research in developing cities, especially in slums where government often has less of a presence,
requires sensitivity to social relations. Political scales such as urban, region, and national are not
necessarily the scales at which health processes operate, so there is an implicit need to be flexible
when defining scale of the neighborhood of interest.
Keeping this flexibility in mind, Entwisle (2007) advocates for greater attention to human
agency in neighborhood health effects studies and points to four forms of agency that are thought
to be obfuscating contextual effects on health. These include (1) individual choice about where
to live; (2) the effects of residential mobility on the neighborhoods themselves; (3) presence of
local interventions which do not appear at random, but in places of need; and (4) individual
engagement in activity spaces that may not map directly to our preconceived notion of
neighborhood (Entwistle, 2007). These realities implicate new methodological challenges in the
way we conceptualize neighborhoods and residence, the way we collect data (with emphasis on
longitudinal data), and the way we measure social phenomenon and parameterize human
nature. Newer, more cutting-edge simulation models such as agent-based modeling attempt to
bridge these views by modeling the interdependencies of social interaction and place. Entwisle
(2007) joins the chorus of interest in agent-based models for their ability to capture the richness
of spatially explicit interactions between individuals and their environment in both a top-down
and bottom-up manner.
Still another framework borne of the human-environment tradition is vulnerability
analysis. Vulnerability analysis examines social vulnerability to stressors or shocks that may be
caused by man-made social factors or natural disasters, and has recently advanced within
geographys human-environment tradition (see for example Cutter et al., 2003; Turner et al.,
2003). Vulnerability analysis also features a renewed emphasis on equity and social justice
(Eakin & Luers, 2006). This framework has been expanding to fill gaps in research and action,
and has been adapted to explore health outcomes in low-income populations (Grzywacz et al.,
2004). While there has not yet been tremendous cross-over to the epidemiology literature,
vulnerability analysis roots in 1960s and 1970s social science radicalism have instilled the
values of advocating and intervening on behalf of vulnerable populations regardless of whether
we theorize place discretely or as a social construction.

Final thoughts
The notion that place influences health originated thousands of years ago, though most of
the progress in theory development has taken place in the last 150 years. The intersection of
geography and public health continues to produce a rich body of literature that is used to inform

7
public policy rather than just be science for sciences sake. Public health research exists in many
forms today; it has become increasingly complex and interdisciplinary as we generate ever-
increasing data sources and analytical techniques. Scales of analysis now range from global to
intracellular, yet the notion of place has not been lost on todays medical geographer or spatial
epidemiologist who carries the torch lit by Hippocrates prescient linkage of health and
geography. Most health professionals are driven to serve or advocate for those most in need
within the boundaries of our communities, whether global or local. By incorporating a
geographic perspective and drawing upon the human-environment tradition, we exploit those
boundariesboth figuratively and literallyto further improve our world.

References
Blakely, T. A., Lochner, K., & Kawachi, I. (2002). Metropolitan area income inequality and self-
rated health a multi-level study. Social Science & Medicine, 54, 65-77.
Bongaarts, J., Frank, O., & Lesthaeghe, R.. (1984). The proximate determinants of fertility in
sub-Saharan Africa. Population and Development Review, 10(3), 511-537.
Cutter, S. L., Boruff, B. J., & Shirley, W. L. (2003). Social vulnerability to environmental
hazards. Social Science Quarterly, 84(1), 242-261.
Davis, K., & Blake, J. (1956). Social structure and fertility: An analytic framework. Economic
Development and Cultural Change, 4(4), 211-235.
Del Casino, V. J., & Stevens, J. P. (2007). Space for social inequality researchers: A view from
geography, in L. Lobao, G. Hooks, & A. R. Tickamyer, (Eds). The sociology of spatial
inequality. Albany, NY: State University of New York Press.
Eakin, H., & Luers, A. L. 2006. Assessing the vulnerability of social-environmental systems.
Annual Review of Environmental Resources, 31, 365-394.
Entwisle, B. (2007). Putting people into place. Demography, 44(4), 687-703.
Goldstein, H. (2003). Multilevel statistical models. London, UK: Arnold.
Golledge, R. (2002). The nature of geographic knowledge. Annals of the Association of
American Geographers, 92(1), 1-14.
Graham, E., Boyle, P., Curtis, S., & Moore, E. (2004). Health geographies in the developed
world, in P. Boyle, S. Curtis, E. Graham, & E. Moore (Eds), The geography of health
inequalities in the developed world: Views from Britain and North America (pp. 3-33).
Burlington, VT: Ashgate Press.
Grzywacz, J. G., Almeida, D. M., Neupert, S. D., & Ettner, S. L. (2004). Socioeconomic status
and health: A micro-level analysis of exposure and vulnerability to daily stressors.
Journal of Health and Social Behavior, 45, 1-16.
Hartshorne, R. (1939). The nature of geography: A critical survey of current thought in the light
of the past. Lancaster, PA: Association of American Geographers.
Huntington, E. (1921). The relation of health to racial capacity: The example of Mexico.
Geographical Review, 11(2), 243-264.
Livingstone, D. N., (1992), The geographical tradition: Episodes in the history of a contested
discipline. Oxford, UK: Blackwell Publishers.
Lobao, L., Hooks, G., & Tickamyer, A. R. (2007). Advancing the sociology of spatial inequality,
in L. Lobao, G. Hooks, & A. R. Tickamyer, (Eds). The sociology of spatial inequality.
Albany, NY: State University of New York Press.
Massey, D. (1994). Space, place, and gender. Minneapolis, MN: University of Minnesota Press.

8
Meade, M. S. (1977). Medical geography as human ecology: The dimension of population
movement. Geographical Review, 67(4), 379-393.
Meade, M. S., & Earickson, R. J. (2000). Medical geography. New York: The Guildford Press.
Oakes, J. M. (2004). The (mis)estimation of neighborhood effects: causal inference for a
practicable social epidemiology. Social Science & Medicine, 58, 1929-1952.
Sauer, C. O. (1925). The morphology of landscape. University of California Publications in
Geography, 2(2), 19-53.
Schaefer, F. K. (1953). Exceptionalism in geography: A methodological examination. Annals of
the Association of American Geographers. 43(3):226-249.
Semple, E. C. (1910). The Anglo-Saxons of the Kentucky Mountains. Bulletin of the American
Geographical Society, 42(8), 561-594.
Stoler, J., Weeks, J. R., Getis, A., & Hill, A. G. (2009). Distance threshold for the effect of urban
agriculture on elevated self-reported malaria prevalence in Accra, Ghana. American
Journal of Tropical Medicine & Hygiene, 80(4), 547-554.
Stoler, J., Fink, G., Weeks, J. R., Otoo, R. A., Ampofo, J. A., & Hill, A. G. (in review). When
urban taps run dry: Sachet water consumption and health effects in low income
neighborhoods of Accra, Ghana. Health & Place.
Turner, B. L. II, Kasperson, R. E., Matson, P. A., McCarthy, J. J., Corell, R. W., Christensen, L.,
Schiller, A. (2003). A framework for vulnerability analysis in sustainability science.
Proceedings of the National Academy of Sciences of the United States of America,
100(14), 8074-8079.
Weeks, J. R., Hill, A. G., Getis, A., & Stow, D. (2006). Ethnic residential patterns as predictors
of intra-urban child mortality inequality in Accra, Ghana. Urban Geography, 27(6), 526-
548.
Weeks, J. R., Getis, A., Hill, A. G., Agyei-Mensah, S., & Rain, D. (2010). Neighborhoods and
fertility in Accra, Ghana: An AMOEBA-based approach. Annals of the Association of
American Geographers, 100(3), 558-578.
Woods, R. (2004). The origins of social class mortality differentials, in P. Boyle, S. Curtis, E.
Graham, & E. Moore (Eds), The geography of health inequalities in the developed world:
Views from Britain and North America. Burlington, VT: Ashgate. 37-52.

Satellite Remote Sensing for Forecasting and Mapping Mosquito-Borne
Disease Risk
Michael C. Wimberly
1
, Yi Liu
2
, Ting-Wu Chuang
1
, Alemayehu Midekisa
1
, and Aashis
Lamsal
1
1
Geographic Information Science Center of Excellence, South Dakota State University
2
Department of Electrical Engineering and Computer Science, South Dakota State University
Mosquito-borne diseases such as malaria, West Nile virus, and dengue hemorrhagic fever
are globally important public health threats. Many of these diseases are currently emerging in
new locations or resurging in areas where they were previously eliminated. Satellite remote
sensing can be used to map the geographic distributions of infectious diseases and associated
vector and host species, monitor the health impacts of climatic variability and land use change,
and forecast future health risks. Remote sensing data provide information on spatially and
temporally variable environmental characteristics of the Earths surface that influence pathogen
development, habitat suitability for vector and host species, and disease transmission to humans.
However, a major limitation of this approach is that satellite-derived variables are typically
indirect measurements of the proximal environmental factors influencing mosquito populations
and disease risk. Furthermore, most public health organizations lack the technical knowledge and
specialized tools needed to utilize remote sensing data effectively.
To address these issues, we designed the EASTWeb software system to integrate satellite
remote sensing with other sources of geospatial data for modeling mosquito-borne disease risk.
The graphical user interface and overall system controls are programmed in Java, with Python
scripts used to drive the main geoprocessing subroutines. The system can acquire data from
multiple online archives, convert and reproject these data into a consistent format, compute a
variety of environmental indices, and integrate these remotely-sensed metrics with
epidemiological data in a PostgreSQL database. The environmental variables produced include
MODIS land surface temperature (LST) and normalized difference vegetation index (NDVI),
TRMM precipitation, and actual evapotranspiration (ETa) computed using data from MODIS
and FEWSNet.
We used this system to develop forecasts of West Nile virus risk for the northern Great
Plains region of the United States. Statistical models were fitted to historical WNV using
deviations of cumulative NDVI, LST, and ETa from their long-term means for each county.
Early-season models were driven by cumulative NDVI and had the lowest predictive
capabilities. Predictions of WNV risk improved when additional environmental data from late
spring and early summer was incorporated into the models. The influences of LST and ETa were
greater in these late-season models. We developed a web atlas software framework to assist in
disseminating these forecasts along with other relevant geospatial datasets. These data can be
accessed at our project website: http://globalmonitoring.sdstate.edu/eastweb. Our current
research efforts include developing a version of the EASTWeb software system that can be
disseminated to public health agencies and research labs, and improving our forecasting models
for WNV risk in the northern Great Plains and malaria risk in the Amhara region of Ethiopia. 7/11/2011
1
Satellite Remote Sensing for
Forecasting and Mapping
Mosquito-Borne Disease Risk
Mi chael C. Wi mber l y
Geogr aphi c I nf or mat i on Sci ence Cent er of
Excel l ence
Sout h Dakot a St at e Uni ver s i t y
Br ooki ngs , SD 57007
URISA 2011 GIS in Public Health Conference June 27-30, 2011 - Atlanta, Georgia - www.urisa.org
Overview
y Background
{ Disease Ecology and Forecasting gy g
{ Remote Sensing
y Research Project Overview
{ Computer System Development
{ Predictive Modeling WNV Example
{ Web-Based Dissemination of Results
y Summary
Summary
y Follow-up Talks
{ Chuang Mosquito Populations
{ Midekisa Malaria Epidemics 7/11/2011
2
Background Disease Forecasting
y Public health strategies for g
disease control and prevention
y Challenge of planning for
disease epidemics and other
unpredictable events
{ Risk of not responding
i k f di
{ Risk of overresponding
y Importance of accurately
forecasting future disease
outbreaks
Background Disease Ecology
URISA 2011 GIS in Public Health Conference June 27-30, 2011 - Atlanta, Georgia - www.urisa.org 7/11/2011
3
Background Disease Forecasting
Higher
Seasonal
Uncertainty
Seasonal
Forecasts
Case
Environment
Monitoring
Early Warning
Early Detection
Years Months Weeks
Lead Time
Lower
Surveillance
Adapted from Da Silva et al. (2004) Improving
epidemic malaria planning, preparedness and
response in Southern Africa. Malaria Journal 3:37
Early Detection
Background - Remote Sensing
True color (red, green, and blue) False color (near and middle infrared) 7/11/2011
4
MODIS land surface temperature (red=warm, blue=cool)
MODIS (1000 m)
Landsat (30 m)
Quickbird (2 m) 7/11/2011
5
Remote Sensing Background
y Land surface phenology (MODIS, AVHRR)
Remote Sensing of Mosquito-Borne Disease
Control
Measures
Vegetation
Indices
Land Surface
Temperature
Precipitation
Vegetation
Structure
Ambient
Temperature
Near-Surface
Humidity
Human
Disease
Human
Behavior
Measures
Vector
Populations
Pathogens
Land Cover/
Land Use
Remote Sensing Zoonotic System
Standing
Water
Local Environment
Human
Biology
Host
Populations
Human System 7/11/2011
6
Project Overview
y An integrated system for the epidemiological
application of earth observation technologies application of earth observation technologies
y Specific Aims:
{ Develop a computer application for automated processing of
satellite remote sensing data to generate environmental
measurements for early warning of disease outbreaks
{ Analyze the predictive capabilities of these remote sensing
products using retrospective datasets of human disease cases
products using retrospective datasets of human disease cases
and vector populations
{ Develop web-based tools to facilitate visualization and analysis
of these forecasts and other geospatial data products by public
health practitioners
Computer Application Development
y Automatically acquire, process, and summarize data
y Generate the following environmental variables y Generate the following environmental variables
{ Vegetation indices (MODIS)
{ Precipitation (TRMM)
{ Land surface temperature (MODIS)
{ Actual Evapotranspiration (FEWSNet ETo and MODIS)
y Integrate environmental and epidemiological
g p g
datasets
{ Areal data (county-level summaries)
{ Point data (geocoded cases, mosquito traps) 7/11/2011
7
Water Mask
Zone
Shapefile
Download HDF Files
Reproject, Mosaic,
Convert to GeoTiff
Compute LST
Summary Grids
PostgreSQL
Database
Download NetCDF
Files
Reproject/Convert to
GRID
Download ETo Data
Reproject, Composite,
Convert to GRID
Compute ETf and ETa
D l d HDF Fil
Reproject, Mosaic, Compute Vegetation LP DAAC
LP DAAC
MOD11A2
FEWS Net
GES DISC
Daily TRMM
Compute
Precipitation
Summary Grids
Zonal Statistics
Zonal
Statistics
Data
Storage
Download HDF Files
p j , ,
Convert to GeoTiff
p g
Index Grids MCD43B4
Tabular
Summaries
Map
Summaries
Download Convert/
Reproject
Compute
Environmental
Indices
Database Queries
8
West Nile Virus Background
Mosquito Vectors Wild Bird
Mosquito Vectors
Culex pipens (eastern US)
Culex quinquefasciatus
(southern US)
Culex tarsalis (Northern
Great Plains)
Wild Bird
Reservoir Hosts
Northern cardinal
American robin
Common ground dove
House sparrow
Northern mockingbird
Dead-End
Hosts
West Nile Virus Results
Regional Patterns of
WNV Incidence
9
Ecological Forecasting of WNV
Mosquito
Populations
Bird
Populations
Mosquito
Infection Rate
Mosquito
Populations
Amplification Transmission
Start of
Spring
May-June
Weather
July-Aug
Weather
Winter
Weather
West Nile Virus Methods
y Dependent Variable Log relative risk (LRR)
computed for each county in each year computed for each county in each year
y Independent Variables Anomalies of
cumulative RS indices computed at different days
of the year (DOY 105-217 at 8-day intervals).
y Fit generalized additive models (GAMs) to
historical environmental and case data
y Make predictions using 2011 environmental data 7/11/2011
10
Cumulative ETa (DOY 208)
Cumulative ETa Anomalies
11
C l i LST d ET
0
.
3
0
.
4
0
.
5
0
.
6
A
d
j
u
s
t
e
d

R
2
Cumulative NDVI, measuring
the timing of spring greenup is
Cumulative LST and ETa,
measuring accumulated degree
days and moisture, become
more important in the late-
season models
120 140 160 180 200 220
0
.
2
DOY
the timing of spring greenup, is
the most important predictor in
the early season models
DOY 137 DOY 209
0
.
5
1
.
0
1
.
5
2
.
0
O
b
s
e
r
v
e
d
2004
2005
2006
2007
2008
2009
2010
0
.
5
1
.
0
1
.
5
2
.
0
O
b
s
e
r
v
e
d
2004
2005
2006
2007
2008
2009
2010
0.0 0.5 1.0 1.5 2.0
0
.
0
0
Predicted
0.0 0.5 1.0 1.5 2.0
0
.
0
0
Predicted 7/11/2011
12
Web Atlas
y Implemented as a
software framework software framework
y XML database of
digital map products
y Includes a content
management system
for easy update y p
y Goal is to allow
automatic updates
13
Testing Use Cases for Web Visualization
Key Points - WNV
y WNV epidemic in 2003 followed by high incidence in
2005, 2006, and 2007 2005, 2006, and 2007
y Relative low incidence in 2008, 2009, and 2010
y Some of this decline can be explained by climatic
variability
{ Late onset of spring
{ Cool, wet conditions in late spring and early summer
, p g y
y Suggests a potential for resurgence if climatic
conditions change
y But there are other important drivers besides climate 7/11/2011
14
Application of RS for Disease Forecasting
y Need for translational applications of disease
forecasting forecasting
y Potential for applying new types of information
{ ETa (actual evapotranspiration) - sensitive to both soil
moisture and temperature
{ New portions of the electromagnetic spectrum passive
microwave
I t f i t t d h
y Importance of integrated approaches
{ Early warning plus early detections
{ Weather forecasting
Acknowledgements
y Coauthors and
Collaborators
y Funding Sources
{ NIH/NIAID An Integrated System for
Collaborators
{ Ting-Wu Chang
{ Alemayehu Midekisa
{ Yi Liu
{ Aashis Lamsal
{ Geoffrey Henebry
{ NIH/NIAID, An Integrated System for
the Epidemiological Application of Earth
Observation Technologies (R01-
AI079411).
{ NASA Applications Feasibility Study,
Enhanced forecasting of Mosquito-Borne
Disease Outbreaks Using AMSR-E
(NNX11AF67G)
NASA E th d S S i F ll hi
{ Mike Hidreth
{ Gabriel Senay
{ NASA Earth and Space Science Fellowship
to Alemayehu Midekisa, Integrating
Multi-Sensor Satellite Data for Malaria
Early Warning in the Amhara Region of
Ethiopia.
http://globalmonitoring.sdstate.edu/eastweb 7/15/2011
1
Environmental Modeling of Mosquito
Population Dynamics: A Comparison of
Ground-Based Measurements and
Satellite Remote Sensing
TI NG- WU CHUANG, PH. D.
GEOGRAPHI C SCI ENCE CENTER OF EXCELLENCE,
SOUTH DAKOTA STATE UNI VERSI TY
Satellite Images
URISA 2011 GIS in Public Health Conference - Atlanta, Georgia - www.urisa.org
Image of global observations from MODIS 7/15/2011
2
Satellite Images
y Digital elevation
model (DEM) of
Sierra Nevada
Satellite Images
Hurricane Katrina near Florida
URISA 2011 GIS in Public Health Conference - Atlanta, Georgia - www.urisa.org 7/15/2011
3
Remote Sensing Techniques and Environmental
Indicators
y Satellite remote sensors record the electromagnetic
radiation reflected or emitted by the Earth surface y
y Multiple environmental indicators can be derived from
remote sensing image products
--land cover/land use
--climate: land/sea surface temperature, precipitation
--hydrology: soil moisture, water fraction
--vegetation: NDVI, EVI
Pros (a) Contin o s and a tomated data collection y Pros (a) Continuous and automated data collection
(b) Multiple environmental indicators
y Cons (a) Complicated data processing
(b) Cloud cover
Remote Sensing Technique and Environmental
Indicators
y Spatial resolution
--the size of pixels that recorded in a raster image
e.g. 30-meter, 1KM
y Temporal resolution
--the frequency of flyovers by the satellite
e.g. daily, 8-day composite
y Spectral resolution
--the wavelength width of the different frequency bands
recorded (e.g. visible light, infrared, or microwave etc) recorded (e.g. visible light, infrared, or microwave etc)
4
Remote Sensing, Climate Variations and Mosquito-
Borne Diseases
y Monitoring mosquito abundance plays an important role
on vector-borne disease prevention and vector control p
y Climate changes/variations alter the ecological niches and
population dynamics of disease vectors
y Remote sensing has been applied to study vector-borne
disease since 1985 (Hayes, 1985)
Using Ikonos Landsat
Using Ikonos, Landsat
7, and aerial photos to
predict vector habitats
in the western Kenya
highlands (Emmanuel
et al. 2006, Malaria J)
Remote Sensing, Climate Variations and Mosquito-
Borne Diseases
Using Landsat ETM+
to predict An. to predict An.
Hyrcanus density in
Southern France (Tran
A. et al, 2008,
International J Health
Geographics)
5
Remote Sensing and Weather Stations
y Weather station data provides direct measurement of
environmental variables (air temperature, precipitation, p , p p ,
and relative humidity) which can be used to forecast
mosquito population dynamics
y However, weather stations are not ubiquitous particularly
in rural areas
y Satellite remote sensing technique could compensate for
th i d i f th t ti d t d id the inadequacies of weather station data and provide more
diverse environmental information
Study Aim
y The advanced microwave scanning radiometer- y The advanced microwave scanning radiometer-
earth observing system (AMSR-E) will be
compared with weather stations and the moderate
resolution imaging spectroradiometer (MODIS) to
predict the population dynamics of Culex tarsalis
in Sioux Falls, SD, 2005-2008
6
The Advanced Microwave Scanning Radiometer-
Earth Observing System (AMSR-E)
y AMSR-E is the total power passive
microwave radiometer system on the Aqua
S t llit (l h d i ) hi h d t t Satellite (launched in 2002) which detects
Earth-emitted microwave radiation
y The long wavelength of microwaves can
penetrate through cloud, haze, dust, and
heavy rainfall
y Environmental parameters derived from
AMSR-E include near-surface air
temperature soil moisture water fraction temperature, soil moisture, water fraction,
atmospheric water vapor, and vegetation
index
y Temporal resolution: Daily
y Spatial resolution: 0.25 degree ~(25 KM)
Moderate Resolution Imaging Spectroradiometer
(MODIS)
y MODIS is the key instrument on the
Terra (launched in 1999) and Aqua
t llit hi h i 6 t l satellites which acquires 36 spectral
bands ranging in wavelength from 0.4 m
to 14.4 m (visible light, Near IR, Short
Wave IR, and Mid Wave IR)
y Multiple variables can be derived from
MODIS (e.g. land surface temperature,
normalized difference vegetation index,
and actual evapotranspiration (ETa)) and actual evapotranspiration (ETa))
y Temporal resolution: 8-day or 16-day
y Spatial resolution: 250m, 500m, and 1KM
Land surface temperature in early July in
SD, 2010 (MOD11A2 product) 7/15/2011
7
Vegetation Index
y Normalized Difference Vegetation Index (NDVI)
--indicators of plant growth and green vegetation cover p g g g
--linkage between start-of-season (SOS) and mosquito
emergence
y Enhanced Vegetation Index (EVI)
--remove atmosphere contamination caused by smoke or
clouds
NDVI in January NDVI in July
Evapotranspiration
y The sum of evaporation and plant
transpiration from the land surface to p
atmosphere
y An important component of water cycle
which accounts the water movement,
soil, and canopy conditions
y Potential indicator of mosquito habitat
availability availability
y Actual evapotranspiration (ETa) vs.
potential evapotranspiration (PET)
--surface energy balance (SEB) model
8
Study LocationSioux Falls
y The largest city in
South Dakota, ,
population 157,935
y Urban/Suburban
landscape
y Surrounded by
cultivated croplands,
pastures, and hayfields pastures, and hayfields
y Big Sioux river passes
through the city
Study Designdata collection
y 24-30 CDC-CO
2
Light Traps were set within or near Sioux Falls from
Mon to Thu and mosquito samples were collected the next day
l li h k f l { Culex tarsalisthe key vector of West Nile virus (WNV) transmission in
the Northern Great Plains
y Ground-based weather data was collected from three weather stations
in Sioux Falls
{ Daily average temperature, precipitation, and relative humidity
y Remote sensing based weather data were from the
{ AMSR E Daily air temperature water fraction soil moisture and { AMSR-E Daily air temperature, water fraction, soil moisture, and
vegetation index
{ MODISLand surface temperature, NDVI, and ETa (generated from
daytime land surface temperature and potential evapotranspiration using
the surface energy balance (SEB) model)
9
Study designdata analysis
y Study period was from 2005-2008
y Environmental parameters of AMSR-E and weather Environmental parameters of AMSR E and weather
stations were averaged on 8-day basis, which is concordant
with the temporal resolution of MODIS
y 3 temporal lags were generated to capture the lagged effects
of environmental influences on mosquito populations
y Polynomial distribution lag (PDL) models were used to deal
with the time series data with high collinearity between with the time series data with high collinearity between
independent variables
y Corrected Akaike information criterion (AICc) was used to
compare the influences of environmental variables in
different models
10
Comparisons of Temperature Measurements from
Three Different Sources
PDL Models for AMSR-E
Parameters Environmental Variables AICc
AICc
Difference
4
Air Temperature
Water Fraction
95 32 0
The best model contained air
temperature water fraction 4
Soil Moisture
Vegetation Index
95.32 0
3
Air Temperature
Water Fraction
Soil Moisture
Vegetation Index
122.86 27.54
3
Air Temperature
Water Fraction
Soil Moisture
Vegetation Index
111.42 16.1
Ai T t
temperature, water fraction,
and vegetation index
Vegetation index was the
most important factor
AMSR-E model had lower
AICc than the models based
on MODIS and weather
3
Air Temperature
Water Fraction
Soil Moisture
Vegetation Index
93.28 -2.04
3
Air Temperature
Water Fraction
Soil Moisture
Vegetation Index
157.12 61.8
stations 7/15/2011
11
PDL Model for MODIS
AICc
Difference
The best model contained all
Difference
3
LST
NDVI
ETa
118.99 0
2
LST
NDVI
ETa
120.97 1.98
2
LST
NDVI
ETa
129.19 10.2
LST
The best model contained all
the variables
ETa was the most important
variable to predict mosquito
abundance
2
S
NDVI
ETa
183.78 64.79
PDL Models for Weather Station
AICc
Difference
The best model contained
Difference
3
Temperature
Relative Humidity
Precipitation
196.24 0
2
Temperature
Relative Humidity
Precipitation
310.17 113.93
2
Temperature
Relative Humidity
Precipitation
227.67 31.43
The best model contained
all the variables
Temperature was the most
important variable
2
Temperature
Relative Humidity
Precipitation
223.42 27.18
12
Model performance-AMSR-E
The AICc of the best
model=93.28
Variable effects:
--Air temperature (+) at lag=0-1
--Water fraction (+/-)
--Vegetation index (+) at lag=0-1
The seasonal pattern of
mosquito abundance was
captured by the AMSR-E model
Vegetation index was the most
important factor
Model performance-MODIS
model=118.99
Variable effects:
--LST (-) at lag=0
--NDVI (+/-)
--ETa (+) at lag=0-1
Both the seasonal mosquito
abundance and inter-annual
variations were capture by
the MODIS model
the MODIS model
ETa was the most important
factor 7/15/2011
13
Model performance-Weather Stations
model=196.24
Variable effects:
--Temperature (+) lag=0-3
--Precipitation (-) lag=0-3
--Relative humidity (+) lag=0-2
The seasonal mosquito
abundance and was captured
by weather stations but the
k l i d
peak values were missed
except for 2006
Temperature was the most
important parameter in the
model
Summary
y Both AMSR-E and MODIS data have better performance on
predicting Cx. tarsalis abundance than weather stations p g
y The land surface temperature (LST) from MODIS is a good
proxy of air temperature (r>0.9)
y AMSR-E successfully predicted the seasonal patterns of
mosquito abundance and had lower AICc than other
models
(1) the prediction curve captured seasonal but not (1) the prediction curve captured seasonal, but not
inter-annual variations
(2) the data quality of AMSR-E during winter is not
reliable
14
Summary
y MODIS data also predicted the mosquito abundance well
-- the inter-annual variations of mosquito q
abundance were captured by the MODIS data
y Weather station data can predict the seasonal pattern of
mosquito abundance but the peak of values was missed
y Vegetation indices and evapotranspiration are the key
variables in the satellite remote sensing models that are not
measured by weather stations measured by weather stations
Summary
y Satellite remote sensing data can be used to model and
forecast mosquito populations when local weather stations q p p
are not available
y The diversity of environmental parameters generated from
remote sensing data can improve the prediction power of
statistical models
y RS is not simply a proxy for WS measurements
y RS provides measurements that are different and y RS provides measurements that are different and
potentially more relevant for mosquito ecology (surface
moisture vs. precipitation, more localized temperature
measurements)
15
Future Study
y Apply the statistical models to multiple locations to verify
the transferability of the needs to different landscape y p
environments
y Evaluate both disease incidence and mosquito
infection/abundance by using remote sensing products to
enhance disease prevention and vector control in public
heath practice
Acknowledgements
y People
{ Dr. Michael C. Wimberly, Professor, South Dakota State University
{ Dr Geoffrey M Henebry Professor South Dakota State University { Dr. Geoffrey M. Henebry, Professor, South Dakota State University
{ Dr. Michael B. Hildreth, Professor, South Dakota State University
{ Dr. John S. Kimball, Professor, the University of Montana
{ Denise L. VanRoekel, Health Program Coordinator, Sioux Falls Health Department
y Institutions
{ Geographic Information Science Center of Excellence, South Dakota State University
{ The University of Montana
{ Sioux Falls Health Department
F di S t y Funding Support
{ NIH/NIAID, An Integrated Systems for the Epidemiological Application of Earth
Observation Technologies (Ro1-A1079411)
{ NASA Applications Feasibility Study, Enhanced Forecasting of Mosquito-Borne
Disease Outbreaks Using AMSR-E (NNX11AF67G)
7/18/2011
1
Remotely Sensed Environmental Correlates of Malaria
Risk in a Highland Region of Ethiopia
Alemayehu Midekisa, PhD Student
Geographic Information Science Center of Excellence,
South Dakota State University
Outline of Talk
Background
Malaria Early Warning Malaria Early Warning
Environmental Indicators
Remote Sensing
Cross Correlation Analysis
Time Series Model
Discussion
7/18/2011
2
Background
Global Burden of Malaria
Globally there are 500 million cases
every year
Globally there are 2 million malaria
related deaths
Sub-Saharan Africa
Among all cases globally, 90% are in
the Sub-Saharan Africa
Malaria affects economic growth in
developing countries
Malaria transmission often coincides
with harvesting season
(Source: RBM/WHO/UNICEF, 2005)
Source: Roll Back Malaria, WHO
Background: Ethiopia
Ethiopia - Malaria Burden
Almost 50 million (68%) of the Almost 50 million (68%) of the
population at risk of malaria
Malaria is the number one cause of
illness & death (2003/04)
Admissions 20.4%
Hospital Deaths 27.0%
Malaria Transmission season:
September - December
April- May p y
Malaria Transmission is seasonal and
unstable
7/18/2011
3
Study Site: Amhara Region
9Located in the NW of Ethiopia
9Population of 17, 214,056
9Total area of 156 960 km
2
MODIS Satellite image taken 81 DOY , 2011
9Total area of 156,960 km
9Equivalent to half the size of
New Mexico
Malaria Early Warning
y Early warning can be assessed using environmental
indicators that are favorable for malaria transmission
y Early warning can gives us predictions of epidemics
months in advance
y It will allow time to make public health decision making
for control and prevention measures
y This lead time can be used to :
y Plan control and prevention strategy Plan control and prevention strategy
y Manage logistics
y Distribute sprays, drugs & bed nets
7/18/2011
4
Environmental Indicators
Malaria is a disease of tropical and temperate
countries between the latitudinal limits of 64 countries between the latitudinal limits of 64
North and 57 South (Gill, 1921)
Environmental factors affect malaria
transmission.
Temperature affects parasite development ( 25
- 30C no development <16C and > 40C - 30 C, no development <16 C and > 40 C
(Gilles, 1993).
Rainfall provides surface water in which female
anopheles can lay eggs.
Global Distribution of Malaria Transmission
Source: Simon I. Hay et al., 2007 The Spatial Distribution of P. falciparumMalaria Endemicity
7/18/2011
5
Remote Sensing
Environmental indicators can be measured Environmental indicators can be measured
using earth observing satellites sensors
Satellite data provides us a continuous
observation
Detailed environmental data in space
and time helps us to generate predictive
models.
Hi t i l t llit d t ill id Historical satellite data will provide a
background on retrospective case studies
Source: CCRS/CCT
Earth Observation Satellites for Monitoring Vector Born Disease
TRMM
TRMM (Tropical Rainfall Measuring Mission)
MODIS Aqua
MODIS Terra
MODIS (Moderate Resolution Imaging
Spectroradiometer
Landsat
ASTER ASTER
ASTER (Advanced Spaceborne Thermal
Emission and Reflectance Radiometer)
SRTM
SRTM (Shuttle Radar Topography Mission)
7/18/2011
6
Data
Monthly reported outpatient malaria cases at
district (woreda) level district (woreda) level
Time Series Satellite-derived indices:
{ Rainfall Estimates from TRMM sensor
{ Land Surface Temperature (LST) from MODIS sensor
{ Normalized Vegetation Index (NDVI) from MODIS sensor
{ Enhanced Vegetation Index (EVI) from MODIS Sensor { Enhanced Vegetation Index (EVI) from MODIS Sensor
{ Actual Evapotranspiration (ETa) modeled from LST
Satellite Derived Environmental Indicators
April (DOY 97, 2010)
May (DOY 121, 2010) June (DOY 153, 2010) July (DOY 185, 2010) August (DOY 217, 2010)
NDVI (Normalized Difference Vegetation Index
May (DOY 121, 2010) June (DOY 153, 2010)
July (DOY 185, 2010) August (DOY 217, 2010)
April (DOY 97, 2010)
LST (Land Surface Temperature)
7/18/2011
7
Research Questions
1. Are the major recent malaria epidemics in the 1. Are the major recent malaria epidemics in the
Amhara region triggered by environmental
factors?
2. Which environmental factors are contributing
more importantly to malaria outbreak and at what
specific lead time? specific lead time?
Time Series Modeling Approach
y ARIMA (Autoregressive Integrated Moving Average) Model
accounts for temporal autocorrelation p
y In Multivariate ARIMA models the response series
depends on both its past values and input environmental
parameters as predictors
y ARIMA Models use both case data and environmental
variables
y SARIMA (Seasonal ARIMA) Models control for seasonality y SARIMA (Seasonal ARIMA) Models control for seasonality
y Helpful to combine Early Detection and Early Warning
Approaches
7/18/2011
8
1500
2000
2500
3000
3500
4000
4500
South Achefer District
Incidence rate
0
500
1000
1500
2001 2002 2003 2004 2005 2006 2007 2008 2009
3000
3500
4000
4500
Mecha District
Incidence rate
Recent epidemics
recorded in 2002,
2003, 2005, and 2009.
Outpatient monthly
reported case data at
district level (2001-2009).
0
500
1000
1500
2000
2500
2001 2002 2003 2004 2005 2006 2007 2008 2009
South Achefer District South Achefer District Mecha District Mecha District
Temporal Association of Rainfall with Malaria Cases
in 2003 Outbreaks
150
200
250
300
2000
2500
3000
3500
4000
4500
2003
Rainfall
Incidence Rate
150
200
250
300
2000
2500
3000
3500
4000
4500
2003
Rainfall
Incidence Rate
C
a
s
e
s
a
l
l

(
m
m
)
C
a
s
e
s
R
a
i
n
f
a
l
l

(
m
m
)
0
50
100
0
500
1000
1500
1 2 3 4 5 6 7 8 9 10 11 12
0
50
100
0
500
1000
1500
1 2 3 4 5 6 7 8 9 10 11 12
R
a
i
n
f
a
Month
R
Month
7/18/2011
9
0 1
0
0.1
0.2
0.3
0.4
0 1 2 3 4 5 6 7 8 9 10 11 12
ETa
EVI
NDVI
Rainfall
2SD
S. Achefer
The cross correlation function between monthly values of Incidence
-0.4
-0.3
-0.2
-0.1
Lags
-2SD
LST
y
rate and environmental variables for the period 2001 2009
Rainfall estimate has significant positive correlation at lag 3
EVI has significant positive correlation at lags 2 and 3 months
-0.1
0
0.1
0.2
0.3
0.4
0.5
0 1 2 3 4 5 6 7 8 9 10 11 12
ETa
EVI
NDVI
Rainfall
2SD
-2SD
Mecha
The cross correlation function between monthly values of Incidence
rate and environmental variables for the period 2001 - 2009
-0.4
-0.3
-0.2
Lags
LST
9Rainfall has significant positive correlation at a lag of 2, 3 & 4 months
9EVI has significant positive correlation at lags of 2 and 3 months
9NDVI has significant positive correlation at lags of 1, 2, and 3 months
7/18/2011
10
S. Achefer District S. Achefer District Mecha District Mecha District
SARIMA Model Result
y Autoregressive term was
statistically significant
= 0.05 level
y Rainfall at lag 2 and 3
was statistically significant
0 0 l l
y Autoregressive term was
= 0.05 level
y Rainfall at lag 2 and 3
was statistically significant
0 0 = 0.05 level
y EVI at lag 1 and 3 was
= 0.05
= 0.05
1000
1500
2000
2500
3000
3500
4000
4500
S. Achefer
Observed
Predicted
0
500
2001 2002 2003 2004 2005 2006 2007 2008 2009
3000
3500
4000
4500
5000
Mecha
Predicted
Observed
0
500
1000
1500
2000
2500
2001 2002 2003 2004 2005 2006 2007 2008 2009
7/18/2011
11
Recap Research Questions
1. Are the major recent malaria epidemics in the
Amhara region triggered by environmental Amhara region triggered by environmental
factors?
2. Which environmental factors are contributing
more importantly to malaria outbreak and at
what specific lead time?
Result
Model results indicate environmental factors
contributed significantly to malaria outbreaks in
the two sites. the two sites.
Rainfall at lag 2 and 3 was found to be a
significant predictor of malaria Incidence in both
sites.
EVI was significant predictor in at lag 1 and 3.
This lead time of 1 3 months can help early This lead time of 1-3 months can help early
warning efforts.
7/18/2011
12
Summary
Accumulated rainfall was
one of the significant
en ironmental triggers of environmental triggers of
malaria
Vegetation indices such as
EVI are also proxy
indicators of rainfall
Malaria outbreaks in the
i i d b region are constrained by
presence and absence of
surface water pools
Conclusion
We found relationships between remotely sensed
climate variables and temporal patterns of malaria climate variables and temporal patterns of malaria
cases.
These results will serve as a starting point for
future modeling efforts.
These preliminary results can provide the basis for
Malaria Early Warning System for the Amhara Malaria Early Warning System for the Amhara
region.
The current modeling approach also integrates
early detection and early warning
7/18/2011
13
Institutes Institutes
People People
Acknowledgments
y Geographic Information Science Center of
Excellence, South Dakota State University,
USA
y Anti Malaria Association (AMA), Ethiopia
y Amhara Regional Health Bureau, Ethiopia
y National Aeronautics and Space
Administration (NASA) , USA
y Dr. Michael C. Wimberly, Professor, South Dakota
State University, USA
y Dr. Geoffrey M. Henebry, Professor, South Dakota
State University, USA
y Dr. Gabriel Senay, Professor, South Dakota State
University, USA
y Dr. Asrat Genet, Director, Amhara Regional Health
Bureau, Bahir, Ethiopia
y Abere Mihretie, Director, Anti Malaria Association,
Addis Ababa, Ethiopia
y Paulos Semunigus Malaria Project Officer Anti
y Paulos Semunigus, Malaria Project Officer, Anti
Malaria Association, Addis Ababa, Ethiopia
Funding Source:
o NIH/NIAID, An Integrated System for the Epidemiological Application of Earth
Observation Technologies (R01-AI079411).
o NASA Earth and Space Science Fellowship, Integrating Multi-Sensor Satellite
Data for Malaria Early Warning in the Amhara Region of Ethiopia.
Alcohol Outlets and Assaultive Violence in Philadelphia

Tony H. Grubesic, Geographic Information Systems and Spatial Analysis Laboratory, College of Information Science
and Technology, Drexel University, Philadelphia, PA 19104. (email: grubesic@drexel.edu)

Loni Philip Tabb, Department of Epidemiology & Biostatistics, School of Public Health, Drexel University,
Philadelphia, PA 19102

William Alex Pridemore, Department of Criminal Justice, Indiana University, Bloomington, IN 47405

1
AlcoholOutletsandAssaultiveViolenceinPhiladelphia
Abstract
Alcoholrelatedmorbidity,includingviolence,representsaseriouspublichealththreatthatcreates
othersocialburdensintheUnitedStates.Althoughasignificantbodyofevidencesuggeststhat
neighborhoodlevelsocialdisorganizationandalcoholoutletdensitiesarestronglyrelatedtohigher
ratesofassaultiveviolence,lessisknownaboutthespecificgeographicdistributionofviolencearound
theoutlets.Thepurposeofthispaperistoexplorethespatiallinkagesbetweenalcoholoutletsandthe
distributionofassaultiveviolenceinthecityofPhiladelphia.Usingacombinationofexploratoryspatial
dataanalysis(ESDA)methods,wefoundclustersofassaultiveviolencearoundalcoholoutletsandwere
abletodeterminethedistancesoftheseclustersfromoutletagglomerations.Wediscussthetheoretical
andpolicyimplicationsoftheseresults.
Keywords:Alcoholoutlets;Assault;GIS;SpatialAnalysis;Policy;PublicHealth
1.Introduction
Hazardous drinking presents a constellation of public health threats and social problems to
public health agencies and law enforcement officials. Consider, for example, both heavy drinking and
binge drinking in the United States. The Center for Disease Control (CDC) defines heavy drinking as
more than two drinks per day (on average) for men or more than one drink per day (on average) for
women.TheCDCdefinesbingedrinkingasdrinkingfiveormoredrinksduringasingleoccasionformen
and four or more drinks during a single occasion for women. Recent statistics from the Center for
Disease Controls Behavioral Risk Factor Surveillance System (2009) suggest that approximately 5% of
the population drinks heavily and 15% of the population binge drinks. This is significant for several
reasons.First,Kannyetal.(2011)notethatexcessivealcoholuseisthethirdleadingpreventablecause
ofdeathintheUnitedStates,responsiblefornearly79,000deathsduringeachyearbetween2001and
2005. Second, over 50% of all alcohol consumed in the United States is in the form of binge drinking
(Brewer and Swahn, 2005). From the perspective of law enforcement, binge drinking is particularly
problematicbecauseitoftenbegetsriskybehaviorresultinginoutcomessuchasmotorvehiclecrashes
andviolencenearalcoholoutlets(BrewerandSwahn,2005;Irvine,2008).
Recentacademicresearchthatfocusesontherelationshipbetweenalcoholoutletsandviolence
confirmsasignificantandpositiverelationshipbetweenthetwo.Thatis,higherdensitiesofoutletscan
belinkedtohigherlevelsofassaultiveviolence(Cunradietal.,2001;Branasetal.,2009;Franklinetal.,
2010; Livingston, 2011; Pridemore and Grubesic, 2011). This body of research typically utilizes general
linear models, such as regression, to control for local ecological characteristics (e.g., demographic and
socioeconomic), land use, outlet type or some combination of the three in an effort to isolate the
factors which promote violent outcomes (Bromley and Nelson, 2002; Gruenewald et al., 2006;
GruenewaldandRemer,2006;Livingston,2008;Zhuetal.,2004).
3
Whilethesetypesofstudiesareimportantfordevelopingabetterunderstandingoftheimpacts
of alcohol outlets and associated violence on neighborhoods and communities, one important facet of
the relationship between availability and violence remains obfuscated: the explicit geographic
connectionbetweenalcoholoutletsandassaultlocations.Inpart,thisgapinempiricalevidenceisdue
totherelianceonaggregatedataandestablishedgeographicunits(e.g.,blockgroups,censustracts)for
analysis. The resulting analyses, while valid, often suggest a coexistence of elevated levels of violence
and alcohol outlets for an entire study area. While explicitly spatial approaches (Pridemore and
Grubesic, 2011) are better structured for isolating and identifying spatial dependencies in these data
and the modeled results, a fundamental ability to pinpoint the exact geographic relationship between
outletsandassaultlocationsremainsdifficultwhenusingregressionbasedapproaches.
The purpose of this paper is to deepen our understanding of the geographic relationship
between alcohol outlets and assaultive violence. Using a combination of exploratory spatial data
analysis approaches, including proximity analysis and spatial clustering techniques, this paper explores
thegeographicnatureofcrimedispersionaroundalcoholoutletsinPhiladelphia,Pennsylvania.Building
upon the previous literature on crime displacement (Bowers and Johnson, 2003; Ratcliffe, 2005) and
dispersion (Murray and Roncek, 2008; Roncek and Maier, 1991; Scott, 2001), this paper provides a
unique, spatially focused approach for more clearly delineating statistically significant distance
relationshipsbetweenoutletsandassaults.
2.Background
The linkages between alcohol and violence are well established in criminology, sociology,
epidemiology,publichealth,psychology,neurobiologyandelsewhere(Giancolaetal.,2003;Koobetal.,
1997; Parker and Rebhun, 1995; Pernanen, 1991). From a pharmacological perspective, the effects of
alcohol are generally viewed as being disinhibitory, smoothing the way for participation in a broad
4
spectrum of risky behavior and its consequences, ranging from dangerous driving patterns to sexual
aggression and other forms of violence (Lane et al., 2004). Recent work suggests a significant dose
responserelationshipbetweenbloodalcohollevelandviolence(Macdonaldetal.,2005).
Given the known effects of alcohol on humans, the resulting social interaction between
inebriated individuals varies widely. Again, these span a continuum, ranging from relatively normal
socialrelationstoviolenceandotherformsofcriminalactivity(Pernanen,1991).Whilethetriggersfor
alcohol related violence are diverse, including mental disorders, neuropsychological vulnerability,
personality factors or combinations and interactions of other drugs (HaggardGrann et al., 2006), the
impactofsocialcontrolsonalcoholandriskybehaviormustalsobeconsidered.Forexample,incertain
settings, the presence of a capable guardian (Cohen and Felson, 1979) to help thwart alcoholrelated
criminal activity and violence is an important consideration. That said, because bars and offpremise
alcohol outlets are also locations where motivated offenders and potential victims converge, those
lackingadequateguardianshiporpossessingotherriskycharacteristicsmayprovidepermissivesettings
wheredeviantbehaviorcanbloom.Attheinterfaceofthesetwoconditions(i.e.,guardianshiporlackof
it), Murray and Roncek (2008) note that some alcohol outlets, particularly bars, encourage patrons
involved in confrontations to leave the premises. As a result, violence can spill into surrounding area
(e.g.,alleysandnearbystreets)escalatingthepotentialforinjuries.
Thisspillovereffectintoaneighborhoodisalsoknownasdisplacement(Green,1995).Ineffect,
violent confrontations that would have occurred in a bar if no capable guardian been present are
displaced. Instead, the confrontation is hosted in a location where no guardianship is present (e.g., a
nearby vacant lot). A related concept is that of dispersion, where multiple, unrelated incidents are
distributedaroundanalcoholoutlet(MurrayandRoncek,2008),astheoutletsmayactashoststosuch
behavior for several reasons due to its use as a social gathering spot and because patrons and others
5
maybeundertheinfluence.Inthiscontext,anabilitytoidentifythegeographicrangeoftheeffectof
alcohol outlets on assaultive violence is important. For example, liquor control boards may use this
informationtomanagethespatialdispersionofoutletsandliquorpermitsinanefforttoavoidpotential
agglomeration effects of outlets and violence (Grubesic and Pridemore, 2011). This information may
alsobeusedtohelpcapturethesecondaryeffectsofalcoholrelatedcrimeandviolenceonsurrounding
businesses,whereissuesofvandalismortheperceptionofsafetyimpactslocalretailers.
Measuring the displacement and dispersion of alcoholrelated crime and violence is typically
doneinoneoftwoways.Thefirstapproachutilizesproximityanalysis(Murrayetal.,2001)tocapture
spatialdisplacement.Specifically,radialbufferzones,orsomeversionofthemwithalternativeshapes,
are used to capture and summarize the distributions of crime around an area of interest (e.g., bar,
school, park) (Bowers and Johnson, 2003; Roncek and Lobosco, 1983). In addition to issues associated
with buffer shapes, identifying the appropriate buffer distances (i.e., size) is important. Depending on
the morphological characteristics of neighborhoods, cities, or regions, a onesizefitsall approach to
buffer size may not be appropriate. In many instances, existing public policies may also dictate buffer
size selection and use. For example, the use of spatial restriction zones that limit the locations where
sex offenders can establish a permanent residence typically range between 500 and 2,500 feet from a
sensitivefacilitysuchasaschool,busstoporpublicpark(Grubesicetal.,2007).Regardless,theuseof
radialbuffersiscommoninawidevarietyofcriminologicalstudies(MurrayandRoncek,2008).
A second approach focuses on the relationships between crime and its distribution via the
geographic contiguity of areal units. Often referred to as spatial adjacency (Anselin, 1988), this
approach is able to capture a wide variety of geographic trends, including crime displacement
(Gruenewald et al., 2006). Although adjacencybased approaches are not as geographically flexible as
radial buffers, different definitions of adjacency are possible. For example, both rook and queen
6
contiguity matrices can be constructed, as well as distancebased or knearest neighbor adjacency

measuresamongothers(Anselinetal.,2006).
In this paper we opt for an alternative approach for capturing the geographic relationship
between alcohol outlets and assaultive violence. Building upon the work of Grubesic and Pridemore
(2011), we utilize a focused test for spatial clustering (Waller et al., 1992) for exploring the geographic
distribution of violence around outlets. Unlike the use of adjacency techniques, radial buffers, or more
generalizedclusteringapproaches,thescoretest(Walleretal.,1992)utilizespredefinedlocales(inour
case, alcohol outlets) for exploring increased risks of particular outcomes (in our case, assaults).
Further, because observed distributions are tested against a benchmark distribution (e.g., Poisson),
hypothesistestingispossible.
3.StudyArea,DataandMethodology
3.1StudyArea
ThestudyareaforouranalysiswasthecityofPhiladelphia.In2010,Philadelphiawashometo
approximately1.5millionpeoplelivingwithin17,227cityblocks(EASI,2011).Theaveragepopulation
perblockwas88residents.Demographically,Philadelphiaissomewhatdiverse,withanethnic
distributionthatisprimarilywhite(41%),black(43.4%)andHispanic(12.2%)(Census,2010).The
medianageofaresidentinPhiladelphiaisapproximately34years,whichisslightlylowerthanthe
nationalmedianof35.4years(Census,2010).ThemedianhouseholdincomeforPhiladelphiais
$37,090,whichissignificantlybelowthestatemedianof$50,702(EASI,2011).Theviolentcrimeratein
Philadelphiafor2009was1,238per100,000,withanaggravatedassaultrateof576per100,000.Both
ofthesemeasuresaresignificantlyabovethenationalaverageof429.4and262.8,respectively(FBI,
2009),butnotdramaticallydifferentfromotherlargecitiesintheUnitedStates.Forexample,violent
7
crimeratesinMinneapolis,MN(1,109),Miami,FL(1,189),Chicago,IL(1,212)andWashington,DC
(1,265)areallcomparable.
3.2Data
Theunitsofanalysisforthisstudywerecensusblocks,whicharethesmallestunitsofgeography
forwhichtheCensusBureaupublicallyreleasesdatafromthedecennialsurvey.Bydefinition,city
blocksareboundedonallsidesbyeitherpolitical,transportationorothergeographicfeatures(Census,
2000).ForthecityofPhiladelphia,17,227blocksexistedintheyear2000.Unfortunately,blocksfrom
the2010CensuswerenotyetavailableforPhiladelphiaatthetimethispaperwaswritten.Thatsaid,it
isunlikelythattheoverallresultswouldbeimpactedfrommodestchangesinblockaggregationsor
theirassociatedgeometries.Fromamodelingperspective,unlikeotherstudieswhichexclude
nonresidentialblocks,thisanalysisincludesallcityblocksforPhiladelphia.Therearetworeasonsfor
this.First,wewerenotconcernedinthisstudywiththedemographicandsocioeconomicconditions
withinblockswhenanalyzingthelinkagebetweenalcoholoutletsandassaultiveviolence,asthese
relationshipsarealreadylargelyestablishedintheexistingliterature(BromleyandNelson,2002;
GrubesicandPridemore,2011;Gruenewaldetal.,2006;Livingston,2008;Zhuetal.,2004).Instead,we
focusondeterminingtheexplicitgeographicrelationshipbetweenoutletsandclustersofassaultive
violence.Therefore,theproblemscreatedwhenattemptingtocaptureandcontrolforthedemographic
andsocioeconomicconditionsfornonresidentialareasareanonissue.Second,becauseaportionof
thisanalysisisconcernedwiththeriskofassaultiveviolence,heterogeneitiesinpopulationdistributions
betweenresidentialandnonresidentialareasofthecityareofinterestbecausepopulationwillbeused
asacontrolvariable.Thus,theuseofallcityblocksinPhiladelphiaiswarrantedandareaswithasmall
populationbuthighlevelsofassaultiveviolencecouldbeparticularlyinteresting.Populationestimates
for2010foreachblockwereobtainedfromEASI(2010).
8
Weareinterestedspecificallyinseriousviolence,andthuscollecteddataonaggravatedassaults
only,assimpleassaultencompassesabroadrangeofbehaviorthatisnotnecessarilyviolentinnature.
AllaggravatedassaultsdocumentedbythePhiladelphiaPoliceDepartment(PPD)for2010were
collectedfromtheppdonline.orginterfacemaintainedbythePPD.Allreportedaggravatedassaults
(withandwithoutweapons)wereutilizedforanalysis,amountingto8,704for2010.ThePhiladelphia
PoliceDepartmentusesUniformCrimeReporting(UCR)definitionsforallreportedcrimesinthecity.In
thisinstance,aggravatedassaultisdefinedasanunlawfulattackbyonepersonuponanotherforthe
purposeofinflictingsevereoraggravatedbodilyinjury.Thistypeofassaultisusuallyaccompaniedby
theuseofaweaponorbyothermeanslikelytoproducedeathorgreatbodilyharm(FBI,2010).These
dataweregeocodedandassignedlatitudeandlongitudecoordinatesbythePPD.Publishedreportsand
PPDdisclaimersindicatethatgeocodeddataexhibitanaccuracyrateof96%98%,whichismorethan
sufficientforthisanalysis.
Finally,alcoholoutletdatawereobtainedfromthePennsylvaniaLiquorControlBoard(PLCB)
permitsystemforNovember2010.Insum,atotalof2,000establishmentswereactivelysellingalcohol
inPhiladelphiaduringthatmonth.SubsequentinquiriestothePLCBsuggestthatthisisarelatively
stablefigure,althoughtherearesomeminorvariationsinpermitseachmonthwhenoutletseither
permanentlycloseoropenforbusiness.TherearefourmajoroutlettypesinPhiladelphia:clubs(5.9%),
distributors(5.6%),eatingplaces(6.5%)andrestaurantsandbars(75%).Itisimportanttonotethatall
hotels,restaurantsandeatingplacesarepermittedtosellsixpacksofbeertogo(limitedto192fluid
ounces),whiledistributorsmayonlysellcasesofbeer(24count)orsinglecontainersofatleast128
ounces(i.e.,minikegsorlarger)forwalkincustomers.Locationscorrespondingtoairports,distilleries,
largescaledistributorsandvenuesservingsacramentalwinewereeliminatedfromtheanalysis.Asa
result,1,964outletswereretainedandgeocodedusingtheArcGISv.10geocodingengine.Onlyoutlets
9
assignedastreetlevelmatchwereutilizedinthefinalanalysis.Theseareconsideredrooftophitsand
representthebestpossibleoutcomefromageocodingalgorithm.
3.3Methods
Threegroupsofexploratoryandconfirmatoryapproacheswereusedtohighlightthespatial
distributionsofalcoholoutletsandassaultiveviolenceinPhiladelphia.Thefirstgroup,leveragesbasic
cartographicanalysis(e.g.,visualizationofassaultdensities)andsimplemeasuresofassaultrisk.Where
riskisconcerned,consideraregion,
,andthetotalpopulationwithintheregion,
.Althougha
proportionalmeasureofrawrisk(
)canhighlightdiscrepanciesinassaultsbyblockthrough
theuseoftotalblockpopulationasacontrol,measuresofrelativeriskcanprovidemorecontextby
comparingtherateateachlocationtotheoverallcitywidemean:

Inthiscase,nisthenumberofblocksinthestudyregion.Oncethisaveragemeasureofriskis
calculated,itcanbeusedastheexpecteddistributionforcalculatingrelativerisk,where
.
The second group of approaches utilizes local spatial statistical analysis to help uncover
geographicconcentrationsofalcoholoutlets.Specifically,thelocalMoransIstatistic(LISA)wasusedto
identify locations within Philadelphia where agglomerations of alcohol outlets exist. It is specified as
follows(Anselin,1995):

where
and
areobservationsforlocationsiandj(withmean)
and
10
spatialweightsmatrixwithvaluesof0or1,basedonknearestneighbors(k=4)
There are several reasons for selecting this local approach. In effect, the local Morans I
identifies statistically significant ( clusters of blocks that display a high density of alcohol
outlets.Weareparticularlyinterestedinthehighhighcategorygeneratedinthisanalysis,whereblocks
with a high density of alcohol outlets are surrounded by blocks with similarly high densities. In this
application, a knearest neighbors (KNN) approach is used to delineate local neighborhoods for each i.
KNNusesdistance,ratherthancontiguity,todefineneighbors.ItiscomputedastheEuclideandistance
between block centroids, where k refers to the number of neighbors of a location. It is a particularly
effective approach when observational units have varied sizes and shapes. While this is not a major
concern for Philadelphia, which is one of the original cities in North America to use a grid plan (Grant,
2001), there is enough diversity in block morphology throughout the city to merit the use of a KNN
weights matrix. Also, where the weights election is concerned, a sensitivity analysis of the KNN
approach was conducted for Philadelphia, where values of k = 4 through k = 10 were evaluated. K = 4
wasultimatelyselectedastheoperationaldefinitionfordelineatingagglomerationsofalcoholoutletsin
this study for two reasons. First, it provided the most compact and parsimonious agglomerations of
highhighblocks.Second,ithadthemoststatisticalpower,withI=0.2346.Asecondfacetofthislocal
analysis was the use of Empirical Bayes (EB) rates in the LISA statistic (Assuncao and Reis, 1999; Bailey
and Gatrell, 1995). This standardizes raw rates to obtain a constant variance through a rescaling
procedure. Its purpose is to minimize the chances of spurious inferences with the Morans I test.
Finally, there is the nagging question associated with the control variable used for estimating the LISA
with EB rates. In previous work, roadway miles has been used to better account for areas (e.g.,
entertainment districts) that attract a substantial share of nonresidential traffic (Grubesic and
Pridemore, 2011; Gruenewald et al., 2006; Livingston, 2008). While this is a viable option, Livingston
11
(2008)notesthatbothpopulationandroadwaymilestypicallyfindsimilarresults.Further,becausethis
analysis makes use of census blocks, there would be difficulties associated with aggregating roadway
mileage to polygonal units (i.e., blocks) that are frequently delineated and/or bounded by actual
roadways. In essence, deciding which blocks to assign road mileage to, without double counting or
undercountingwouldbegeometricallycomplex.Forthesereasons,weoptforblockpopulationasour
controlvariable.
Thethirdapproachusedinthisanalysiswasconfirmatoryinnatureandtestedthepotential
impactsofalcoholoutletagglomerationsonviolence.Thisfocusedtestofspatialclustering(Walleret
al.,1992)utilizespredefinedlocationstodetermineifthereisanincreasedriskforaparticular
outcome.Wetestedthenullhypothesisthat:
H
0
Therearenoclustersofaggravatedassaultsaroundthealcoholoutletagglomerations.
Althoughinformationregardingthemagnitudeofexposuretotheoutletagglomerationsisnonexistent,
exposurecanbeoperationalizedasafunctionofdistance.Thus,asdistancefromtheoutlet
agglomerationsincreases,exposuredecreases.Becauseoutletagglomerationsaregeographically
definedasablockorgroupsofblocksinthisstudy,arepresentativepointwillbeusedtohelp
operationalizethefocusedclusteringtest.Specifically,agglomerationcentroidswillbeusedforthis
purpose.TheScoreStatistic(Walleretal.,1992)isspecifiedasfollows:

Thescoreteststatistic,
sc
T ,representsthesumofthedifferencebetweentheobserved(
i
O )and
expected(
i
E )assaultcountsateachlocation( n i,..., ),weightedbytheexposuretotheoutlet
agglomeration.Inversedistance,
i
d 1 ,isusedastheweightforeachobservation.Again,thenull
12
hypothesisforthistestisthattheobservednumberofcasesineachblockisindependent,distributedas
Poissonrandomvariableswithacommonassaultfrequency.Thealternativehypothesisisthatthe
observednumberofcasesineachregionisindependent,distributedasPoissonrandomvariableswhere
theassaultfrequencyisaproportionallyincreasingfunctionofexposure.Underthenullhypothesis,
sc
T shouldequal0.SignificanceisobtainedthroughMonteCarlosimulation(n=999).
4.Results
Figure1displaysthestudyarea,Philadelphia,PA,andthespatialdistributionofalcoholoutlets
forNovember2010.Asnotedpreviously,1,964outletsareincluded,rangingfromclubsanddistributors
toeatingplaces,restaurantsandbars.ThedenseclusterofoutletsonthemapcorrespondstoCenter
City(i.e.,thecentralbusinessdistrict),whichisoneofseveralmajordiningandentertainmentareasin
Philadelphia.
Table1andFigure2provideasuiteofdescriptive,distributionalstatisticsassociatedwith
aggravatedassaultsandassaultdensitypersquaremileinPhiladelphiafor2010.Theaverageassault
densityforPhiladelphiablocksisapproximately137persquaremile,orabout0.5assaultsperblock.Itis
interestingtonotethat95%ofPhiladelphiablockshadtwoorfeweraggravatedassaults,with5%of
blockscontaining42.1%ofaggravatedassaultsinPhiladelphiain2010.Figure2highlightsthe
geographicdistributionofaggravatedassaultdensitiesforthecitythroughastandarddeviationmap.
Areasingrayrepresentlocationswhereaggravatedassaultsaresignificantlybelowaverage,whilethose
representedinyelloware0.51.5standarddeviationsaboveorbelowthemeanof137.Areas
representedwithorangeandredarePhiladelphiablockswhereassaultsarerelativelyhigh,with
standarddeviations1.5orgreaterthanthemean.ManyoftheseareasarefoundinNorthPhiladelphia,
anotoriouslyviolentsectionofthecity,withseveralothersinWestPhiladelphiaorSouthPhiladelphia.
13
AfinaloverviewofassaultsispresentedbyFigure3whichdisplaystheexcessriskofassaultfor
eachblock.Fromaninterpretiveperspective,thelegendhighlightsareaswheretheobservedratioof
assaultdensitytopopulationdensityisxtimeshigherorlowerthanwouldbeexpectedbasedonthe
ratioforthecityofPhiladelphia.Forinstance,areashighlightedinredrepresentblocksthathavean
excessriskatleastfourtimesthatofthecityasawhole.Again,thesemaybeproblematiclocales,but
sincethereisnostatisticalsignificanceassignedtothesederivedrates,amorerigorousanalysisis
required.
4.1LocalStatisticalAnalysis
Figure4displaysthespatialdistributionofalcoholoutletagglomerationsinPhiladelphiathat
weregeneratedwiththelocalMoransIapproachwithEBweights.Recallthatk=4wasultimately
selectedastheoperationaldefinitionfordelineatingagglomerationsofalcoholoutletsinthisstudy.Not
onlydiditprovidethemostcompactandparsimoniousagglomerationsofhighhighblocks,italsohad
themoststatisticalpower,withI=0.2346.Shadingissimplyusedtodifferentiatetheblocksfromtheir
neighboringagglomerations.Therearetwoimportantfacetsoftheseresultstokeepinmind.First,
whilethevastmajorityoftheagglomerationsconsistofmorethanoneblock,therearesingleblocks
(e.g.,Agglom3,Agglom12)thatcompriseacompleteagglomeration.Again,theseareareaswhere
multipleoutletsarepresentandtheiroutletdensitiesaresignificantlyhigherthantheirneighboring
areas.Forthepurposesofdisplay,theinsetmapforFigure4isfocusedonCenterCityandits
surroundingneighborhoods,buttherearenumerousagglomerationsofalcoholoutletsthatarelocated
outsidethisregion.Thereareseveralreasonsthatalargenumberofalcoholoutletagglomerationsare
foundinCenterCity.First,thepopulationofCenterCityhasincreased27%since1990(CCD&CPDC,
2011).Secondthenumberofdowntownhotelroomsincreased95%since1991,from5,677toover
11,000,reflectingincreasedlevelsoflocaltourismandconventionactivity(CCD&CPDC,2011).When
14
thesetwofactorsarecombined,thedemandforestablishmentsservingbothfoodandalcoholis
dramaticallyaffected.Forexample,in1992only65finediningrestaurantsoperatedinCenterCitybut
bytheendof2010,therewere278.Moreimpressively,althoughoutdoorcafeswerenonexistent
priorto1995,213nowexistandofferover3,500seatstopatrons(CCD&CPDC,2011).Whilenot100%
oftheseestablishmentsarelicensedtosellalcohol,mostare.
Table2providesaslightlyalternativeviewoftheoutletagglomerationsdelineatedforanalysis,
providingsomebasicsummarystatisticsforeach.Perhapsthemostnotableaspectofthe
agglomerationcompositionsisthedominanceofrestaurants.Again,unlikemanyplacesintheUnited
States,beerandwinearenotavailableingrocerystoresorlocalconveniencestoresinPennsylvania.
Instead,beerandwinearesoldbystatelicenseddistributorsorsoldforcarryout(inlimitedquantities)
byrestaurants.Asaresult,theoverwhelmingpresenceofrestaurantsintheagglomerationsistobe
expected.
4.2AgglomerationsandClustersofViolence
Table3providestheresultsofthefocusedtestforspatialclusteringbetweenaggravated
assaultsandalcoholoutletagglomerationsforPhiladelphia.Thereareseveralfacetstothistablethat
needtobehighlighted.First,theteststatisticforeachagglomerationanditsassociatedpvalueare
reported.Highervaluesoftheteststatisticsuggeststrongerclustersofviolenceneartheoutlet
agglomeration.Second,threedistancebandsarepresentedforconsideration.Thesedistances
representthegeographicrangeofclustersofassaultiveviolenceforeachstatisticallysignificant
agglomeration.Forexample,Agglomeration7,whichislocatedintheChinatownneighborhoodof
Philadelphia,hasateststatisticof20.467(p=0.001)anddisplayshigherlevelsofobservedassaults(vs.
expected)between268and586feet.Further,therearetwoadditionaldistancebandsdisplayedwhere
15
observedassaultsexceedexpectedassaultsforAgglomeration7,onebetween657and821feetand
anotherbetween909and1,166feet.
AbriefexaminationofTable3suggestsamarkeddisparityinboththeteststatisticandthe
distancerangeswhereviolenceclustersformanyoftheoutletagglomerations.Forexample,
Agglomerations7and20displaythemoststatisticalstrength,yetthethirddistancerangeassociated
withAgglomeration9hasthelargestspatialextentofanyrangenoted.Figure5providesagraphical
summaryofthegeographicdistributionofviolencearoundtheoutletagglomerations7and9for
Philadelphia.Distance,upto1/2mile,isdisplayedalongthexaxisandcumulativeassaultsare
displayedontheyaxis.Locationsonthegraphwheretheredline(cumulativeobservedassaults)rides
abovetheblueline(cumulativeexpectedassaults)arethedistanceswhereviolenceclustersaround
eachagglomeration.Moreimportantly,thesearealsothelocationsdenotedinTable3asthedistance
rangeswhere,O>E.
Afinalwaytographicallysummarizethedifferencesinassaultdistributionsaroundeachalcohol
outletagglomerationisbyvisualizingthedistancebandsinamoreexplicitgeographiccontext.Figure6
highlightsthedistancerangeswhereobservedaggravatedassaultsexceedexpectedaggravatedassaults
(Table3)forAgglomerations5,6,7,8and9.Inordertoeaseinterpretation,onlythefirsttwodistance
rangesfromTable3aredisplayedforeachagglomeration.Thickerbandscorrespondtoalarger
distancerangewhereobservedassaultsexceedtheexpecteddistribution.Forexample,asmentioned
earlier,theseconddistancebandforAgglomeration9isthelargestintheanalysis,extendingfrom1,031
ft.to1,505ft.Ofparticularinterestarethelocationswherebandsintersectinthestudyarea.For
example,boththefirstandsecondbandsforAgglomeration6andAgglomeration7intersect.While
individualassaultsarenotexplicitlyassignedtoeachoutletagglomeration,thegeographicproximityof
16
assaultclustersandtheirassociatedoverlapsmayrepresentparticularlyproblematicordangerous
placesinPhiladelphia,atleastwherealcoholrelatedassaultiveviolenceisconcerned.
DiscussionandConclusion
Thereareseveralfacetsoftheresultsthatmeritfurtherdiscussion.First,alcoholoutletsarenot
randomlydispersedthroughouttheurbanenvironment,resultinginagglomerationsofoutletsinvarious
partsofthecity.Thisresultsfrompolicy(e.g.,zoninglaws,plannedentertainmentdistricts),business
decisionsbasedonmarketforces,andsocialfactorslikecommunitydisorganization(Nielsenetal.
2005).Second,thefocusedclustertestssuggestthatassaultiveviolenceclustersaroundmanyofthe
alcoholoutletagglomerationsinPhiladelphia.However,boththestrengthofgeographicclusteringand
itsoverallfrequencyvaryconsiderably.Forexample,ofthe23outletagglomerationsdetectedinthe
city,onlynineofthem(39%)wereassociatedwithclustersofviolence.Thisisnearlyidenticaltothe
ratefoundbyGrubesicandPridemore(2011)inthecityofCincinnati,where5of14outlet
agglomerations(35%)exhibitedclustersofaggravatedassaultiveviolence.Further,itisinterestingto
notethattheninestatisticallysignificantclustersforPhiladelphiaexhibitedarelativelybroadrangeof
distanceswhereviolenceclustered.Forexample,whileassaultsbegantoclusterat198feetfor
Agglomeration16,theydidnotbegintoclusterforAgglomeration6until362feet(Table3).Asimilar
variationwasfoundinCincinnati.Intandem,whiletheseresultsdonotprovideenoughevidenceto
makeanyfirmgeneralizationsaboutallurbanareas,theresultscertainlysuggestthatoutlet
agglomerationsareoftenassociatedwithclustersofviolence,evenifthecriticaldistancesassociated
withassaultiveviolencevaryacrosslocales.
Anothermajorfacetofthisstudyworthnotingistheuseofthenewvisualizationapproachfor
delineatingstatisticallysignificantdistancebandsthatareassociatedwithassaultiveviolence.
Representingamajordeparturefromthesimple,albeiteffectivelinegraphstypicallyassociatedwith
17
focusedclusteringtests(Figure5),thedistancebandsallowanalyststoexplorehowproximal
agglomerationsofalcoholoutletsmayinteractinageographiccontext.Again,ifoneconsidersthe
overlapexhibitedbyAgglomeration6andAgglomeration7(pleaseseemycommentearlier)(Figure6),
itisclearthattheirproximitymayleadtounwantedinteractionsbetweeninebriatedbarpatronsor
provideatargetrichenvironment,freefromcapableguardians,whereoffendersmaylookforvictims
(e.g.,robberiesandassaults).Inadditiontothesepotentialinteractionsinspace,thesynchronous
closingofrestaurantsandbarscanleadtointensespatiotemporalinteractionsofpatronsandpotential
offendersinareasexhibitingclustersofalcoholoutlets.Theeffects,therefore,couldbemultiplied
whenthespheresofinfluenceofagglomerationsinteractwitheachother.Clearly,moreresearchis
neededtoisolateanddeterminetheseeffects.However,thisresearchprovidesafirststepin
highlightinghowthegeographicproximityofoutletagglomerationsandtheirassociatedlevelsof
violencemayinteract.
Theanalyticalapproachoutlinedinthispaperpresentsaviablealternativetoproximityor
adjacencybasedtechniquesforexploringthediffusionordisplacementofassaultsaroundalcohol
outlets.Infact,itisahybridofbothtechniques,whereadjacencymeasureshelpeddefineclustersof
alcoholoutletagglomerationsandproximitywasusedtomeasurethedistributionofassaultsaround
eachagglomeration.Themajordepartureofthisworkfrompreviouseffortsistheuseofthefocused
basedclusteringtesttodetermineifassaultiveviolenceclustersaroundeachagglomeration.Thechief
advantagetothishybridsuiteofmethodsforexploringtheconnectionbetweenoutletsandviolenceis
theabilitytoquantifythegenerallocationswhereobservedviolenceexceedsexpectations.Thisaddsa
layerofgeographicspecificitytothelinkbetweenoutletsandassaultsthathasbeenlargelymissingin
previousresearch.Thatsaid,thereareseverallimitationstotheapproachworthnoting.Forexample,
theuseofblockcentroidstodefinerepresentativelocationsforoutletagglomerationsthatconsistof
multipleblockshasthepotentialtounderestimatetheclusteringofassaults.Simplyput,thisprocedure
18
caninflatedistancesbetweenagglomerationblocks(especiallythosewithalinearorirregular
morphology)andassaultobservations.Anotherproblemwiththisapproachistheinabilitytoconfirm
thatobservedassaultswerealcoholrelated.Unfortunately,theabsenceofthisinformationissharedby
nearlyallpreviousstudiesofthistypeandrepresentsalimitationofthisgeneralresearcharea.Lastly,
thefocusedclusteringtestcanbeinfluencedbypopulationcompositionanddensityduringthe
generationofexpecteddistributionsundertheconstantriskassumption.Asaresult,thepowerofthe
focusedtestcanexhibitspatialheterogeneities.
Severalconclusionsandavenuesforfutureresearchcanbedrawnfromourresearch.First,
furtherstudiesarerequiredtounderstandwhycertainalcoholoutletagglomerationsexhibithigher
ratesofassaultandsomedonot.Thisislikelytheresultnotonlyofthecharacteristicsoftheoutlets
withintheagglomerationsbutalsoofthelocalneighborhoodandspatialenvironment.Systematicsocial
observationofoutletsandsurroundingareaswouldaidinthisendeavor.Second,abetter
understandingofwhythedistancebandsvaryaroundagglomerationsisnecessary.Thisisalsolikely
influencedbyoutletandneighborhoodcharacteristics.Third,ourapplicationofthesetechniquesto
understandbetterthespatialassociationbetweenalcoholoutletsandassaultissimplyoneillustrative
exampleoftheirpower.Thesetechniquescanbeappliedbroadlytoanswerimportanttheoretical
questions,andtheyrepresentpromisingtoolsforgeographiccriminology,epidemiology,andrelated
disciplines.Finally,ourresearchandtheinformationthatcanbegainedfromthesetechniques
representpowerfultoolstopolicymakers.Thistypeofinformationcanbegeneratedtoanswer
questionsabouttheefficacyofgrantingliquorlicensestoproposedoutletsincertainareas,the
increasedriskincertainareasduetooverlappinginfluenceoftroublesomeagglomerations,andthe
moreefficientuseoflimitedpublicandpolicingresourcesbytargetinghumanresourcesandrelated
techniquesmeanttoincreasepublicsafety.
19
References
Anselin,L.(1988).SpatialEconometrics:MethodsandModels.Boston:KluwerAcademicPublishers.
Anselin,L.(1995).LocalIndicatorsofSpatialAssociationLISA.GeographicalAnalysis.27(2):93115.
Anselin,L.,Syabri,I.,Y.Kho.(2006).GeoDa:AnIntroductiontoSpatialDataAnalysis.Geographical
Analysis.38:522.
AssociatedForeignPress[AFP].(2011).Australiaarrests1,600boozedupidiots.
http://tinyurl.com/6hddgjz
Assuncao,R.M.andE.A.Reis.(1999).AnewproposaltoadjustMoransIforpopulationdensity.
StatisticsandMedicine.18:21472162.
Bailey,T.C.andA.C.Gatrell.(1995).InteractiveSpatialDataAnalysis.Harlow,Essex:Longman
ScientificandTechnical.
Bowers,K.J.,andS.D.Johnson.(2003).Measuringthegeographicaldisplacementanddiffusionof
benefiteffectsoncrimepreventionactivity.JournalofQuantitativeCriminology.19:275302.
Branas,C.C.,Elliiott,M.R.,Richmond,T.S.,Culhane,D.P.andD.J.Wiebe.(2009).AlcoholConsumption,
AlcoholOutletsandtheRiskofBeingAssaultedwithaGun.Alcoholism:ClinicalandExperimental
Research.33(5):906915.
Brewer,R.D.andM.H.Swahn.(2005).BingeDrinkingandViolence.JAMA.294:6168.
Bromley,R.D.FandA.L.Nelson.(2002).Alcoholrelatedcrimeanddisorderacrossurbanspaceand
time:EvidencefromaBritishCity.Geoforum.33(2):239254.
CenterforDiseaseControl[CDC].(2009).BehavioralRiskFactorSurveillanceSystem.URL:
http://www.cdc.gov/BRFSS/
Cohen,L.E.andM.Felson.(1979).SocialChangeandCrimeRateTrends:ARoutineActivitiesApproach.
AmericanSociologicalReview.44:588607.
EasyAnalyticSoftwareIncorporated[EASI].(2011).Blockleveldemographicdata.
URL:http://www.easidemographics.com/
FederalBureauofInvestigation[FBI].(2009).CrimeintheUnitedStates.
URL:http://www2.fbi.gov/ucr/cius2009/index.html
FederalBureauofInvestigation[FBI].(2010).AggravatedAssault.
URL:http://tinyurl.com/44mho9u
Franklin,F.A.,LaVeist,T.A.,Webster,D.W.andW.K.Pan.(2010).AlcoholOutletsandViolentCrimein
WashingtonD.C.WesternJournalofEmergencyMedicine.11(3):283290.
20
Giancola,P.R.,Saucier,D.A.andN.L.GusslerBurkhardt.(2003).Theeffectsofaffective,behavioraland
cognitivecomponentsoftraitangeronthealcoholaggressionrelation.Alcoholism:Clinical&
ExperimentalResearch.27:19441954.
Grant,J.(2001).Thedarksideofthegrid:powerandurbandesign.PlanningPerspectives.16(3):219
241.
Green,L.(1995).CleaningupdrughotspotsinOakland,California:Thedisplacementanddiffusion
effects.JusticeQuarterly.12:737754.
Grubesic,T.H.andW.A.Pridemore.(2011).AlcoholOutletsandClustersofViolence.International
JournalofHealthGeographics.10:30.
Grubesic,T.H.,Mack,E.A.andA.T.Murray.(2007).GeographicExclusion:SpatialAnalysisforEvaluating
theImplicationsofMegansLaw.SocialScienceComputerReview.25(2):143162.
Gruenewald,P.J.,Freisthler,L.,Remer,E.,LaScala,A.andA.Treno.(2006).EcologicalModelsofAlcohol
OutletsandViolentAssaults:CrimePotentialsandGeospatialAnalysis.Addiction.101:666677.
Gruenewald,P.J.andL.Remer.(2006).ChangesinOutletDensitiesAffectViolenceRates.Alcoholism:
ClinicalandExperimentalResearch.30:11841193.
HaggardGrann,U.,Hallqvist,J.,Langstrom,N.andJ.Moller.(2006).Theroleofalcoholanddrugsin
triggeringcriminalviolence:acasecrossoverstudy.Addiction.101:100108.
Kanny,D.,Liu,Y.andR.D.Brewer.(2011).BingeDrinkingUnitedStates,2009.Morbidityand
MortalityWeeklyReport.60(Supplement):103.
Koob,G.F.andE.J.Nestler.(1997).TheNeurobiologyofDrugAddiction.TheJournalof
NeuropsychiatryandClinicalNeurosciences.9:482497.
Lane,S.D.,Cherek,D.R.,Pietras,C.J.andO.V.Tcheremissine.(2004).Alcoholeffectsonhumanrisk
taking.Psychopharmacology.172:6877.
Livingston,M.(2008).AlcoholOutletDensityandAssault.Addiction.103:619628.
Livingston,M.(2011).Alongitudinalanalysisofalcoholoutletdensityanddomesticviolence.
Addiction.DOI:10.1111/j13600443.2010.03333.x
Murray,A.T.,McGuffog,I.,Western,J.S.andP.Mullins.(2001).ExploratorySpatialDataAnalysis
TechniquesforExaminingUrbanCrime.BritishJournalofCriminology.41:309329.
Murray,R.K.andD.W.Roncek.(2008).MeasuringDiffusionofAssaultsAroundBarsThroughRadius
andAdjacencyTechniques.CriminalJusticeReview.33:199.
21
Nielsen,A.L.,Martinez,R.andM.T.Lee.(2005).Alcohol,ethnicityandviolence:Theroleofalcohol
availabilityforLatinoandBlackaggravatedassaultsandrobberies.TheSociologicalQuarterly.46:479
502.
Parker,R.N.andLA.Rebhun.(1995).Alcohol&Homicide:ADeadlyCombinationofTwoAmerican
Traditions.Albany:StateUniversityofNewYorkPress.
Pernanen,K.(1991).AlcoholinHumanViolence.NewYork:Guilford.
Ratcliffe,J.H.(2005).Detectingspatialmovementofintraregioncrimepatternsovertime.Journalof
QuantitativeCriminology.21:103123.
Roncek,D.andA.Lobosco.(1983).Theeffectofhighschoolsoncrimeintheirneighborhoods,Social
ScienceQuarterly.64:598613.
Roncek,D.W.andP.A.Maier.(1991).Bars,blocksandcrimesrevisited.Linkingthetheoryofroutine
activitiestotheempiricismofhotspots.Criminology.29:725753.
Scott,M.S.(2001).Assaultsinandaroundbars:Problemorientedguidesforpoliceseriesno.1.
Washington,DC:OfficeofCommunityOrientedPoliceServices,U.S.DepartmentofJustice.
CenterCityDistrict&CentralPhiladelphiaDevelopmentCorporation[CCD&CPDC].(2010).Stateof
CenterCityPhiladelphia.URL:http://www.centercityphila.org/docs/SOCC2011.pdf
PennsylvaniaLiquorControlBoard.(2010).URL:http://tinyurl.com/ygo4x5n
Pridemore,W.A.andT.H.Grubesic.(2011).AlcoholOutletsandCommunityLevelsofInterpersonal
Violence:SpatialDensity,OutletType,andSeriousnessofAssault.JournalofResearchinCrimeand
Delinquency.DOI:10.1177/0022427810397952.
UnitedStatesCensusBureau[Census].(2000).http://www.census.gov.
UnitedStatesCensusBureau[Census].(2010).http://www.census.gov.
Waller,L.A.,Turnbull,B.W.,Clark,L.C.andP.Nasca.(1992).Chronicdiseasesurveillanceandtestingof
clusteringofdiseaseandexposure:ApplicationtoleukemiaincidenceandTCEcontaminateddumpsites
inupstateNewYork.Environmetrics.3:281300.
Zhu,L,Gorman,D.M.andS.Horel.(2004).AlcoholOutletDensityandViolence:AGeospatialAnalysis.
AlcoholandAlcoholism.39:369375.

22
FigureCaptions
Figure1:AlcoholOutletsintheCityofPhiladelphia:November2010
Figure2:AggravatedAssaultDensity(byblock)inPhiladelphia,PA(2010)
Figure3:ExcessRiskofAggravatedAssaultbyBlock:Philadelphia,PA(2010)
Figure4:AlcoholOutletAgglomerations:Philadelphia,PA(2010).
Figure5:ObservedVersusExpectedAssaultCountsbyDistanceFromAlcoholOutletAgglomerations
Figure6:AlcoholOutletAgglomerationsandViolence:StatisticallySignificantDistanceBands.
Table 1: Aggravated Assault Distributions by Block and Associated Descriptive Statistics
Assault Count Frequency by Block Cumulative Percent
0 12401 72
1 2834 88.4
2 1097 94.8
3 460 97.5
4 224 98.8
5 91 99.3
6 48 99.6
7 27 99.7
8 20 99.9
9+ 25 100
N 17,277
Mean 0.5
Median 0
Standard Deviation 1.097
Variance 1.204
T
a
b
l
e

2
:

A
l
c
o
h
o
l

O
u
t
l
e
t

A
g
g
l
o
m
e
r
a
t
i
o
n
s

f
o
r

P
h
i
l
a
d
e
l
p
h
i
a
A
g
g
l
o
m
e
r
a
t
i
o
n
O
u
t
l
e
t

C
o
u
n
t
T
o
t
a
l

B
l
o
c
k
s
A
v
g
.

p
e
r

B
l
o
c
k
A
v
g
.

O
u
t
l
e
t

D
e
n
s
i
t
y
N
e
i
g
h
b
o
r
h
o
o
d
P
r
i
m
a
r
y

T
y
p
e
(
s
)
1
8
7
2
4
3
.
6
2
5
1
,
3
1
1
R
i
t
t
e
n
h
o
u
s
e
;

W
a
s
h
i
n
g
t
o
n

S
q
u
a
r
e
R
e
s
t
a
u
r
a
n
t

(
8
7
%
)
2
4
5
1
2
3
.
7
5
1
,
1
3
3
R
i
t
t
e
n
h
o
u
s
e
R
e
s
t
a
u
r
a
n
t

(
7
7
%
)
3
2
1
2
1
,
0
0
0
R
i
t
t
e
n
h
o
u
s
e
R
e
s
t
a
u
r
a
n
t

(
1
0
0
%
)
4
4
2
2
1
0
0
0
S
W

C
e
n
t
e
r

C
i
t
y
R
e
s
t
a
u
r
a
n
t

(
1
0
0
%
)
5
5
2
2
.
5
1
,
3
7
5
W
a
s
h
i
n
g
t
o
n

S
q
u
a
r
e
R
e
s
t
a
u
r
a
n
t

(
1
0
0
%
)
6
2
1
2
2
8
6
C
o
n
v
e
n
t
i
o
n

C
e
n
t
e
r

D
i
s
t
r
i
c
t
H
o
t
e
l

(
1
0
0
%
)
7
1
1
4
2
.
7
5
1
,
1
0
8
C
h
i
n
a
t
o
w
n
R
e
s
t
a
u
r
a
n
t

(
8
1
%
)
8
4
2
2
6
6
7
W
a
s
h
i
n
g
t
o
n

S
q
u
a
r
e
;

S
o
c
i
e
t
y

H
i
l
l
R
e
s
t
a
u
r
a
n
t

(
7
5
%
)
9
5
6
1
3
4
.
3
1
,
7
1
7
O
l
d

C
i
t
y
R
e
s
t
a
u
r
a
n
t

(
9
6
%
)
1
0
1
0
5
2
1
,
1
3
3
S
o
c
i
e
t
y

H
i
l
l
;

B
e
l
l
a

V
i
s
t
a
R
e
s
t
a
u
r
a
n
t

1
0
0
%
1
1
4
2
2
8
3
3
S
o
c
i
e
t
y

H
i
l
l
;

B
e
l
l
a

V
i
s
t
a
R
e
s
t
a
u
r
a
n
t

1
0
0
%
1
2
2
1
2
2
0
0
P
e
n
n
s

L
a
n
d
i
n
g
R
e
s
t
a
u
r
a
n
t

1
0
0
%
1
3
2
1
2
1
0
0
0
B
e
l
l
a

V
i
s
t
a
R
e
s
t
a
u
r
a
n
t

1
0
0
%
1
4
2
1
2
2
8
5
W
h
a
r
t
o
n
R
e
s
t
a
u
r
a
n
t

1
0
0
%
1
5
2
1
2
1
0
0
0
W
h
a
r
t
o
n
R
e
s
t
a
u
r
a
n
t

1
0
0
%
1
6
2
1
2
5
0
0
H
a
d
d
i
n
g
t
o
n
R
e
s
t
a
u
r
a
n
t

5
0
%
1
7
1
0
3
3
.
3
3
5
1
6
U
n
i
v
e
r
s
i
t
y

C
i
t
y
R
e
s
t
a
u
r
a
n
t

9
0
%
1
8
1
4
6
2
.
3
3
1
,
6
9
4
M
a
n
a
y
u
n
k
R
e
s
t
a
u
r
a
n
t

9
2
%
1
9
2
1
2
3
3
3
L
o
w
e
r

K
e
n
s
i
n
g
t
o
n
R
e
s
t
a
u
r
a
n
t

1
0
0
%
2
0
2
1
2
6
6
7
K
e
n
s
i
n
g
t
o
n
;

U
p
p
e
r

K
e
n
s
i
n
g
t
o
n
R
e
s
t
a
u
r
a
n
t

1
0
0
%
2
1
4
2
2
6
6
7
R
i
c
h
m
o
n
d
R
e
s
t
a
u
r
a
n
t

7
5
%
2
2
2
1
2
5
0
0
O
x
f
o
r
d

C
i
r
c
l
e
R
e
s
t
a
u
r
a
n
t

1
0
0
%
2
3
2
1
2
3
3
3
M
a
y
f
a
i
r
R
e
s
t
a
u
r
a
n
t

1
0
0
%
Table 3: Clustering of Aggravated Assault and Associated Distance Ranges
Agglomeration Test Statistic p-value
1
Distance O > E
2
Distance O > E
3
Distance O > E
1 -25.039 0.498 ----- ----- -----
2 0.513 0.293 ----- ----- -----
3 -1.282 0.900 ----- ----- -----
4 -0.259 0.551 ----- ----- -----
5 1.969 0.039 222-377 437-595 627-947
6 9.244 0.001 362-613 740-1000 1071-1405
7 20.467 0.001 268-586 657-821 909-1166
8 6.335 0.007 348-558 682-886 923-1357
9 6.923 0.001 221-585 649-930 1031-1505
10 -1.583 0.920 ----- ----- -----
11 0.753 0.218 ----- ----- -----
12 1.304 0.105 ----- ----- -----
13 -0.886 0.950 ----- ----- -----
14 0.906 0.260 ----- ----- -----
15 0.554 0.286 ----- ----- -----
16 3.964 0.007 198-571 617-810 936-1302
17 -0.935 0.853 ----- ----- -----
18 -0.809 0.730 ----- ----- -----
19 7.778 0.002 300-514 556-751 787-1149
20 17.836 0.001 259-528 639-835 855-1237
21 2.604 0.025 234-502 558-885 894-1349
22 -1.910 0.971 ----- ----- -----
23 -2.297 0.989 ----- ----- -----
NOTES:
1
First distance range (in feet) where cumulative observed assaults exceeds cumulative expected assaults
2
Second distance range (in feet) where cumulative observed assaults exceeds cumulative expected assaults
3
Third distance range (in feet) where cumulative observed assaults exceeds cumulative expected assaults
!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!!!!
!!
!
!
!
! !
!
!
!
!
!!
!
!
!
!
!!
!
!
!!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!!!
!
!
!!!!
!!
! !
!
!
!!
!!
!
!
!
!
!
!
!
!
!
!! !!
! !
!!
!!
! !!
!
!!
!
!
!
!
!!
!
!
!!
!!
!
!
!!!
!
!
!!
!
!
!!
!
!!!
!
!!
!
!!
!
!
!
!!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!!!
!
!
!!
!
!
!
!
!
!
! !
!
!
!
!
!!!!
!
!
!
!!
!
!
!
!
!
!!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!! !
!
!
!
!
!!!
!
!
! !!
!
!!!!!!!
!!
!!!
!!
!
!
!
!!
!!
! !!
! !
!
!!
!
!
!
! !
!
!
!
!
!
!!
!
!
!
!
!! !
!
!!
!!
!!
!
!
!!!!
!!
!!!
!!
!!
!
!!
!!
!!
!
!!
!
!
! !
!!
!!!!!
!!!
!
!
!
!!! !!!!
!
!
!
!!!
!
!
!
! !!!
!
!
!
!!
!
!
!
!
!!
!
!
!
!
!
! !
!
!
!!
!
!
!
!
!
!
!
!
!!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!!
!!
!
!!
!
!
!!
!
!
!
! !
!
!
!! !
!
!!
!!!
!
!
!
! !
!!
!
!
!!! !
!
!!
!
!
!!
!
!
!!
!
!
!
!!!
! !
!!!
!!!
!
! !
!!
!
!
!
!!
! !!!
!
!!
!
!!
!
!
!
!
! !
!
!
!
!
!!
!!
!
!
!
!
!
!
!
!
!
!
! !
!
!!
!
!
!!!
!
!
!!
!
!
!
!!! !
!
!!!!! !
!
!
!
!!
!!!!!! !!!!!!!!!!!!! ! !!
!!!!!
!!
!
!!
!
!!
!!
!
!
!
! !
!
! !!
!
!! !!!
! !!! ! !!!
!
!!
!!!!!
!!!!!
!
!
! !!
!!!!!
!
!!!
!!
!!! !!
!
!
!
!
!
!!!
!!
!!
!
!!!!
!
!!
!!
!!
!!
!!
!!
!!!
!!
!
!!
!! !!!!! !! !
!!!!
!
!
!
!
!
!
!!!
!
!
!
!!!!
!
!
!!
!
!!
!
!
!
!
! !!!
!
!
!
!
!
! !
!
!
! !!
!
!
! !!!! !!
!!
!!!
!
!!
!!
!
!
!
!! !
!!
!
!
!!!
!
!!
!!!
!
!!!
! !!
!!
!!! !
!
!!!! !!
!
! !
!!
!!
!!
!
!
!
!
!
!
!
!!
!
!!
!
!
!
!
!
!
!
!
!
! !!!!! !!!
!
!
!!!!!
!!
!
!!!!!!
!
!
!!!
! !
!
!
!
!
!!!
!!!!!!!!!
!
!
!
! !
!!!!!!!!
!!!!!!
!
!!!!!
!
!
!!!
!!!!
!!
!
!!
!!
!
!!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!!
!!!
!
!
!
!
!
!!!
!
!
!
!!!
!!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!!
!
!
!
!
!
!
!
!
!
!!
!!
! !
!
!
!
!
!
!
!
!
! !
!!
!
!
!!
!
!!
!
!
! !!!
!
!!!! !!
!! !
!
!
!
!
!!!!!
!
!
!!
!
!!
!!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
! !!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!!
!
!
! !
!
!
!
!
!! !!
!
!
!
!
!
!
!! !
!
!
!
!!!!!
!
!!
!
!
!!
!
!
!
!
!
!
!!
! !
!
!!! !
!
!
!
!! !
!
!
!
!
!
!! ! !!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!!
!
!
!
!
!
!
!
!!
!!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!
! !
!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!!
!
!
!
!!
!!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!!!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!!
!!!
!!
!
!
!
!
!
!
!
!
!
!
!
!!
!!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!!
!!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!!
!
!
!
!!
!!
!
!
!
!
!
!!
!
!
!
!!
!
!
!
!
!
!
! !
!
!
!
!
!
!
! !!
!
!
!
!
! !!
!
!!!
!!
!
!
!
!
!
!
!!!
!!!
!
!
!
!
!!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!! !!
!
!
!
! ! ! !
!
!
!
!!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!!!!
!
!
!
!
!!
!! !
!
!
!
!
!
! !
!
!
!
!
!
!!
!
!!
!
!
!
!
!
!
!
!
!
!!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!!
! !
!
!!!
!
!
!
! !
!
!
!!
!
!
!
!
!
!
!!
!
!!!
!!
!!
!!
!
!
!
!
!
! !
!
!!!
! !
!
!
!
!
!
!
!!
!
!
!
!
!
!!!
!!!!
!
!
!
!
!
! !
! !!
!
!
!!!
!
!
! !!
!
!!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!!
!!
!
!
!
!
!!
!!
! !!
!
!
!
!!
!!!
!
!
! !
!
!
!!
!
! !!
!
!
!
! !
!
!!!
!
!
!
!
!
!
!!
!
!
!
!!
!
!!!
!!
!!
!!
!
!
!
!
!!
!
!
!!
!
!
!!!
!
!
!
!
!
!
! !
!!
!
!
!
!
!!
!
!!
!
!
!
!
!!
!
!
!!!!
! !
!!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!!
!
!!!
!
!!!
!
!!!
!
!
!!
!
!
!
!!
!
!
!
!
!!
!
!
!
!
!!! !
!!
!
!
!
!
!
0 1 2 3 4 0.5
Miles
Neighborhood
!
Alcohol Outlets
Streets
City Limits
Center City
0 1 2 3 4 0.5
Miles
Assault Density
Neighborhood
< 0.50 SD
0.50 - 1.5 SD
1.5 - 2.5 SD
> 2.5 SD
0 1 2 3 4 0.5
Miles
Neighborhood
Excess Risk
< 0.25
0.25 - 0.50
0.50 - 1.00
1.00 - 2.00
2.00 - 4.00
> 4.00
City Limits
Alcohol Outlet Agglomeration
0 0.3 0.6 0.9 1.2 0.15
Miles
Alcohol Outlet Agglomeration
Agglom 1
Agglom 2
Agglom 3
Agglom 4
Agglom 5
Agglom 6
Agglom 7
Agglom 8
Agglom 9
Agglom 10
Agglom 11
Agglom 12
Agglom 13
Agglom 14
Agglom 15
Agglom 17
Cumulative Observed Cumulative Expected
Agglomeration 7 (Chinatown)
0
20
40
60
80
100
120
140
160
180
0
5
6
1
7
6
6
9
0
9
9
9
6
1
0
9
6
1
1
8
9
1
2
9
6
1
3
5
6
1
4
9
3
1
5
4
9
1
6
6
0
1
7
3
6
1
8
1
8
1
9
0
5
2
0
0
9
2
0
9
5
2
1
7
1
2
2
1
8
2
2
8
9
2
3
6
9
2
4
2
7
2
4
8
6
2
5
3
7
2
6
1
3
Distance (in feet)
A
s
s
a
u
l
t

C
o
u
n
t
Cumulative Observed Cumulative Expected
Distance (in feet)
A
s
s
a
u
l
t

C
o
u
n
t
0
20
40
60
80
100
120
0
3
2
6
5
0
5
6
5
3
7
6
7
8
9
3
1
0
3
2
1
1
0
1
1
1
7
6
1
3
3
1
1
4
3
9
1
4
8
3
1
5
3
7
1
5
8
4
1
7
5
0
1
8
3
5
1
8
8
8
1
9
3
9
1
9
8
5
2
0
0
7
2
0
4
2
2
1
3
5
2
1
6
5
2
1
9
9
2
2
7
2
2
3
0
5
2
3
5
8
2
3
7
5
2
4
0
2
2
4
3
5
2
5
1
2
2
5
7
0
2
5
9
8
2
6
3
8
Agglomeration 9
Agglomeration 7
Aggravated Assault
Block
Agglomeration 8
Agglomeration 9
Agglomeration 7
Agglomeration 6
Agglomeration 5
1

Geocoding Uniform Crime Report data and modeling neighborhood violent crime to
investigate correlates of intimate partner violence in Jefferson County, Alabama
Qing Li, MD, DrPH
Center for Social Medicine and Sexually Transmitted Diseases, Department of Sociology,
University of Alabama at Birmingham, AL
Russell S. Kirby, PhD, MS, FACE
Department of Community and Family Health, College of Public Health
University of South Florida, Tampa, FL
Author for correspondence: Qing Li, MD, DrPH, 2029B Vestavia Park Court, Birmingham,
Alabama, 35216, telephone: 205-4277556; Email: youliqing@hotmail.com
Keywords: Geocoding, Geographic Information Systems (GIS), Bias, Violent crime, Intimate
Partner Violence, Pregnancy
Acknowledgement --- Financial support for undertaking the survey was provided by the
National Institute of Child Health and Human Development (grant P01HD033927-05). This
research was partially supported by a dissertation award to Q. Li from the National Center for
Injury Prevention and Control (grant R49-CE000556-01) and a travel award from The Thompson
Foundation, Birmingham, Alabama. The authors gratefully acknowledge the assistance on
geocoding process from the supportive GIS community in Birmingham including Akhlaque
Haque, Chad Landgraf, Williams Allen, Leonid Mazur, and Ellen Meadows. The authors
gratefully acknowledge the assistance to the access of Uniform Crime Reports by Jennifer
Kilburn, Michael Wood, Esther Callens, Denise Corbin, and Libby Zanthos.

Human Participation Protection --- The original survey protocol and this study proposal to
include Uniform Crime Reports data were reviewed and approved by the institutional review
board of the University of Alabama at Birmingham. Pregnant women provided informed consent
to take part in the survey.

2

Abstract

Objective. We examined neighborhood violent crime influence on intimate partner violence
(IPV) prior to/during pregnancy among low-income women in American Deep South. Previous
studies did not use raw data from official crime reports or report the quality of the crime data.
Methods. Multilevel modeling was used to investigate IPV among 2,887 second-trimester
pregnant women residing in 112 census tracts in Jefferson County, Alabama during 1997-2001.
Data were collected from the Perinatal Emphasis Research Center (PERC2) project, the 2000
Census, and the local Sheriff and Police Department Uniform Crime Reports for 1997-2001. We
geocoded addresses to latitude-longitude coordinates and classified them into census tracts using
ARCView GIS 3.1, MapMarker Plus 9.0, and the Federal Financial Institutions Examination
Council's Geocoding Systems. Neighborhood violent crime was calculated by classifying 62,504
geocoded violent crime events annually per 1,000 census tract residents.
Results. The quality of geocoding varied across data sources. About 98% of home addresses of
women were geocoded. We investigated the quality and edited addresses of violent crime events
to conform to the United States Postal Service address standards. We geocoded 58% of County
data to specific latitude-longitude coordinates and 27% geocoded to ZIP Code centroid. We
geocoded 88% of Birmingham data and 83% of Bessemer data to specific latitude-longitude
coordinates. After we added neighborhood variables in multilevel modeling, neighborhood
violent crime exhibited insignificant effects on IPV at the individual level.
Conclusions. The insignificant analytical findings may result in part from limited
standardization of addresses in the Uniform Crime Report and unavoidable bias of geocoding.
Trainings to law enforcement staff could enhance quality of data reporting and geocoding.

3

Introduction
The relationship between neighborhood crime and intimate partner violence (IPV) was
investigated in several studies recently (Block & Skogan, 2001; Browning, 2002; Burke, 2003;
Li et al., 2010; O'Campo et al., 1995) but excluded in a review on area-level measures and
maternal and child health (Rajaratnam et al., 2006). As shown in Table 1 and described below,
each study measured neighborhood crime and conceptualized its own influence on IPV.
Neighborhood violent crime was hypothesized to be an indicator of environmental stress (Burke,
2003; O'Campo et al., 1995). Based on 1990 data from Baltimore City's Division of Planning,
OCampo et al. (1995) used per capita overall crime, including robbery, homicide, theft, burglary,
rape, and aggravated assault. Burke (2003) used neighborhood personal crime density, which
was collected from Baltimore City Police Crime report and caculated for each census block
group as total number of personal crimes devided by populations. Block & Skogan (2001) used
rates of violent crime and drug-related offenses; and crime data were from Chicago Police
department, and sorted by the geographic unit police beat and specific types including each index
crime, gun crime, drug street crime, school crime, etc. According to a qualitative study, mothers
living in crime-ridden, inner-city neighborhoods cited family violence as a carry-over from street
violence (Edin, 2000). Occurrence of crime may reflect a community norm, that violence is an
acceptable behavior at home (Browning, 2002). Neighborhoods in which violence is prevalent
may transmit this behavioral orientation to multiple contexts, including intimate relationships
(Browning, 2002; Massey & Denton, 1993; Miles-Doan, 1998). Therefore, neighborhood violent
crime was hypothesized to be positively associated with the prevalence of IPV among low-
income pregnant women due to the acceptance of violence as a social norm in their
neighborhoods, thus making it acceptable at home and within intimate relationships (Li et al.,
4

2010). Applying multilevel analysis, these five studies reported insignificant association between
neighborhood violent crime and IPV (Block & Skogan, 2001; Browning, 2002; Burke, 2003; Li
et al., 2010; O'Campo et al., 1995) (Table 1).
To explain the insignificant findings between neighborhood violent crime and IPV, two
studies considered the reporting bias in the operationalization of this measure (Browning, 2002;
Li et al., 2010). The reporting bias in Uniform Crime Reports was gained attention because the
rates of officially reported domestic assaults are different due to differential reporting to criminal
justice systems, based on individual and neighborhood characteristics (Miles-Doan, 1998).
Researchers in their investigation of IPV or other violent behaviors have limited their analysis to
homicides as a more reliable measure for neighborhood crime (Browning, 2002; Molnar et al.,
2003; Sampson et al., 1997). Our empirical analysis (Li et al., 2010) followed Morenoff (2003)
in measuring the violent crime rate rather than the total crime rate because violent crimes are
likely to be more widely publicized and be seen as serious threats to personal safety.
Additionally, violent crimes have more chances to happen in intimate relationship. The violent
crime events tabulated included murder, rape, robbery, aggravated assault, and domestic assaults.
Potential biases and methodological details in the process of geocoding were reported in
the spatial analyses of data recently (Hay et al., 2009; Krieger et al., 2001). Previous studies on
IPV did not use raw data from official crime reports or report the quality of geocoded crime data
(Block & Skogan, 2001; Browning, 2002; Burke, 2003; O'Campo et al., 1995). Though a
dissertation reported these methodological details (Li, 2006), this was not a focus of our
empirical paper (Li et al., 2010). We intend to provide methodological detail in this paper to
assess possible biases during geocoding and modelling the influence of neighborhood violent
crime on IPV prior to/during pregnancy among low-income women in American Deep South.
5

The objectives of this study were 1) to assess the quality of the location data and the accuracy of
neighborhood violent crime measure based on geocaded data from Uniform Crime Reports; 2) to
investigate if neighborhood violent crime is positively associated with the prevalence of IPV at
the individual level. We discuss possible biases that may be introduced by the geocoding.

Methods
A study of 2887 pregnant women in the second trimester in prenatal care settings was
conducted in Alabama. The details of this sample were reported previously (Li et al., 2010) and
are summarized here. Multilevel modeling was used to investigate IPV among 2,887 second-
trimester pregnant women residing in 112 census tracts in Jefferson County, Alabama during
1997-2001. Data were collected from the Perinatal Emphasis Research Center project, the 2000
Census, and the local Sheriff and Police Departments Uniform Crime Reports for 1997-2001.
Uniform Crime Reports at the census tract level were not publicly available. Efforts were
made to collect the data from by far the largest two sources (Jefferson County Sheriffs Office
and Birmingham City Police Department) as well as Bessemer City Police Department serving
an area with high crime rate. However, crime data were not requested or collected from 7 cities
(i.e., Midfield, Irondale, Tarrant, Leeds, Hoover, Homewood, and Brighton) within Jefferson
County. These data were accessed with permission from these law enforcement agencies under a
study protocol approved by the Institutional Review Board at University of Alabama at
Birmingham. The crime data were collected from the Sheriffs incident/offense reports for the
address of the location of each incident.
Access to crime data was facilitated by law enforcement staff, who identified appropriate
measures and developed routines for pulling the data from their management information
6

systems managed by individual law enforcement offices (Li et al., 2010). Dr. Robert Sigler, a
dissertation committee member with expertise in the field of criminal justice wrote a formal
response letter to the Sheriff in Jefferson County, which guided an interest in forming academic-
community partnerships in data sharing to inform evidence-based effective planning for law
enforcement in a long-term agenda. Through this collaborative process, we involved partners
from academics and law enforcement in the process to combine knowledge and action
collectively for the under-researched prevalent crime in Jefferson County, Alabama.
The data were then geocoded to classify events by census tract. We calculated
neighborhood violent crime by classifying geocoded violent crime events annually (i.e., murder,
rape, robbery, aggravated assault, and domestic assaults) per 1000 census tract residents. The
addresses of the violent crime events were edited to conform to United States Postal Service
(USPS) address standards, geocoded, and grouped by census tracts. We geocoded addresses to
latitude-longitude coordinates and classified them into census tracts using ARCView GIS 3.1,
MapMarker Plus 9.0, and the Federal Financial Institutions Examination Council's Geocoding
Systems. Neighborhood violent crime was calculated by classifying 62,504 geocoded violent
crime events annually per 1,000 census tract residents. Annual neighborhood crime rates were
calculated for each census tract as the total number of violent crime events in the 5-year period
divided by five and the population as reported in the 2000 census, reported as an annual rate per
1000 persons (Li, 2006).
Despite potential multicollinearity (correlation coefficients =0.687) (Bonate, 1999), both
neighborhood concentrated disadvantage and violent crime were simultaneously entered into
multilevel analysis, because no literature about suitable cut-off points for correlation between
two variables from neighborhood level was found (Li et al., 2010).
7

Results
Figure 1 shows 5 categories of neighborhood (census tract) violent crime rates in the
study neighborhoods for 1997-2001. Some census tracts cross minor civil division boundaries
(e.g., Birmingham City, Bessemer City) in Jefferson County. Total 0.635 million records in 5
years included 151726, 478298 and 5087 respectively from Sheriff Office, Birmingham and
Bessemer City Police Departments (Table 2). Records with reports of violent crimes (i.e., murder,
rape, robbery, aggravated assault, and domestic assaults) were included for geocoding.
Over 98% of home addresses of 2887 women from the research data were successfully
geocoded and classified by census tracts using three geocoding softwares. Specifically, the
geocoding tool available in ARCView GIS 3.1 was used to link addresses to specific latitude-
longitude coordinates and with the census tracts in which they reside. A total of 3,160 cases were
matched in batch mode in ARCView using additional data from ESRI StreetMap 2000 (1997) in
a score ranging from 0-100 indicating the accuracy or validity of each address match, with 100
being a certain match. A score of 75 or above is generally regarded as a good match
(Environmental Systems Research Institute, 1999). Only good matches were accepted for this
study. A total of 2,573 cases (81%) were successfully matched with scores of 75-100. The
remaining 587 cases with scores less than 75 were regeocoded interactively using MapMarker
Plus 9.0 (MapInfo Corporation, 2004) by a geocoder with extensive experience working with
Jefferson County data (William Allen, Jefferson County Department of Health). The acceptable
match quality was a single close match, which indicated that the record was matched to a single
address candidate. Only 42 cases could not be matched because their addresses did not exist in
USPS databases. A total of 16 records were geocoded initially to invalid census tracts, and were
later interactively geocoded using the Federal Financial Institutions Examination Council
8

website (http://www.ffiec.gov/geocode/default.htm), resulting in valid census tracts for 7 records.
Therefore, 51 cases (42 and 9) with addresses could not be geocoded to valid census tracts. The
total successful match rate was 98% (3109/3160). The output 15-digit Federal Information
Processing Standards U.S. Bureau identifier specified the census tract.
We investigated issues in crime administritive data sources, and edited addresses of
violent crime events to conform to the USPS address standards. The initial match rate in batch
mode was 26%. 48123 (31.7%) records did not have either house number or zip code. 7334
(4.8%) records did not have house number but Zip code. 940 records were about Jefferson
County Jail. The data from Sheriff did not include the city name but some did provide ZIP Code.
Therefore, the ZIP Codes with the same addresses were used to substitute the one without ZIP
Code. Eventually, we geocoded 58% of addresses in Jefferson County Sheriff Office data to
specific latitude-longitude coordinates and 27% geocoded to ZIP Code centroid.
The data from Birmingham City police department did not include the ZIP Code. It was
safe to use Birmingham as the city name. The key issue was to ensure correct formatting of
data in the of Sttype and Address fields. For example, after sorting by street, we deleted all
"EN" in the stdirect for later cases in the sequence for Ensley area, and fully spelt out the "EN"
as Ensley for the previous cases in the sequence. The decisions were confirmed through
checking each case during geocoding. Several groups of cases with the same street names did not
geocode in batch mode but were successfully geocoded manually. We geocoded 88% of
Birmingham data to specific latitude-longitude coordinates. Furthermore, after standardizing the
spelling of addresses and deleting the duplicate counts for the same incidents in the data from
Bessemer City Police Departments, the final successful geocoding rate with 83%. This smallest
9

data were provided by Sergeant Wood, who closely used past crime address in guiding the
current law enforcement service.
Violent crime events per census tract were summed to get 5-year total of crime events.
Annual neighborhood crime rates were calculated for each census tract as the total number of
violent crime events in the 5-year period divided by five and the total population as reported in
the 2000 census, reported as an annual rate per 1000 persons and measured as a continuous
variable. Annual neighborhood crime rates of 112 census tracts ranged from 0.0001 to 0.1494
with a mean of 0.0296 and standard deviation of 0.028 (Li et al., 2010).
After we added neighborhood variables in multilevel modeling, neighborhood violent
crime exhibited insignificant effects on IPV at the individual level (OR=17.8; 95% confidence
interval=0.01- ) (Li et al., 2010). After dropping tracts covered Homewood and Hoover City
limits (with low crime) and Bessemer City limits (with high crime), the study sample was
so homogeneous that multilevel analysis was not necessary.

Discussion
The accuracy of neighborhood violent crime measure varied across data sources since the
quality of geocoding depended on the quality of location data. The accuracy of neighborhood
violent crime measure varied by the quality of geocoded address data. Compared with 98%
geocoding rate of the research data, geocoding rates of adminstritive data from Birmingham and
Bessemer Cities to specific latitude-longitude coordinates were 88% and 83%, better than 58%
of Jefferson County Sheriff Office data and 27% was geocoded to ZIP Code centroid.
Neighborhood violent crime measure was not as accurate as possible because addresses for crime
locations in 1997-2001 were not recorded uniformly by the police officers or entered consistently
10

by public safety staff into the computer systems. This may cause differential bias and
underestimate or over-estimate the effect of neighborhood violent crime on IPV.
Neighborhood violent crime was not significantly associated with the prevalence of IPV
at the individual level. This finding may be explained by the specific neighborhood processes
and accuracy of neighborhood violent crime measurement. Additional efforts explored potential
under-reporting issues of crime events per census tracts. In addition to the quality of address data,
census boundaries and city limit in the geocoding software differed from the jurisdiction
boundaries from the police offices. The Sheriffs Office has jurisdiction anywhere in the county,
even if some area within has its own police department; and the custom is to take reports,
regardless of where they occur, and then to forward the report to that citys police department.
For example, law enforcement services for the town of Maytown were performed by the
Sheriffs Office. The City of Brighton has its own police department, which is small and the
Sheriffs Office takes reports when their officer is busy and serious or complicated crimes occur.
This research has public health implications for further monitoring and intervening
neighborhood violence cime in the American Deep South. Academic-community partnership
was the key way to ensure the successful access to Uniform Crime Reports data and the
community-based participatory research. The methodological detail and possible biases during
geocoding and modelling neighborhood violent crime documented daily and rountine work that
people in law enforcement agencies already know. The built trust and mutual knowledge can
influence directions of future research and insure that law enforcement agencies get the type of
information for effective planning. In order to improve the efficiency of use of this
administrative data source for the purpose of crime assessment and prevention, law enforcement
agencies need to use the United States Postal Service standard addressing conventions for
11

completing Uniform Crime Reports and update computer systems for data entry and geocoding.
Trainings to law enforcement staff could enhance quality of data reporting and geocoding.
Furthermore, integrated and coordinated documents, technology, and training could better
enhance the local capacity across jurisdiction boundaries. Funded by the National Institute of
Justice, the Community Block Grant Administration, and the Community Oriented Policing
Services, the Milwaukee COMPASS site led that shared data is integral to good collaboration
and problem solving. The American Deep South might have more need in similar collaboration
and system integration for a healthy community.
This study is subject to several limitations. First, the findings may have limited
generalizability to other metropolitan areas or settings with own history, personals and dynamics
of crime report. Second, the geocoding experience here is applicable for address matching rather
than relational joins for spatially aggregated data, global positioning system, and other
alternatives. Third, MapMarker Plus 9.0 was not available for the administrative data with
matching scores less than 75. Fourth, variance in accuracy of geocoding across softwares and
bias due to spatiotemporal mismatches between zip codes and US census-defined geographic
areas (Krieger et al., 2002; Krieger et al., 2001) were important but could not be assessed. Fifth,
potential recall bias could not be ruled out when this manuscript was not written prospectively.
Sixth, we lack updated street maps of newly developed areas for better quality of geocoding.
The study findings indicate that neighborhood violent crime measure and the quality of
geocoding depend on the quality of location data in Uniform Crime Reports and reporting bias
exisit. The insignificant finding may be explained by limited standardization of addresses in the
Uniform Crime Report and unavoidable bias of geocoding. The results suggest trainings to law
enforcement staff and coordinated effort could enhance quality of data reporting and geocoding.
12

13

References

Block C, Skogan W. (2001) Do collective efficacy and community capacity make a difference behind
closed doors. Available at http://www.ncjrs.gov/pdffiles1/nij/grants/194711.pdf Accessed May 1,
2004
Bonate PL (1999) The effect of collinearity on parameter estimates in nonlinear mixed effect models.
Pharmaceutical Research 16: 709-717.
Browning CR (2002) The Span of Collective Efficacy: Extending Social Disorganization Theory to
Partner Violence. Journal of Marriage and Family 64: 833-850.
Burke J. (2003) Intimate partner violence among low income women: associated individual and
contextual risk factors. Baltimore: [doctoral dissertation, John Hopkins University, 2003].
Dissertation Abstracts International. 2003;64(02):659B. University Microfilms No. 3080630.
Available at Accessed
Edin K (2000) What do low-income single mothers say about marriage? Social Problems 47: 112-133.
Environmental Systems Research Institute (1999) Getting to know ArcView GIS: the geographic
information system (GIS) for everyone. Redlands, California: Prentice Hall College Div.
Hay G, Kypri K, Whigham P et al. (2009) Potential biases due to geocoding error in spatial analyses of
official data. Health Place 15: 562-567.
Krieger N, Waterman P, Chen JT et al. (2002) Zip code caveat: bias due to spatiotemporal mismatches
between zip codes and US census-defined geographic areas--the Public Health Disparities
Geocoding Project. Am J Public Health 92: 1100-1102.
Krieger N, Waterman P, Lemieux K et al. (2001) On the wrong side of the tracts? Evaluating the
accuracy of geocoding in public health research. Am J Public Health 91: 1114-1116.
Li Q. (2006) Intimate partner violence among low-income pregnant women: multilevel analysis
Birmingham: [doctoral dissertation, University of Alabama at Birmingham, 2006]. Dissertation
Abstracts International. ProQuest document ID: 1280149701. University Microfilms No.
3253060. Available at Accessed
Li Q, Kirby RS, Sigler RT et al. (2010) A multilevel analysis of individual, household, and neighborhood
correlates of intimate partner violence among low-income pregnant women in Jefferson County,
Alabama. Am J Public Health 100: 531-539.
MapInfo Corporation. (2004) MapMarker plus 9.0 user guide. In M. Corporation (Ed.). Available at
http://extranet.mapinfo.com/documentation/software/mapmarker_plus/english/9_2/MapMarkerPl
us92.pdf Accessed March 28, 2005
Massey DS, Denton NA (1993) American Apartheid: Segregation and the Making of the Underclass.
Boston: Harvard University Press.
Miles-Doan R (1998) Violence between Spouses and Intimates: Does Neighborhood Context Matter?
Social Forces 77: 623-645.
Molnar BE, Buka SL, Brennan RT et al. (2003) A multilevel study of neighborhoods and parent-to-child
physical aggression: results from the project on human development in Chicago neighborhoods.
Child Maltreat 8: 84-97.
Morenoff JD (2003) Neighborhood mechanisms and the spatial dynamics of birth weight. AJS 108: 976-
1017.
O'Campo P, Gielen AC, Faden RR et al. (1995) Violence by male partners against women during the
childbearing year: a contextual analysis. Am J Public Health 85: 1092-1097.
Rajaratnam JK, Burke JG, O'Campo P (2006) Maternal and child health and neighborhood context: the
selection and construction of area-level variables. Health Place 12: 547-556.
Sampson RJ, Raudenbush SW, Earls F (1997) Neighborhoods and violent crime: a multilevel study of
collective efficacy. Science 277: 918-924.

14

Figure 1, 5 categories of neighborhood (census tract) violent crime rates in the study
neighborhoods in 1997-2001.

15

Table 1. Summary of Studies on Neighborhood Crime Measures and Intimate Partner Violence in the United States
Author
Year
Violence Sample & Data Neighbor-
hood
Neighborhood
factors
Individual & household
Factors
Analysis Findings
OCamp
o et al.,
1995
Moderate to
severe physical
violence,
Conflict
Tactics Scale
157 women in 3
rd

trimester & 6
months
postpartum 93%
AA, 20% 20
years
76 census
tract in
Baltimore,
Maryland
Crime rate, per capita
income, unemployment
rate, ratio of home
owners to renters

friend support, confidant male
partner, confidant relative, partner
drug use, age, marital status,
education, employment status,
income, parity and race
GEE,
Logistic
&
multilevel
Neighborhood variables were
significant, and modified the
individual-level risk.
Block &
Skogan,
2001
help seeking in
previous year;
cession &
severity of
violence
Chicago
Womens Health
Risk Study,
210 abused
women
Alternative
Policing
Strategy
census data
Informal social control;
organizational
involvement; downtown
connections
Crime; relative
disadvantage;
neighborhood stability,
disorder and stress
Past IPV; physical &mental
health, pregnancy & childbirth.
Interpersonal factors: womens
informal social network, material
resources, household situation,
partners controlling behaviors,
relationship with abuser,
characteristics of abuser
Multiple
regression
Neighborhood variables did not
greatly contribute to the degree of
variance explained (beyond
individual-level factors)
Burke,
2003
Physical and
sexual
43%
Women, AIDS
and the Violence
Epidemic
352 (96% AA)
women in
Baltimore
212 census
block groups
1993 policy
crime report
Neighborhood wealth, %
male unemployment,
personal crime density,
residential mobility
Intrapersonal factors: HIV status,
age income, employment,
education, drug, alcohol, self-
esteem, childhood abuse
Interpersonal factors: size of
social network, partner drug use,
HIV status, age, employment,
education
HLM Womens experiences of IPV are
influenced by a multitude of
factors including intrapersonal,
interpersonal, and neighborhood
characteristics.
Li et al.,
2010
Physical and
sexual
7.4%
2887 low income
pregnant women
in Birmingham,
Alabama
(85% AA)
112 census
tract in
Birmingham,
Alabama
Neighborhood
residential stability,
neighborhood
concentrated
disadvantage and violent
crime
women performing most of the
housework, being unmarried,
alcohol use, older age at first
vaginal intercourse and a
greater sense of mastery
Multilevel
modeling
Both neighborhood contextual
and individual and household
compositional effects are
associated with IPV.
Note. Significant factors in bold. IPV=Intimate Partner Violence. AA= African American. SDT=Social Disorganization Theory.
GEE=Generalized Estimate Equation Model. ICC=Intra-Class Coefficient
16

TABLE 2- Geocoding of the addresses of the crime events and research data in 2005-2006

Address Source N Data Quality Geocoding Methods Geocoding
Rate
Sheriff Office 151726 No house number or zip code
(32%)
ARCView GIS 58 %
Zipcode
27%
Birmingham City 478298 Format did not follow USPS ARCView GIS 88 %
Bessemer City 5087 Format did not follow USPS;
duplicate counts
ARCView GIS 83 %
Research data 2887 Excellent ARCView GIS,
MapMarker Plus 9.0,
on-line source
98 %
7/11/2011
1
Creating a GIS Application for
Health Care Facilities Planning at Jeddah City
PROF. ABDULKADER A MURAD
DEPARTMENT OF URBAN AND REGI ONAL
PLANNI NG,
FACULTY OF ENVI RONMENTAL DESI GN,
KI NG ABDULAZI Z UNI VERSI TY KI NG ABDULAZI Z UNI VERSI TY,
JEDDAH, SAUDI ARABI A.
E- MAI L: GI S_PLANNI NG@YAHOO. COM
Contents
1- Introduction
2 Background 2- Background
3- Methods
4-Results & Discussion
5-Conclusion 7/11/2011
2
Introduction
y The purpose of this paper is to discuss a GIS
application created for health care planning at application created for health care planning at
Jeddah city, Saudi Arabia. The application covers
important health care facilities planning issues
including:
y Defining accessibility to health care facilities,
y Identifying and classifying the distribution of health y g y g
demand at Jeddah city, and
y Modeling spatial variation of patient locations
Location of Jeddah, Saudi Arabia 7/11/2011
3
Location of Jeddah, Saudi Arabia
Location of Jeddah, Saudi Arabia 7/11/2011
4
Location of Jeddah, Saudi Arabia
Background
GIS and Health Care Planning
y GIS has been widely used in health care planning
due to the following 4 factors: due to the following 4 factors:
y 1- The increasing availability of geo-coded health
data that lead to having health information systems. 7/11/2011
5
Background
y 2- The availability of digital geographic data at
micro and/or micro scale that has several GIS
coverage with enormous attribute data such as
land use, ownership, etc.
Background
y 3- GIS software such as ArcGIS produced by
ESRI become inexpensive and easier to use and
runs on a wider range of platforms, and
y 4- The availability of spatial analysis tools, as
separate software modules or embedded is GIS. 7/11/2011
6
Background
GIS tools for health care planning
y 1- Geo-coding
y This function is used by several applications to create
points on a map from a table of addresses
Background
y 2- Overlay Analysis
y The concept of overlay analysis is one of the major The concept of overlay analysis is one of the major
GIS procedures that are used by several studies. It
manipulates spatial data organized in different layers
to create combined spatial features
y Union, Interact and Identity are the major polygon
overlay functions. 7/11/2011
7
y 3- Network Analysis
y This type of analysis can be used to find the shortest y This type of analysis can be used to find the shortest
routes or to find the service area of any facilities.
y It uses network data model to produces the analysis
outputs.
Background
The potential applications of GIS in health studies
are: are:
1: disease mapping and geographical correlation
studies,
2: patterns of health service use and access,
3: environmental hazards and disease clusters, and
d li h l h i f i l 4: modeling health impacts of environmental
hazards 7/11/2011
8
Background
Th ill d it i f t 5:The surveillance and monitoring of vector
borne disease,
6: Water borne diseases,
7: Environmental health,
8: Modeling exposure to electromagnetic fields,
9: Quantifying lead hazards in a neighborhood, 9 Q y g g ,
10: Predicting child pedestrian injuries, and
11: The analysis of disease policy and planning.
Background
GIS could also be used for
1:exposure assessment,
2:identification of study populations,
3:disease mapping, and
4:public health surveillance. 7/11/2011
9
Incidence of severe injury by Health Service Delivery Area (HSDA)
Nadine Schuurman, et al, 2008, The spatial epidemiology of trauma: the potential of geographic information science to
organize data and reveal patterns of injury and services, Canadian Journal of Surgery, Vol. 51, No. 5, 389-395
Methodologic Issues and Approaches to Spatial Epidemiology
Linda Beale et al, 2008, Environmental Health Perspectives vol .116 , No. 8 , 1105-1110
Risk analysis using the Rapid Inquiry Facility RIF 7/11/2011
10
A Geospatial Analysis of CDC-funded HIV Prevention Programs for African Americans in the
United States,
G. Aisha Gilliam, 2008, Journal of Health Disparities Research and Practice, Volume 2, Number 2, 39-60
Cumulative HIV/AIDS rates, Alabama
Counties.
Referral from secondary care and to aftercare in a tertiary care
university hospital in Japan
Shin-ichi Toyabe* and Akazawa Kouhei, 2006, Health Services Research
6:11
Distribution of residence of inpatients. (A) All inpatients (n = 8,177) admitted to Niigata University
Hospital during the period from April 2003 to March 2005 were geocoded to a GIS map of Niigata
Prefecture. (B) All inpatients under 15 years of age (n = 1,323) were also plotted on a map. 7/11/2011
11
Referral from secondary care and to aftercare in a tertiary care
university hospital in Japan
Shin-ichi Toyabe* and Akazawa Kouhei, 2006, Health Services Research
6:11
Referral to aftercare. (A) Kernel density plot of patients who were referred to the Outpatient Department of
Niigata University Hospital. (B) Kernel density plot of patients who were referred to outpatient
departments of other hospitals. Dots represent the locations of hospitals with paediatricians.
Background
Maps of health statistics can be invaluable in Maps of health statistics can be invaluable in
understanding local patterns of disease and their
geographical associations. They have the advantages
of conveying instant visual
information accessible to non-experts as well
as public health professionals though their as public health professionals though their
interpretation. 7/11/2011
12
Background
All of the above applications indicate
that GIS has very useful tools and
functions for any health care study
METHODS
The data base of this application covers:
Location of health centers at Jeddah city
which is created as a point feature and
all attribute data about health centers
are saved at the attribute table of this
fil file. 7/11/2011
13
METHODS
The second main GIS data feature is the line feature
which has one dimensional shape that represent which has one dimensional shape that represent
geographical features too narrow to depict as area.
GIS software store lines as a series of ordered x,y
coordinates with relevant attributes. For the
presented application, road network of Jeddah city
is represented as line feature with attributes about p b b
length and type of each road at this city.
RESULTS & DISCUSSION
Spatial distribution of health
services services
There are 39 health centers distributed at Jeddah city.
y The location of health centres in Jeddah city were
plotted on GIS as point features map and classified
based on their supply or service.
y There were 11 types of health services supply for each
centre: 1 - physicians (general practitioners, GPs), 2 -
famil ph sicians 3 dentists 4 n rses 5 family physicians, 3 - dentists, 4 - nurses, 5 -
midwives, 6 - pharmacists (pharmacologists), 7 -
laboratory technicians, 8 - x- ray technicians, 9 -
administrators, 10 - servants, and 11 - others. 7/11/2011
14
Spatial distribution of health
i services
Figure 1 shows a classification model for a health
centre based on the number of physicians .
Figure 2 demonstrates health centresbased on the
number of dentists number of dentists.
Fig. 1 Classification of health centers based on number of Physicians 7/11/2011
15
Fig. 2 Classification of health centers based on number of Dentists
y Each health centre has different amount of
h i i d i physicians or dentists.
y Alsafa Centre and Gulail Centre engage more
physicians than Alsharafia Centre and Althayuar
Centre.
y In Alazizia Centre there are two dentists while in
l l h d l h h Alselamainah Centre and Almarwah Centre there
are no dentists at all.
y One reason for this difference is related to the
size of existing demand. 7/11/2011
16
GIS was applied to classify more types of supply
known as multiple data classification the so-called known as multiple data classification, the so called
multiple queries by using logical operations that deal
directly with the database and allow the user to
identify and select features by a special set of criteria
with more than one parameter.
One could find out health centres with many
physicians but without any dentists or with less physicians but without any dentists or with less
servants.
The features were identified and selected from the
database and highlighted on the map according to a
combination of several conditions.
These features can be saved in a new coverage for g
further analysis. 7/11/2011
17
The presented paper has detected which centres are
owned by the health authority and engage more owned by the health authority and engage more
than 3 physicians
Several health centres in Jeddah city are located on
rented buildings and engage numerous physicians
(Figure 3). e.g. Ghulail and Aljameaa centres are in
the southern area while Alzahraa and Alsalamah
centres are in the northern area.
Fig. 3 Rented health centers with more than 3 Physicians 7/11/2011
18
Several health centres such as Alrabwa, Alzahra
and Alnaeem providing large and various types of and Alnaeem providing large and various types of
health services to the public by family physicians
and dentists (Figure 4).
Alsafa and Alsohaifa centres, however, do not
engage dentists and family physicians at all.
The local health authority can therefore use this The local health authority can therefore use this
technique to find quick and clear answers about
any issues related to quality and quantity of health
services delivered at Jeddah city.
Fig. 4 Health centers with family doctors and dentists 7/11/2011
19
Accessibility to Health centers
y The presented application has selected distance to
provider method and produces accessibility provider method and produces accessibility
indicators to health centers in Jeddah city.
y One way of defining accessibility to health centers is
by knowing how far patients live from their nearest
centers.
y In order to define the level of accessibly to health
centers GIS proximity analysis was used and the centers, GIS proximity analysis was used and the
output of this model classifies the city into deferent
zones based on the distance between clinic location
and city districts.
y Based on this output, several parts of the city are
located at areas with more than 2 KM accessibility b y
zone. These areas are mainly situated north and
east of the city with some to the west. 7/11/2011
20
#
Fig. 5 Spatial Accessibility to Health Centers
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
# #
#
#
#
#
# #
#
#
#
#
#
#
#
10000 0 10000 20000 Meters
Priximity zones
0 - 2000
2000 - 4000
4000 - 6000
6000 - 8000
Road
# Health centers
N
E W
S
Presenting Health Demand/Supply Data
There are 39 health centers distributed at Jeddah There are 39 health centers distributed at Jeddah
city.
Every one of those centers has records about the size
of registered patients.
These records can be used in GIS to define the
spatial pattern of patients in Jeddah city. p p p y
The results of these data are very useful for
identifying pattern and location of patients in
Jeddah city. 7/11/2011
21
Presenting Diabetes Data
Fig. 6 shows the output of the spatial distribution of g p p
diabetic disease at Jeddah city.
The resulted distribution indicates that diabetes
patients are concentrated mainly at Al-Rabwah, Bani
Malik and Al-Sabail districts.
These areas are covering north, central and southern
city districts city districts.
#
Fig. 6. spatial distribution of diabetic disease
#
#
#
#
#
#
#
#
# #
#
# #
#
#
#
# # #
#
#
#
#
#
#
#
#
#
#
#
# #
Al-Rabwah
Bani Malik
Red Sea
Al-Rowais
#
#
#
Al-Sabail
10 0 10 20 Kilometers
City Districts
Road
Diabetes Patiants
# 0
# 1 - 189
# 190 - 290
#
291 - 745
#
746 - 2572
N
E W
S 7/11/2011
22
I dditi t l if i di b ti ti t d t In addition to classifying diabetic patient data,
GIS is used at the presented application to
make a spatial comparison (Based on mean
value) between health centers to find out
centers that are having large amount of
diabetes patients.
This study calculates standard diabetic rates (SDRs) This study calculates standard diabetic rates (SDRs)
for Jeddah city.
Fig 7 shows SDR values and indicates that the areas
with the highest rates trend to be in the central and
southern parts of the city. In addition, two main
areas (Al Rabwa and Al bawadi) located north of
Jeddah city are also having high SDR rates Jeddah city are also having high SDR rates. 7/11/2011
23
#
Fig. 7. standard diabetic rates (SDRs) for Jeddah city
#
#
#
#
#
#
#
#
# #
#
# #
#
#
#
#
# #
#
#
#
#
#
#
#
#
#
#
# #
Al-Rabwah
Bani Malik
Red Sea
Al-Rowais
#
#
#
#
#
Al-Sabail
City Districts
Road
SDR
# 0
# 0 - 58.9
# 58.9 - 90.3
#
90.3 - 232.1
#
232.1 - 801.2
N
E W
S
Finding Relationships
GIS is used in this application to find out
relationships between locations of diabetic patients
and population density at Jeddah city.
The main objective of this comparison is to find out
if areas that have large size of diabetic patients are if areas that have large size of diabetic patients are
also having large population density. 7/11/2011
24
The first step of this part of the application was to
create a GIS coverage which divides the city in to
different population density zones.
E h i d fi d b th b d f it l t Each zone is defined by the boundary of its relevant
city district.
GIS feature classification function is applied on the
created coverage and the results of this step show
that there are three main density groups at Jeddah
city.
These are known as low density (less than 50 These are known as low density (less than 50
person/hectare), medium density ( 50-100
person/hectare ), and high density districts ( more
than 100 person/hectare). 7/11/2011
25
The results of this performed spatial query (Fig 8) The results of this performed spatial query (Fig. 8)
indicate that high population districts such as Al-
Rabwah Bani Malik, and Al-Nuzlah are also having
large sizes of diabetes patients.
Accordingly, it can be safely said that spatial
patterns of diabetes patients follows the patterns of diabetes patients follows the
pattern of population density at Jeddah city.
#
Fig. 8. Finding Relationships
#
#
#
#
#
#
#
#
# #
#
# #
#
#
#
#
#
# #
#
#
#
#
#
#
#
#
#
#
#
#
#
#
Al-Rabwah
Bani Malik
Al-Nuzlah
Al-Jamah
Red Sea
Al-Shatea
Population Density
0 - 23.3
23.3 - 63.88
63.88 - 133.23
133.23 - 198.57
198.57 - 316.07
Diabetes Patiants
# 0
# 1 - 189
#
190 - 290
#
291 - 745
#
746 - 2572
N
E W
S 7/11/2011
26
Modeling Health demand Market
Geographical information systems are developed
d l d l f l d l h b
g p y p
today to include several useful models that can be
used for defining spatial variations of health data.
One of these models is known as Kriging models,
which are also known as geastatistical models, and
considered as optimal interpolators that produce
estimates which are unbiased and have known
minimum variance minimum variance.
The presented study has selected this technique to
model the spatial variation of diabetic disease at
Jeddah city.
Modeling Health demand Market
The Kriging model ass mption indicates The Kriging model assumption indicates
that spatially distributed objects are
spatially correlated, in other words,
things that are close together tend to
have similar characteristics have similar characteristics. 7/11/2011
27
Modeling Diabetes Variation
Fig. 9 shows the output of this model and it defines
h d b d ll
g p
how diabetes patients are spread out at all city
parts but with different amounts.
It also shows how the highest concentrations of
diabetes which are found at Al-Rabwah, Bani
Malik, Al-Jamiah, and Al-Rowais, are also spread
at the areas closer to them.
F l ll d Al S li i h i h For example, an area called Al-Syliamaniah is have a
remarkable diabetes patients because it is very
close to a higher diabetes location called Al-
Jamiah.
Fig. 9. Modeling Diabetes Variation
Red Sea Red Sea
Al-Rabwah
Bani Malik
Al-Rowai s
Al-Jamah
20 0 20 Kilometers
Predicted Diabetes
0.101 - 106.196
106.196 - 212.292
212.292 - 318.387
318.387 - 424.483
424.483 - 530.579
530.579 - 636.674
636.674 - 742.77
No Data
Road
N
E W
S 7/11/2011
28
Modeling Diabetes Variation
This model can be used by health planners at Jeddah This model can be used by health planners at Jeddah
city to define city parts that are more likely to have
more diabetic patients.
These parts can be reached for health protection
purposes and for the management of patients living
within these areas.
CONCLUSION
This paper has demonstrated that GIS can be used to
explore the patterns of health demand and to p p
predict the spatial variation of health demand at
Jeddah city.
The outputs of this application indicate that diabetes
disease is located mainly at Al-Rabwah, Bani
Malik, Al-Jamiah, and Al-Rowais at Jeddah city.
It is also founded that there is a direct relationship It is also founded that there is a direct relationship
between health demand location and population
density. 7/11/2011
29
Thank You
FAZLAY FARUQUE( PHD) , S. CHATTERJEE,
DAVI D DZI ELAK( PHD) , WARREN JONES( MD)
IDENTIFYING GAPS IN AVAILABLE
HEALTHCARE RESOURCES IN THREE
SOUTHERN STATES MS,LA,AL
y OUTLINE
{ INTRODUCTION
PROBLEM STATEMENT
OBJECTIVE
{ LITERATURE REVIEW
{ IMPLEMENTATION
DATA DESCRIPTION
METHODOLOGY
RESULTS
{ CONCLUSIONS
{ REFERENCES
IDENTIFYING GAPS IN AVAILABLE HEALTHCARE
RESOURCES IN THREE SOUTHERN STATES MS,LA,AL
INTRODUCTION
Delta and Non-Delta Counties in AL, LA and MS with
Population Density
PROBLEM STATEMENT
The Delta areas of MS, AL and LA have the lowest rankings in the
field of HealthCare.
The death rate from heart disease in the MS delta is more than 45%
greater than the US average. The teen birth rate is the highest in the
country. (obtained from www.mississippimedicalnews.com/delta-
task-force-recommends-eliminating-health-disparities-cms-1303)
In recent decades, the United States has made substantial progress
in improving the residents health and reducing disparities, but
ongoing economic, racial/ethnic and other social disparities in
health still exist.(obtained from Fact Sheet CDC Health
Disparities and Inequalities Report US , 2011)
DATA GENERATED
State Healthcare Resource Data Source
Mississippi
Healthcare Facilities
(n = 1,119)
Directory of MS Health Facilities
MS State Department of Health
Healthcare Providers
(n = 6,026 )
-Dentist, -Nurse, -Pharmacy
Licensure boards
Louisiana
(n = 3,547) LA Department of Health & Hospitals
(n = 28,541)
Licensure boards
Alabama
(n= 2,572)
Healthcare Facilities Directory
AL Department of Public Health
(n = 15,509)
Licensure boards
DATA GENERATED ( HEALTH CARE FACILITIES)
Healthcare Facilites
LA AL MS
Adult Residential Care Abortion or Reproductive Health Centers Abortion
Ambulatory Surgical Centers Ambulatory Surgical Centers Ambulatory Surgery Facilities
Community Mental Health Center(Enrollment Only) Assisted Living Facilities Community Health Centers-Satellite Facilities
Comprehensive Outpatient Rehab Facilities Assisted Living Facilities (Specialty Care) End Stage Renal Disease Facilities
Emergency Medical Transportation Cerebral Palsy Centers Home Health Agencies-Hospital Based
End Stage Renal Disease Community Mental Health Centers Home Health Agencies-Private Freestanding
Home Health Agencies End Stage Renal Disease Treatment Centers Home Health Agencies-State Department of Health
Hospice Federally Qualified Health Centers Hospices
Hospitals Home Health Agencies Hospitals
Intermediate Care Facilities for Dev. Disabilty Hospices ICF/MR Facilities and Community Homes
Mental Health Clinic (State) Hospitals Long Term Care Facilities
Nurse Aide Training Schools Independent Clinical Laboratories Outpatient Rehab Facilities
Nursing Home Independent Physiological Laboratories Personal Care Home Facilities
Outpatient Physical Therapy Nursing Homes Psychiatric Residential Treatment Facilities
Pain Management Clinics Organ Procurement Centers Rehabilitation Facilities
Personal Care Attendant Portable X-Ray Suppliers Rural Health Clinics
Substance Abuse Rehabilitation Centers Certified X-Ray
Residential Treatment Facilities
Rural Health Clinics
Sleep Disorders Centers
3547 2572 1119
OBJECTIVE
To identify:
Whether any Pattern exists between racial/ethnic disparities in
income
Whether there is any spatial dependency of the variables.
Pattern in the distribution of Geriatric Population.
The accessibility to hospitals classified as General, Acute Care and
Critical Access.
Finding the Service Area around a hospital within 30 min drive time
Population Distribution in block groups which are more than 30
minutes away from a nearest hospital where they can go for
stabilization.
y OUTLINE
{ INTRODUCTION
PROBLEM STATEMENT
OBJECTIVE
{ LITERATURE REVIEW
{ IMPLEMENTATION
DATA DESCRIPTION
METHODOLOGY
RESULTS
{ CONCLUSIONS
{ REFERENCES
LITERATURE REVIEW
HEALTHPREDICTORS
M.A. Winkleby, D.E. Jatulis et. al (1992) in their study on socioeconomic
status (SES) and health found that higher education is the best SES
predictor of good health.
Williams, David R. (1999) stated that Racism can also directly affect
health in multiple ways. According to him - Residence in poor
neighborhoods, racial bias in medical care, the stress of experiences of
discrimination and the acceptance of the societal stigma of inferiority
can have deleterious consequences for health.
LITERATURE REVIEW(contd.)
Waitzman NJ and Smith,(1998), showed that residing in a poverty area
was positively associated with elevated risk of cardiovascular and
cancer mortality over a 13 to 16 year period, independent of individual
level factors including household income and baseline health status.
Studies by Dowler,2001 and Lawlor et al. ,2005 also suggested that
regions with high rates of socioeconomic deprivation are associated
with increased risk of cardiovascular disease.
SCALE OF MAPPING
Nadine Schuurman, Nathaniel Bell et al. (2007) highlighted the influence
of spatial extent and scale on mapping population health. Deprivation
indices are susceptible to spatial granularity and the Modifiable Aerial
Unit Problem (MAUP) effect is best ameliorated by using large scales
with higher resolution.
EFFECTS OF SPATIAL AUTOCORRELATION
Vincent Lorant, Isabelle Thomas et al. (2001) in their study on
Deprivation and mortality in UK concluded that spatial autocorrelation
has a significant impact on the relationship between mortality and
socioeconomic variables.
If found it suggests that many statistical tools and inferences are
inappropriate: correlation coefficients or least square estimators are
biased and overly precise (Tiefelsdorf ,2000).
As stated by Timothy S. Hare(2007) Access to medical services is a
multidimensional variable. Beyond distance to services, other factors like
insurance, employment, income, education jointly affect the accessibility and
utilization of the facilities.
MEASURES OF GEOGRAPHIC ACCESSIBILITY:
1) simple ratio of supply to demand in an area
2) distance to the closest provider
3) Gravity method a) a two-step floating catchment area (2SFCA) method and
(b) a kernel density (KD) method.
Duck-Hye Yang, Robert Goerge and Ross Mulner(2005) found that accessibility
ratios using kernel method is extremely volatile and are not suitable in
identifying low accessibility areas.
y OUTLINE
{ INTRODUCTION
PROBLEM STATEMENT
OBJECTIVE
{ LITERATURE REVIEW
{ IMPLEMENTATION
DATA DESCRIPTION
METHODOLOGY
RESULTS
{ CONCLUSIONS
{ REFERENCES
SPATIAL DISTRIBUTION OF MINORITY POPULATION
MINORITY = TOTAL
POPULATION - WHITE
The frequency distribution plot
shows a distinct bimodal
distribution indicating
clustering of the communities
Data Source: ESRI Business Analyst
Datum : WGS 1984
Projection :Web_Mercator_Auxiliary_Sphere
POPULATION WITH AGE>25 YRS AND EDUCATION
LESS THAN 9
TH
GRADE
The crescent shaped area
with low income group
and minority population
also faces problems in
cultivating a safe
learning environment
Datum : WGS 1984
SPATIAL DISTRIBUTION OF MEDIAN HOUSEHOLD
INCOME
Skewed distribution with the
peak in the low income group
Datum : WGS 1984
Projection :Web_Mercator_Auxiliary_Sphere here
MINORITY DISTRIBUTION AND HOUSEHOLD INCOME
WHEN VIEWED SIDE BY SIDE
Datum : WGS 1984
Blockgroups with minority greater than 83% has a
leptokurtic distribution in income with low
amount of variance
MINORITY DISTRIBUTION AND HOUSEHOLD INCOME
WHEN VIEWED SIDE BY SIDE
Datum : WGS 1984
Blockgroups with minority less than 16.6% has a
platykurtic distribution in income with high amount
of variance
MORANS I SCATTER PLOT and HYPOTHESIS TESTING
MINORITY DENSITY
Reference Distribution for computing the
significance of Morans I
Randomization histogram
There is a statistically significant spatial autocorrelation in
distribution of minority population
LISA MAP FOR MINORITY DENSITY
High value of Local Morans I a
statistic used for spatial
autocorrelation - indicates a very
strong positive relationship.
These are areas that have higher or
lower rates than is to be expected by
chance
The value at each block group is compared with the weighted
average of the value of its neighbors. The weights file is
constructed based on queens contiguity.
Datum : WGS 1984
MORANS I SCATTER PLOT and HYPOTHESIS TESTING
MEDIAN HOUSEHOLD
INCOME
Reference Distribution for computing the
significance of Morans I
Randomization histogram
There is a statistically significant spatial autocorrelation in
distribution of Median Household Income also.
LISA MAP FOR MEDIAN HOUSEHOLD INCOME
Local Morans I 0.48
indicates spatial
clustering
Datum : WGS 1984
BIVARIATE LISA CLUSTER MAP of MINORITY DENSITY w/
MEDIAN HOUSEHOLD INCOME
A significant and negative
Morans I (-0.334) indicates block
groups with high minority density
are adjacent to block groups with
low median household income .
Datum : WGS 1984
ANALYZING PATTERNS IN DISTRIBUTION OF GERIATRIC
POPULATION
Getis Ord General G statistic (0.026) for geriatric
population indicates a low cluster pattern
Given the z-score of -4.24, there is a less
than 1% likelihood that this low-
clustered pattern could be the result of
random chance.
Datum : WGS 1984
SERVICE AREA AROUND HOSPITALS BASED ON 30
MINUTES DRIVE TIME
Hospitals with General,
Acute Care and Critical
Access facilities taken for
calculation
It shows a considerable
lack in coverage in Delta
areas.
Datum : WGS 1984
DETERMINING ACCESSIBILITY RATIO USING 2 STEP FLOATING CATCHMENT
AREA METHOD
They have found the two step floating
catchment area method ( originally proposed
by Luo and Wang (2003) for giving better
accessibility ratios than the kernel density
method.
Source: Duck-Hye Yang et
al. (2005)
STEPS:
1) First a service area is defined.
2) A ratio is computed by dividing the
number of hospitals by the population
values within the service area
3) In places where the service areas
overlap the ratios are added together
to give the new accessibility ratio.
27
DETERMINING ACCESSIBILITY RATIO USING 2 STEP FLOATING CATCHMENT AREA
METHOD
It suggests what Gormick
has observed in 2003
that those who are
economically and socially
advantaged use more
type of services that
prevent illness and
improve health than do
other minority groups,
who are less advantaged
and in poorer health.
Datum : WGS 1984
Data Source: ESRI Business Analyst D WGS 8 D S ESRI B i A l
POPULATION WITHOUT ANY ACCESS TO A HOSPITAL
WITHIN 30 MINUTES

As seen the areas outside
30 min drive time coverage
lies mostly in areas where
minority density is high
Datum : WGS 1984
CONCLUSIONS
y Block groups with minority > 83% also have low average household income.
y Distribution of average household income and minority density shows positive
spatial autocorrelation.
y Block groups with high minority density are adjacent to block groups with low
household income.
y The distribution of geriatric population also shows a low clustered pattern which
cannot be the result of random chance.
y The areas outside 30 min drive time coverage lies mostly in areas where
minority density is high.
y High accessibility ratio in the delta areas compared to non delta visa-vis the
lowest ranking in the field of healthcare in delta areas only goes on to support
the fact already stated by Timothy S. Hare that besides the distance to services,
other factors like insurance, employment, income, education jointly affect the
accessibility and utilization of facilities.
UU
REFERENCES
y Duck-Hye Yang, Robert Goerge and Ross Mulner (2005). Comparing GIS Methods of Measuring Spatial Accessibility
to Health Services. Journal Med Sys 30(1):23-32
y Dawler, E.(2001). Inequalities in diet and physical activity in Europe. Public Health Nutrition, 4(2B), 701-709
y Gormick, M.E.(2003). A decade of research on disparities in medicare utilization : Lessons for the health and
healthcare of vulnerable men. American Journal of Public Health, 93(5), 753-759
y Hare, T.S., Holly R. Barcus(2007). Geographical accessibility and Kentuckys heart related hospital services. Applied
Geography 27(2007) Pages 181-205
y Luo W, Wang F (2003) Measures of Spatial accessibility to health care in a GIS environment: Synthesis and a case
study in Chicago region. Environment and Planning B 30(6):865-884
y Lawlor, D.A., OCallaghan, M.J., Mamun, A.A., Williams, G.M., Bor,W., & Najman, J.M.(2005). Socioeconomic
Position, Cognitive function, and clustering of cardiovascular risk factors in adolescence: Findings from the master
University study of Pregnancy & its outcomes. Psychomatic Medicine, 67(6), 862 868.
y Nadine Schuurman, Nathaniel Bell, James R. Dunn, and Lisa Oliver(2007).Deprivation Indices, Population Health
and Geography: An evaluation of the spatial effectiveness of Indices at Multiple Scales. Journal of Urban Health:
Bulletin of the New York Academy of Medicine, Vol 84, No 4
y Tiefelsdorf, M., 1997. Modelling spatial processes: The identification and analysis of spatial relationships in
regression residuals by means of MoranI, Springer, Berlin.
y Vincent Lorant, Isabelle Thomas, Demise Deliege and Rene Tonglet(2001). Deprivation and mortality: the
implications of spatial autocorrelation for health resources allocation. Social Science & Medicine, Volume 53, Issue 12,
December 2001, Pages 1711-1719
U
REFERENCES(Contd)
y Winkleby MA, Jatulis DE, Frank E, Fortmann SP.(1992) Socioeconomic status and health:
how education, income, and occupation contribute to risk factors for cardiovascular disease.
AmJ Public Health. 82(6):816-20.
y Williams, David R. (1999) Race, Socioeconomic Status, and Health New York Academy of
Sciences, 896. pp. 173-188
y Waitzman NJ, Smith KR (1998) Phantom of the area: Poverty-area residence and mortality
in the United States. American Journal of Public Health. 1998, 88:973976.
THANK YOU
SUGGESTIONS ?
7/1/2011
1
Accessible, Online Tools for
Health Policy and Planning
JENNIFER L. RANKIN, PHD
JUNE 28, 2011
Agenda
y About the Robert Graham Center
y www.udsmapper.org
y www.medschoolmapper.org
y HealthLandscape
7/1/2011
2
Robert Graham Center- Charge from the
American Academy of Family Physicians
y The Center would be responsible for research and
analysis to inform the deliberations of the Academy analysis to inform the deliberations of the Academy
in its public policy work and provide a family
practice perspective to policy deliberations in
Washington
y The Center's work would include:
{ research to support the Academy's policy development
and advocacy efforts (research done at the direction and and advocacy efforts (research done at the direction and
request of the Academy)
{ Center-initiated research to explore policy issues affecting
the ability of family physicians to provide their services to
the public at a maximum level of effectiveness
Robert Graham Center Staff
y 10 full-time staff including health geographers,
primary care doctors, health economists, and primary care doctors, health economists, and
sociologists
y 1 two-year fellow
y 10-12 scholars annually
y 2-3 part time research assistants
y Partners with HealthLandscape LLC and numerous y Partners with HealthLandscape, LLC and numerous
other collaborators
y Blue Raster, LLC, does most programming
7/1/2011
3
www.graham-center.org
www.udsmapper.org
y Funded by the Health Resources and Services
Administration Bureau of Primary Health Care Administration Bureau of Primary Health Care
y Uses Uniform Data System data- annually collected
data from federally funded Community Health
Centers
{ Very dense report- uses only one table, Patients by ZIP Code
{ First time that this dataset has been displayed at a local level
y Displays the footprint of the federal health center
program nationally
7/1/2011
4
Penetration of the Low Income Population
Local Example- Woonsocket, Rhode Island
7/1/2011
5
Additional Layers in the UDS Mapper
Download Data
7/1/2011
6
Analysis Results- User Input
New Features of the UDS Mapper
y Interactive data tools (Analysis)
{ User entered data is combined with data from UDS Mapper { User entered data is combined with data from UDS Mapper
{ Provides language that gives context to the calculations
y Interactive geo-data tools (Sliders)
{ Allows the users to interact with a dataset and select
thresholds for display
{ Can be used for continuous or categorical data
{ Can be programmed for single datasets or composite indices
7/1/2011
7
Single Dataset- Poverty
Poverty Slider- 15% Threshold
7/1/2011
8
Composite Index- HPSA Scores
Composite Index Sliders
7/1/2011
9
Data Used in the UDS Mapper
y Data come from many sources and are updated as
frequently as monthly but we will only update the q y y y p
data as frequently as the source updates their data
{ UDS is annual
{ HGDW is monthly
{ US Census currently decennial
y Dates for main data sources:
{ UDS is from 2009 calendar year; 2010 data expected in July 9 y ; p y
{ HGDW- varies depending on when they upload data which is
on varying cycles depending on the data source
{ US Census is currently from 2000; will use estimates based on
2005-2009 ACS and new ZCTA boundaries
Data Considerations
y UDS Mapper was built on a single use-case, but
could be expanded to meet broader needs cou d be e pa ded to eet b oade eeds
{ JSI provides data calculated for this use-case- Robert
Graham Center does not receive raw health center data
y U.S. Census Data
{ The decennial census no longer contains income data
{ American Community Survey to take its place
A d t ith th d t lit h A new product with those new product glitches
y HRSA Geospatial Data Warehouse
{ Major source for provider location data
Spotty reliability for downloading data
Limited to no metadata
7/1/2011
10
www.medschoolmapper.org Footprint of
All Medical Schools in Single State
Graphic Data Incorporated into Map View
7/1/2011
11
Single Medical School Footprinting
County-Level Data Display via Rollovers
7/1/2011
12
Dynamic Footprint Comparison
Advanced Analytic Tools
7/1/2011
13
New HealthLandscape Platform (State Poverty)
Dynamic Geography Selection
7/1/2011
14
Modular Tools
Easily Add Your Own Data
7/1/2011
15
HealthLandscape Australia Default Map Total
Healthcare Workers per 10,000
Melbourne (Inner) Rollover Total Healthcare
Workers per 10,000
7/1/2011
16
Sydney vs. Melbourne Health Professional and
Population Data
Contact Information
y Robert Graham Center:
1350 Connecticut Ave NW Ste 201 1350 Connecticut Ave NW, Ste. 201
Washington, DC 20036
202-331-3360
y Jennifer Rankin: jrankin@aafp.org
Text udsmapper to 50500 for contact
details
1

Online geographic information systems for improving health planning practice:
lessons learned from the case study of Logan-Beaudesert, Australia

Ori Gudes

Griffith Health Institute
Griffith University
University Drive, Meadowbrook, Queensland, Australia, 4131
61-7-33821539, 61-7-33821414, o.gudes@griffith.edu.au

School of Urban Development
Queensland University of Technology
2 George St, Brisbane QLD 4000, GPO Box 2434, Australia
61-401197227, ori.gudes@student.qut.edu.au

Elizabeth Kendall

Griffith Health Institute
Griffith University
University Drive, Meadowbrook, Queensland, Australia, 4131,
61-7-33821202, 61-7-33821414, e.kendall@griffith.edu.au

Tan Yigitcanlar

61-7-31382418, 61-7-31381170, tan.yigitcanlar@qut.edu.au

Virendra Pathak

61-7-31386801, virendra.pathak@qut.edu.au

2

ABSTRACT
This study examines the impact of utilising a Decision Support System (DSS) in a
practical health planning study. Specifically, it presents a real-world case of a
community-based initiative aiming to improve overall public health outcomes.
Previous studies have emphasised that because of a lack of effective information,
systems and an absence of frameworks for making informed decisions in health
planning, it has become imperative to develop innovative approaches and
methods in health planning practice. Online Geographical Information Systems
(GIS) has been suggested as one of the innovative methods that will inform
decision-makers and improve the overall health planning process. However, a
number of gaps in knowledge have been identified within health planning
practice: lack of methods to develop these tools in a collaborative manner; lack of
capacity to use the GIS application among health decision-makers perspectives,
and lack of understanding about the potential impact of such systems on users.
This study addresses the abovementioned gaps and introduces an online GIS-
based Health Decision Support System (HDSS), which has been developed to
improve collaborative health planning in the Logan-Beaudesert region of
Queensland, Australia. The study demonstrates a participatory and iterative
approach undertaken to design and develop the HDSS. It then explores the
perceived user satisfaction and impact of the tool on a selected group of health
decision makers. Finally, it illustrates how decision-making processes have
changed since its implementation. The overall findings suggest that the online
GIS-based HDSS is an effective tool, which has the potential to play an important
role in the future in terms of improving local community health planning practice.
However, the findings also indicate that decision-making processes are not merely
informed by using the HDSS tool. Instead, they seem to enhance the overall sense
of collaboration in health planning practice. Thus, to support the Healthy Cities
approach, communities will need to encourage decision-making based on the use
of evidence, participation and consensus, which subsequently transfers into
informed actions.
3

INTRODUCTION
In the last few decades, the focus on building healthy communities has grown significantly
(Ashton, 2009). This trend is the result of an international initiative to create the broad conditions
that contribute to health rather than simply to continue to treat burgeoning levels of disease
(Otgaar et al. 2011). As part of those efforts, the process of developing healthy communities has
become an important focus for health planners (Otgaar et al. 2011). There is growing evidence
that new approaches to planning are required, based on timely use of local information,
collaborative health-planning, and the engagement of the communities in local decision-making
(Murray, 2006; Scotch & Parmanto, 2006; Ashton, 2009; Kazada et al., 2009). However, there is
little research in relation to the methods that support this type of responsive, local, collaborative
and consultative approach to health planning (Northridge et al., 2003).
Some research justifies the use of Decision Support Systems (DSS) in planning for healthy
communities in that they have been found to increase collaboration between stakeholders and
communities, improve the accuracy and quality of the decision making process, and improve the
availability of data and information for health decision-makers (Nobre et al., 1997; Cromley &
McLafferty, 2003; Waring et al., 2005). Geographical Information Systems (GIS) has been
suggested as an innovative method by which to implement DSS. Furthermore, literature has
indicated that online environments have a positive impact on decision-making by enabling access
by a broader audience (Kingston et al., 2001).
However, only limited research has been conducted about how to implement online DSS or
evaluating its impact on decision-makers. Previous studies have emphasised that due to the lack
of effective information, systems and an absence of frameworks for making informed decisions
in health planning, it has become imperative to develop innovative approaches to, and methods
for, health planning practice (Higgs & Gould 2001). Researchers have identified a number of
gaps in our knowledge (Kazada et al., 2009; National Health and Hospitals Reform Commission,
2008), including, a lack of methods to develop DSS tools in a collaborative manner; lack of
knowledge about GIS applications among health decision-makers; and limited understanding
about the potential impact of DSS on decision-making processes. Thus, this study focuses on
developing a DSS, and a method of evaluating its impact on health planners and decision-
makers. Specifically, the study examines the development and implementation process, the usage
and response to the intervention, and its impact on decision-making processes in a particular case
study, the Logan-Beaudesert Health Coalition.
In response to the growing level of health risk factors in the last five years in the Logan-
Beaudesert area, it was identified that the cost of chronic disease to society remains significant
and current management and planning methods do not appear to be having sufficient impact.
Consequently, collaborative planning was suggested as a method for improving outcomes. As a
result, the Logan Beaudesert Health Coalition (LBHC) was established in 2006. The LBHC aims
to deliver innovative services that focus on broader determinants of health framework (i.e.,
Schulz & Northridge, 2004) to reduce risk factors, thus reducing the incidence of chronic disease
4

in the area (Kendall et al., 2007). However, it quickly became clear that the LBHC did not have
access to new methods or tools for undertaking collaborative planning.
This study focused on the development, implementation and evaluation of a practical
method for decision-makers to participate in collaborative health planning that can encourage the
creation of the local conditions necessary to promote health in their region. It culminated in an
innovative tool (i.e., HDSS
1
) that aimed to enhance collaborative mechanisms by facilitating
decision-making based on evidence, participation and consensus.
USE OF EVIDENCE, PARTICIPATION AND CONSENSUS: THE WAY HEALTHY
COMMUNITIES MAKE DECISIONS
Although more than 20 years have passed since the initiation of the Healthy Cities
movement, there is some evidence that it has not yet achieved its full potential (Ashton, 2009).
Recently, the founder of the Healthy Cities movement (i.e., Kickbusch) called for a renewal of
the commitment (Ashton, 2009) on the basis that the urban agenda has become even more
relevant. Trends such as rapid urbanisation, unsustainable development, and global warming
have highlighted the necessity of a focus on urban health. Towns, cities and communities
committed to promoting health and sustainability now face two key challenges: how to move
health promotion from the margins to the mainstream; and how to integrate multiple forms of
information and sectors in such a way that planning can contribute to the development of
Healthy Cities (Dooris, 1999).
The promotion oI healthy` public policy has been noted as being central to the Healthy
Cities approach (Flynn, 1996). However, the Healthy Cities concept necessitates planning that
moves beyond current approaches. It requires planning that focuses on the whole community and
the promotion of health, rather than being confined to the development of responses to one or
more specific health problems based on a narrow body of knowledge. The Healthy Cities
concept is based on models of city governance in which public authorities recognise the need to
work with and support a range of actors who are either fully committed to health, or play a
significant role in contributing to the conditions that promote health (WHO, 1999). The Healthy
Cities concept suggested the need to restructure health decision-making processes, shifting
power to the local level, and basing decisions on a localised but comprehensive body of
knowledge. Planning for Healthy Cities requires collaboration between different groups in the
community that can contribute to health-promoting conditions, such as local government,
community organisations, universities, private organisations, and health services. The literature
suggests that stronger collaborations between urban planners and public health practitioners may
prove effective in designing and planning for Healthy Cities (Northridge et al., 2003). Given this,
the process of decision-making in health planning should be based on a structured model that
draws together multiple forms of knowledge and increases the possibility of coherent localised
and responsive solutions (Scotch & Parmanto, 2006).

1
HDSS denotes the name of the system prototype, whereas DSS is a term which represents the decision support
systems concept
5

Flynn (1996) suggests the following steps for developing Healthy Cities: establishing a
broad structure for the community, encouraging community participation, assessing community
needs, establishing priorities and strategic plans, soliciting political support, taking local action,
and evaluating progress. Despite the presence of these guidelines for creating Healthy Cities,
there is little consensus about how health planning can best be practiced (Duhl & Sanchez,
1999). Thus, to support the Healthy Cities approach, communities will need to encourage
decision-making based on the use of evidence, participation and consensus, which subsequently
transfers into actions.
POTENTIAL OUTCOMES OF DECISION SUPPORT SYSTEMS IN HEALTH
PLANNING
The role of DSS in health planning practice continues to evolve, with the application of
this technology being an important step towards better understanding of public health issues and
their inherent complexities (Waring et al., 2005). The literature identifies a number of potential
outcomes of DSS, including increased collaboration or participation, trust, increased satisfaction
in decision-making, user satisfaction, construction of knowledge, and increased use of evidence
in decision-making processes (Igbaria & Guimaraes, 1994). Even if the system (i.e., DSS) does
not address all users` needs, the fact that users have played an important role in designing the
system and its constant refinement process, contributes to the overall notion of collaboration
reflected by participants (Omar & Lascu 1993; Murray, 2006).
DSS is perceived to have a role in a number of settings for health planning. These include
identifying service health barriers and health needs for particular populations or regions,
supporting strategies to address gaps, facilitating multi-directional communication channels, and
re-affirming transparent communication and decision-making processes (Phillips et al., 2000).
To encourage community engagement and reduce health inequalities, DSS may be used as an
outreach vehicle for community-based public health empowerment. In turn, this 'may help our
understanding oI the complex relationship between socioeconomic Iactors and health status
(Phillips et al., 2000, p. 976).
One contemporary method for implementing a DSS is to use GIS. Research has indicated
that GIS has the potential to be used in a range of decision-making tasks. The use of analysis and
visualisation capacities (e.g., spatial aspect) within GIS provides an opportunity to use this tool
to create an innovative DSS. For example, through GIS, users can visualise the effects of
healthcare delivery strategies (Higgs & Gould, 2001). If GIS is to be integrated into the decision-
making mechanism, then several prospective improvements could be accrued, particularly in the
context of the local government public health sector.
Significantly, research indicates that online environments have a positive impact on
decision-making (Kingston et al., 2001).The ultimate technical goal of online DSS is to ensure
that information is made available for end-users to perform analyses and represent their own
results within the system (Yigitcanlar & Gudes 2008). Contrary to static presentations, online
information becomes dynamic when users are allowed to access or interact with the database
from their own computer (Croner, 2003). The number of online DSS is increasing rapidly as the
technology becomes more readily available and more industries realise their potential (Su et al.,
2000). Indeed, Richards et al. (1999) emphasized that the application of GIS techniques in an
6

online DSS allows decision-makers to ask questions of maps and to quickly, clearly, and
convincingly show the results of complex analyses. However, unless health planning is also
practiced in a collaborative manner, simply increasing access to effective information through
DSS may not be sufficient to generate the type of decision-making that can lead to healthy cities.
CASE STUDY
The LBHC partnership was established to address the growing level of chronic disease risk
factors in the Logan-Beaudesert region of Queensland. This initiative aimed to enhance existing
services and infrastructure, establish formal partnerships, improve existing resources, and
implement additional services and strategies. Its ultimate goal was to improve the health capacity
of the region at multiple levels through enhanced and responsive localised planning. The LBHC
has a central committee (LBHC board), which oversees six health programs, each with an
advisory group drawn from the relevant sector. Each program addresses a specific area identified
as needing attention in the region, early childhood (0 to 8 years of age), multicultural health, the
prevention and management of existing chronic diseases, the integration between general
practices and acute settings, efficient health information management, and health promotion. By
providing recommendations and information, the programs assist the LBHC board to make
decisions and develop policies and strategies. The LBHC board coordinates and directs the
coalition as a whole. The Queensland State Government funds the LBHC, and has given its
board the mandate to modify, alter or adapt any of the current programs in response to evidence
and performance data, with the scope to design and implement new health initiatives as required.
The decisions of the LBHC board are reflected back to the six health programs for
implementation. Thus, the LBHC was an ideal platform for designing, implementing and
evaluating the DSS, arising from the need to help LBHC board members make better decisions
that would contribute to the development of a healthy communities in the Logan-Beaudesert
area. Figure 1 illustrates the Area of Interest (AOI).
7

Figure 1. Map of the Logan Beaudesert region

METHOD
PARTICIPATORY ACTION RESEARCH
Collaborative planning approaches are increasingly being advocated and implemented in
Healthy Cities initiatives due to the demonstrated benefits of these approaches (Murray, 2006).
One approach for facilitating collaboration that has been used for some time in many fields is
Participatory Action Research (PAR). PAR is being increasingly used as an overarching name
for an orientation toward research practice that places the researcher in the position of co-learner,
and puts a strong emphasis on input from participants or end-users as well as the ongoing
translation of research findings into action (Minkler, 2000). This approach has gained attention
in the health planning field, particularly in the public health context (Minkler & Wallerstein,
2003). One of the most important characteristics of PAR is the fact that participants or
stakeholders, whose lives are affected by the research initiative, take an active role in its design,
implementation and evaluation. It was anticipated that the application of PAR to the
development, implementation and evaluation of the HDSS tool would predispose the LBHC
board to engage in collaborative processes, actively participate in decisions and take collective
responsibility for the outcome of the study. By exposing the board members to the DSS in this
manner, it was hypothesised that these same qualities might be reflected in their decision-making
once the DSS was established (i.e., use of evidence, participation and consensus in decision-
making).
The PAR approach also addressed an important requirement of the DSS literature, namely
flexibility. Specifically, research has emphasized that one of the key requirements of a
collaboration-based system is its Ilexibility to adapt to users` needs, thereby increasing planning
8

efficiency. Thus, by applying PAR, the DSS was more likely to respond appropriately to users`
needs, resulting in greater engagement in the long-term and, presumably, better decision-making.
There is evidence in the literature that decision-making satisfaction in the context of a decision
support system is likely to be associated with the perceived quality of the system, information
and presentation. Indeed, the literature emphasises that the best predictor of effective decision-
making is satisIaction with one`s decision-making (Bharati & Chaudhury, 2004).
Finally, PAR enabled the DSS to be applicable to the local circumstances. The DSS
literature has emphasized the fact that health planners do not have at hand the local knowledge
needed to determine the type of information required for good decisions (Gudes et al., 2010).
However, it has also highlighted that the development of information frameworks is not a simple
matter. As Flynn (1996) has argued, every community is unique, with different physical, social,
political and cultural contexts that must be understood in the planning process. For this reason,
health planners must develop a thorough understanding of the community health profile and the
structural features that influence its health. The framework used to structure information should
organise that information in a way that directs the attention of decision-makers to the entire
range of conditions influencing health (Gudes et al., 2010). By using PAR, collective agreement
was reached on a suitable framework (i.e., Schulz and Northridge, 2004) to guide the GIS data
collection efforts. In addition, the participants were able to prioritize each layer of information
according to their local requirements.
Participatory Action Research Intervention
The PAR design incorporated both quantitative and qualitative techniques of data
collection and analysis to engage board members in the design, development and implementation
of the HDSS. Our PAR approach was implemented in three cycles, namely: PAR 1 (i.e.,
Introduction Stage); PAR 2 (i.e., Interaction Stage), and PAR 3 (i.e., Trialling Stage). Figure 2
illustrates the PAR Intervention.
PAR cycle 1: Introduction Stage
The Introduction Stage was associated with the early days of the study, where the concept
of GIS was first introduced to LBHC board members and included several introductory
presentations to raise their awareness. The PAR intervention phase commenced with a series of
GIS introductory presentations to the LBHC board and other advisory groups that took place in
March and April 2010. The primary purpose of this cycle was to raise awareness of the GIS and
DSS as tools to support decision-making. To raise the awareness of the LBHC board, this cycle
included a number of demonstrations of GIS, as well as discussion about its impact and potential
application to local decision-making.
PAR cycle 2: Interaction Stage
The Interaction Stage was associated with the period of time between the Introduction
Stage and Trialling Stage, where LBHC board members were engaged (e.g., via consultation
meetings and workshops) in the collaborative process of designing and developing the HDSS.
In line with the recommendation of Maeng and Nedovic-Budic (2010), PAR 2 consisted of
a series of consultation meetings to obtain input from end-users about the prospective GIS
information items, its features and functionality, and health scenarios (i.e., workflows) to be
9

included in the HDSS. To assist in identifying the relevance and urgency of including particular
types of information in the HDSS prototype, a Data Priority Survey was conducted with the
HDSS end-users (i.e., LBHC board members). The information obtained in this survey was
based on the Schulz and Northridge`s (2004) Iramework which was agreed by LBHC board
members as being a suitable foundation for the management of information. The data collected
from the survey was categorised into three groups according to the level of priority (1 = essential
now; 2 = could be included in next phase; and 3 = not necessary at all). The total score was
calculated for each item across the board members (See Appendix 1), which was then used to
determine the final level of priority. This ranking system made it possible to ascertain which GIS
information items to include in the HDSS prototype. The data was presented back to the board
members and the information items which were not considered to be essential were excluded
from the current version of the system.
In addition to selecting information items, board members participated in discussions about
the inclusion of features and functionality in the HDSS. The LBHC members were provided with
examples of potential features and functionality using demonstrations of other GIS applications.
A discussion was held about each feature, and LBHC board members were asked to determine its
inclusion and priority until a final list was constructed.
Based on the board`s decisions, a list of workflows was suggested. HDSS users were
guided through a series of structured workflows that identify the subsequent spatial output that
might be generated, based on a group of predefined information items. The workflows were
designed to demonstrate functional capability of the proposed HDSS prototype, based on real
health data. LBHC board members commented on the suggested workflows, in particular, what
data (i.e., GIS layers) to include in each workflow. Thorough discussion was facilitated to
determine the level of priority of specific data (i.e., GIS layers) within the proposed HDSS
prototype to support their day-to-day planning and decision-making practice. After a fruitful
discussion, two workflows were carefully chosen to be part of the HDSS scope. The revised
workflows were disseminated among the LBHC board members for received final endorsement
prior to the engagement of a web-based GIS developer who created the prototype. In summary,
throughout the PAR Cycle 2 (i.e., Interaction Stage), feedback and information was collected and
analysed collectively providing an invaluable opportunity to design and develop the HDSS in a
collaborative manner.
PAR cycle 3: Trialling Stage
The Trialling Stage encompassed a period of three months from when the HDSS prototype
was officially deployed (March, 2011), and LBHC board members began using the system. The
primary purpose of this stage was to implement and trial the system, while simultaneously
collecting evidence about the extent of usage and degree of user satisfaction. In keeping with the
PAR method, feedback collected during consultations and training sessions was incorporated
into the prototype during this cycle. To collect usage and satisfaction information, two
instruments were used:
Google Analytics script to monitor the number of unique visits, views, and the average
time users used the HDSS; and
10

A User Satisfaction survey to explore and understand the experiences of the LBHC
board members in using the HDSS. This survey was also an important tool for
continual refinement of the system.
Omar and Lascu (1993) identify a five-construct (23 items) scale for measuring user
satisfaction that has been validated and used in a range of contexts. The survey consisted of the
following constructs: information quality (characteristics of information in terms of currency,
accuracy, relevance, flexibility, ease of use and access - 9 items), planning (characteristics of
planning, whether the system was developed as part of a broader planning agenda - 6 items),
staff and services (staff competence and the quality of services supporting the system - 3 items),
system support for decision-making (ability of the information system to support decision-
making processes - 2 items), and user involvement (attributes that generate and encourages user
involvement and participation - 3 items).
Data Analysis
The User Satisfaction survey was utilised to identify the perceived levels of HDSS
satisfaction experienced by LBHC board members. Given that only 17 LBHC board members
participated in this survey, the data was used descriptively to improve the HDSS in accordance
with the PAR method (i.e., as part of PAR Cycle 3). Derived from Omar and Lascu`s (1993)
recommendations, 23 items were identified. These items were associated with five constructs:
Information quality, Planning, Staff and services, Systems supports for decision-making, and
User involvement. The items were then divided into two main groups: importance and
performance. As suggested by Omar and Lascu (1993), the 23 performance items were
multiplied by the importance items, yielding weighted perIormance items`. To measure the
statistical dependence between each of Omar and Lascu (1993) five constructs and a broad
question asking respondents to rate their overall level of satisfaction with the HDSS (See Item 24
in the User Satisfaction survey, Appendix 4), Spearman's correlation test was utilised. Therefore,
the 23 items were cumulated to the five constructs, and were then correlated with the overall
satisfaction construct. This has revealed which construct attained the highest level of correlation
with the overall satisfaction construct.
1
1

P
A
R
C
y
c
l
e
2
(
i
.
e
.
,
I
n
t
e
r
a
c
t
i
o
n
s
t
a
g
e
)
P
A
R
I
n
t
e
r
v
e
n
t
i
o
n
*
In
f
o
r
m
a
t
io
n
it
e
m
s
S
u
r
v
e
y
*
F
u
n
c
t
io
n
a
lit
y
a
n
d
f
e
a
t
u
r
e
s
*
S
y
s
t
e
m
w
o
r
k
f
lo
w
s
P
A
R
C
y
c
l
e
1
(
i
.
e
.
,
I
n
t
r
o
d
u
c
t
i
o
n
s
t
a
g
e
)
P
A
R
C
y
c
l
e
3
(
i
.
e
.
,
T
r
i
a
l
l
i
n
g
s
t
a
g
e
)
P
r
e
P
A
R
I
n
t
e
r
v
e
n
t
i
o
n
P
h
a
s
e
A
u
g
u
s
t
2
0
0
8
(
A
R
C
a
n
d
P
h
D
p
r
o
j
e
c
t
s
c
o
m
m
e
n
c
e
d
)
M
a
r
c
h
2
0
1
0
(
I
n
t
e
r
v
e
n
t
i
o
n
c
o
m
m
e
n
c
e
d
)
M
a
r
c
h
2
0
1
1
(
H
D
S
S
p
r
o
t
o
t
y
p
e
d
e
p
l
o
y
e
d
[
H
D
S
S
p
h
a
s
e
1
]
)
H D S S D E V E L O P M E N T P R O C E S S
H
D
S
S
P
h
a
s
e
2
M
a
r
c
h
2
0
1
2
(
H
D
S
S
P
h
a
s
e
2
)
S
y
s
t
e
m
d
e
s
i
g
n
,
d
e
v
e
l
o
p
m
e
n
t
a
n
d
t
r
i
a
l
l
i
n
g
P
o
s
t
P
A
R
I
n
t
e
r
v
e
n
t
i
o
n
P
h
a
s
e

F
i
g
u
r
e

2
.

P
A
R

I
n
t
e
r
v
e
n
t
i
o
n

12

Decision-making impact evaluation
To understand the potential role of HDSS in improving decision-making processes,
observational data was collected. This method was employed prior and subsequent to the
PAR intervention. Two waves of data collection were used, one prior to the beginning of the
PAR intervention and one following completion of the PAR intervention. This, in turn,
helped exploration and understanding of the decision-making experiences of the LBHC board
members.
Observational data
LBHC board meetings were recorded and transcribed from the outset of this study until
the establishment of the HDSS prototype. Our data collection activities involved listening to
these LBHC board meetings as well as reading through minutes of meetings and summary
notes. Data collected was used to measure and analyse the actual decision-making of the
LBHC board members. To identify trends in the number and, more importantly, the nature of
the decisions made by the LBHC board due to the HDSS intervention, two meetings were
selected in each year as the sample, commencing from the outset of this study (i.e., 2008) to
the Post-Intervention Phase (i.e., 2010 and 2011). When trying to examine whether any
change has occurred in the way actual decisions were made, analysed meetings were
clustered into two groups. Specifically, four analysed meetings were associated with the
period before the intervention (i.e., Pre-Intervention Phase) and four meetings after (i.e., Post-
Intervention Phase). The scale embraced the following dimensions: use of evidence,
participation, and consensus. The response rate was determined by the researcher`s
observation and included the following rates: limited use (e.g., limited use of evidence in the
actual decision), moderate use, and high use.
FINDINGS
PAR CYCLE 1: INTRODUCTION STAGE
Although the GIS concept was introduced informally on several occasions throughout
2008-2009, it was formally presented to the LBHC board members at a meeting in April 2010
after baseline data has been collected. During this meeting, details and a variety of maps were
presented to explain and clarify the potential role of GIS in health planning. LBHC board
members were encouraged to think about their required data needs. In one of these
presentations, a participant stated: ZH QHHG to know what information should be included
in the system. As a result of the initial GIS interaction, some LBHC board members
requested additional GIS information. During the presentations, one participant noted: <HV,
agree this is an important marker in the development of evidence used in the LBHC.
These reactions implied an evolving awareness of the use of GIS in the LBHC board`s
decision-making processes.
PAR CYCLE 2: INTERACTION STAGE
During the Interaction Stage, the LBHC board members collaboratively defined the key
components for designing the HDSS: Information items, features and functionality, and
health scenarios. The following provides more information about the instruments used to
design and develop the system. Appendix 1 presents the main findings from the Information
Items survey. The findings indicate that the most essential information items included:
socioeconomic, demographic, public transportation, shops, roads, recreation, community
facilities, education facilities, health facilities and disease data. Two data items (health
behaviours and hospital admissions) were indicated as being essential, but due to difficulties
13

accessing these datasets, this data was not used in the HDSS prototype. Appendix 2 presents
the final list of selected features and functions which were included in the HDSS prototype,
along with a description of the purpose of each. Based on the information items selected and
the defined features and functionality, the LBHC board members were consulted to articulate
the details of the two workflows of the HDSS prototype (i.e., proximity and accessibility to
health facilities). One of the designated workflows is illustrated in Appendix 3.
PAR CYCLE 3: TRIALLING STAGE
During the Trialling Stage we used two instruments to understand the extent of usage
and degree of satisfaction the HDSS attained. The first instrument was Google Analytics
script which monitored the systems logs. Findings indicate that throughout the three months
of trialling the system, it was visited more than 100 times by 33 unique users (excluding the
admin group). On average, users spent four minutes in using the system. Also, evidence
indicates that some users were using the systems from different computers (e.g., office, home
etc.). Given that only 17 LBHC board members had access to the system and the time of
implementation was short (three months), the extent of usage was considered to be good.
As for the degree of satisfaction, we utilised a User Satisfaction survey to understand
user`s experiences with the system. Twelve LBHC board members completed the
questionnaire, and given that there were approximately 17 HDSS users at the time, this
response rate was considered to be good (i.e., 70%). As suggested by Omar and Lascu
(1993), 23 items were grouped into two major groups: importance items and performance
items. As for the importance constructs, findings indicate that System supports for decision-
making in addition to Staff and services constructs rated the highest score (i.e., 6.4), whereas
the User involvement construct yielded the lowest score (i.e., 5.6). In the Performance
constructs, Staff and services rated the highest with a score of 6.1, while System supports for
decision-making and Planning constructs rated the lowest (i.e., 5.0 and 4.9 respectively).
Derived from Omar and Lascu`s (1993) recommendations, the five Performance
constructs (Omar & Lascu, 1993, p. 8; Table 3.2) were multiplied by the Importance
constructs to yield weighted perIormance constructs`. The weighted performance constructs
were then correlated to the Overall satisfaction variable (See item 24 in the User Satisfaction
survey, Appendix 4). The Spearman's correlation test shows that Information quality and
System supports for decision-making constructs attained the highest level of correlation (0.62
and 0.59 respectively) with the Overall satisfaction construct. The Spearman's correlation test
shows also that this correlation was significant. The Planning construct was rated 0.37 with
trending towards significance. User involvement attained the lowest level of correlation (i.e.,
0.28); however, this score was not significant. Interestingly, although the Staff and services
construct yielded the highest weighted mean (i.e., 39.7), the Spearman's correlation test
shows it was less correlated (i.e., 0.37) to the Satisfaction construct. However, this was found
to be non-significant.
In summary, the quantitative and qualitative findings of the User Satisfaction survey
confirm that overall there was high level of satisfaction with the HDSS (Mean=5.8, SD=1.0,
N=12) by its users. Findings indicate that items associated with system supports for decision-
making and the information quality constructs were highly important to participants.
However, these constructs were only rated moderately by HDSS users. This was also
supported by the correlation findings which point out that system supports for decision-
making and information quality planning were perceived as important elements for the
overall satisfaction of HDSS.
14

OBSERVATIONAL FINDINGS
The actual LBHC decisions were aligned with two phases (i.e., Pre PAR Intervention
Phase and Post PAR Intervention Phase). Once decisions were evaluated, it was possible to
examine whether there was any distinction between the two phases. The Pre PAR
Intervention Phase (four meetings) included seven decisions. Five of these decisions were
characterised by limited use of evidence, six decisions were characterised by limited level of
participation, and five decisions were characterised by a low level of consensus. Thus, only a
few decisions were characterised by moderate or high levels in any of the key dimensions. In
the Post PAR Intervention Phase (four meetings), 14 decisions were observed. Table 1
summarises the Pre and Post PAR Intervention decisions by key dimensions. The findings
indicate that ten of these decisions were characterised by moderate use of evidence, ten
decisions were characterised by high level of participation and 11 decisions were
characterised by high level of consensus. Furthermore, only three decisions were
characterised by limited level of evidence.
Furthermore, findings show that more decisions were characterised by either a
moderate or a high level in any of the key dimensions in the Post PAR Intervention Phase.
This implies that the decision-making process of the board changed over time towards greater
use of evidence, participation and consensus. It was observed that the LBHC board has been
through a cultural change. For instance, less negative comments were observed in the Post
PAR Intervention Phase about the board`s practice and the Iact that decisions were made out
of meetings. To support this, more positive comments were observed in the LBHC board
meetings about the level and thoroughness of discussions. For instance, one of the
participants noted: 'There was a cultural shift in the LBHC while another participant
stated: 'I think now, there is a greater level of confidence in the boardThus, the evidence
suggests a shift in the way discussions and decisions were made throughout the study.
Table 1. Pre and Post PAR Intervention Phases summary of decisions by key dimensions

Pre PAR Intervention
Phase
Use of evidence Level of participation Level of consensus
Limited level 5/7 6/7 5/7
Moderate level 1/7 ----- 1/7
High level 1/7 1/7 1/7
Post PAR Intervention
Phase
Use of evidence Level of participation Level of consensus
Limited level 3/14 ----- -----
Moderate level 10/14 4/14 3/14
High level 1/14 10/14 11/14
15

DISCUSSION AND CONCLUSIONS
This study suggested a collaborative-based planning method (i.e., PAR Intervention) to
design the HDSS. Data were collected with a PAR approach that informed the development
and conceptualisation of the HDSS. The PAR approach consisted of three cycles that were
executed:
PAR cycle 1: Introduction Stage;
PAR cycle 2: Interaction Stage; and
PAR cycle 3: Trialling Stage.
In PAR cycle 1 the primary purpose was to raise awareness of the GIS concept for
decision-making, and that was implemented by a series of GIS introductory presentations
with the LBHC board members. In PAR cycle 2 we scoped the HDSS and its technical
requirements in a collaborative manner. While in PAR cycle 3 the system was deployed and
trialled for three months by LBHC board members. Findings indicate that although the
system was designed in a collaborative manner and in accordance with the LBHC board
needs, substantial development and expansion was still required. This was particularly
pertinent in terms of information items, which were likely to improve HDSS application in
LBHC board`s day-to-day role. Furthermore, findings suggest that more analytical tools are
required to improve the use of evidence in decision-making and make the HDSS more
applicable.
As for the decision-making impact, the PAR Intervention was embedded within a
longitudinal Pre and Post research design aimed at determining the impact of the PAR
intervention on decision-making processes within the LBHC. Two waves of data collection
were used - one prior to the beginning of the PAR intervention and one following completion
of the PAR intervention. Findings show that more decisions were characterised by either a
moderate or high level of participation, consensus, and use of evidence in the Post PAR
Intervention Phase. This implies that the decision-making process of the board and LBHC
changed and improved over time. Further, evidence suggests that knowledge was created by
the PAR Intervention rather than just as a result of the HDSS technical design and
development process. For example, findings show that the process helped to create the notion
oI collaboration` in the planning process. This, in turn, positively contributed to the overall
impact of the HDSS, as LBHC participants sensed they were contributing in the planning
process and played an important role in developing the system. In addition, evidence suggests
that the board had gone through a cultural shift throughout the study. Therefore, it is
concluded that HDSS can produce the type of information and effectiveness that facilitates
collaborative planning. Thus, it improved the way decisions were made in terms of: use of
evidence, consensus, and participation. However, some questions were raised about testing
the HDSS framework in the longer term, and clarifying whether it could achieve a positive
impact, not only at the decision-making processes level, but also in the long term Health
Outcomes level in the community (see the framework suggested by Gudes et al. 2010, p. 26).
These questions remain unanswered and form the basis of future study.

16

REFERENCES
Ashton, J. R. 2009. From Healthy Towns 1843. Public Health, 123, 11-13.

Bharati, P., & Chaudhury, A. 2004. An empirical investigation of decision-making
satisfaction in web-based decision support systems. Decision Support Systems, 37(2),
187-197.

Cromley, E. K., & Mclafferty, S. L. 2003. GIS and public health. Health & Place, 9, 279.

Croner, C. M. 2003. Public Health, GIS and the Internet. Public Health, 24, 57-80.

Dooris, M. 1999. Healthy Cities and Local Agenda 21: the UK experiencechallenges for
the new millennium. Health Promotion International, 14(4), 365-375.

Duhl, L. J., & Sanchez, A. K. 1999. Healthy cities and the city planning process. A
Background Documentation On Links Between Health And Urban Planning,
Retrieved 10 October, 2008, from
http://www.euro.who.int/__data/assets/pdf_file/0009/101610/E67843.pdf

Flynn, B. C. 1996. Healthy cities: toward worldwide health promotion. Public Health, 17,
299-309.

Gudes, O., Kendall, E., Yigitcanlar, T., Pathak, V., & Baum, S. 2010. Rethinking health
planning. Health Information Management Journal, 39(2), 18-29.

Higgs, G., & Gould, M. 2001. Is there a role for GIS in the 'New NHS'? Health & Place, 7,
247-259.

Igbaria, M., & Guimaraes, T. (1994). Empirically Testing the Outcomes of User Involvement
in DSS Development. Omega, 22(2), 157-172.

Kazada, M. J., Beel, E. R., Villegas, D., Matinez, J. G., Patel, N., & Migala, W. 2009.
Methodological complexities and the use of GIS in conducting a community needs
assessment of a large U.S. municipality. Community health. 34 (3), 210-215.

Kendall, E., Muenchberger, H., Baum, S., & Yigitcanlar, T. 2007. Coalitions for community
health [LP0882066]. ARC Linkage Round 2: Australian Government.

Kingston, R., Carver, S., Evans, A., & Turton, I. 2001. Public participation, GIS and
cyberdemocracy: evaluating on-line spatial decision support systems. Environment
and Planning B: Planning and Design, 28(6), 907-921

Maeng, D. M., & Nedovic-Budic, Z. 2010. Relationship between ICT and urban form in
knowledge-based development: empirical analysis of Washington, DC metro region.
International Journal of Knowledge-Based Development, 1(1), 97-117.

Minkler, M. 2000. Using participatory action research to build healthy communities. Public
Health Reports, 115(2-3), 191.

17

Minkler, M. E., & Wallerstein, N. E. 2003. Community based participatory research for
health. San Francisco, CA, USA: Jossey-Bass.

Murray, D. J. 2006. A Critical Analysis of Communicative Planning Theory as a Theoretical
Underpinning for Integrated Resource and Environmental Management. Unpublished
PhD, Griffith University, Brisbane.
National Health and Hospitals Reform Commission. 2008. A healthier future for all
Australians- Interim report December 2008. National Health and Hospitals Reform
Commission.

Nobre, F., Braga, A., Pinheiro, R., & dos Santos Lopes, J. 1997. GISEpi: a simple
geographical information system to support public health surveillance and
epidemiological investigations. Computer Methods and Programs in Biomedicine,
53(1), 33-45.

Northridge, M. E., Sclar, E. D., & Biswas, P. 2003. Sorting Out the Connections Between the
Built Environment and Health: A Conceptual Framework for Navigating Pathways
and Planning Healthy Cities. Journal of Urban Health: Bulletin of the New York
Academy of Medicine, 80, 556-568.

Omar, M. H., & Lascu, D. N. 1993. Development of a user information satisfaction scale: an
alternative measure with wide applicability. Journal of Information Technology
Management, 4(2).

Otgaar, A., Klijs, J. & van den Berg, L. 2011. Towards Healthy Cities: Comparing
Conditions for Change (EURICUR Series (European Institute for Comparative Urban
Research)). Surrey, United Kingdom: Ashgate Publishing Limited.

Philips, R. L., Kinman, E. L., Schnitzer, P. G., Lindbloom, E. J., & Ewigman, B. 2000. Using
Geographic Information Systems to Understand Health Care Access. Archives of
Family Medicine, 9, 971-978.

Richards, T., Croner, C., Rushton, G., Brown, C., & Fowler, L. 1999. Geographic
Information Systems and Public Health: Mapping the Future. Public Health Reports,
114(4), 359-360.

Schulz, A., & Northridge, M. E. 2004. Social Determinants of Health: Implications for
Environmental Health Promotion. Health Education & Behavior, 31(4), 455-471.

Scotch, M., & Parmanto, B. 2006. Development of SOVAT: A numerical-spatial decision
support systems for community health assessment research. Medical Informatics, 75,
771-784.

Su, Y., Slottow, J., & Mozes, A. 2000. Distributing proprietary geographic data on the World
Wide WebUCLA GIS database and map server. Computers and Geosciences,
26(7), 741-749.

Waring, S., Zakos-Feliberti, A., Wood, R., Stone, M., Padgett, P., & Arafat, R. 2005. The
Utility of geograophic information systems (GIS) in rapid epidemiological
18

assessments following weather-related disasters: Methodological issues based on the
Tropical Storm Allison experience. International Journal of Hygiene and
Environmental Health, 208, 109-116.

World Health Organisation (Ed.). 1999. "Health: Creating Healthy Cities in the 21st
Century" in The Earthscan reader in sustainable cities. London: Earthscan
Publications.

Yigitcanlar, T., & Gudes, O. (Eds.). 2008. Encyclopedia of Decision Making and Decision
Support Technologies. New York: IGI Global.

ACKNOWLEDGMENT
This study is part of a broader ARC project entitled Coalitions for Community Health:
A Community based Response to Chronic Disease. The authors would like also to
acknowledge two other investigators involved in this research, Scott Baum and Heidi
Muenchberger, for their contribution.
19

Appendices
Appendix 1. Summary of information items survey

Please rate your level of requirement for each of the
following information items. For example, tick the
cell that best represents how important you think
each type of information is for inclusion in the
HDSS prototype. Please add any comments you
think may be relevant to our decisions about
information
This group of
information
items is
essential now

N (%)
This group
of
information
items could
be included
in phase 2 of
the HDSS

N (%)
This group
of
informatio
n items is
not
necessary
at all

N (%)
Demographic (Population, Projected population
(2007-2027), Mortality rate, Indigenous, Multicultural
(Clustered Nationalities), Nationalities and Population
density)
10 (100%)
___ ___
Socio Economic (SEIFA Index, Employment and
Unemployment rate, Income average and financial
resources, Internet access, Education, Businesses by
Industry Division, and Public Housing
9 (90%) 1 (10%)
___
Sustainable Built and Natural Environments
(Environmental hazards, Biodiversity and
Contaminated land)
2 (20%) 8 (80%)
___
Terrain (Aerial images, Topography and Contour) 1 (10%) 7 (70%) 2 (20%)
Public transportation (Bus stations, Bus routes,
Railway Stations and Railway routes)
10 (100%)
___ ___
Recreation (Parks, City swimming pools, Sporting
facilities and Cycling paths)
10 (100%)
___ ___
Emergency (Police, Fire station and Ambulance
station)
4 (40%) 6 (60%)
___
Shops (Shopping centres, Fast food outlets) 8 (80%) 2 (20%) ___
Roads (Major roads and Streets) 9 (90%) 1 (10%) ___
Health facilities (Pharmacies, Aged care, Breast
Screen, Child Health, Medical Services, Mental health,
Oral health, Public hospitals, Private hospitals, GP`s
and Medicare)
10 (100%)
___
Education Facilities (Child community Services,
Higher education, Libraries, Schools, Special
education, State Pre School, Youth clubs, Play groups
and Universities / TAFE)
9 (90%) 1 (10%)
___
Community facilities (Non profit organisations,
Community centres, Community facilities, Community
Welfare, Employment services, Religious institutions,
Services clubs, Social clubs Sporting clubs, Youth
clubs, Schools, State, Non-state schools and Centre
link offices)
9 (90%) 1 (10%)
___
20

Appendix 2. Features and functionalities selected by LBHC board members for
the HDSS prototype

Feature / Function Purpose
User Login Screen for user to log into system
Map Navigation Basic Map Navigation, including zooming and panning
Base Map/ Imagery
View
Ability to select aerial imagery or street maps as a base view
Layers Ability to view health and demographic layers of the LBHC
Layer list Ability to turn layers on or off
Identify attributes Ability to view details of attributes found at a certain location
Online Help Accessibility to text on help notes for using the system
Print Map Ability to print a map
Map Legend Ability to view an image indicating symbology used in the map
Layer Metadata Ability to view metadata (i.e., data on data) for each of the layers used
in the system
Spatial Bookmarks Ability to store the extent of a view for quick zoom in
Simple Search Ability to undertake a simple geographical search of a name field on
two spatial layers: SLAs (Statistical Local Areas) and community
health centres
Redlining and Measurements Ability to draw points, lines, polygons and text on the map
User Feedback Ability for users to submit feedback regarding data set issues, updates
or any other requirements of the system.
Accessibility analysis Ability to compute the service area of two layers (public hospitals and
GPs) based on driving or pedestrian travel time
Proximity function Ability to find features in specified layers (public hospitals and GPs)
within a specified buffer distance of a point entered by the user
*** Health Behaviours (Obesity [BMI]) 10 (100%) ___ ___
*** Hospital admissions (summary by year of the
total number of separations by SLA for the following
admitted diseases: Depression, Cardiovascular,
Diabetes
and Asthma)
10 (100%)
___ ___
Health data (Avoidable mortality, Chronic disease,
Composite indicators chronic diseases, Health Risk
Factors, Premature mortality by selected cause, Private
health insurance and Self assessed health)
8 (80%)
___
2 (20%)
21

Appendix 3. Proposed workflow for accessibility function

Workflow
Name
Accessibility Function
Description The literature emphasises that accessibility to health facilities has been
identified as a key determinant of health
Objective To test the effect of travel time to health facilities
Suggested
End Users
LBHC members, Logan and Scenic Rim planners
Anticipated
Outcome
To Identify gaps in the provision of health facilities in the community
Suggested
Workflow
1. User logs into HDSS Prototype.
2. A map view is presented showing SLA boundary suburb
names.
3. The user zooms in to a specific area.
4. The user selects a button on the interface to calculate
service area catchments for a facility layer.
5. A form appears in which the user has the option to:
6. Pick a facility layer which may be one of three types:
Public Hospitals (default)
GP Clinics
Chronic Disease Centres

7. Pick a transport mode:
Pedestrian
Private Car (default)

8. Enter in travel time, (5,10, or 20 minutes)
9. Click on a button to show the service area. The system
processes the request and updates the map to show travel
time from the selected facility in the map view as
polygons.
10. The user can visualise gaps between polygons which
highlight areas not serviced.
11. The user sends the map to the printer.
Optional
Workflow
The user turns on a layer of population statistics to compare demographic
data to the accessibility to facilities.
Suggested
GIS Data
Street map/aerial imagery
SLA
Suburbs
Public hospitals
GP Clinics
Chronic diseases centres
Population statistics (optional)

2
2

A
p
p
e
n
d
i
x

4
.

U
s
e
r

s
a
t
i
s
f
a
c
t
i
o
n

s
u
r
v
e
y

P
l
e
a
s
e

t
i
c
k

t
h
e

r
a
t
i
n
g

y
o
u

f
e
e
l

m
o
s
t

r
e
p
r
e
s
e
n
t
s

y
o
u
r

e
v
a
l
u
a
t
i
o
n

o
f

t
h
e

H
D
S
S

f
e
a
t
u
r
e

b
o
t
h

p
e
r
f
o
r
m
a
n
c
e

a
n
d

i
m
p
o
r
t
a
n
c
e

r
e
s
p
o
n
s
e
s

n
e
e
d

t
o

b
e

g
i
v
e
n

f
o
r

e
a
c
h

i
t
e
m
.

I
m
p
o
r
t
a
n
c
e

P
l
e
a
s
e

p
r
o
v
i
d
e

y
o
u
r

r
a
t
i
n
g

o
f

t
h
e

i
m
p
o
r
t
a
n
c
e

y
o
u

a
t
t
a
c
h

t
o

e
a
c
h

f
e
a
t
u
r
e
,

o
n

a

s
c
a
l
e

o
f

1
-
7

w
h
e
r
e

1

i
s

l
o
w

i
m
p
o
r
t
a
n
c
e

a
n
d

7

i
s

h
i
g
h

i
m
p
o
r
t
a
n
c
e

P
e
r
f
o
r
m
a
n
c
e

P
l
e
a
s
e

p
r
o
v
i
d
e

y
o
u
r

r
a
t
i
n
g

o
f

t
h
e

p
e
r
f
o
r
m
a
n
c
e

o
f

t
h
e

H
D
S
S

o
n

e
a
c
h

f
e
a
t
u
r
e
,

o
n

a

s
c
a
l
e

o
f

1
-
7

w
h
e
r
e

1

i
s

p
o
o
r

p
e
r
f
o
r
m
a
n
c
e

a
n
d

7

i
s

e
x
c
e
l
l
e
n
t

p
e
r
f
o
r
m
a
n
c
e
.

L o w

M e d i u m

H i g h

P o o r

M e d i u m

E x c e l l e n t
1
.

A
v
a
i
l
a
b
i
l
i
t
y

a
n
d

t
i
m
e
l
i
n
e
s
s

o
f

i
n
f
o
r
m
a
t
i
o
n

p
r
o
v
i
d
e
d

b
y

t
h
e

H
D
S
S

1

2

3

4

5

6

7

1

2

3

4

5

6

7

2
.

A
b
i
l
i
t
y

t
o

a
c
c
e
s
s

t
h
e

s
y
s
t
e
m

w
i
t
h
o
u
t

s
u
p
p
o
r
t

f
r
o
m

t
h
e

s
y
s
t
e
m

a
d
m
i
n
i
s
t
r
a
t
o
r

1

2

3

4

5

6

7

1

2

3

4

5

6

7

3
.

A
c
c
u
r
a
c
y

a
n
d

c
o
m
p
l
e
t
e
n
e
s
s

o
f

t
h
e

i
n
f
o
r
m
a
t
i
o
n

p
r
o
v
i
d
e
d

b
y

t
h
e

s
y
s
t
e
m

1

2

3

4

5

6

7

1

2

3

4

5

6

7

4
.

F
l
e
x
i
b
i
l
i
t
y

o
f

t
h
e

d
a
t
a

a
n
d

i
t
s

a
p
p
l
i
c
a
b
i
l
i
t
y

t
o

a

r
a
n
g
e

o
f

s
c
e
n
a
r
i
o
s

1

2

3

4

5

6

7

1

2

3

4

5

6

7

5
.

U
s
e
r

c
o
n
f
i
d
e
n
c
e

i
n

t
h
e

s
y
s
t
e
m

1

2

3

4

5

6

7

1

2

3

4

5

6

7

6
.

E
a
s
e

o
f

a
c
c
e
s
s

f
o
r

u
s
e
r
s

t
o

t
h
e

H
D
S
S

1

2

3

4

5

6

7

1

2

3

4

5

6

7

7
.

C
u
r
r
e
n
t

a
n
d

u
p
-
t
o
-
d
a
t
e

i
n
f
o
r
m
a
t
i
o
n

p
r
o
v
i
d
e
d

b
y

t
h
e

s
y
s
t
e
m

1

2

3

4

5

6

7

1

2

3

4

5

6

7

8
.

E
f
f
i
c
i
e
n
c
y

o
f

t
h
e

s
y
s
t
e
m

i
n

s
e
t
t
i
n
g

u
p
,

u
p
d
a
t
e

a
n
d

m
a
i
n
t
e
n
a
n
c
e

1

2

3

4

5

6

7

1

2

3

4

5

6

7

9
.

R
e
l
e
v
a
n
c
e

o
f

t
h
e

s
y
s
t
e
m

o
u
t
p
u
t
s

t
o

L
B
H
C

1

2

3

4

5

6

7

1

2

3

4

5

6

7

1
0
.

S
y
s
t
e
m

p
r
i
o
r
i
t
i
e
s

t
h
a
t

r
e
f
l
e
c
t

t
h
e

o
v
e
r
a
l
l

L
B
H
C

o
b
j
e
c
t
i
v
e
s

1

2

3

4

5

6

7

1

2

3

4

5

6

7

1
1
.

D
e
f
i
n
i
n
g

a
n
d

m
o
n
i
t
o
r
i
n
g

i
n
f
o
r
m
a
t
i
o
n

s
y
s
t
e
m
s

p
o
l
i
c
i
e
s

f
o
r

t
h
e

H
D
S
S

1

2

3

4

5

6

7

1

2

3

4

5

6

7

1
2
.

L
e
v
e
l

o
f

L
B
H
C

i
n
v
o
l
v
e
m
e
n
t

i
n

d
e
f
i
n
i
n
g

a
n
d

m
o
n
i
t
o
r
i
n
g

t
h
e

s
y
s
t
e
m

1

2

3

4

5

6

7

1

2

3

4

5

6

7

1
3
.

E
x
i
s
t
e
n
c
e

o
f

a

p
l
a
n
n
i
n
g

a
g
e
n
d
a

t
o

d
e
v
e
l
o
p

t
h
e

s
y
s
t
e
m

1

2

3

4

5

6

7

1

2

3

4

5

6

7

1
4
.

I
m
p
r
o
v
e
m
e
n
t
s

t
o

t
h
e

s
y
s
t
e
m

1

2

3

4

5

6

7

1

2

3

4

5

6

7

1
5
.

S
y
s
t
e
m

r
e
s
p
o
n
s
i
v
e
n
e
s
s

t
o

c
h
a
n
g
i
n
g

u
s
e
r

n
e
e
d
s

1

2

3

4

5

6

7

1

2

3

4

5

6

7

1
6
.

Q
u
a
l
i
t
y

a
n
d

c
o
m
p
e
t
e
n
c
e

o
f

t
h
e

s
y
s
t
e
m

1

2

3

4

5

6

7

1

2

3

4

5

6

7

1
7
.

T
e
c
h
n
i
c
a
l

c
o
m
p
e
t
e
n
c
e

l
e
v
e
l

o
f

t
h
e

s
y
s
t
e
m

a
d
m
i
n
i
s
t
r
a
t
o
r

1

2

3

4

5

6

7

1

2

3

4

5

6

7

1
8
.

C
o
m
m
u
n
i
c
a
t
i
o
n

b
e
t
w
e
e
n

u
s
e
r
s

a
n
d

t
h
e

s
y
s
t
e
m

a
d
m
i
n
i
s
t
r
a
t
o
r

1

2

3

4

5

6

7

1

2

3

4

5

6

7

1
9
.

D
a
t
a

a
n
a
l
y
s
i
s

c
a
p
a
b
i
l
i
t
i
e
s

o
f

t
h
e

s
y
s
t
e
m

t
o

s
u
p
p
o
r
t

t
h
e

d
e
c
i
s
i
o
n
-
m
a
k
i
n
g

p
r
o
c
e
s
s

1

2

3

4

5

6

7

1

2

3

4

5

6

7

2
0
.

A
v
a
i
l
a
b
i
l
i
t
y

o
f

t
o
o
l
s

i
n

t
h
e

s
y
s
t
e
m

t
o

a
n
a
l
y
s
e

i
s
s
u
e
s

r
e
l
a
t
e
d

t
o

t
h
e

L
o
g
a
n
-
B
e
a
u
d
e
s
e
r
t

a
r
e
a

1

2

3

4

5

6

7

1

2

3

4

5

6

7

2
3

2
1
.

U
s
e
r
`
s

I
e
e
l
i
n
g

o
I

p
a
r
t
i
c
i
p
a
t
i
o
n

i
n

t
h
e

H
D
S
S

1

2

3

4

5

6

7

1

2

3

4

5

6

7

2
2
.

U
s
e
r

i
n
f
l
u
e
n
c
e

o
n

t
h
e

d
e
v
e
l
o
p
m
e
n
t

o
f

t
h
e

s
y
s
t
e
m

1

2

3

4

5

6

7

1

2

3

4

5

6

7

2
3
.

H
e
l
p
f
u
l
n
e
s
s

o
f

t
h
e

s
y
s
t
e
m

a
d
m
i
n
i
s
t
r
a
t
o
r

1

2

3

4

5

6

7

1

2

3

4

5

6

7

P
l
e
a
s
e

t
i
c
k

t
h
e

r
a
t
i
n
g

y
o
u

f
e
e
l

m
o
s
t

r
e
p
r
e
s
e
n
t

y
o
u
r

e
v
a
l
u
a
t
i
o
n

o
f

t
h
e

f
o
l
l
o
w
i
n
g

q
u
e
s
t
i
o
n

S
a
t
i
s
f
a
c
t
i
o
n

P o o r

M e d i u m

H i g h
2
4
.

O
v
e
r
a
l
l
,

h
o
w

w
o
u
l
d

y
o
u

r
a
t
e

y
o
u
r

s
a
t
i
s
f
a
c
t
i
o
n

w
i
t
h

t
h
e

H
D
S
S

s
y
s
t
e
m
?

1

2

3

4

5

6

7

Evaluating Geospatial Visualization Methods for West Nile Virus Risk Mapping
Aashis Lamsal

and Michael C. Wimberly
Geographic Information Science Center of Excellence, South Dakota State University, Brookings, SD

Good quality epidemiological maps assist in effective decision making to mitigate
disease risks and to minimize associated losses. Although there has been considerable research
addressing the design and usability of maps for public health applications, there has been less
work on methods for visualizing complex spatiotemporal phenomena.

To address these issues, we developed static, animated, and interactive visualization
methods to evaluate the usability of alternative visualization methods for Web-based
epidemiological mapping for discerning spatiotemporal patterns of West Nile virus (WNV)
human cases and to discover how well end users identify disease patterns by using the preferred
visualization method for WNV risk mapping across the conterminous United States. The
usability evaluation was comprised of the usability test and the usability survey. On the usability
test, upon first visiting the site, the participants were randomly assigned to one of the
visualization methods for the usability test among static, animated, and interactive visualizations,
and each end user was asked to respond the quiz questionnaire. After completing the usability
test, study participants were next allowed to view all three different visualization methods on the
same shared location for usability survey, and each user asked to fill out a survey questionnaire
as a subjective evaluation of their preferences for the geospatial visualization methods.
During usability evaluation, different WNV maps were presented to different sets of subjectively
chosen test subjects representing public health workers, academicians, and students in order to
assess their abilities to discern spatiotemporal patterns, determine their opinions about the
effectiveness of different geo-visualization methods, and deduce the most effective visualization
method for epidemiological mapping.

The results from this study found that there are very different requirements and needs
from various user groups to be fulfilled by Web-based GIS applications used in spatial
epidemiology, and there were mixed opinions regarding the map preferences and suitability
among the end users with their experience and expertise. None of the visualization methods
emerged as clearly the most suitable, although there was evidence that animated maps are less
effective than other maps types for visualization purely spatial patterns. Moreover, no matters of
the visualization methods, the end users are prone to give correct responses to questions about
static patterns than questions about spread direction and rate. Epidemiological risk mapping
application has to focus not only on describing patterns of historical epidemics, but also on
increasing risk awareness in public health professionals, policy makers, and other end users. The
findings form this study will help public health organizations and policy makers to design more
effective mapping tools to support the development of health plans, improve resource allocation,
enhance the understanding and control of disease outbreaks, and ultimately creating more
disease-resilient communities.

7/15/2011
1
Evaluating Geospatial
Visualization Methods for
West Nile Virus Risk Mapping
AASHIS LAMSAL AND MICHAEL C. WIMBERLY
GEOGRAPHI C I NF ORMATI ON S CI ENCE CENTER OF EXCELLENCE, S OUTH
DAKOTA S TATE UNI VERS I TY, BROOKI NGS , S D 5 7 007 .
C ONT A C T I NF ORMA T I ON: A A S HI S . L A MS A L @S DS T A T E . E DU.
Outline
y Background
y Introduction y Introduction
y Objectives
y Geospatial Visualization Methods
y Methodology
y Results
y Discussion
y Future Work
7/15/2011
2
Background
What is West Nile virus (WNV)?
y WNV is the most widespread arbovirus in y WNV is the most widespread arbovirus in
the world.
y Zoonotic disease, wild birds are primary
host.
y WNV originally emerged in Uganda in
1937
Year 1999
1937.
y 1996-1999 Southern Romania, southern
Russia, & northeastern US.
Year 2003
Background contd.
y Why geospatial visualization is necessary?
y In times of war the accurate mapping of enemy y In times of war, the accurate mapping of enemy
positions can be the key to victory.
y Mapping the distribution and habitat of mosquitoes
can play a crucial role in combating mosquito borne
epidemics at the ground level.
y The mapping of WNV estimates the necessary The mapping of WNV estimates the necessary
allocation of public health resources, detects
hotspots for health surveillance systems, and
analyzes the environmental causes for its spread.
7/15/2011
3
Introduction
y GIS have been widely used for disease exposure
assessment for individuals and other public health
applications.
y Limited ability to deals with multi-temporal datasets
where exposure sources and epidemics change
through time.
y Relatively little consideration to reconstruct disease
b d hi i exposure based upon histories.
Introduction
y Internet and WWW provide new ways of mapping and
visualizing geospatial data- Good consolidation of GIS and
visualization tools to view geospatial data spatiotemporally visualization tools to view geospatial data spatiotemporally.
y However, the design of Web-based visualization may always
not be appropriate without focusing on a particular user
group.
y Effective mapping methods that support user interactive
visualizations for the extraction of well-organized
spatiotemporal epidemic patterns and knowledge discovery spatiotemporal epidemic patterns and knowledge discovery
are highly desired.
Visualization Technique
Environmental and Life Sciences
Geospatial Data Interface
7/15/2011
4
Objectives
y Compare the suitability of alternative visualization
methods for Web-based epidemiological mapping. methods for Web based epidemiological mapping.
y Discover how well end users discern disease patterns
using the proposed visualization technique.
Static Visualization
y The maps are either viewed
onscreen or printed as a hard onscreen or printed as a hard
copy
y These maps allow limited
opportunities for
customization but are
provided in formats such as
PDF,PNG, or JPEG.
7/15/2011
5
Animated Visualization
y Animated visualization uses a
time series of static maps time series of static maps
covering a long time period.
y It is effective for revealing
dynamic geographical
phenomena through its ability
to display interrelations
among location, attribute,
and time.
y Carried out onscreen only.
Interactive Visualization
y Interactive visualization
provides a simple and provides a simple and
intuitive interface for the
visualization and analysis of
geographic data in Web.
y Visualization provides a wider
range of options for
visualization, analysis, and
creation of new map displays
than the others
y Layer change, pan, zoom etc.
7/15/2011
6
Methodology
A Fundamental Architecture of Web GIS
Conceptual Diagram
Database US Shape file
CGI / Web-Server
Users
7/15/2011
7
Usability Evaluation
U
s
Static
Visualization
Interactive
Visualization
Animated
Visualization
START
s
a
b
i
l
i
t
y

T
e
s
t
OR
Questionnaire
Static
Vis ali ation
Interactive
Vis ali ation
Animated
Vis ali ation
END
y
U
s
a
b
i
l
i
t
y

S
u
r
v
e
y
AND
Questionnaire
Visualization Visualization Visualization
Demo
y http://globalmonitoring.sdstate.edu/eastweb/maps/
7/15/2011
8
Results
Quiz Questions
7/15/2011
9
Quiz Results
80
100
Static
Animated
%
O
f

C
0
20
40
60
1 2 3 4 5 6
Interactive
Question
C
o
r
r
e
c
t

A
n
s
w
e
r
s
Question 1 2 3 4 5 6
X-squared statistic 1.2224 0.1517 2.2342 1.1096 3.27 0.4667
p value 0.5427 0.927 0.3272 0.5742 0.195 0.7919
Significance level Degree of Freedom Critical Value
0.05 2 5.99
Survey Questions
S
c
o
r
e
Strongly Agree = 5
Strongly Disagree = 1
7/15/2011
10
Survey Results
4
4.5
5
Static
Animated
1
1.5
2
2.5
3
3.5
4
6
Interactive
S
c
o
r
e
s
1 2 3 4 5 6
Question
Question 1 2 3 4 5 6
Kruskal statistic 0.2022 2.3875 6.274 3.2998 1.0137 4.3871
p value 0.9038 0.3031 0.04341 0.1921 0.6024 0.1115
Significance level Degree of Freedom Critical Value
0.05 2 5.99
Survey Feedbacks
y The application is very useful. As a professional I would prefer interactive
map but for general public (lay people) the static maps would be best since they p g p ( y p p ) p y
might not have basic knowledge of maps and confused by how to choose
provided options.
y The animated map could give a good quick progression of WNV, but the
interactive map would be easier to look at specific dates to determine specific
information. The static map allows the same thing, but with my aging eyes, it is
more difficult to view without clicking to enlarge. This adds more mouse
clicks.
y The static map would have been more useful with larger images. The animated The static map would have been more useful with larger images. The animated
map needs a slider or arrow bars so that one can go backward or forward in the
animation.
7/15/2011
11
Discussion
y There is not much difference in suitability and
pattern identification among three different pattern identification among three different
visualization methods.
y Based upon this result, we can recommend to give
people a choice of different geovisualization products
because different people will have different
preferences for geovisualization products.
Future Works
y A more detailed follow-up study would be required
to determine why different people have different to determine why different people have different
preferences.
y The follow-up study should probably try to target a
larger population and obtain a larger sample size.
y Probably need to ask a larger number of more
detailed and more difficult questions on the test to
q
determine whether these is an effect of map type.
7/15/2011
12
Acknowledgement
y NIH/NIAID, An Integrated System for the
Epidemiological Application of Earth Epidemiological Application of Earth
Observation Technologies (R01-AI079411).
y Dr. Ting-Wu Chuang WNV Datasets
1
Healthcare Demand Estimator
A Per i met er - Bas ed Mappi ng Fr amewor k f or
Vi s ual i zat i on of Di s eas e St at e
Pei c hung Shi h and Dav i d Sc hol l
Si emens Heal t hc ar e, Mal v er n, PA
Background
y Customer is looking to open up a new imaging center
y Factors that need to be taken into consideration: y Factors that need to be taken into consideration:
{ Perimeter - size and shape of the service area
{ Patient population that falls into perimeter
Demographics
Disease incidence/prevalence rates
{ Competition in the area:
H it l d/ di ti i i t
Hospitals and/or diagnostic imaging centers
List will change depending on the imaging modality (e.g., CT,
MRI, PET, etc.), since not all facilities offer every modality.
7/1/2011 7/1/2011
2
Perimeter Standard Web Form
y Geographic Criteria
{ Site address { Site address
{ Radius
{ ZIP-code List
{ County List
y Limitations
{ No address available
{ How big is the 15-mile radius?
{ Highway directions
{ Natural barriers, e.g., mountains, lakes, etc.
{ Keep up to date with ZIP-code changes
Perimeter Spatial Form
y Implementation
{ Microsoft Bing Map { Microsoft Bing Map
{ JavaScript library
{ Map tile caching
y Geographic Criteria
{ Place a pushpin on map
{ Radius
{ User defined polygons
{ Drive-time zone
URISA 2011 GIS in Public Health Conference - Atlanta, Georgia - www.urisa.org 7/1/2011 7/1/2011
3
Tile Caching & Static Maps
System Architecture
4
New Cancer Cases by Perimeter
Goals of Site Selection
1. Reach highest number of prospective patients
{ Of the possible sites which has a service area that has the { Of the possible sites, which has a service area that has the
highest number of prospective patients?
2. Find underserved areas/Avoid redundancy
{ Are there currently alternative facilities in the area?
{ Is your new site more conveniently located compared to
current facilities?
{ Can current facilities already handle patient population? { Can current facilities already handle patient population?
5
Goals of Site Selection Data Required
1. Reach highest number of prospective patients
{ Estimation of patient population (e g new cancer cases) { Estimation of patient population (e.g., new cancer cases)
{ Knowledge of where these populations are located
2. Find underserved areas/Avoid redundancy
{ Identify potential current facilities in area:
Location relative to proposed sites
Identify their capabilities to serve patients (e.g., identify all sites
that have a PET scanner for imaging tumors) that have a PET scanner for imaging tumors)
{ Estimation of anticipated market share based on site proximity
relative to current facilities. i.e., the percent of patients are
likely to come to you versus current facilities.
Simplistic Equal Share Solution
y Required data:
{ Total population for each { Total population for each
sites service area.
{ Count of facilities that
could compete against
each site.
y Assumptions:
{ Residents equally likely to { Residents equally likely to
go to any site.
{ Only facilities within
radius compete with site.
6
Share Based on Proximity: Steps
1. Identify all the block groups within perimeter of
each site each site.
2. Calculate market share estimate for each block
group, which requires:
a) Identifying competitors within maximum distance
(distance patients are willing to travel).
b) Calculate distance from block group to each identified
i d i f i
competitor and to site of interest.
3. Multiply market share against population of
interest and sum up for all block groups within
each site radius.
Share Based on Proximity: Notes
y Required data:
{ Total population and distance { Total population and distance
to each BG within service area.
{ Count of competitors* per BG.
{ Distance of each competitor to
BG.
*Competitors are located within x miles from BG, where x is size of site radius.
y Notes:
1. Calculation is done per site per
BG.
2. Multiply share time BG pop
and sum for likely population.
1. Blue square represents a block group
2. Blue circle shows max distance patient
from BG is willing to travel.
3. Green area shows area of all potential
competitors
7
Patient Population Totals
Share Based on Proximity: Illustration
Estimated Patient Share Potential Patients per Site
X
=
Share Based on Proximity: Assumptions
Assumptions:
{ Residents willing to travel { Residents willing to travel
X distance to your site
would travel X distance to
competitor too.
{ The closer a facility is the
more likely a patient is to
go there.
Blue square represents a block group
Blue circle is the catchment area for the block group
Green area shows area of all potential competitors
8
Share Based on Proximity: Example Output
Conclusion
1. Visual feedback of the area analyzed is key to
understanding understanding.
2. A simple GUI is worth its weight in gold.
3. Need to take current services available to patient
population.
4. Location mattersIdeally should be close to most
patients with least amount of current facilities
patients with least amount of current facilities
serving them.
5. Data analysis should be done at lowest geographic
level feasibly possible.
7/14/2011
1
E ti ti f S ll A P l f
Joseph Ahern and Terese Lenahan
The Center for Community Solutions
Estimation of Small-Area Prevalence of
Mental Disorders in Cuyahoga County,
Ohio
John Garrity, Ph.D.
Alcohol, Drug Addiction, and Mental Health Services
Board of Cuyahoga County
Alcohol Drug Addiction and Mental Health
Services Board of Cuyahoga County
ADAMHS Board is responsible for the planning, funding
and monitoring of public mental health and substance
b ti d li d t i di t id t abuse prevention delivered to indigent residents.
The ADAMHS Board has a staff of 44, including
behavioral health programming, quality improvement,
evaluation and research, and grants management.
The ADAMHS Board does not provide direct services,
but contracts with 55+ mental health providers to deliver
services to 50,000 low-income residents per year, using
federal, state, and county funding.
2
7/14/2011
2
ADAMHS Boards
Need for Planning Data
Accurate data are required for needs
assessment and allocation of scarce assessment and allocation of scarce
resources of the public health system.
Cuyahoga County, Ohio faces significant
state and federal funding cuts compared to
th i ti i Ohi tl d other growing counties in Ohio, partly due
to a migration trend to outlying counties.
3
Data Needs (continued)
The core population left behind is
disproportionately poor with more service needs.
Cuyahoga County was compared to six other
urban OH counties in terms of demographics,
risk factors and prevalence of mental disorders.
Small areas analyses estimate true prevalence Small areas analyses estimate true prevalence,
and provide a more accurate estimate of real
funding needs.
4
7/14/2011
3
Project Objectives
To calculate local population in need for mental
health services, especially those with low incomes. ea t se ces, espec a y t ose t o co es
Sources: Population data from 2006-08 American
Community Survey (ACS) and national mental
disorder prevalence rates.
Using utilization and claims data fromADAMHS Using utilization and claims data from ADAMHS
mental health providers, compare estimated local
prevalence to local service data to establish unmet
needs.
5
Sources of National Prevalence Rates
of Mental Disorders
Adults: National Institute of Mental Health
The Numbers Count: Mental Disorders in America (2010)
Website. Sources of data:
National Comorbidity Survey Replication
Epidemiological Catchment Survey
Children: Mental Health: A Report of the Surgeon
General (1999)
6
7/14/2011
4
County-Level Prevalence
Estimates
NIMH prevalence rates for mental disorders:
15.6% of adults 18 and over
20.9% of children 5 to 17
Applied to Cuyahoga County population:
201 800 at all income levels 201,800 at all income levels
64,300 at incomes under 200% of Federal
Poverty Level (FPL) This is target population.
7
Problems with Estimating Prevalence for
Municipalities and Neighborhoods
Mental disorders and po ert are not Mental disorders and poverty are not
distributed uniformly across the county.
For smaller populations, MOEs are larger,
resulting in greater uncertainty than for resulting in greater uncertainty than for
county-level populations.
8
7/14/2011
5
Method for Sub-County Estimation
Apportion share of estimated county-wide prevalence
between Cleveland and suburbs using 2008 ACS between Cleveland and suburbs using 2008 ACS
(population 65,000+).
Apportion share of prevalence for Cleveland
neighborhoods and suburbs using Census 2000
(population < 65,000).
Caveat: Census cognitive difficulty variable different
from clinical diagnosis criteria for mental disorders.
Census data is only used to find distribution of
population in need in small geographies.
9
10
7/14/2011
6
11
A ti i C t P l Apportioning County Prevalence
Between Cleveland and Suburbs
7/14/2011
7
Cleveland and suburban residents with cognitive
difficulty are basically evenly distributed
at all income levels
Cleveland
35,500 (50.2%)
Suburbs
35,200 (49.8%)
All Persons with
Cognitive Difficulty:
70,700
13
Source: 2008 Census ACS
Disproportionate % in
Cleveland because
city is about 1/3 of
county population.
Apply same percents to mental disorders
prevalence at all income levels
Cleveland
101,300 (50.2%)
Suburbs
100,500 (49.8%)
All Persons with
Mental Disorders:
201,800
14
Source: 2006-2008 Census ACS
7/14/2011
8
Now, apply 2008 Census ACS estimate of
33% of persons in county
under 200% of FPL:
Census Cognitive Difficulty Census Cognitive Difficulty
70,700 (all income levels) x 33% = 23,500 <200% FPL
Apply 33% to NIMH prevalence rate estimate:
Mental Disorders
201,800 (all income levels) x 33% = 64,300 <200% FPL
15
Venn diagram shows subset of persons with
cognitive difficulty that are <200% of FPL
AND proportion between city and suburbs.
Suburbs
All Persons
U d 200% f FPL
3,500 (15.1%)
Under 200% of FPL
Cleveland
20,000 (84.9%)
All Persons with
Cognitive
Difficulty in
county
16
Source: 2008 Census ACS
Cleveland has 85% of
Pop. < 200% of FPL
7/14/2011
9
Now doing same city/suburb split applied to NIMH
mental disorders. This is target population for
ADAMHS.
Suburbs
All Persons
U d 200% f FPL
9,700 (15.1%)
Under 200% of FPL
Cleveland
54,600 (84.9%)
All Persons with
Mental Disorders
in county
17
Source: 2006-2008 Census ACS
Cleveland has 85% of
Pop. < 200% of FPL
Apportioning Cleveland Prevalence Apportioning Cleveland Prevalence
Among Individual Neighborhoods
(Statistical Planning Areas or SPAs)
7/14/2011
10
Estimated Prevalence of Mental Disorders (Age 5+)
in Selected Cleveland Neighborhoods
All Income Levels
Neighborhood
(SPA)
Cognitive
Difficulty
(2000 Census Tracts)
Percent of
Cleveland Total
NIMH Estimated
Prevalence 2006-
08
08
Brooklyn Centre 630 2.1% 2,130
Buckeye Shaker 820 2.7% 2,740
Central 970 3.2% 3,240
.. .. .. ..
West Boulevard 1,030 3.4% 3,440
Woodland Hills 550 1.8% 1,820
... .. . .
Cleveland Total 30,300 100.0% 101,300
19
By applying each neighborhoods proportion of < 200% FPL
to previous tables data, enables ADAMHS to apply
resources according to concentration of need.
Neighborhood
(SPA)
Cognitive
Difficulty
Percent of
Cleveland Total
NIMH Estimated
Prevalence 2006-08
Brooklyn Centre 340 2.1% 1,150
Buckeye Shaker 420 2.6% 1,420
Central 820 5.1% 2,780
. .. ..
West Boulevard 430 2.7% 1,470
Woodland Hills 370 2.3% 1,260
.......... . .. ..
Cleveland Total 16,100 100.0% 54,600
20
7/14/2011
11
Geographic Distribution Geographic Distribution
of Persons in Need
in Neighborhoods and Suburbs
22
7/14/2011
12
23
Mental Health Service Consumers Mental Health Service Consumers
and Percent of Need Met
in Cleveland Neighborhoods and Suburbs
7/14/2011
13
25
26
7/14/2011
14
Geographic Distribution Geographic Distribution
of Service Consumers
by Major Diagnosis
28
7/14/2011
15
29
30
7/14/2011
16
Benefits of Geocoding Analyses
for ADAMHS Board
Comparing the geocoded addresses of mental
health service consumers with the estimated health service consumers with the estimated
prevalence yields estimates of persons with
unmet needs.
Overlaying the location of mental health service
delivery sites onto the map of met and unmet delivery sites onto the map of met and unmet
need helps improve delivery of mental health
services.
31
Community Benefit
Mapping resonates with funders who find it much
clearer to work with than traditional data tables.
Helps guide decisions on how to invest resources in
the areas of greatest need, strengthen safety net
services, and support use of evidence-based
practices.
Helps create a system of superior services which is
consumer- focused cost efficient and improves the consumer focused, cost efficient, and improves the
lives of residents.
Local foundations are also interested in the results of
the needs assessment for their community planning,
particularly in times of reduced funding.
7/14/2011
17
33 URISA 2011 GIS in Public Health Conference June 27-30, 2011 - Atlanta, Georgia - www.urisa.org 7/14/2011
1
The Effects of Changing Spatial
Extents When Deriving
Environmental Risk Factors of
Canine Leptospirosis
RAM RAGHAVAN, KAREN BRENNER, JAMES
HI GGI NS, KENNETH HARKI N
COLLEGE OF VETERI NARY MEDI CI NE
COLLEGE OF ARTS & SCI ENCES
KANSAS STATE UNI VERSI TY
Disease risk factor
y Risk factor
{ Event (rainfall flooding) condition (environmental health) { Event (rainfall, flooding), condition (environmental, health)
character (age, sex)
{ Only association not causation
{ Clues to molecular epidemiologists, other researchers
{ Environmental risk factors
Environmental variables
GIS/RS datasets (land-use/land-cover, soil)
GIS/RS datasets (land use/land cover, soil) 7/14/2011
2
Case-control study
y Case-control
{ Cases (positive for a disease condition) { Cases (positive for a disease condition)
{ Controls (negative for a disease condition, all else same)
{ Representativeness, comparability
{ Rare disease studies
Less data
Retrospective data
Environmental variables
y Environmental variables
{ Land-cover/land-use (NLCD GAP datasets) { Land cover/land use (NLCD, GAP datasets)
{ Soil (SSURGO, STATSGO)
{ Wetland (NWI)
{ Hydrologic parameters (NHD, NHD-Plus)
y Variable extraction
{ Single circular buffer
{ Multiple circular buffers
Irregular distances, reflect animal behavior
3
Logistic regression
y Logistic regression
{ Binary (0,1) dependent variable { Binary (0,1) dependent variable
{ Screening
Bivariate logistic regressions
Liberal cut-off (P-value < 0.1)
{ Multicollinearity
Variable Inflation Factor (VIF) > 10
{ Most parsimonious model { Most parsimonious model
Multivariate logistic regression (P-value <0.05 = accept, > 0.1 =
reject)
Stepwise modeling, AIC vale
{ Model evaluation
Odds ratios
y Odds ratios
{ p/1-p { p/1 p
{ logit (p/1-p)
y P-value
4
Landscape characteristics
y Heterogeneous
{ Distance { Distance
{ Direction
y Spatial scale
{ Spatial resolution (grain size)
{ Spatial extent (total study area size)
Focus mainly on spatial resolution, less attention to extent
Changing spatial extents
y Land-cover classification methods
{ Turner et al 1989a 1989b { Turner et al., 1989a, 1989b
{ Hunsaker et al., 1994
y Land-cover pattern metrics
{ Saura and Martinez-Millan, 2009
Patch density
Edge density
5
Research question
y Does changing spatial extents change the types and
significance of derived risk factors? significance of derived risk factors?
Canine leptospirosis
6
Free living leptospira
in contaminated
water, moist soil
Peridomestic mammals
Domestic maintenance hosts
Wild mammals
y KSVDL leptospirosis medical records
{ February 2002 December 2009 { February 2002 December 2009
y Cases (n = 94)
{ Positive urine PCR or
{ Negative urine PCR and one of the following
Isolation of leptospires in the urine culture
Single reciprocal serum titer 12,800
F f ld i i l t tit Four-fold rise in covalescent serum titer
y Controls
{ Negative urine PCR
{ Convalescent serum titer < 400
7
Geocoding
y Address information
{ Street level { Street level
{ Clients at the time of sample submission
{ Verification
Google Earth, MapQuest
Phone call(s)
{ US Census 2000 TIGER Line Street Data
URISA 2011 GIS in Public Health Conference -
Atlanta, Georgia - www.urisa.org
Case/control distribution
8
Materials and methods
y National Land Cover Dataset (NLCD)
y Kansas GAP analysis dataset y Kansas GAP analysis dataset
{ Circular buffers at 500 m interval up to 5000 m
{ Extract land cover polygons
{ Estimate % land cover types within each buffers
{ Logistic regression
Bivariate screening
l i i d li Multivariate modeling
Model strength
Results (NLCD)
Distance (m) NLCD Land cover risk P value OR 95% CI
500 De eloped high intensit 0 027 1 491 1 186 1 876 500 Developed high intensity 0.027 1.491 1.186, 1.876
1000 Developed high intensity 0.029 1.494 1.188, 1.880
2500 Developed medium intensity 0.016 1.866 1.443, 2.412
3500 Developed medium intensity 0 024 1 645 1 331 2 033
4000 Evergreen forest 0.022 1.692 1.366, 2.095
4500 Evergreen forest 0.021 1.693 1.144, 2.506
5000 Evergreen forest 0.020 1.704 1.146, 2.531 7/14/2011
9
Results (KS GAP)
Distance (m) KS GAP Land cover risk P value OR 95% CI
500 Urban areas 0 015 2 036 1 373 3 019 500 Urban areas 0.015 2.036 1.373, 3.019
1000 Urban areas 0.017 2.044 1.378, 3.031
1500 Urban areas 0.012 2.060 1.389, 3.055
2000 Urban areas 0.018 2.056 1.384, 3.055
2500 Urban areas 0.026 2.013 1.355, 2.991
3000 Forest/woodland 0.006 1.873 1.449, 2.422
3500 Forest/woodland 0 006 2 009 1 551 2 603
3500 Forest/woodland 0.006 2.009 1.551, 2.603
4000 Forest/woodland 0.004 2.010 1.555, 2.599
4500 Forest/woodland 0.000 2.013 1.557, 2.603
5000 Forest/woodland 0.001 2.013 1.551, 2.613
Conclusions
y Changing spatial extents changed the types and
significance of derived risk factors significance of derived risk factors
y Selection of spatial extents should be made carefully
{ Host behavior
{ Vector home-ranges
y Potentially similar implications for other research
areas
{ Non-infectious diseases
y Reporting in literature
{ Why certain spatial extent
10
Acknowledgements
y KSVDL
y NSF EPSCoR y NSF EPSCoR
1
SmartMaps and Built Environment
Exposure: Extending Measurement
Paradigms
2011 URISA GIS IN PUBLIC HEALTH
CONFERENCE
JUNE 27-30, 2011
ATLANTA, GA, USA
1
1 of 27
Philip M. Hurvitz*, Anne V. Moudon
phurvitz@uw.edu, moudon@uw.edu
Urban form Lab, Box 354802, College of Built Environments
University of Washington, Seattle, WA 981954802
Overview
y Introduction to SmartMaps
{ Background/Relevance
2
{ Expanding the conceptual model
{ Operationalization
{ Summary
{ Preliminary Q&A
y Proof of concepts studies
{ Hurvitz: Built Environment exposure (GPS traces)
{ Lin: Transportation (Schools)
{ Jiao: Transportation (Shopping behavior)
2 of 27 8/5/2011
2
Background/Relevance
y Definitions:
{ Built Environment = features in the environment that were
3
{ Built Environment features in the environment that were
created by people (streets, buildings, etc.)
{ Exposure = a state of being in the presence of or subjected to a
force or influence (e.g., viral exposure, heat exposure).
Mosby's Medical Dictionary, 8th edition. 2009, Elsevier.
URISA 2011 GIS in Public Health Conference June 27-30, 2011 - Atlanta, Georgia - www.urisa.org 3 of 27
Background/Relevance
y Environmental exposures occur continuously
through space and time
4
g p
y Current technologies allow space-time continuous
measurement of location
{ location in space
{ location in time
y New approaches are needed for integration of space-
time location data and environmental exposure
time location data and environmental exposure
{ Environment behavior (urban planning, transportation)
{ Environment health outcomes (epidemiology, public
health)
4 of 27 8/5/2011
3
Background: Theoretical model
y Social Ecologic Model of Behavior (SEM)
(Bronfenbrenner, Lewin)
5
(Bronfenbrenner, Lewin)
Built/Natural
Individual
Characteristics
Social
/
Enfvironment
Behavior/
Health
outcomes
Environment
Background: Limitations of current
approaches
y Census unit is commonly used as a proxy for home
neighborhood (public health/epidemiology/nutrition
examples):
6
examples):
{ Franco, M., Diez Roux, A. V., Nettleton, J. A., Lazo, M., Brancati, F., Caballero,
B., et al. (2009). Availability of healthy foods and dietary patterns: the Multi-
Ethnic Study of Atherosclerosis Neighborhood characteristics and availability of
healthy foods in Baltimore. Am J Clin Nutr, 89(3), 897-904.
{ Zenk, S. N., & Powell, L. M. (2008). US secondary schools and food outlets.
Health Place, 14(2), 336-46.
{ Moore, L. V., & Diez Roux, A. V. (2006). Associations of neighborhood
characteristics with the location and type of food stores. Am. J. Public Health,
96(2), 325-331.
M l d K Wi S & R A D ( ) Th C l Eff f h L l
{ Morland, K., Wing, S., & Roux, A. D. (2002). The Contextual Effect of the Local
Food Environment on Residents Diets: The Atherosclerosis Risk in
Communities Study. American Journal of Public Health, 92(11), 1761-1768. doi:
10.2105/AJPH.92.11.1761.
6 of 27 8/5/2011
4
Background: Limitations (MAUP)
y Tract level
exposure measures
7
exposure measures
may be bunk
o
n
p
n
Background: Limitations of current approaches
y Recent studies have moved away from aggregate area
measures to individually-based measures
8
measures to individually based measures
y Individually defined neighborhoods may circumvent
ecologic fallacy problems
y Home location used with detailed environmental GIS
databuffers rather than predefined administrative
boundaries
URISA 2011 GIS in Public Health Conference June 27-30, 2011 - Atlanta, Georgia - www.urisa.org 8 of 27 8/5/2011
5
Measuring BE with GIS
y Point-buffer selection & analysis
9
Home/Store location Network characteristics 0.5 mile buffer
(Anne V. Moudon, 2007)
y Conceptualization of environmental exposure is
limited:
10
limited:
{ Only home environment has been considered
y Planning/transportation examples:
R3
#
#
Origin
Destination
Destination
R2
Area
Area
R1
Lee & Moudon (2004) Frank et al. (2005) 8/5/2011
6
Saelens, B. E., & Handy, S. L. (2008). Built environment correlates of walking: a
review. Medicine and Science in Sports and Exercise, 40(7 Suppl), S550-566. doi:
10 1249/MSS 0b013e31817c67a4
11
10.1249/MSS.0b013e31817c67a4.
y 32 studies that correlated walking with environment
(2005-2006)
{ 12 used exclusively subjective data
{ 17 used objective data
7 used administrative unit (range in size: BG to country)
10 used individual household location
{ No individual-level studies examined environmental
properties of non-home locations
11 of 27
y Problem with individually defined home-based
neighborhoods: people roam about freely and are
12
neighborhoods: people roam about freely and are
exposed to multiple neighborhoods (Galster
2008):
O = outcome of interest
i = individual
i = individual
t = time period
P = personal characteristics
N = neighborhood characteristics
Galster, G. C. (2008). Quantifying the Effect of Neighbourhood on Individuals: Challenges, Alternative Approaches, and Promising Directions. Schmollers
Jahrbuch, 128(1), 748. doi: 10.3790/schm.128.1.7. 8/5/2011
7
Overview
13
{ Summary
{ Preliminary Q&A
13 of 27
Expanding the conceptual model
y Over relatively short time scales, we can expect the
individual and social characteristics of the Social
14
individual and social characteristics of the Social
Ecologic Model to become space-time constant
background conditions, while built and natural
environmental effects maintain variability
Individual
Characteristics
Individual
Built/Natural
Environment
Behavior
Social
Environment
Built/Natural
Environment
Behavior
Characteristics
Social
Environment
Built/Natural
Environment
Behavior 8/5/2011
8
Expanding the conceptual model
y A trace through space and time conceptualized as a
series of localized environmental exposures that
15
p
may affect space-time localized behavior
!(
!(
!(!( !(!(!(!(!( !(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!( !(!(!(!(!(!(!( !(!(!(!(!(!(!(!(!( !( !(!( !(!( !(!( !( !(!( !( !( !( !(!( !( !(!( !(!( !( !(!( !(!( !(!(!(!( !( !(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!( !(!(!(!( !(!(!(!(!(!( !(!( !(!(!(!(!( !(!(!(!(!(!( !(!( !(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!( !(!(!(!(!(!(!(!(!(!(!(!(!( !(!( !(!(!( !(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!( !(!( !(!(!(!(!(!(!(!(!(!(!(!(!(!( !( !( !(!(!(!( !(!(!(!( !(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!( !(!(!(!(!(!( !(!(!( !(!( !(!(!(!(!(!(!(!( !(!( !( !(!( !( !(!(!( !(!( !( !( !(!(!( !(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!( !(!(!(!(!(!(!(!(!(!(!(!(!( !(!(!(!(!(!( !(!( !(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!( !(!(!(!(!(!(!( !(!(!(!(!( !(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!( !( !( !( !( !( !( !( !( !(!(!(!(!(!(!(!( !(!( !( !(!( !(!(!(!(!(!(!(!(!( !(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!( !(!(!(!( !(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!( !(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!( !(!( !(!( !( !(!(!(!( !(!( !(!(!( !(!(!(!(!(!(!( !(!(!(!( !(!(!(!(!(!(!(!(!(!(!(!(!( !(!(!(!( !( !(!(!(!( !( !(!(!(!(!(!( !(!(!(!( !( !(!( !(!( !( !(!(!( !(!( !( !(!(!(!(!(!(!(!(!(!( !( !(!(!(!(!(!(!(!(!(!(!(!(!( !(!(!(!(!(!(!(!(!(!( !(!(!(!( !(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!( !( !(!(!(!( !(!( !(!( !(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!( !(!( !(!(!( !( !(!( !(!(!(!(!(!(!(!(!(!(!(!( !( !( !( !(!(!( !(!( !(!(!(!(!(!(!(!(!(!(!( !(!( !(!( !(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!( !(!( !( !(!( !(!(!(!( !(!(!( !(!(!(!( !( !( !(!( !(!( !(!( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!( !(!( !(!(!(!( !( !( !( !( !( !( !( !(!( !(!(!(!(!(!(!(!(!(!(!(!(!(!( !(!( !(!(!( !( !(!( !(!(!( !(!(!(!(!(!(!(!(!(!(!(!(!( !(!(!( !(!(!(
!(!(!(!(
!(!( !(
!(
I
n
d
i
v
i
d
u
a
l

c
h
a
S
o
c
i
a
l

E
n
v
i
!( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !( !(!( !(!(
!(!( !( !(!( !(
!(!(!(
!(!(
!(
( ( !(
!(!(
!(!(!(!( !(!(!( !(!( !(!(!(!(!( !( !( !(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!( !(!(!( !(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(!(
0 0.5 1
km
[ 0 20 40 60 80 100
m
[
a
r
a
c
t
e
r
i
s
t
i
c
s
i
r
o
n
m
e
n
t
Overview
16
{ Summary
{ Preliminary Q&A
16 of 27 8/5/2011
9
Operationalization: problem
y Major challenge: how to measure local environment
for large point data sets?
17
for large point data sets?
y Consider 1000 subjects, 30 s interval, 16 hours per
day, 1 week:
2 p/min 60 min/h 16 h/d 7d 1000 = 13,440,000
points
y Common point-centric customized GIS
y Common point-centric customized GIS
measurement routines are impractical
17 of 27
#
#
Origin
Destination
Destination
R3
R2
Area
Area
R1
Operationalization: what we did
y Analytical data sets are prepared in advance
(SmartMaps)
18
( SmartMaps )
y SmartMap rasters are the output from focal
processesto capture local summaries of
neighborhood BE characteristics continuously
across the study area
y Typical focal processes as well as GIS process models
yp p p
18 of 27 8/5/2011
10
Operationalization: SmartMap construction
2
% park within bandwidth
High : 100

Low: 0
! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !
^ 30 mmesh selection
^_
0 200 400 600 100 m [
local neighborhood
BE measurement
19
5
6
12
parcels
! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !
^_
^_
^_
^_
0 200 400 600 8001,000 100 m
[
_ 30 mmesh selection
! 30 mmesh
commercial
institutional
other
park
residential
transportation
vacant
water
^_
^_
0 200 400 600 100 m [
^_
^_
12
0 200 400 600 100 m
[
19 of 27
0 200 400 600 100 m [
^_
0 200 400 600 100 m [
30 m grid
focal processing and interpolation
continuous values across XY space
Operationalization: detailed example
Value and number
of units are tallied
Parcels around focal
grid point are selected
833 m radius =
10 minute walking distance
20
Areaweighted
apportioning
Focus moves to
next grid point
Parcel values are summed
Sum placed on focal grid point
$1,546,913,200 8/5/2011
11
Measurement & data: Residential density
21
0 10 20 30 40 50
km
du/km
2
High : 8012
Low : 0
Measurement & data: Traffic density
22
0 10 20 30 40 50
km
vehicles/km
2
High : 422700
Low : 0 8/5/2011
12
Measurement & data: Point measures
y Any arbitrary location can be queried for local BE
characteristics
23
characteristics
y Raster values are extracted for each point in the data
set (Arc: SurfaceSpot; PostGIS WKT Raster:
ST_Intersects)
Overview
24
{ Summary
{ Preliminary Q&A
24 of 27 8/5/2011
13
Summary
y A method for generating raster data sets
representing local neighborhood BE characteristics
25
representing local neighborhood BE characteristics
at any/all location in the study area (SmartMaps)
y SmartMaps can be used in multiple projects (rather
than needing new GIS based
measurements/analytics for new point data)
y Wide range of applicability (next 3 presentations)
g pp y ( 3 p )
25 of 27
Preliminary Q&A
y Questions about SmartMaps?
26
14
Overview
27
{ Summary
{ Preliminary Q&A
27 of 27 8/5/2011
1
Built Environment and
Travel Behavior to Grocery
Stores
1
CONFERENCE
JUNE 27-30, 2011
ATLANTA, GA, USA
1 of 29
Junfeng Jiao*, Anne V. Moudon
hkuffj@uw.edu ,moudon@uw.edu
y Research Background
y Research Objectives
Overview
2
y Research Objectives
y Research Questions
y Conceptual Framework
y Research Data
y Research Methods
URISA 2011 GIS in Public Health Conference June 27-30, 2011 - Atlanta, Georgia - www.urisa.org URISA 2011 GIS in Public Health Conference June 27-30, 2011 - Atlanta, Georgia - www.urisa.org
y Research Findings
y Conclusions & Limitation
2 of 29 8/5/2011
2
y The relationship between peoples Socioeconomic Status
(SES), Built Environment (BE) and the Non-Work Travel
Research background
3
( ), ( )
Activity (NWTA) is an important research area in the
transportation planning field
y Shopping travel as a major NWTA accounts for 19.6% of
all trips in the US (NHTS, 2009). Grocery shopping
consumption made up about 15% of all retail sales in 1997
(US C 2000)
(US Census, 2000)
y Little is known about grocery shopping travel behavior
and its relationship to SES and BE.
3 of 29
y Investigate how BE influences peoples travel mode
choice and travel frequency to grocery stores
Research objectives
4
choice and travel frequency to grocery stores
URISA 2011 GIS in Public Health Conference June 27-30, 2011 - Atlanta, Georgia - www.urisa.org URISA 2011 GIS in Public Health Conference June 27-30, 2011 - Atlanta, Georgia - www.urisa.org 4 of 29 8/5/2011
3
y Controlling for SES, how does the BE around
peoples homes and grocery stores affect travel to
Research questions
5
people s homes and grocery stores affect travel to
grocery stores?
{ travel modes (driving vs. non-driving)
{ travel frequency (once a week vs. twice or more per week)
URISA 2011 GIS in Public Health Conference June 27-30, 2011 - Atlanta, Georgia - www.urisa.org URISA 2011 GIS in Public Health Conference June 27-30, 2011 - Atlanta, Georgia - www.urisa.org 5 of 29
Conceptual framework
6
BE around
grocery
store
BE around
home
Distance SES
Mode choice and travel frequency
6 of 29 8/5/2011
4
y Seattle Obesity Study (SOS)
R01 DK076608PI: Adam Drewnowski
Research data
7
R01 DK076608PI: Adam Drewnowski
y 2,001 respondents were sampled from 97 zip codes
in King County, WA
y Food shopping behavior and demographic data were
collected through telephone surveys and finished in
2009
9
y 1,994 homes and 1,910 primary grocery stores were
geocoded
7 of 29
Home (n = 1994)
8
Primary Grocery Store
(n = 1910)
5
SOS population demographics and socioeconomic status
9
variable SOS sample King County
age 54.53 46.90
household income 50k75k 70k
household size 2.31 2.37
gender (male) 38% 50%
# of Children 012yrs old 0.32 0.39
# of Cars 1.75 1.67
education (13 y college education) 80.8% 73.5%
years at the current residence 16.19 12.85
living in a singlefamily house 77% 62%
y Driving vs. non-driving
SOS population travel mode to primary grocery stores
10
travel mode n %
y Non-driving modes
travel mode n %
walking 145 7 3
travel mode n %
driving 1751 87.7
nondriving 243 12.2
walking 145 7.3
biking 2 0.1
transit 52 2.6
other* 44 2.2
*skateboarding, online shopping,
carpooling, provided by friends or
relatives 8/5/2011
6
SOS population travel frequency to primary grocery stores
11
ti k n % times per week n %
01 950 49.9
2+ 946 49.7
total 1896 99.6
missing 7 0.4
U.S. households primary shoppers made 2.44 grocery
p y pp 44 g y
shopping trips per week (Progressive Grocer report 2002)
y Measure the built environment around homes and
primary grocery stores using SmartMaps
Research methods
12
primary grocery stores using SmartMaps
y Measure the network distance from homes to
different stores
y Statistical modeling
7
BE variables
13
Network distance between home and grocery store
14
To the closest
supermarket
To the closest non
chain grocery store
P. Hurvitz
To the primary
grocery store
14 of 29 8/5/2011
8
home to: travel mode N mean median SD p value
Network distance summary
15
home to: travel mode N mean median SD pvalue
primary store
of choice
nondriving 219 1.52 .47 2.58 .001
driving 1672 3.33 2.23 4.55
all 1891 3.12 2.04 4.40
closest
supermarket
nondriving 240 .74 .37 .65 .001
driving 1751 1.29 1.01 .90
all 1991 1.22 1.02 .89
closest non non driving 240 .52 .29 .66 .001
chain
grocery store
driving 1751 1.50 .91 1.43
all 1991 1.38 0.87 1.39
15 of 29
y Bivariate analysis
{ T-test for continuous variables
Statistical methods
16
{ Chi-square for categorical variables
y Base model: A binary logit model based on important SES
variables
y One by one analysis: Use the base model to test the
significance level of distance variables and BE variables
around home and grocery store, respectively
around home and grocery store, respectively
y Final model: Include the significant distance and BE
variables from the one by one analysisto the base model
and run the final analysis
16 of 29 8/5/2011
9
Model structure
17
Model 1
(base model)
Model 2
(+distance)
Model 3
(+home BE)
Model 4
(+store BE)
SES SES SES SES
distance variables distance variables distance variables
home BE variables home BE variables
grocery store BE variables
Nested binary logistic models
Model fit evaluated by AIC, BIC, and log likelihood test
17 of 29
y Strongest predictors of driving:
18
variable OR 95% CI
Conclusion for travel mode choice model
y Strongest predictors of non-driving
number of cars per
household adult
13.3 7.9 22.4
living in a single family house 2.2 1.3 3.7
number of adults in household 2.1 1.4 3.0
variable OR 95% CI
number of quick service restaurants
near homes
0.97 0.94 0.99
street density near homes 0.92 0.86 0.97
having a nonchain grocery store
around primary grocery stores
0.53 0.30 0.91 8/5/2011
10
y Strongest predictors of more shopping trips per week:
19
Conclusion for travel frequency model
variable OR 95% CI
shopping at highcost grocery stores 1.64 1.08 2.51
number 1218 years old children in
the household
1.59 1.29 1.96
having an indoor fitness facility
around store
1.30 1.01 1.65
household income 1.06 1.02 1.10
y Strongest predictors for fewer grocery shopping trips per
week:
20
Conclusion for travel frequency model
variable OR 95% CI
having a golf or tennis court around
home
0.58 0.38 0.87
driving to grocery store 0.60 0.41 0.87
thinking food should be inexpensive 0.85 0.78 0.93
g p 8/5/2011
11
y SOS respondents were different from the KC general
population
Limitations
21
{ older
{ more females
{ more single family house dwellers
y Travel paths were estimated
{ actual travel from home to store was not known
y King County is unlike many other areas
{ income, education, race/ethnicity, segregation, urban form, etc.
y Spatial dependence and autocorrelation
{ individuals living close to each other may have similar behaviors due
to unmeasured variables
21 of 29
y SmartMaps allowed efficient measurement of
multiple BE variables describing home and grocery
Conclusions
22
multiple BE variables describing home and grocery
store locations
y Both environmental and individual characteristics
affected grocery store shopping travel mode and
frequency
12
y Questions and Suggestions.
Questions and Suggestions
23
y Detailed model result tables follow this slide
24
13
Variable Measurement N Zero Value% Minimum Maximum Mean Median S.D.
Coffee a) Counts within the bandwidth (833m) 1910 37.4% 0.00 95.47 5.10 1.00 13.83
Convenience Same as a) 1910 35.1% 0.00 22.00 2.47 1.00 3.66
Ethnic Dinning Same as a) 1910 44.0% 0.00 67.59 4.22 1.00 8.99
Ethnic Grocery Same as a) 1910 60.9% 0.00 20.00 0.92 0.00 1.96
Food Drug Combo Same as a) 1910 73.9% 0.00 11.00 0.55 0.00 1.39
Non-chain Grocery Same as a) 1910 54.8% 0.00 14.00 1.16 0.00 2.25
Quick service Same as a) 1910 28.6% 0.00 198.90 11.12 3.00 28.83
Descriptive analysis of BE around grocery stores
25
Supermarket Grocery Same as a) 1910 64.2% 0.00 5.20 0.50 0.00 0.82
Traditional Restaurant Same as a) 1910 44.2% 0.00 118.37 5.47 1.00 16.61
Grocery, supermarket,
warehouse
Same as a) 1910 41.1% 0.00 14.00 1.65 1.00 2.54
Ethnic Combination* Same as a) 1910 31.0% 0.00 131.64 8.56 2.00 17.28
Fast food Same as a) 1910 52.8% 0.00 26.72 1.90 0.00 4.04
Fast food Convenience Same as a) 1910 31.5% 0.00 47.00 4.37 2.00 7.46
Fast food Quick Service Same as a) 1910 26.5% 0.00 223.42 13.00 3.32 32.53
Limited selection& Service Same as a) 1910 21.8% 0.00 200.37 14.05 4.48 30.76
Meat Fish Same as a) 1910 48.7% 0.00 15.00 1.27 0.82 1.95
Convenience and food drug
combo
Same as a) 1910 33.7% 0.00 30.00 3.02 1.00 4.84
Produced Food Same as a) 1910 35.6% 0.00 25.00 2.64 1.00 3.86
Parking Count (at ground) # of at ground parking stalls within the bandwidth 1910 3.8% 0.00 22075.66 3815.84 2769.28 3580.89
Golf Tennis Same as a) 1910 92.4% 0.00 2.00 0.08 0.00 0.29
Indoor Fitness Same as a) 1910 46.3% 0.00 16.07 1.70 1.00 2.84
Leisure Sports Same as a) 1910 89.6% 0.00 2.00 0.10 0.00 0.30
Outdoor Sports Same as a) 1910 87.5% 0.00 3.00 0.14 0.00 0.38
Private Facility Same as a) 1910 33.7% 0.00 18.07 2.22 1.00 3.37
Public Facility Same as a) 1910 75.8% 0.00 5.00 0.35 0.00 0.72
Sports Club Same as a) 1910 91.2% 0.00 2.00 0.09 0.00 0.29
Swim Skating Same as a) 1910 89.2% 0.00 2.00 0.11 0.00 0.31
Park area (acre) Park area in acre within the bandwidth 1910 4.6% 0.00 294.82 29.73 19.60 32.02
Percentage park Percent of cells covered by park within the bandwidth 1910 4.0% 0.00 57.06 6.29 4.12 6.65
Job Density Average job per acre within the bandwidth 1910 0.8% 0.00 273.67 11.84 1.86 38.58
Residential Unit Density Average res-unit per acre within the bandwidth 1910 0.0% 0.03 28.84 4.61 3.26 4.41
25 of 29
Model 1: Base Model , 2LL:878.749
Base Model
26
Variable name Pvalue O.R. 95%CI O.R.
1 Gender (Male) 0.067 0.715 0.499 1.024
2 Income per adult 0.578 1.023 0.943 1.110
3 Number of adults 0.000 2.174 1.564 3.020
4 Number of 012 yrs old children 0.035 1.519 1.030 2.241
5 Number of cars per adult 0.000 17.899 11.525 27.796
6 Time at the current residence 0 079 1 014 0 998 1 031
6 Time at the current residence 0.079 1.014 0.998 1.031
7 Live in a single family house 0.000 2.808 1.860 4.239
26 of 29 8/5/2011
14
Distance Model
Model 2: Distance Model 2LL:759.023
Variable name pvalue O.R. 95%CI O.R.
1 Gender (Male) 0.078 0.707 0.480 1.040
27
2 Income per adult 0.117 1.071 0.983 1.167
3 Number of adults 0.000 2.084 1.465 2.965
4 Number of 012 yrs old children 0.025 1.606 1.061 2.431
5 Number of cars per adult 0.000 15.190 9.285 24.853
6 Time at the current residence 0.118 1.014 0.997 1.031
7 Live in a single family house 0.000 2.324 1.499 3.602
8 Net. distance from home to primary
grocery store
0.000 1.290 1.154 1.441
grocery store
9 Net. distance from home to the closest
supermarket
0.000 2.073 1.487 2.891
10 Net. distance from home to the closest
nonchain grocery store
0.000 1.718 1.286 2.297
27 of 29
Domain Variable name P-value O.R. 95%CI O.R.
SES
Gender (Male) 0.123 0.722 0.477 1.092
Income per adult 0.022 1.113 1.015 1.220
Number of adults 0.000 2.097 1.445 3.043
Number of 0-12 yrs old children 0.053 1.491 0.995 2.235
Number of cars per adult 0.000 13.300 7.906 22.372
Travel Mode Choice--Final Model Results
28
p 3 3 7 9 37
Time at the current residence 0.492 1.006 0.989 1.024
Live in single-family house 0.003 2.199 1.297 3.729
Distance
Net. distance from home to primary grocery store 0.000 1.236 1.099 1.390
Net. distance from home to the closest supermarket 0.036 1.482 1.026 2.142
Net. distance from home to the closest non-chain grocery
store
0.750 1.054 0.762 1.459
BE
around
home
Employment density 0.006 1.029 1.008 1.050
Number of quick service restaurants 0.043 0.970 0.941 0.990
Have a Liquor & Tavern 0.054 0.599 0.356 1.009
Street density 0.005 0.915 0.860 0.974
Employment Density 0.148 0.975 0.942 1.009
BE
around
grocery
store
Employment Density 0.148 0.975 0.942 1.009
Have a non-chain grocery store 0.022 0.525 0.302 0.912
Number of coffee shops 0.358 1.038 0.958 1.125
Number of quick service restaurants 0.854 1.004 0.963 1.046
Number of traditional restaurants 0.130 0.944 0.877 1.017
BusRd07 (per thousand rider) 0.687 1.011 0.960 1.064
Number of liquor & tavern 0.960 0.998 0.929 1.072
Have an Indoor fitness 0.114 0.628 0.353 1.117
Number of traffic signals 0.523 1.013 0.975 1.052
Number of at-ground parking (In) 0.006 1.590 1.139 2.220
Pedestrian collisions in 2001-04 0.687 1.003 0.990 1.015
28 of 29 8/5/2011
15
Domain Variable name P -value O.R. 95%CI O.R.
SES Income 0.002 1.062 1.023 1.104
Gender (Male) 0.314 1.120 0.899 1.395
Travel Frequency--Final Model Results
29
Number of cars per adult 0.059 1.233 0.992 1.532
#12-18 years old children 0.000 1.589 1.290 1.957
Having a moderate activity per day
0.233 1.218 0.881 1.685
Driving to grocery store 0.007 0.595 0.407 0.870
Think food should be inexpensive 0.000 0.851 0.779 0.930
Avg. Money Spent ($) 0.000 0.985 0.983 0.988
Avg. Time Spent(minute) 0.000 0.980 0.973 0.987
Distance Network distance home to grocery store 0.000 0.857 0.818 0.897
BE Have a golf or tennis court around home 0.008 0.576 0.383 0.865
BE
around
home
Have a golf or tennis court around home 0.008 0.576 0.383 0.865
At ground parking around home (LN) 0.000 0.875 0.825 0.927
BE
around
grocery
store
Store type (low-cost)
0.000
Store type (medium-cost)
0.776 0.966 0.762 1.224
Store type (high-cost)
0.021 1.644 1.078 2.508
Have an indoor fitness around store 0.038 1.295 1.014 1.654
29 of 29 8/5/2011
1
Using SmartMaps to Capture
Environmental Characteristics
of Daily Travel Patterns
CONFERENCE
JUNE 27-30, 2011
ATLANTA, GA, USA
1
1 of 21
Lin Lin*, Anne V. Moudon
ll3@uw.edu, moudon@uw.edu
Outline
y Background
2
y Research Questions
y SmartMap application
y Results
2 of 21 8/5/2011
2
Background
3
4
Background
3
Research Questions
y How was residential density associated with daily
travel time expenditure of individual adults?
5
travel time expenditure of individual adults?
y How are environments around home and school
associated with childrens active commuting to
school?
Question One
y How was residential density associated with daily
travel time expenditure (DTTE) of individual adults?
6
travel time expenditure (DTTE) of individual adults?
4
Data
y 2006 Puget Sound
Regional Council
7
Regional Council
Household Activity
and Travel Survey
{ 4746 Households
1322 households with children
(ages <=18)
{ 8213 Persons (ages > 18)
i h hild
2673 persons with children
{ Geocoded origin and
destination activity locations
7 of 21
Daily Travel Time Expenditure
y Total time used for daily activity travel in a weekday,
e.g.,
8
e.g.,
{ going to work
{ dropping off
{ picking up
{ going shopping
{ going home
y DTTE Mean: 83 minutes; SD: 57 minutes
8 of 21 8/5/2011
5
SmartMap Residential Density
9
homes schools
Model Results
Variables Measure Model 2
Constant 82.656***
n
d
iv
id
u
a
l C
h
a
r
a
c
t
e
r
is
t
ic
s
Age Age 0.903***
Squared of age -0.009***
Sex
1. Woman -4.527**
0. Man (ref)
Education Education level 1.109**
Average Time spent at
work per survey day (work)
13 full time 18.537***
12 part time 31.498***
11 no work (ref)
10
I
n
d
e
p
e
n
d
e
n
t
V
a
r
i
a
b
l
e
s
I
n 11 no work (ref)
Sex work Woman 13 full time -0.835
Woman 12 part time -10.811**
H
o
u
s
e
h
o
ld

S
E
S
Household income 13 Above $100,000 7.827***
12 $50,000 to $100,000 4.384**
11 Below $50,000 (ref)
Number of adult household
members -2.464*
Vehicles
Having children younger
than 18 in the household
(kid)
1. live with children younger than 18 in
the household -17.931***
0. do not live with children younger than
18 in the household (ref)
Sex kid Woman kid 18.209***
Work kid Work 13 kid 17.150***
Work 12 kid 21.352***
Sex work kid
Sex work kid
Woman 13 full time kid -21.340***
Woman 12 part time kid -21.574**
H
o
u
s
e
h
o
ld

E
n
v
ir
o
n
m
e
n
t
Residential density
Residential density within 10 minute
walk of a household logged transformed
(density) -14.953***
Squared of Density 0.785**
T
r
a
v
e
l M
o
d
e All trips done by driving or
being a passenger (car) -43.736***
Car kid 7.186**
Car density 11.769***
Loglikelihood -41397
-2 log likelihood 82794
# of individuals 7709
# of households 4469 10 of 21 8/5/2011
6
11
a
v
e
l

T
i
m
e

E
x
p
e
n
d
i
t
u
r
e
Residential density
D
a
i
l
y

T
r
a
7.6 units/acre
12
7-8 units/acre >20 units/acre
7
Question Two
y How are environments around home and school
associated with children active commuting to school
13
associated with children active commuting to school
(ACS)?
Data
14
Elementary
school
children
(511)
Middle
school
children
(1215)
High
school
children
(1618)
Total
Driving to
school
43 43
Being driven to
school
205 96 58 359
Taking bus 123 79 33 235
ACS 68 25 19 112
Total 396 200 153 749
14 of 21 8/5/2011
8
SmartMap Traffic Volume
15
homes schools
SmartMap Percentage of Parkland
16
homes schools
9
Results
Elementary
school
hild ACS
Middle school
children ACS
High school
children ACS
17
children ACS
Distance Network
distance
Route
directness
from home to
school
Home
neighborhood
environment
Park land
environment

School
environment
Traffic volume
Food store and

restaurant
17 of 21
Limitations
y Self-reported survey diary data
18
y The sample size of high school children was
relatively modest.
y Information on the availability of school buses was
not included in this study
not included in this study
18 of 21 8/5/2011
10
Contributions
y Increased accuracy on travel demand forecasting.
19
y Included both home and school environment
characteristics
y Different environmental attributes were associated
with different travel modes by different age groups
with different travel modes by different age groups.
19 of 21
Acknowledgements
20
11
Questions?
21
r
a
v
e
l

T
i
m
e

E
x
p
e
n
d
i
t
u
r
e
Residential density
D
a
i
l
y

T
r
7.6
units/acre 8/5/2011
1
Using SmartMaps to Characterize
Home and Non-home Built
Environment Exposure
CONFERENCE
JUNE 27-30, 2011
ATLANTA, GA, USA
1
1 of 26
Philip M. Hurvitz*, Anne V. Moudon
{phurvitz,moudon}@uw.edu
Overview
y Proof of concepts study
{ Measurements & data
2
{ Statistical analysis
{ Results
{ Challenges/limitations
y Contributions/summary/conclusion
2
Proof of concepts study
y Aim 1: Examine BE properties for a set of GPS traces
collected from free-roaming individuals
3
collected from free roaming individuals
y Aim 2: Determine whether BE properties (i.e.,
environmental exposures) differ for home and non-
home locations for the study population
y 41 subjects (UW students staff faculty)
41 subjects (UW students staff, faculty)
y 1 week GPS data collection
3 of 26
Measurement & data: SmartMap variables
y SmartMap (BE) variables used (more in
development):
Neighborhood composition
4
development):
Count of jobs 100
Number of residential units 100
Destinations
Count of coffee shops
Count of fast food restaurants
Count of fitness facilities
Area of park (ha)
Count of parks (any overlap)
Percent of area of bandwidth covered by park
Count of supermarkets
Count of traditional restaurants
Transportation/traffic
p / ff
Bus ridership 1000
Intersection density (n/km
2
)
Density of streets (km/km
2
, no freeways)
Estimated traffic volume (1000 vehicles/km
2
)
Trail density (km/km2)
Indices of land use composition & configuration (from FRAGSTATS)
Aggregation index (AI)
Interspersion & juxtaposition index (IJI)
Shannons diversity index (SHDI also known as entropy) 8/5/2011
3
Measurement & data: Point measures
y Output: a set of records representing a GPS trace, each with
multiple marks representing different localized BE
5
p p g
characteristics
y Excerpt from 100k+ records for 1 subject
SID SEQ_ID UTC x y % park shdi iji int_dens
s03 5637 11/29/2007 21:54 1271066 228945 2 1.44 63.8 321.62

s03 5638 11/29/2007 21:54 1271066 228937 2 1.44 61.93 321.52
s03 5639 11/29/2007 21:54 1271065 228935 2 1.44 61.34 321.6
s03 5640 11/29/2007 21:54 1271064 228937 2 1.44 61.68 321.72
s03 5641 11/29/2007 21:54 1271062 228934 2 1.44 60.95 321.88
s03 5642 11/29/200721 54 1271060 228933 2 1 44 60 59 322 01
s03 5642 11/29/2007 21:54 1271060 228933 2 1.44 60.59 322.01
s03 5643 11/29/2007 21:54 1271057 228929 2 1.44 59.63 322.37
s03 5644 11/29/2007 21:54 1271056 228930 2 1.45 59.75 322.52
s03 5645 11/29/2007 21:54 1271058 228934 2 1.44 60.72 322.18
s03 5646 11/29/2007 21:54 1271060 228941 2 1.44 62.64 321.9
s03 5647 11/29/2007 21:54 1271060 228944 2 1.44 63.33 321.87
s03 5648 11/29/2007 21:54 1271061 228950 2 1.44 64.21 321.76
s03 5649 11/29/2007 21:54 1271061 228952 2 1.45 64.39 321.74
s03 5650 11/29/2007 21:54 1271065 228957 2 1.45 64.69 321.94
Overview
6
{ Results
4
Statistical analysis
y Processing was done at an individual level
y Records were dichotomized
7
Records were dichotomized
into bins of home (0-833 m)
and non-home
(>1666 m) to clearly
delineate different
environment locales
y Each GPS location
Each GPS location
had 18 SmartMap
BE measurements
7 of 26
Statistical analysis
y Median values from SmartMaps were generated for
each environmental variable for home and non-
8
each environmental variable for home and non
home location bins
y To avoid artificial inflation of significance, bootstrap
sampling was employed (samples of 6 records per
hour, with 10,000 iterations)
y If the 95% CI of (median
home
median
non-home
)
95 (
home non home
)
included zero, it was concluded that home and non-
home environments were similar in BE
characteristics
8 of 26 8/5/2011
5
Overview
9
{ Results
Results
y 3.9 million GPS locations
18 S tM i bl
Age(y)
c
o
u
n
t
5
10
15
10
20
Gender
5
10
15
20
25
30
10
20
30
40
50
60
%
10
x 18 SmartMap variables
= 70,200,000 BE
measurements
y Sample demographics
{ Fairly young
{ More male
20 30 40 50 60 70
0 0
female male
0 0
HS SC CG PG
Education(level completed)
c
o
u
n
t
0
5
10
15
20
25
30
0
10
20
30
40
50
Asian Hispanic White
Race/Ethnicity
0
10
20
30
40
0
10
20
30
40
50
60
70
%
{ More male
{ Highly educated
{ Mostly white
{ Poor students, rich(er)
staff/faculty
10 of 26
<25 2530 5075 >=75
Income(1000USD$/y)
c
o
u
n
t
0
5
10
15
20
25
30
0
5
10
15
20
25
30
0
10
20
30
40
50
% 8/5/2011
6
Results
y Distribution of difference of medians from one BE
variable for one subject
11
variable for one subject
95% CI contains 0:
Conclude that home TD is not
different from nonhome TD
Results
y Much intra and
inter-subject
r
12
j
variation
h
o
m
e

v
a
l
u
e

h
i
g
h
e
r
h
i
g
h
e
r
n
o
n
h
o
m
e

v
a
l
u
e
8/5/2011
7
13
Results: inter and intra-subject variation
Results
y Summary view of home & non-home differences
14
Nonhome > home
Home > nonhome 8/5/2011
8
Results
y BE variables with more subjects with greater values away
from home
15
8 subjects with
greater
employment
near
home
32 subjects with
greater
employment
away
from home
Results
y BE variables with more subjects with greater values close to
home
16
26 subjects with
more
res. units
near
14 subjects with
fewer
res. units
near
home from home 8/5/2011
9
Results
y BE variables with equal number of subjects with home >
non-home
17
& vice-versa
16 subjects with
more
coffee shops
21 subjects with
more
coffee shops
near
home
away
from home
Results
y BE variable with greatest number of subjects where home
non-home
18
16 subjects with
the same number
of supermarkets
d
near and away
from home 8/5/2011
10
Overview
19
{ Results
Challenges/limitations
y Convenience sample of UW students, staff, faculty
y Basic data problems
GPS b tt lif (l ith Li i b tt i )
20
{ GPS battery life (less so with Li-ion batteries)
Scanning for satellites kills electrons
{ Acquisition times may vary
{ Multipath errors create erroneous traces
y Statistics are murky
{ Spectre of autocorrelation is ever-present
{ Endogeneity (selection of environment often made before exposure
occurs)causality cannot be inferred
S ifi i
y Specification
{ What variables are important?
{ What bandwidth(s) should be used, and for which variables?
{ What temporal sampling rates should be used? Over what duration
should measurements be made?
20 of 26 8/5/2011
11
Overview
21
{ Results
Contributions/summary/conclusion
y SmartMaps allowed estimation of local
neighborhood BE characteristics for any/all
22
g y/
locations in the study area
y GPS tracking coupled with SmartMaps allowed
estimation of BE characteristics/exposures for all
places where individuals were located
y Home and non-home environments differed
significantly for many BE variables
significantly for many BE variables
y Studies that attributed behavior to BE but that only
focused on the home locale may have erroneous
results
22 of 26 8/5/2011
12
Contributions/summary/conclusion
y A new way to measure environmental exposure in a
continuous space-time framework
23
continuous space time framework
y A space-time framework for inclusion of other high
frequency ubiquitously sensed data (e.g., biometrics,
noise, toxins)
y Potential for investigating associations between
environment and behavior or health outcomes
Powered By
24
13
Acknowledgements
y UW } College of Built Environments } Urban Design
& Planning } Urban Form Lab
25
& Planning } Urban Form Lab
y UW Graduate School Hall-Ammerer Fellowship
y UW Royalty Research Fund
y Intel Seattle Labs
y NIH
{ 1 P20 RR020774 01
{ 1-P20-RR020774-01
{ 5R01DK076608-03
{ 5R01HL091881-03
{ 5R21AG032232-03
25 of 26
Questions?
26
((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((
((((((((((((((((((( ((((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((( (((((((( (( (( ( (((((((((((((((((((((( ((( (( ((( ((( ((((((((((((((((((((((((((((((((((( ((((( (((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((( ((((((((((((((((((((((((((((( ((( ((((((((((((((((((((((((((((( ((((( ((( (((( (((( ((((((((((((((( (( (((( ((((( (((((( (((((((((((( (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((( ( ((( (((((((((((((((((( (((((((((((((((((((((((((((( ( ( ( ( ( ( ( ( ( ( ( ((( (( (( ((((((( ((((
( GPS location
Residential value
High : 885008
Low : 96742
(
(
(
(
(
(
(
(
(
(
(
(
(
81283
81282
81281
81280
81279
81278
81277
81276
81275
81274
81273
0 20 40 60 80 100
m
[
( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( (
( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( (( ( (( ((( (( ( ( ( ((((((((((( ((( ((((((( ((((((((((((((((((( (( (((((((((( (((( (( ((((((((((((( (((( (( ((( ( (((((((((((((((( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( (((((((((( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( (
( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( (( ( ( ( ( ( ( ( ( ( ( (
(((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((((((((((((((((((((((((( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( (( ((( ((((( (((((( (((((((((((((((((((((((((((((((( ((((((((((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (( (( ((((((((( ((((((((((((((((((((((((( ((((((((((((((((((((((((((((((((((((((( ( ((((((((((((((((((((((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((( (((((((((((((((( ((((((((((((((( ( ( (( ( (( ((((( (((((((((((((( ((((((((((( ((((((((( ((((((((((((((((((( ((((( ((((((((((((((((((( (((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((((( ((((((((((((((((((((((( (( (((((( (((( ((((( ((((((((((((( ((((((((((((((((((((((((((((((( (( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( (((((((((((((((((((((((((((((( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ((( ((((((((((( ((((((((( ((((((((((((((((((((((((((((((((((((((((((( ( ( (( (( (( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ((((((((((((((((((((((((((((( (((( (( (( (( ( (( ( ( (( ( ( (( ( ( ( (( ( (( (( ((( ( ( (( (( (( (( ((( ((((((((((((((((((((((((((( ((((((((((((((((((((((((((((((((((((( (((((((( (((( ((((( (((( ( ( ( ((((((((((((((( (((((((((((( ((((((((((((((( ((((((((((((((( (((((((((((((((((((((((((((((((((((((((( ((((((((((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((((((((((((((( ((((((( ((((((( (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((((((((((((((((((( ((( (((( ((((( (((( ((( (((((((((((((( (((((( (((((((((((((((((((((((((((((((((((((((((( ((((((( ((((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((((((((((((((((((((( ((((((((((
(( ( ( (((( ( ( ( ( ( ( ( ( ( ( ( ( (
((((( ((((((((((((((((((((((((((((( ((((((((((((( (((((((((( ((((((((((((((( (((((((((((((((( (((((((( (((((((((((((((((((((((((((((((((( (( ( ( ( (( ((((((((((((((((( ((((((((((( ((((( (((( ((( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((( ((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((( ((((((((((((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((( ((((((((((( ((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((((((((((((((((((((((((((( (((((((( ((((( ((((((((((((( ((((((((((((((((( ((((((( (((((((((((((((((((((((((((( (((((((((((((( ((((((((((((((((((((( (((((( (((((((((((((((((( (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((((((((((((( (( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( (( (((((((((( ((((((( ( ( ( ( ( ( ( ( ( ( ( ( (( (((((( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ((((( (( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( (( ( ( ( ( ( ( ( ( (( (((( ( ( ( ( ( (((((((((((((((((( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ((((( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( (( ( (( (( (( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( (((((( (( ( ( ( ( ( ( ( ( ( ( ( ( ( ( (((((((( ((( (( (( ((( (((((((((((((((((((((( (((((((((((((((( ((( (( ( (( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( (((( ((((((((((((((((((((((((((((( (((((((((( (((((((((((((((((((( ((((((((( ((((( (((((( ((((((((((( (((((((( (((((( ((((((((((((((((( ((((((((((((((((((((((((((((( ((((( ((((((( ((( ((( (((((( ((( ((( ((((((((((( ((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((( (((((((((((((((((((((((( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( (( (((((((((((((((((((((((((( ((((( ( ( ( ( ( ( ( ( ( ( ( ( (((((((((((((((((((((((((((((((((((((((((((( (((( (( ( ( ( ( ( ( ( ( ( ( (( ( (((( (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((( ((((((((((((((((((( (((((((((((((((((((( ((( (((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((((((((((((((( ((((((((((( ((((((((((((((((( ((((((((((((((((((((((((((((((( ((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((((( ((((((((((( (((((((((( ((((((((((( ((((((( (((((((((((((((((((((((((((( ((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((((( ((((((((((((((((((((((((((((((((((((((( (((((( (((( (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((( (((((((((((((( (((((((((((((((((((((((((((((((((((((((((((((( (((((((( (((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((( (((
(((((((( (((((((((((((((( ((( (((( (((((( ((( (((( ((((((( ( (( ( (( (((( (((((((((((( ( ( (( (( (( ( (( ( ( (( (( ((( ((( (((( (((( ((((( (((( ((( ((( (( ((( ((((( (((( ((((( (((( ((( (((( (( ((( (( (( (( (( (( ((( ( ((( (( ((( ((( (( (((( ((( (( ((( ((((( ((( ((( ((( ((( (((( ((( (( ((( ((( ((( ((( ((( (((((((((((((((( ((((((((((((((((((((( (( (( (( ((( ((( ((( (( (( ((( (((( (((((((((((((((((((
((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((( ( ( ( ((((((((((((((( (( ( (( ((((( (((( ((( (( (( (( (((( ((( ((( ((( ((((((((((( ((((((((((((((((((((((( ( (( ( ( ( ( (( (( ( ( ( ( (( (( (( (( (( ((( (( (( (( ( (( ((( (( (( (( (( (( (( (( (( (( ((( ( (( (((( ((( (( (( (( (((( ((( (( (( ((( (( ((( (( ((( (( ((( ( (( ((( ((( ((( (((( ((( ((( ((( ((( ((( (( ((( (( ((( ((( (( (( (( (( (( ((( ( (( (( (( (( ((( ((( (( (((( ((( ((( (( ((( ((( (( (( ((( ((((((( (((((((( ((((( (((( (((( (((( (((( ((( (((( (( ((((((((((((( ((((( (((((((( ((((( ((((( (((((( ((((( (((((((((( ((((((( ((((((((((((((( (((( (((((
(( (( (((( (((((((((((((((((((((((((((((((((((((((((((((((( (((((( ((((((((((( (((((((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((((((( (((((((((((((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((( (((((((( ((((((((((((((( ((((((((((((( ( (((((((((( ( ( ( (((((((((( ((( (( (((( ( (( ( ((((((((((((((((((((((((((((((((((((((((((((( (((( (( (( ( (( ( ( ( ( ( ( ( ( ( ( ( ( (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ( ((( (((((((( (((((( ( ( (( (( ( (( ( ( ( ( ( ( ( (( (( (( ((( (((((((((((((((((((( (( (( ((( (( (( ((( ( ( ( ( ( ( ( ( ( (( ( (( ((((((( (((((((((((( ((( ((( ((((( (((((( (((( (((((((((((( (((((( ( (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((( ((( (( (( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((((((((((((((((((((((((((((((( (((((( ((( (((((((((((((((((((((((((((((((((((( ((( (((((((((((((((((((((((((((((((((((((((( ((((( ((((((((( ((((((((((((((((((((((((((( ((((((((((( (((( ((((((( ((((((((((((((((((((((((((( ((((((((((((((( (((((((( (((((((((((((((((((((((((((((((((((((((((((((( (((( (((((((((((((((((((((((((((((((((((((((((( (((((((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((( ((((((((((((((((((((((((((((((((((( ((((( ((((((((( ((((((((((((((((((((((((( (( (( ((((((( ((((( (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((( (( ((((((((((((((((( ((((((( ((((( (((((((((((((((((( (((((((((((((((((((((((((((((((((((((((((( ((( ((((((( ((((( (((((((((((((((((((((((((((((((((((((((( (((((((((((((((((( ((((((((
((((((( ((((((((((((( (((( (((((((((((((((( ((((((((((((((((((((((((((( (( ((((((((((((((((((((((((((((((( (((((((((((((((((((((( ((((( (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ( ((( ( (( ( ( ( ( ( ( (( (((((((( ((((((((((((((((( (( (( (( (( (( ((( ((( ((( ( (( ( ( (( ( ( (( ( ( (( (( (((( ((((((((
(((((((((((((((( ((((((((((((((((((((((((( ((((((((((((((((((((((((((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((( (((((((((((((((((((((((((( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( (( ((((((((( ( ( ( ( ( ( ( ( ( ( ( ( ( ( (((((( (((((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((((((((((((((((((((((( ((((((((((( (( ((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((( ((((( (((((((((((((((((((((((((((((((((((((( (( ( ( ( ( ( ( (( ((((((((((( ((((((( (( ( ( ( ( ( ( ( ( ( (( ( ( ( ( ( ( ( ( ( ( ( ( ( ( (((((((((((((((((((((((((((((((((((((((((((((((( (((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((((((((((((((((((((((((((((((((((((((((((( ((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((( ((((((((((((((((((((( (((((((((((((((((((((((( (((( (((( (((( (((( ((((((((((((((((((((((((((((( ((((((((( (((((((((((((((((((((((((((((((((( ((((((( ((((((((((((((((((((((((((((((((((( ((((((((((((((( ((((((((((((((((((((( ((((((((((((((((((((((((((( ((((((((((((( ((((((((((((((((((((( ((((((((((((((((((((((((( ((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((((((((((((( ( ( ( ( ( (( (((((((( (((((((((( ((((((( (((((((((((( (((((((((((((((((((((((((((((((( ((( ( ( ( ( ( (( ((((((( (( ( ( ( ( (( ((((((((((((((((((((( ((((( (((((((((( (((((((((((((((((((((((((((((( ((((((((((((((((
((((( ((((((((((((((((((( ((((((( (((((( (((( (((((( (((((( (((( (( (((((((( ((((((( ((((((((((((((((((((((((( (((((((((((((((( (((( (((((((((((((((( (((((((((((((((((((((( (((((((((((( ((((((((((((((((((((( (((((((((((((((((((((((((((((((((((((( (((((((( (((((((((((((((((((((((((((((((((((((((((((((((((((( (( (((( ((((((((((((((((((((((((((((((((((((((((((((((((( (((((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((( (((((( ( ((( ((((((((((((((((((((( ((( (( ( ( (( ((((( (((( (( ((( (((((((((((((((((((((((((((((( ((((( ( (( ( (( (((((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((((((( (((( ((((( (((((( ((((((((((((((((((( ((((((((( (((( (( (( ((((( ((((( (((((((( ((( (((
( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ((((((( ( ( ( ( ( ( (( (((((
((
(((( (( ((((((((((((((((((((((
((( ((((((((((((((((((((((((((( ((( ((((
(
((
( (((( (((( (((((((((((((((((( (
(( (( (((((((((( (((((((((((( (((((((((((((( (( (((((((( ((((((((((((((((((((((((((((((( (((( (( ((((( (((( ((((( (( (((((((((
((((( ( (( (((( ( ((((((((((( (( ((((((((((((( (((((((((((( (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((( ((((((((((((((((((((((( ((( ((( (((((( ((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((
(((((((( ((( ((( ((( ((((( (((( ((( (( (( (( (( ( (( (( ( ( (( (( (( ((( ((( ( (((((( ((( (((((((((((((((((((((((( (( (((( (((( ((( (((((((((((((((((((((((((((((((( (((((((((((((((( (((( (((( (((((((((((( ((((((((((((((((((((((((((((((((((( (( (( (( ( ( (( (( (( ((( (( ( (( (( (( (( (( (( ((( ((( ((( ((( (( (( (( ((( (( (( (((( (( ((( ((((( ((( ((((( ((( (((((((((((((( ((((((( (( (( (( ( (( (( (((( (((((( (((((( ((((((((((((( ((( ( (( ((((( (((( (((( ((((( ((((((( ( ( ( ( ( (( (( (( (((( ( (((( ((((((((((((( (((( ( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((((( (((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((( (((((((((( (((((((((((((((((((((((((((( ((((((((((((( ((((((((((((((((((( (((((((((((((((((((((((((((((( ((( ((( ((( ((( (((( ( (( (((((((( (( (( (((( (( (( ( ( (( ((((( (((((((((( ((( ((((( ((((( ((((((((( (
((((((((((((((((( (((((
( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((((( ( (((((((( (((((((((((((((( (((((((((((( (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((((( (((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((((((((((((((((((( ((((((((((((((((((((((( (((((((((( (((((((((((((((((((((((((((((((((( ((((((((( ((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((( (((((((((((((((((((((((( (((((((((((((((((((((((((((((( ((((((((((((((((((((((((((((((
((((((((((((((((((((((( ( ( ( ( ( ( ( ( ( ( (( (((((((((((((((((((((( ((((((((((( ((((((((((((((( (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((( (((((((((((((((((( ((((((((((( ((( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( (( ( ( (( (((((((((((((((((((( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( (((((((((((((( (((((((((((((((((((((((( (((((( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ((( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( (( (( ( ( ( ( ( ( ( ( ( ( ( (((((((((((((( ((((((((((( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( (( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ((((((((( (((( ((((((((( (((((((((((( (((((( (((((((( (((( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ((( ((((( (((((( (((((((((((((((((((((((((((((((((((((((((( ((((((((((((((((( ((((((((((((((((((((( (((( ((((( (((( ((((( (((((((( ((((( (((((((((((((((((((((((( ((((((((((((((((( ( (((((( (((( ((((((((((((((((((((((((((((((( ((((((( (( ( ( (((((( ((((((((((((( ((((((((((((((((((((((((((((((((((((((( ((((((((((((((((((((( ((((( (((( (((((( (( (((((( ((((((( ((( (((( ((((( (( (( (((((( (((( (((((((( (((((((( ((((((((((((((((((((((((((( ( (((((((((((((( (( ( (( ((((( ((((((( (((((((((((( (((( (((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((((((((((( ((((((((((((((((((( ((((((((((((((((((((((((((((( (((((((((((((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((((((( (( ( ( (( ((((((((((((((((((((((((((((((((((((((((( (((((((((((((((((((((((((((((( (((( ((( (( ((((( ((( (((( (((((((((( ((((( (((((((((((((((((((((((((( (((((((((((((((((( ((( (((((((((((((( (((((((((((((((((((((((((((((((( ((((((((((((((((((((((((((((((((((((((((((((((((((( (((((((( ((((((( ((((((((((((((( (((((((((((((((((((( (((((((((
(
(
(
(
(
(
(
(
(
(
(
(
(
(
(
(
(
(
(
(
(
(
(
(
(
(
( ( ((((( ( ( ( ( (
(
(
75342
75339
75336
75334
75332
75330
75328
75326
75324
75322
75320
75318
0 1 0.5 km 0 100 50 m 7/21/2011
1
Working People:
Place based Effects on Hypertension,
Diabetes-2, and Obesity occurrence
ALBERTO COLOMBI MD MPH ( 1 ) ,
JUNA PAPAJORGJI ( 2)
1 ) P P G I N D U S T R I E S I N C 1 ) P P G I N D U S T R I E S , I N C .
2 ) U N I V E R S I T Y O F F L O R I D A
PPG is
A global maker of paints, coatings, optical products,
specialty materials, chemicals, glass, and fiber glass
Atlanta, Georgia - www.urisa.org 7/21/2011
2
Geographic Study Area: 37 Work Sites in 24 Counties across 19 States
DOMAINS AND VARIABLES DOMAINS AND VARIABLES
Four Domains: Four Domains:
Food Environment
Health and Behavior Environment
Socioeconomic Environment
Three Variables:
P f All Ob k i Percentage of All Obese per work site
Diabetes Medical Episodes per 1,000 Medical Episodes
Hypertension Episodes per 1,000 Episodes of Care 7/21/2011
3
Populations
y 37 sites
y 24 Counties
y 19 States
y For Each worksite:
{Health Risk Profile aggregate end points
{Medical Claim experience- end points
{Geographic data based on the county where the
g p y
worksite resides
Worksite Population Health Risk Profile
Prevalence of risk elements summarized according to the , 10 Keys to Healthy Aging (10 Ks )
University of Pittsburgh . Source: Wellness Checkpoint ( 7/1/2005- ending: 6/30/2008).
y Health Risk Appraisal participation
y % at low/no health risk (0-2 risk factors)
y % aged 50+ (Male Female-Total)
y % knowing their numbers: blood pressure, glucose, cholesterol tot
and LDL
y % smokers and % ready to quit
% i i d y % immunized
y % participating to age related cancer screening
y % at risk for lack of physical activity
y % with strong social ties
y % obese (Body Mass Index > 30)
4
Geographic Variables (County based)
y POP2000 population in the county based on Census 2000
y Charact level of urbanization of the county (see report for how it was calculated)
y CountyName county where ppg site is located (with 1 or 2 exceptions there is 1 site per 1 county, y y ppg p p y,
see report for specifics)
y FIPS unique county code defined by census (did not use, but left it in place)
y INC_MEDIAN median household income based on census 2009 by county
y INC_AVERAG average household income based on census 2009 by county
y DI_MEDIAN median household disposable income based on census 2009 by county
y DI_AVERAG average household disposable income based on census 2009 by county
y NW_MEDIAN median household net worth based on census 2009 by county
y NW_AVERAGE- average household net worth based on census 2009 by county
y VAL_MEDIAN- median home value based on census 2009 by county
y VAL_AVERAG- average home value based on census 2009 by county
y CIAPI Composite Index Air Pollution Intensity (CIAPI) (see report for how it was developed)
y LStore10kP Composite Index for Local Food Availability per 10,000 population (see report on how
it was calculated)
y CHILD_OBES (food atlas), low income preschool obesity rate
y NO_CAR nr of housing unit without car who reside more than 1 mile from a grocery store (food
atlas)
Geographic variables (continued)
y PERC_NOCAR percentage of housing units without car who reside more than 1 mile from a
grocery store (food atlas)
y LOWINC_FD nr of people with low income who live more than 1 mile from a grocery store (food
atlas)
y PERC_LOWIN percentage of people with low income who reside more than 1 mile from a grocery
store (food atlas)
y GROCERY nr of grocery stores and supermarkets (food atlas)
y GROCERYPERPOP nr of grocery stores and supermarkets per 1,000 residents (food atlas)
y CONVGAS nr of convenience stores (food atlas)
y CONVPERPOP nr of convenience stores per 1,000 residents (food atlas)
y FF_PERPOP - nr of FAST FOOD restaurants per 1,000 residents (food atlas)
y RESTAURANT - nr of full service restaurants (food atlas)
RESTPERPOP f f ll i id (f d l ) y RESTPERPOP - nr of full service restaurants per 1,000 residents (food atlas)
y SNAPSTORE food atlas variable for the supplemental nutrition assistance program i.e. (food
stamps)
y SNAPPERPOP - food atlas variable for the supplemental nutrition assistance program i.e. (food
stamps)
y SNAPREDEMP - food atlas variable for the supplemental nutrition assistance program i.e. (food
stamps)
5
Geographic Variables (continued)
y WIC_STORES - did not use, food atlas variable for the women, infant, children
program
y WIC PERPOP - did not use, food atlas variable for the women, infant, children WIC_PERPOP did not use, food atlas variable for the women, infant, children
program
y WICPERSTOR - did not use, food atlas variable for the women, infant, children
program
y WIC_REDEMP - did not use, food atlas variable for the women, infant, children
program
y FAccess Composite Index for accessibility to food establishments per 1,000 residents
(see report for how it was calculated)
y BadFood Composite Index for unhealthy food establishments per 1,000 residents (see
t f h it l l t d) report for how it was calculated)
y FAvailab Composite Index for general food availability per 1,000 residents (see report
for how it was calculated)
Epidemiological endpoints
y DM2-Epis/1000 T2 Diabetes Medical Episodes
of care per 1 000 Insured employees per of care per 1,000 Insured employees per
worksite
y HTN-Epis/1000 Hypertension Episodes of
care per 1,000 insured employees per worksite
y Percent ObeseWorksite Percentage of active employees
reporting in their Health Risk Assessment a Body Mass p g y
index (BMI) > 30
6
Overview Overview
Percent active employees who are Obese (BMI > 30) , by PPG work site 7/21/2011
7
T2 Diabetes Episodes of care per 1,000 Insured active employees, by PPG
work site
Hypertension Episodes of care per 1,000 Insured active
employees, by PPG work site 7/21/2011
8
Urban vs. Rural PPG Counties in the North East area (10 urban/17
rural)
Socioeconomic Environment Socioeconomic Environment 7/21/2011
9
Median Household Income by Block Group in Ohio and PPG Sites in red
Median Household Disposable Income by Block Group in Ohio and PPG
Sites in red 7/21/2011
10
Median Household Net Worth by Block Group in Ohio and PPG Sites in
black
Median Home Value by Block Group in Ohio and PPG Sites in
black 7/21/2011
11
Allegheny County, Pennsylvania 6 PPG sites with 5 and 8 mile buffers
overlaid with Median Household Net Worth
Median Household Disposable Income by County and PPG work sites 7/21/2011
12
Median Household Net Worth by County and PPG work sites
Median Household Home Value by County and PPG work sites 7/21/2011
13
Food Environment
Composite Index of Local Food Production per 10,000 population
showing PPG counties in Ohio and Pennsylvania 7/21/2011
14
Density of Farmer Markets and PPG Sites NE
Area
Density of Grocery Co-ops and PPG Sites NE
Areas 7/21/2011
15
Density of Community Supported Agriculture (CSAs) and PPG Sites
NE Areas
Six Spatial Measures from the Food Environment Atlas (USDA
2010)
Grocery Stores /1,000pop Convenience Stores /1,000pop Fast Food Stores /1,000pop
Restaurants /1,000pop No Car Households > 1 ml to grocery Low Income People > 1 ml to grocery 7/21/2011
16
Composite Index of Food Availability per 1,000 population in North East PPG
counties
Composite Index of Food Accessibility per 1,000 population in North East PPG
counties 7/21/2011
17
Composite Index of Unhealthy Food per 1,000 population in North East PPG
counties
Health Environment Health Environment 7/21/2011
18
Five Spatial Rankings from the County Health Ranking Atlas (UW 2010)
Health Outcomes, Health Factors, Health Behavior, Diet Exercise, Socioeconomic
Status
Ohio, Health Outcomes State Rankings and PPG sites. 7/21/2011
19
PPG Counties by Health Outcome State Percentile Interval
Analysis
y JMP-9
y Principal Component Analysis y Principal Component Analysis
y Multivariate Analysis
y Least Square Modeling- Screening
y Prediction Profiling
Question: how much of the variance in the occurrence of
the epidemiological endpoints is explained by population the epidemiological endpoints is explained by population
health risk and how much is explained by geographic
differences ?
20
Hypertension 1
Source DF Sum of Squares Mean Square F Ratio
Model 2 43208.264 21604.1 18.6872
Error 34 39307.248 1156.1 Prob > F
C Total 36 82515 512 < 0001*
Analysis of Variance
Summary of Fit
RSquare 0.523638
RSquare Adj 0.495617
Root Mean Square Error 34.0014
Mean of Response 118 9862
Term Estimate Std Error t Ratio Prob>|t|
Intercept 177.63658 37.02791 4.80 <.0001*
% Low Risk -276.7449 51.55145 -5.37 <.0001*
RESTPERPOP 139.88163 37.82077 3.70 0.0008*
Parameter Estimates
C. Total 36 82515.512 <.0001*
Mean of Response 118.9862
Observations (or Sum Wgts) 37
Hypertension 2
Model 4 42942.729 10735.7 16.7952
Error 31 19815.573 639.2 Prob > F
C. Total 35 62758.302 <.0001*
Summary of Fit
RSquare 0.684256
Intercept 164.82537 34.44112 4.79 <.0001*
% Low Risk -278.3765 43.61131 -6.38 <.0001*
FARM_TO_SC 36.310238 10.93448 3.32 0.0023*
% Aged 50+ 108.50271 33.06784 3.28 0.0026*
FF_PERPOP 84.17493 27.20468 3.09 0.0042*
Parameter Estimates
21
Obesity
Summary of Fit
RSquare 0.789791
Ob ti ( S W t ) 37
Source DF Sum of
Squares
Mean Square F Ratio
Model 5 1807.3823 361.476 23.2944
Error 31 481.0501 15.518 Prob > F
C T t l 36 2288 4324 0001*
Intercept 89.22371 14.27391 6.25 <.0001*
% Low Risk -32.72054 6.631636 -4.93 <.0001*
SNAPREDEMP 5.4824e-5 1.622e-5 3.38 0.0020*
Social ties% strong ties -51.96366 15.29397 -3.40 0.0019*
WICPERSTOR 0.0000355 0.00001 3.52 0.0014*
LDL Chol % who know -7.328688 2.996518 -2.45 0.0203*
Observations (or Sum Wgts) 37 C. Total 36 2288.4324 <.0001*
Parameter Estimates
T2 Diabetes
Model 4 10261.642 2565.41 9.4874
Error 32 8652 846 270 40 Prob > F
Summary of Fit
RSquare 0.542528
Root Mean Square Error 16 44389
Intercept -16.24054 18.97555 -0.86 0.3984
Phys. Activity % at no risk -64.35339 29.23629 -2.20 0.0351*
Blood Glucose % HX elevated 177.26702 60.64752 2.92 0.0063*
% Aged 50+ 65.183661 20.77725 3.14 0.0036*
RESTPERPOP 62.647187 18.31548 3.42 0.0017*
Parameter Estimates
Error 32 8652.846 270.40 Prob > F
C. Total 36 18914.488 <.0001*
22
Conclusions
y The geography surrounding the worksite seems to play
some role in the frequency of occurrence of Hypertension,
Obesity and of Type 2 Diabetes in geographically diverse Obesity , and of Type 2 Diabetes, in geographically diverse
working populations.
y Besides health risk, age and behaviors, the attributable
variance of episodes of care per 1000 active employees
due to geographic factors is :
{ 20 % for Hypertension
{ 17% for Obesity { 17% for Obesity,
{ 17 % for Type 2 Diabetes
y Besides socio-economic variables, food geography seem
to be particularly relevant and deserving further
investigations.
1
Racial Residential
Segregation and Stroke
Mortality in Atlanta
SOPHI A GREER, MPH
CENTERS FOR DI SEASE CONTROL AND PREVENTI ON
EPI DEMI OLOGY AND SURVEI LLANCE BRANCH
SMALL AREA ANALYSI S TEAM
Collaborators
y Michele Casper, PhD (CDC) p , ( )
y Michael Kramer, MS, MMSc, PA-C (Emory
University)
y Greg Schwartz, MS (CDC)
y Jim Holt, MPA, PhD (CDC)
y Elaine Hallisey (ATSDR) y Elaine Hallisey (ATSDR)
y Gordon Freymann, MPH (GA Div Public Health)
y Yuequin Zhou (GA Div Public Health)
2
Overview
y Definition of racial residential segregation
y Study design
y Racial residential segregation measures
y ArcGIS macro
y Methods
y Results
y Conclusions
What is residential segregation?
y The spatial separation of 2 or more racial/ethnic groups in
residential space
y Results from long-term institutionalized racism including housing
policies, educational policies, healthcare discrimination.
y Leads to unequal opportunity for racial/ethnic minorities
y Often referred to as one of the fundamental social determinants
of health
3
Residential segregation and health
Source: Kramer M R , Hogue C R Epidemiol Rev 2009;31:178-194
Study Design
y Objective: to examine the association between racial residential
segregation and stroke mortality in the Atlanta Metropolitan Statistical
Area (MSA) Area (MSA)
y Population: Non-Hispanic Blacks and Whites, ages 35+ (Sources: GA
Div Community Health and US Census Bureau)
y Unit of analysis: census tract
Time period: 2000 2006 y Time period: 2000 2006
y Stroke Definition: Cerebrovascular disease (ICD-10 Codes I60-I69),
GA Div of Community Health
4
Examples of Segregation measures
y Dissimilarity Index
{ Evenness of population { Evenness of population
y Spatial Proximity Index
{ Spatial clustering
y Isolation Index
{ Probability of exposure
Source: Massey and Denton Social Forces 1988 67(2): 282-315
Spatial Isolation Index
y Ranges between 0 and 1 y Ranges between 0 and 1
y Reardon et al. have developed a macro in ArcGIS to
measure segregation on a smaller scale
(http://www.pop.psu.edu/services/GIA/research-
projects/mss/mss-about)
y Incorporates population distribution of neighboring areas
Sources: Reardon et al. Soc Methodology 2004 31(1): 121-162,
Wong 2002 Geo and Environ Modelling 6(1): 81-97,
Massey and Denton Social Forces 1988 67(2): 282-315
5
Calculating spatial isolation index
using ArcGIS macro
Block level population counts
(by race) are used to derive
density surface for each pixel density surface for each pixel
within area of interest (using
pycnophylactic smoothing
techniques)
Neighborhood is defined by
user
Individual indices for each
50m x 50m pixel were
p
averaged across census tract
to create summary isolation
index
Segregation using %black
6
Isolation index and stroke mortality model
y GEE Poisson model
y Exposure: Isolation index y Exposure: Isolation index
y Outcome: Stroke deaths
y Covariates: poverty, education
y Model estimates rate ratios and stratified by race
Source: Greer S, Casper M, Kramer M et al (In press). Residential Segregation and Stroke in Atlanta. Ethn Dis.
Results: Segregation and Stroke Mortality
among Blacks
Segregation by age in years RR 95% CI p RR 95% CI p
35-64
Group 1 (ref ) (predominantly white) (<0 3)
Model A
*
Model B
Blacks
Group 1 (ref ) (predominantly white) (<0.3) --- --- --- --- --- ---
Group 2 (racially mixed) (0.3-0.7) 1.07 (0.82, 1.40) 0.63 0.97 (0.73, 1.28) 0.81
Group 3 (predominantly black) (>=0.7) 1.60 (1.28, 2.01) <.0001 1.49 (1.17, 1.90) 0.001
65
Group 3 (predominantly black) (>=0.7) 0.87 (0.71, 1.07) 0.19 0.78 (0.63, 0.97) 0.03
Percentage of people below poverty level
Low (ref ) (<5.25%) --- --- ---
Medium (5.25 - 11.75%) 0.92 (0.79, 1.06) 0.25
High (>11.75%) 0.88 (0.75, 1.04) 0.13
Percentage of people without high school diploma
Low (ref ) (<=10%) --- --- ---
Medium (10 - 20%) 1.36 (1.18, 1.58) <.0001
High (>=20%) 1.45 (1.22, 1.73) <.0001
*Model A: segregation, age, age x segregation.
Model B: segregation, age, % below poverty, % without high school diploma, age x segregation.
Source: Greer S, Casper M, Kramer M et al (In press). Residential Segregation and Stroke in Atlanta. Ethn Dis. 7/19/2011
7
Results: Segregation and Stroke Mortality
among Whites
Segregation by age in years RR 95% CI p RR 95% CI p
35-64
G 1 ( f ) ( d i tl hit ) (<0 3)
Whites
Model A
*
Model B
Group 2 (racially mixed) (0.3-0.7) 1.37 (1.18, 1.59) <.0001 1.22 (1.05, 1.41) 0.009
65
Group 1 (ref ) (predominantly white) (<0.3) --- --- ---
Percentage of people below poverty level
Low (ref ) (<5.25%) --- --- ---
Medium (5.25 - 11.75%) 1.07 (0.86, 1.01) 0.07
High (>11 75%) 0 97 (0 83 1 14) 0 73
High (>11.75%) 0.97 (0.83, 1.14) 0.73
Percentage of people without high school diploma
Low (ref ) (<=10%) --- --- ---
Medium (10 - 20%) 1.19 (1.09, 1.29) <.0001
High (>=20%) 1.37 (1.26, 1.50) <.0001
*Model A: segregation, age, age x segregation.
Model B: segregation, age, % below poverty, % without high school diploma, age x segregation.
Source: Greer S, Casper M, Kramer M et al (In press). Residential Segregation and Stroke in Atlanta. Ethn Dis.
Strengths/Limitations
Strengths
y ArcGIS macro improves accuracy of the spatial distribution of ArcGIS macro improves accuracy of the spatial distribution of
race/ethnicity on smaller scale
y Measuring segregation on a smaller scale allows more understanding
of how local neighborhood characteristics could be linked to health
Limitations
y Lack of point level data did not allow segregation to be defined for each
individual in this study individual in this study
y Census 2000 data used for modeling rates and calculating segregation
index may not accurately reflect average population counts for study
period
8
Conclusions
y Spatial isolation index provides more precise measure of y Spatial isolation index provides more precise measure of
residential segregation at a local scale
y Residential segregation is associated with stroke mortality
in the Atlanta MSA for younger age groups (35-64) even
after controlling for percent below poverty and percent
without high school diploma
y Results validate prior work on this topic and the
importance of considering residential segregation as a
factor that contributes to disparities in health
Questions?
2011-07-14
1
Avail abil ity to tobacco outl ets
and other GIS at the Swedish
National Institute of Publ ic
Heal th
A K i J h Anna- Kar i n Johansson
The Swedish National Institute of Public Health
y A state agency under
the Ministry of the Ministry of
Health and Social
Affairs.
y Has a special
government
commission to
monitor, evaluate
monitor, evaluate
and coordinate the
implementation of
the national public
health policy.
2011-07-14
2
Overview
1. Tobacco - availability
2. Gambling availability
3. Public health GIS examples at the web
4. Data visualization
Tobacco
What is research telling us?
y Density and availability of tobacco outlets are y Density and availability of tobacco outlets are
important factors influencing to what extent young
people smoke.
y Point of sale tobacco marketing is more common in
neighbourhoods with lower average income and a
large proportion of resident younger than 18 years.
2011-07-14
3
Tobacco
Tobacco act
In Sweden the tobacco act forbid vendors
to sell to minors (under 18 years of age).
Girls
(grade 9)
Boys
(grade 9)
Girls
(grade 11)
Boys
(grade 11)
Cigarettes 35 % 40 % 52 % 67 %
Snus (oral 17 % 36 % 41 % 66 %
Snus (oral
tobacco)
17 % 36 % 41 % 66 %
Despite this legislation, many buy themselves
Tobacco
What is done?
y Each elementary and upper secondary school in
Sweden has been geocoded.
y Tobacco vendors have been digitalized.
y The buffering method has been used.
y Based on earlier studies each school was given a
300 meters buffering zone (walking distance).
y This enables us to sort out vendors within walking
y This enables us to sort out vendors within walking
distance from schools.
2011-07-14
4
Tobacco
300 m walking distance
Point of
sale. Prio.
School
Point of
Point of
sale
Point of
sale
g
Point of
sale. Prio
Point of
sale
Point of
sale
Tobacco
Schools and vendors in an urban area
2011-07-14
5
Tobacco
Availability varies between municipalities
Share of schools
within walking distance within walking distance
from tobacco vendor.
Evaluation
Identify
problem
Planning
Tobacco
Evaluation
Correlations
Facilitator
of debate
problem
areas
Map
14/07/2011 Page 10
Survey
2011-07-14
6
Gambling
What is research telling us?
P bl bli M d l i y Problem gambling: Men, young and low income are
overrepresented.
y The relationship between problem gambling and
availability is complicated.
y Gambling venues locate where there is a demand
y More common in areas with many problem
y p
gamblers.
Gambling
Geodata in Sweden: cells 250*250 meters
2011-07-14
7
Gambling
Availability to gambling relative to income
w Average income/parish
w Average income within 500m
from slot machines
-> Slot machines in low income
areas
Gambling
Gambling relative to inhabitants
2011-07-14
8
Gambling
How to use the information
y The map complement statistical analysis.
y Base for discussions at municipality level 1) when
opinion before licensing,
2) when planning for preventional efforts, 3)
gambling-free zones
y Can be complemented with schools Have
y Can be complemented with schools. Have
youngsters walking distance to gambling during
daytime?
GIS examples at the web
Valls
Snstorp
Andersberg
Vilhelmsflt
Sannarp
Gustavsflt
Frennarp
Furet
Larsfrid
Nyhem
Galgberget rjans vall
Hamnen
Rotorp
Alet Sder
Mickedala
rk
Accessibility to green open space
Secure way to school
Elderly housing
Inventory of obstacles along the walkways
Traveltime by bike
etc
2011-07-14
9
Data visualization
y Plans 2012: Making
i l bli h lth regional public health
data available through
the web. The
application contains i.
e. correation chart
(two variables plus
time) and municipality
map which allow
p
laboration on your
own.
Thank you for your attention!
anna-karin.johansson@fhi.se
www.fhi.se 7/8/2011
1
Developing a
Community Health
Information Web Portal
PRESENTATI ON BY:
PAUL D. JUAREZ, PHD &
WANSOO I M, PHD
DEPARTMENT OF FAMI LY & COMMUNI TY MEDI CI NE
DI VI SI ON OF COMMUNI TY HEALTH
MEHARRY MEDI CAL COLLEGE
NASHVI LLE, TENNESSEE
June 29, 2011
Challenges Facing Public Health
y Chronic Diseases and Injury are the Leading Causes
of Morbidity, Disability and Death
y Chronic Diseases/Injury are Shaped by Environment
y Research Tools are Needed to Examine Relationship
of Physical, Built, and Social Environments & Health
y Trans-disciplinary Investigators are Needed
{ Knowledge of Informatics & Data Base Management
{ Geographic Information Systems / Mapping
{ Social, Behavioral, Environmental, Policy & Health Sciences
{ Statistical Analyses (spatial analysis, throughput analysis, etc)
2
New Health Promotion Strategies are Needed
y Traditional Biomedical Research is Not Decreasing
Health Disparities Health Disparities
y New Strategies are Needed to Improve the Publics
Health & Reduce Disparities
y Interventions Need to be Culturally Appropriate
y Interventions Need to Target At-Risk Communities
y Community Participation is Needed to Identify Local
y Community Participation is Needed to Identify Local
Barriers and Strategies for Improving Health of
Communities
Underlying Assumptions
y Academics dont have a Lock on Knowledge or Ideas
y Health Risk and Outcomes Vary by Age Race y Health Risk and Outcomes Vary by Age, Race,
Gender, and Place of Residence
y Data are Needed at a Sub-county Level
y Community Engagement = Community
Empowerment
y Empowered Communities Increase Likelihood of y Empowered Communities Increase Likelihood of
Sustainability of Efforts
3
Advantages of Community Engagement
y Breaks Dependence on Academic Partners
y Taps Community Knowledge y Taps Community Knowledge
y Promotes Community Buy-in
y Empowers Community to Become Active Partners
y Allows for Targeting of Community Interventions
y Establishes Baseline Data for Evaluating Outcomes
IMNashville.com Web Portal
y Conceptual Framework
{ Use Sub-county Data to Enable Analysis of Factors Affecting { Use Sub county Data to Enable Analysis of Factors Affecting
the Publics Health & Target Interventions that Address them
{ Integrate Large Data Sets
Health Outcomes
Physical, Built, Social & Policy Environments
{ Track Changes Over Time
{ Ensure Confidentiality of Personal Health Data { Ensure Confidentiality of Personal Health Data
{ Enable Easy Access to Data by Communities
{ Provide User Friendly Tools to Analyze the Data
{ Provide Different Ways to Visualize Data
4
Health Outcomes Data Sets
y Examples
{ Death Certificates { Death Certificates
{ Medicare
{ Medicaid
{ Hospital Discharge
{ Center for Health Statistics Data Sets
{ Vital Statistics
P li A t D t { Police Arrest Data
{ Fatal Accident Reporting System (FARS)
{ Motor Vehicle Crash Data
{ Practice-based Research Networks (PBRNs)
Environmental Data Sets
y Physical Environment
{ US Geological Society: Topographic Geologic Hydrologic { US Geological Society: Topographic, Geologic, Hydrologic
Maps
{ Environmental Protection Agency (EPA)
Air, Water, Soil Conditions, Pollutants
y Built Environment (City Planning Departments)
{ Land Use, Zoning
Sid lk t t li ht { Sidewalks, street lights
y Social Environment
{ Community Assets/Services
{ Businesses, Political Districts, etc.
5
Confidentiality & Other Data Use Issues
y Federal, State, and Local Policies and Regulations
{ HIPPA FERPA State Laws { HIPPA, FERPA, State Laws
y Ownership of Data (City Planning Departments)
y Institutional Review Boards (IRB)
y Community Concerns about Portraying their
Neighborhood in a Bad Light
{ Schools { Schools
{ Realtors
{ Businesses
{ Residents of Public Housing Developments
{ Health Disparities Communities
Easy Access to Data by Communities
y Web Portal
{ Descriptive / Easy to Remember Web Address { Descriptive / Easy to Remember Web Address
{ Home Page Needs to Encourage Browsing
{ Intuitive Interface
{ Academic Partners
Data Sources Cited
Description of Data Sets, Data Collection Methods, Validity, etc.
Able to Download Data Print Maps with Legends etc Able to Download Data, Print Maps with Legends, etc.
{ Community Audience
Relevant / Address Neighborhood Concerns
Non-technical Tools
Minimal Computer Resources (CPU, storage, Screen, etc.)
6
User Friendly Tools
y Academic Researchers
{ Standard Definitions { Standard Definitions
{ Calculate Rates per 100,000
{ Present Results by Frequency, Rates, Charts, Graphs
{ Export/Download Data to Spreadsheets/Database
y Community Partners
{ Map / Visualize Data
{ Access to Relevant Shape Files (Neighborhoods, School Zones)
{ Compare Area of Interest with Other Areas
{ Look at Health within Context of Social/Environmental Data
{ Print Maps
www.IMNashville.com Web Portal
y Established to Promote Academic/Community
Partnerships in Addressing Health Disparities Partnerships in Addressing Health Disparities
y Targeted Nashville/Davidson County
y Six Areas of Health Disparities:
{ Cancer
{ Cardio-metabolic Disease (Obesity, Diabetes, CVD)
{ HIV/AIDS /
{ Infant Mortality
{ Intentional & Unintentional Injuries
{ Mental Health/Substance Abuse
y Period: 2005- Present
7
Youth Violence Pilot
y Population Level Health Outcomes Data
{ Death Certificates { Death Certificates
{ Hospital Discharge (Inpatient & Outpatient)
{ Metro Nashville Police Department Arrests of Youth for
Violent Crimes
{ Youth Risk Behavior Survey (Survey)
y Limitations of Each Data Set
G i i A D { Gaining Access to Data
{ Confidentiality
{ Different Geographic Areas
{ Population Sample
Environmental Data
y Physical Environment
{ Google Earth { Google Earth
{ US Geological Society
y Built Environment
{ Google Earth
{ Metro Nashville Planning Department
Shape Files
i l Greenways, Bicycle Lanes
y Social Environment
{ Community Assets Data Base
8
Community Assets Database
y 211 Taxonomy to Standardize Coding of Services
y Identified Multiple Existing Human Service y Identified Multiple Existing Human Service
Directories
y Contacted Each Agency to Verify Data
y Developed Data Management Structure
y Developed Policies for Updating Data
Pil t T t d D t S t y Pilot Tested Data Set
y Future:
{ Hold Community Forums to Review/Modify Data Base
{ Work with the 211 Agency to Integrate Databases
Data Analyses Tools
y Generate Map by County and Sub-county Shape Files
y Calculate Rates per 100 000 by Age Race & Gender y Calculate Rates per 100,000 by Age, Race, & Gender
y Produce Graphs & Charts
y Print Maps
9
Environmental Mapping
y Community Assets Database
Interactive Mapping
y Health Outcomes
y Environmental Data y Environmental Data
y Shape Files
y US Census Data
y Community Assets
Evanilde Maria Martins
Dental School
Pontifical Catholic University of Minas Gerais
Belo Horizonte, Brazil
0115531-3319-4585, evas@pucminas.br

Renato Csar Ferreira
Dental School
0115531-3319-4169, saudecoletiva@pucminas.br

Joo Francisco de Abreu
Geography Graduate School
0115531-3319-4211, jfabreu@pucminas.br

THE ACCESSIBILITY AND UTILIZATION OF ORAL HEALTH
SERVICES IN BELO HORIZONTE, BRAZIL.

Abstract: The objective of the present study is to assess the
accessibility and utilization of oral health services in Belo Horizonte in
order to verify the effectiveness of the strategies proposed by the
model of oral health utilized, which seeks to be universal, regional
and hierarchical. Secondary data from the Information System of the
Municipal Health Department of Belo Horizonte, oral health database,
year 2000, were used, which consisted of 24,972 entries of dental
service users. The use of Geographical Information Systems allowed
the georeference of 11,354 user domiciles. Maps of the spatial
distribution of those domiciles were produced according to
aggregation levels. The information collected showing the place
where the service was offered and the location of the domiciles was
used to create flow maps. It was verified that in the year 2000, there
were 128 units of primary care, 116 of which provided dental care; 15
of them did not supply information to the database. The total
population seen at these units was classified as follows: 72% were
children and teenagers; 27% were adults and less than 1% was
elderly people. The great majority of the users lived in areas with
higher health vulnerability (66.8%). As to the number of patients
seen, we could observe that in 65 units, 50% or more of the users
lived in the area surrounding the health units; however, in the 36
units located in the more central parts of the city, the great majority of
the users came from different areas. Even though universalization of
the service is attempted, there is greater concentration of care for
children and teenagers. As to the regionalization and maintenance of
the health care focus, we can say the number of users seen in the
health units is high in the surrounding areas of the domiciles.

INTRODUCTION

Belo Horizonte is the capital of Minas Gerais state, Brazil. In the year
2000, the population of the city was approximately 2.3 million. Throughout
the years, the city grew in a disorderly manner without either urban
planning or efficient public policies, which led to an equally disorderly
occupation of the urban space and, consequently, the appearance of the
favelas. This lack of planning explains the coexistence of poverty areas
and areas of greater development and higher quality of life. Such a mosaic
of life conditions challenges the municipal government in its ability to face
the health problems of the population. (Martins, 2007, pp. 76-77)

In its endeavours to approach the public health problem, the City has
followed the principles established by the Brazilian Health System, and
has, in the past decades, worked towards structuring the local health
system. In order to do so, factors such as the living conditions of the
population and the inequalities regarding the manifestation of health
problems, are taken into consideration. When we take planning and the
organization of the attention allotted to health, we find that the city is
divided into several levels of spatial disaggregation: there are nine
sanitary districts which are subdivided in areas of scope of the primary
care health facility.

As to what concerns oral health, great focus has been observed on the
construction of a model based on the expansion of health promotion and
the prevention of oral diseases, as well as on the increase of the
accessibility to dental treatment. This model seeks to place the delivery of
oral health within reach of the population in need of dental treatment. It
was seen that the changes in oral health approaches aligned themselves
with the organizational strategy of the health system, which establishes a
regional and hierarchical network. This network seeks to abide by the
principles of the decentralization of dental services, as mentioned above.

The objective of the present study is to analyze the accessibility and
utilization of oral health services in Belo Horizonte in order to verify the
effectiveness of the strategies proposed by the model of oral health
utilized, which seeks to be decentralized.

METHODS

Secondary data from the Information System of the Municipal Health
Department of Belo Horizonte, oral health database, year 2000, were
used, which consisted of 24,972 entries of dental service users.

The database contain information on users identification, such as age,
sex, address, place where they receive dental treatment and a description
of oral diseases. In order to create the georeference, the program
SISGEO developed by PRODABEL, and which is the property of
Municipal Health Department, was used. SISGEO has been used to
understand the geographical references in cases of morbidity and
mortality observed in the network of health service providers that belong to
the Health System of Belo Horizonte. The program makes it possible not
only to locate the georeferenced point of each domicile, but also to find the
reference of this point in the census track sector and in the area of scope
of the primary care health facility.

The use of Geographical Information Systems allowed the
georeference of 11,354 user domiciles. Maps of the spatial distribution of
those domiciles were produced according to aggregation levels, like
Sanitary District and others. The information collected showing the place
where the service was offered and the location of the domiciles was used
to create flow maps. Toblers approach is used for mapping the flows, and
for the identification of movement patterns of public dental service users.
The software ARGIS 9.2 was used to make the maps, specifically the
softwares application Flow Map Model Tools.

RESULTS

It was verified that in the year 2000, there were 128 units of primary
care, 116 of which provided dental care; 15 of them did not supply
information to the database. The total population seen at these units was
classified as follows: 72% were children and teenagers; 27% were adults
and less than 1% was elderly people.

The great majority of the users lived in areas with higher health
vulnerability (66.8%). As to the number of patients seen, we could
observe that 79% of the users have received dental care in the health
center close to where they live. This is a positive result in terms of
decentralization: we see that only 21% of users looked for treatment away
from home. Figure 1 shows the number of residents in each district who
were seen in the public oral health service. The Distrito Centro-Sul and
Nordeste had the highest number of residences located, whereas the
Barreiro area had the lowest.

FIGURE 1
MUNICIPALITY MAP OF BELO HORIZONTE, SANITARY DISTRICT,
TOTAL OF DOMICILES OF DENTAL USERS OF PUBLIC DENTAL
SERVICE, 2000.

Figure 2 illustrates the eleven most significant movements. The
polygon centroid was used due to the fact that users went to different
public health centers to receive oral treatment. In the Centro-Sul area,
45% of the users seen by the dental health service lived in districts other
that their own. The tendency the central region had to attract more users
is clearly seen on the map. It should be added that the central region is
the oldest and the most urbanized part of the city where the most
structured health services are located.
The greatest flow observed was the one from Distrito Oeste to Centro-
Sul, which seems to be caused by the existence of a big slum in Distrito
Oeste and which has enormous health care demands. Such demands
cause an oversaturation of the oral health service in Distrito Oeste forcing
users to migrate to Distrito Centro-Sul, the nearest one available.

Another important flow observed was the one from the Distrito
Noroeste to the Pampulha area, which could be explained by the proximity
of the health centers and by the easy access to public transportation.

FIGURE 2
MUNICIPALITY MAP OF BELO HORIZONTE, SANITARY DISTRICT,
MOST IMPORTANT FLOWS FOR THOSE SEEKING DENTAL
SERVICES AWAY FROM THEIR DOMICILES, 2000.

DISCUSSION

The access to health services encompasses innumerous factors which
can be discussed through several analysis approaches such as
geographical, cultural, economic, and functional. (Unglert et al.,1987,
pp.439-446). The use of the health system depends on several factors,
among which they mention the ones that have to do with the organization
of the health services: resources available; characteristics of the supply;
compensation possibilities; geographical and social access. Among the
factors pertinent to the individuals, they point out: need of treatment
(seriousness and urgency of the disease); gender; age; social, economic
and cultural conditions. (Travassos and Martins, 2004)

In the year 2000, not all primary health care facilities of SUS-BH
(Sistema nico de Sade de Belo Horizonte, the public municipal health
care system) offered dental services; this was due to the fact that the
public health service network was being structured. Oddly enough, some
of the facilities that did offer dental care never submitted any data for
analysis. (Martins, 2007, p. 103)

Despite the restructuring of the dental services provided by the health
system, which sought the inclusion of the population as a whole,
regardless of age, what was noticed is that dental service is still primarily
given to children and adolescents in school age. The results showed the
adult and elderly population as a minority (18%), and we may say that
they are still not a priority in the dental care units of the health system.
These results allowed some hypotheses to be raised: (i) oral health
services were not organized in a way so as to offer access to the adult
and elderly population; (ii) the solutions offered by the dental treatment
proposed did not meet the necessities of the adult and elderly population
who, in this specific phase of life, demanded more complex treatments
and rehabilitation through prosthetics procedures which were not offered
by the basic health units.

As to the regionalization of the dental service, we can say that the
number of users who seek dental treatment in health units in the vicinities
of their domiciles is significant. The decentralization model is, in fact,
allowing the health service units to offer treatment to the population
residing in their surroundings.

It is imperative to keep in mind that, when access and utilization of
health services are involved, there is a dynamic relation in which two
factors intervene: on the one hand, there are the needs and aspirations of
the population expressed through the demand of services; on the other
hand, the available resources and their technological and organizational
configurations which characterize the supply. (Travassos and Martins,
2004)

CONCLUSION

Even though universalization of the service is attempted, there is
greater concentration of care for children and teenagers. As to the
regionalization and maintenance of the health care focus, what was
observed was that the number of users seen in the health units is high in
the surrounding areas of the domiciles. When users sought an oral health
service from a district other than their own, the movement happened
mostly from the periphery to the center of the city where the most
structured public health services are located.

The importance of flow maps in the process of planning and assessing
the organization strategies in public health has to be underscored, and
efforts to disseminate the practice should be the goal of those involved in
public health administration.

It is common ground that public health decisions involving the control
and prevention of diseases should prioritize groups in greater risk. The
present need to rationalize the allotment of resources to public health as
well as the spending of these resources demands health programs that
are both more effective and more efficient. Such programs can only be
created from careful analyses of a database which make it possible to
spot a) the areas and population sector in greater need; b) the type of
services the population needs; c) the potential location of environmental
risk; and d) the areas where vulnerable social conditions are to be found.
Only then can we have health programs which are efficient and which use
public investment in a rational way.

REFERENCES

Abreu, Joo Francisco, 1982. Migration and Money Flows in Brazil. A
Spatial Analysis. University. of Michigan. Press, Ann Arbor, Michigan.

Bailey, T. C and Gatrell, A. C.,1996. Interactive Spatial Data Analysis. W.
S. Press, N. Y.

Travassos, Cludia e Martins, Mnica, 2004 A review of concepts in
health services access and utilization in. Caderno de. Sade Pblica.
vol.20, pp.190-198.
Martins, Evanilde Maria, 2007. Analysis of the Spatial Distribution of the
Dental Caries in Public Dental Service Users in Belo Horizonte, 2000.
Tese (Doutorado). Pontifical Catholic University of Minas Gerais, Brazil.

SISSON, K. L.., 2007. Theoretical explanations for social inequalities in
oral health in Community Dentistry Oral Epidemiology, v.35, pp.81-88.

UNGLERT, Carmen Vieira de Souza et al., 1987. Access to health
service: a geographical approach to public health in Revista Sade
Pblica, So Paulo, v.21, pp. 439-446.

WALLER, Lance A.; GOTWAY, Carol A., 2004. Applied spatial statistics
for public health data. New Jersey: Wiley.
1

Wheres the Care: Mapping Health Care Providers with the CMS-NPI
by Robert Borchers
(presented June 29, 2011 at the URISA GIS in Public Health Conference, Sheraton
Hotel, Atlanta, GA)

Preface
The expression wheres the care? may express or evoke for some a sense of need or
self pity. It is, in another connotation, a seminal, if obvious, issue for public surveillance
of diseases affecting populations. Care, used here to include both diagnostic and well
as therapeutic services provided by medical service professionals to patients, is
followed by documentation, e.g., records of observations, diagnoses , treatments
provided, prescriptions written, referrals made, billings and claims filed with creditors,
and, ultimately, death certificates executed.
Knowledge of those types of health records which provide useful information for public
health surveillance is alone insufficient. A central cancer registry ( the context from
which these notes are shared) may use most of the aforementioned to find reportable
cases of patients either living or deceased. Trained case abstractors are given the task
of using available records both to find (e.g., through coded billing records) cancer
patients and details concerning their condition and treatment. To be sure, records are
made and kept (at least until archiving) where patients are seen. But where are the
patients seen?

1. Problem
The Wisconsin Cancer Reporting System WCRS) , one of the oldest central registries in
the United States, was established in 1976. Until 1992, records were abstracted
exclusively at hospitals. Approximately two decades ago, non-hospital clinics were
recognized as an efficient channel for case reporting on behalf of individual physicians
(note: much medical care is provided on an outpatient/clinical basis within hospitals).
Since that time, it has been observed that the cancer diagnosis and treatment of some
patients sometimes occurs entirely on an outpatient basis. and for some cancers in
particular, increasingly so. Consider malignant melanoma, a particularly deadly disease
closely followed by epidemiologists and policy makers. A review of cases in the WCRS
2

database, diagnosed 2000 through 2007, indicates that information concerning twenty-
seven percent was reported exclusively by non-hospital sources, notably dermatology
clinics. Hence is recognized the importance of clinic surveillance.
Identifying and communicating with hospitals for a public health disease tracking
program is not especially difficult; they are typically licensed and monitored by
government authorities. This is not the case for non-hospital health service providers in
Wisconsin and many other jurisdictions. While individual practitioners (physicians,
nurses, etc.) serving within such facilities are subject to customary licensing, the
facilities themselves are likely to be operated by organizations established under a
considerable variety of partnership, corporation and other legal arrangements not
explicitly subject to government regulation as health care service providers in the
manner hospitals are. Hence is acknowledged the challenge.
Initially, WCRS staff gleaned clinic names and addresses from classified telephone
directories. This proved useful but very labor intensive. An annual commercial
publication, the Wisconsin Medical Directory (Jola Publications, Minneapolis, MN) was
likewise perused to identify informants for the surveillance program. Its clinics listings,
organized by locality, were more efficiently examined and geocoded. Yet, of the
approximately two thousand clinic listings, clinic names provided the only suggestion
(sometimes fortuitously) of the types of care provided. For more than a decade, health
service providers, large and small, have posted home pages detailing, among other
things, outpatient facilities. While some of these are excellent, providing not only
information about location but also something about particular services provided,
information available thusly is neither consistent nor comprehensive.

2. Toward a Solution: Federal Provider Data
Since May 23, 2007, as required by the Health Insurance Portability and Accountability
Act (HIPAA) of 1996, the National Provider Identifier (NPI) has served as the standard
identifier for health service reimbursements and HIPAA-regulated electronic
transactions. NPI numbers are customarily assigned by the Centers for Medicare and
Medicaid Services (CMS) through online registration. The NPI roster of active health
service providers, presently includes approximately 3.3 million individuals (e.g.,
physicians, nurses) and organization (e.g., hospitals and clinics) registrants. A file of all
current NPI providers, stripped of confidential information and updated several times per
year, is available for free download at a CMS website.

3

A Source
The NPI public dissemination file is a large -- approximately 385 MB -- compressed text
file (uncompressing to almost 4 GB) in comma-separated-values (*.csv) format. While
the format is familiar to many who frequently import it to spreadsheets or databases,
such an import may fail not only because of the large number of records but also on
account of the large number of fields (314) and native field names which can exceed
seventy characters (a longstanding problem with many implementations of the *dbf
format.) e.g., the NPI field name Provider Business Practice Location Address
Country Code (If outside U.S.) Staff at WCRS identified a number of ways to re-format
and label the records to enhance versatility and compactness of the data resource
without loss of information

B Enhancements
The content of the provider records may be divided for improved manageability into
individual (NPI Type 1) and organization (NPI Type 2) and further subdivided into three
sets of fields.
1) Basic administrative and contact information,
2) Functional information (taxonomy codes and related professional licensure),
3) Other provider identifiers and organizational associations)
(The tri-part subdivision and operational renaming of the fields (variables) is set forth as
Appendix 1)
Utilizing SAS, these six tables can be exported variously, e.g., to free-standing *.dbf
files or to tables with a relational database (MS Access 2003 was used by WCRS.)
It was observed that the Function (Func) fields, fifteen sets of four associated fields
and the Other provider (Prov) fields were an attempt to facilitate multiple instances of
these types of attributes within a flat file. But they consumed most of the data storage of
each record - a total of 260 of the fields. To normalize the MS Access 2003 database,
macros were written to create multiple Func and Prov records, for single providers, but
only where data was present in the original, in tables which would be related using the
NPI number key. The net effect is to reduce one table of 314 fields into three related
tables with a total of sixty-six fields between them, a reduction in the number of fields
(e.g., fifteen variables which might hold a taxonomy code of interest) and resultant
efficiency for the database user.
4

(The NPI records reorganized by WCRS into a normalized relational database or
geodatabase is illustrated in Appendix 2)
As the webpage excerpt below indicates, the 820 taxonomy codes used by the NPI to
describe the many health service functions performed by providers includes
considerable detail.

Although there is a fairly neat organization of these service type codes into a three-tier
hierarchy, the text used to search for a code may be exclusively present at any (but not
consistently all) of the hierarchical levels of the classification scheme. A facilitate faster
(and safer) searching for taxonomy codes, a utility was developed (taxonomy helper)
whereby the text from all hierarchical levels of a particular taxonomy would be combined
in a single searchable field,
The first image from the WCRS taxonomy helper utility shows that one would normally
have to be searching the Classification level of the scheme to find physicians with the
urologist specialization. The second image, in contrast, illustrates that the oncologist
and related cancer-care specialist would only be found of the lower (i.e.,
Specialization) field were searched.

5

6

The NPI records are blessed with place data. Separate sets of fields distinguish a
business mailing/contact address and physical location address information. Recently,
the CMS has employed a built-in address validator within its on-line NPI registration
application. As a fortunate consequence to GIS users, a relatively high proportion of NPI
physical location addresses can be matched to standard georeferenced standard
address sets. Approximately 98% of the providers addresses in the fifty states and the
District of Columbia (some providers have foreign or overseas US military locations,) In
addition to the two linkable NPI relational databases (one each for Individual and
organization providers) a separate linkable, single-table geocodes database was
created with output from the geocoding process. The geocodes table serves as a bridge
from provider records to a GIS. In addition to a status field indicating whether Centrus
matched at the street address level, only the zip code, or neither, map-able fields for
each NPI record include latitude and longitude coordinates, and county location (not
otherwise an attribute of the raw NPI data file.). The full set of original location-related
fields remains in the Base files for the providers.
3. Progress
Data made more accessible is not necessarily data of practical value. Yet, the initial
findings at WCRS have proved quite valuable pursuant to the objective of improving the
completeness of its surveillance framework. In addition, it became compellingly
apparent that the NPI may serve public health programs and projects in other ways.

A Disease Tracking: Example of non-hospital cancer care providers
In early 2011, a collaborative effort was initiated between Christian Klaus (North
Carolina Central Cancer Registry) and Laura Stephenson (WCRS) to explore the NPI
records for non-hospital providers of cancer care who were not already in active service
as case reporters to central cancer registries. Among types of medical providers,
oncologists, urologists and dermatologists were of particular interest. The NPI
taxonomies were carefully explored and sifted to detect those particular named
specializations which were ostensible markers of outpatient practices commonly
involved in cancer care.
As already noted, non-hospital dermatology services were observed to play an
especially significant role as exclusive informants to the central registries. In Wisconsin,
prior to this investigation, twenty-six (26) dermatology clinics were listed as active
cancer incidence reporters. Among the NPI records, another twenty-eight (28) were
found. Prostate cancer is sometimes diagnosed and treated (or regarded with watchful
waiting) outside the walls of any hospital. The WCRS, prior to this NPI mining exercise,
7

had listed ten (10) urology clinics as active reporters; the total number more than
quadrupled to forty-one (41).
B Other Opportunities for Public Health Utilization of NPI Data
Surely there are numerous ways in which the NPI data will be found to be useful for
public health programs and projects. Consider, for example:
1) Customized provider directories can be produced from NPI records using nested
database reporting functionality. For example, an elder health organization plans a
study and assessment of the availability of physicians, MDs and DOs specializing in
geriatrics. Two taxonomy codes fit the selection criteria (207QG0300X=Family
Medicine Context 207QGR300X=Internal Medicine Context). An easy MS Access
report, nesting providers by place produces a good starting place for their work. Similar
preparation of provider directories can be done with the standard reporting functionality
of other relational databases or a specialty application such a Crystal Reports.

8

2) Public health workers often find it useful to enlist partners for prevention to
assist with educational programming. A forthcoming series of regional
presentations on prevention of childhood diseases might want to know
practitioners working near the program sites who could be solicited to speak of
their personal experience and concern. The program planner queries the
individual providers taxonomically coded

208000000X MD and DO Pediatricians
236LP0222X Nurse Practitioner in Pediatrics
364SP0200X Clinic Nurse Specialist in Pediatrics

Using the geocoded case selections and either a great circle distance calculation or
distance determined by routing software, an initial speaker candidates roster could be
produced by further specifying a proximity of provider-to-site distance of less than 20
miles or less than 30 minutes.

3) Examination of the area supply of health service providers by type.
Knowledgeable and skilled health services are of little use to patients for whom they are
too distant. Analysis of provider availability by place is a central theme of health
disparities research and a priority of the Patient Protection and Affordable Care Act of
2010 (PPACA). For several years, the Area Resource File (ARF) provided by the
Federal Health Resources and Services Administration (HRSA) has served as a key
resource for health service planners. County-level inventories (counts) of medical
providers and related health service resources are thus available, with county
populations, to facilitate density mapping to expose regional variation in health service
availability.
Geocoded NPI records can mapped with additional utility. Centrus geocoding for the
WCRS working database of NPI providers produced county identifiers for the individual
and organization providers. Those, of course, can be used to map providers along the
lines of the county choropleths typically associated with the ARF provider count data,
County boundaries, however, often do not comport spatially with settlement and
resource distributions, Urban regions and concomitant patterns of patient travel to and
utilization of health care providers may have little correspondence with administrative
9

units. Using hospital discharge data by zip codes, the Dartmouth Atlas Project has
delineated hospital service areas and hospital referral regions (which encompass them.)
Even in the absence of coordinate point locations, NPI providers can be area coded
using Dartmouths zip-area walkthrough tables. In this map, Dartmouths 2008
aggregations of hospital service regions to the more extensive hospital referral regions
is indicated by distinctive coloring.

Recall the importance of dermatologists to the surveillance of melanoma incidence
tracking by the central cancer registries. Planners might query, too, about the availability
of these specialists and the role they may be playing in prevention and early detection.
This map shows quintile rates of dermatologists per 100,000 persons (range: 1-24)

Primary care physicians (PCPs) are slated to play a key role case coordination and
related achievement of quality improvement and cost reduction under the PPACA. The
third map shows quintile rates of PCPs per 100,000 persons (range: 35-195)
10

Many other types of regional analysis may be prompted by the availability of the NPI
provider data. Using central place theory and methods, How might the tools of GIS be
used to sift and winnow the grist of our complex multiple systems of health care
service providers pursuant to supporting a geographically intelligent plan for improved
service delivery in the USA? There is certainly much information within the NPI records
to stimulate the curiosities of medical geographers, public health analysts, and others.

4. Acknowledgements
Helpful project support was provided by Christian Klaus of the North Carolina Cancer
Registry (Raleigh, NC) and Laura Stephenson of the Wisconsin Cancer Reporting
Service (Madison, WI) Geocoding of the NPI records was skillfully and generously
provided by Mary Dabo Brantley and James Tobias of the Centers for Disease Control
and Prevention (Atlanta, GA.)

5. For Further Information Contact
Robert Borchers robert.borchers@wi.gov
Wisconsin Division of Public Health
1 W Wilson St Room 172 Madison, WI 53703
11

References
1) Centers for Medicare & Medicaid Services (CMS) NPI Data Dissemination Website:
http://www.cms.gov/NationalProvIdentStand/06a_DataDissemination.asp
Official Readme documentation:
http://www.cms.gov/NationalProvIdentStand/Downloads/Data_Dissemination_File-
Readme.pdf
Codebook::
http://www.cms.gov/NationalProvIdentStand/Downloads/Data_Dissemination_File-
Code_Values.pdf
File download page: http://nppes.viva-it.com/NPI_Files.html
2) Cromley, Ellen K. and McLaffery, Sarah L. GIS and Public Health Guilford Press,
2002
3) Dartmouth Atlas Project http://www.dartmouthatlas.org/
Downloadable population and GIS resources:
http://www.dartmouthatlas.org/tools/downloads.aspx
4) Klaus, Christian and Stephenson, Laura Mining the National Provider Inventory to
Improve Case Ascertainment: Whos Not Yet in the Reporter Roster? presentation at
the Annual Meeting of the North American Association of Central Cancer Registries
June 21, 2011
5) Health Resources and Service Administration (HRSA) US Department of Health and
Human Services: http://www.hrsa.gov/index.html
Area Resource File (ARF) http://arf.hrsa.gov/purchase.htm
6) Kovner, Anthony K and Knickman, James R (editors) Jonas & Kovners Health
Care Delivery in the United States 8
th
Edition Springer Publishing Company 2005
7) Maheswaran, Ravi and Massimo, Craglia (editors) GIS in Public Heath Practice
CRC Press 2004
8) Shi, Leiyu and Singh, Douglas A. Essentials of the US Health Care System Jones
and Bartlett Publishers, 2005
9) Washington Publishing Company Health Care Provider Taxonomy Code Set:
http://www.wpc-edi.com/content/view/793/1
12

Appendix 1 The Layout of the Raw NPI Records with Original and WCRSAbbreviated Field Names

Seq To
Table Original NPI Field Name Shorter
1
Base
Func
Prov
NPI NPI
2 Base Entity Type Code Type
3 Base Replacement NPI RepNPI
4 Base Employer Identification Number (EIN) EIN
5 Base Provider Organization Name (Legal Business Name) OrgName
6 Base Provider Last Name (Legal Name) Pr LN
7 Base Provider First Name Pr FN
8 Base Provider Middle Name Pr MI
9 Base Provider Name Prefix Text Pr NPref
10 Base Provider Name Suffix Text Pr Nsuff
11 Base Provider Credential Text Pr Cred
12 Base Provider Other Organization Name OOrgName
13 Base Provider Other Organization Name Type Code OOrgNameT
14 Base Provider Other Last Name OPr LN
15 Base Provider Other First Name OPr FN
16 Base Provider Other Middle Name OPr MI
17 Base Provider Other Name Prefix Text OPr NPref
18 Base Provider Other Name Suffix Text OPr Nsuff
19 Base Provider Other Credential Text OPr Cred
20 Base Provider Other Last Name Type Code OPr LNT
21 Base Provider First Line Business Mailing Address M Add L1
22 Base Provider Second Line Business Mailing Address M Add L2
23 Base Provider Business Mailing Address City Name M Add Ci
13

24 Base Provider Business Mailing Address State Name M Add St
25 Base Provider Business Mailing Address Postal Code M Add PoCo
26 Base Provider Business Mailing Address Country Code (If outside U.S.) M Add CoCo
27 Base Provider Business Mailing Address Telephone Number M Add Tel
28 Base Provider Business Mailing Address Fax Number M Add Fax
29 Base Provider First Line Business Practice Location Address P Add L1
30 Base Provider Second Line Business Practice Location Address P Add L2
31 Base Provider Business Practice Location Address City Name P Add Ci
32 Base Provider Business Practice Location Address State Name P Add St
33 Base Provider Business Practice Location Address Postal Code P Add PoCo
34 Base Provider Business Practice Location Address Country Code (If
outside U.S.) P_Add_CoCo
35 Base Provider Business Practice Location Address Telephone Number P Add Tel
36 Base Provider Business Practice Location Address Fax Number P Add Fax
37 Base Provider Enumeration Date EnumDate
38 Base Last Update Date EDUpDate
39 Base NPI Deactivation Reason Code DeacReas
40 Base NPI Deactivation Date DeacDate
41 Base NPI Reactivation Date ReacDate
42 Base Provider Gender Code P Gender
43 Base Authorized Official Last Name AO LN
44 Base Authorized Official First Name AO FN
45 Base Authorized Official Middle Name AO MI
46 Base Authorized Official Title or Position AO Title
47 Base Authorized Official Telephone Number AO Tel
48 Func Healthcare Provider Taxonomy Code 1 Taxo 01
49 Func Provider License Number 1 Lic 01
50 Func Provider License Number State Code 1 LicSt 01
51 Func Healthcare Provider Primary Taxonomy Switch 1 PrTax 01
14

/ Repetition of pattern for sets 2 14 /
104 Func Healthcare Provider Taxonomy Code 15 Taxo 15
105 Func Provider License Number 15 Lic 15
106 Func Provider License Number State Code 15 LicSt 15
107 Func Healthcare Provider Primary Taxonomy Switch 15 PrTax 15
108 Prov Other Provider Identifier 1 OpID 01
109 Prov Other Provider Identifier Type Code 1 OpTy 01
110 Prov Other Provider Identifier State 1 OpSt 01
111 Prov Other Provider Identifier Issuer 1 OpIss 01
/ Repetition of pattern for sets 2 49 /
304 Prov Other Provider Identifier 50 OpID 50
305 Prov Other Provider Identifier Type Code 50 OpTy 50
306 Prov Other Provider Identifier State 50 OpSt 50
307 Prov Other Provider Identifier Issuer 50 OpIss 50
308 Base Is Sole Proprietor IsSole
309 Base Is Organization Subpart OrgSubPt
310 Base Parent Organization LBN ParO LBN
311 Base Parent Organization TIN ParO TIN
312 Base Authorized Official Name Prefix Text AON Pref
313 Base Authorized Official Name Suffix Text AON Suff
314 Base Authorized Official Credential Text AON Cred

Appendix 2 The NPI records within a normalized relational database or geodatabase

Fields in Relational Tables Within MS Access 2003 Databases

NPIBase[1,2] NPIFun_[1,2]
Order Field Name Order Field Name Order Field Name
1
NPI
28
M_Add_Fax
1
NPI
2
Type
29
P_Add_L1
2
Seq
3
RepNPI
30
P_Add_L2
3
Taxo
4
EIN
31
P_Add_Ci
4
Lic
15

5
OrgName
32
P_Add_St
5
LicSt
6
Pr_LN
33
P_Add_PoCo
6
PrTax
7
Pr_FN
34
P_Add_CoCo
8
Pr_MI
35
P_Add_Tel
9
Pr_NPref
36
P_Add_Fax
10
Pr_Nsuff
37
EnumDate NPIPro_[1,2]
11
Pr_Cred
38
EDUpDate Order Field Name
12
OOrgName
39
DeacReas
1
NPI
13
OOrgNameT
40
DeacDate
2
Seq
14
OPr_LN
41
ReacDate
3
OpID
15
OPr_FN
42
P_Gender
4
OpTy
16
OPr_MI
43
AO_LN
5
OpSt
17
OPr_NPref
44
AO_FN
6
OpIss
18
OPr_Nsuff
45
AO_MI
19
OPr_Cred
46
AO_Title
20
OPr_LNT
47
AO_Tel
21
M_Add_L1
48
IsSole
22
M_Add_L2
49
OrgSubPt
23
M_Add_Ci
50
ParO_LBN
24
M_Add_St
51
ParO_TIN
25
M_Add_PoCo
52
AON_Pref
26
M_Add_CoCo
53
AON_Suff
27
M_Add_Tel
54
AON_Cred

7/14/2011
1
Interactive Web-based Mapping:
Bridging Technology and
Data for Health
LI NDA HI GHFI ELD ( PRESENTER)
JUTAS ARTHASARNPRASI T ( PRESENTER)
THOMAS REYNOLDS, PATRI CI A GAI L BRAY,
CECELI A OTTENWELLER,
ARNAUD DASPREZ
Who Are We?
y St. Lukes Episcopal Health Charities
{ Linda Highfield Ph D { Linda Highfield, Ph.D
{ Patricia Gail Bray, Ph.D
y HexaGroup Ltd
{ Jutas Arthasarnprasit
{ Cecelia Ottenweller
{ Arnaud Dasprez
y University of Texas School of Public Health
{ Thomas Reynolds, Ph.D.
2
Overview
y Introduction
y Background y Background
y Objectives
y Technical
Implementation
y Future
y Discussion
Introduction
What are the St. Lukes Episcopal Health Charities?
C d i b E i l Di f T d y Created in 1997 by Episcopal Diocese of Texas and
St. Lukes Episcopal Health System
y Largest public charity focused solely on reducing
health disparities in Houston
y Since 1997, awarded over $83 million in 1,649 grants 997, $ 3 , 49 g
to 365 different non-profit organizations in Houston
and the Diocese of Texas
y Key to Success: Research-informed grant making
3
y 1995 1996: UT Health Science Center and St.
Lukes Hospital promoted CHIS
Portal Background / History
Luke s Hospital promoted CHIS
y 1997 1997: Presentations of ideas and concepts
y 1997: First funding
y 1998: First internet presence
y 1999: First use of interactive mapping and data y 1999: First use of interactive mapping and data
retrieval
y 2011: Launch of CHIS 2.0
M it i Pl i A t
Portal Objectives
h
s
e
Monitoring Planning Assessment
o
m
m
u
n
i
t
y

H
e
a
l
t
h
A
d
d
r
e
s
s

N
e
e
d
s
M
e
a
s
u
r
e

C
h
a
n
g
e
C
o
M 7/14/2011
4
Portal Objectives
Portal Audience
Lay Users
Residents
Health Orgs
FSBOs
Research
Policy makers
Planners
Academics
Research Impact
Introduction
What are the SLEHC Interactive Mapping Portals?
y Web based tools that allow users to access health resource y Web-based tools that allow users to access health resource
and health indicator data
y Interactive portals
{ Query data on the fly
{ Mapping interface support
y Interactive website allows users to y Interactive website allows users to
{ link to community public health resources
{ examine community health profiles
{ conduct population health assessments
5
Introduction
Portal Focus:
y Part of Center for y Part of Center for
Community Based
Research
y Currently focused on
the needs of Greater
Houston Area
{ 10 counties area
{ > 6 mil population
Introduction
The System will
eventually cover the 57 eventually cover the 57
counties of the Episcopal
Diocese of Texas
6
Introduction: DEMO - BHP
Development & Technical: Data Sources
y Vital Statistics (Birth, Mortality, Cancer Incidence)
{ Department of State Health Services p
y Demographic Data
{ US Bureau of the Census
{ State Data Center
{ Commercial
y Community Resources and Assets
{ St. Lukes Episcopal Health Charities { St. Luke s Episcopal Health Charities
{ Health agencies and committees
{ Texas Education Agency Data
{ FQHC
{ Clinic and Health Facility Data
7
The St. Lukes Episcopal Health Charities Interactive
Mapping Portals
Development & Technical: History
y Technical:
{ AutoDesk implementation { AutoDesk implementation
Only viewable in MS Internet Explorer
Provider search via mapping difficulties for Navigators/public
Required Viewer download.
{ Access to Data:
Pre-defined queries only
Multiple large databases Maintenance issues p g
8
Development & Technical: Current
Features:
y On the fly interactive Web based mapping y On-the-fly interactive Web-based mapping
{ One-stop report generation
y Navigator interface with print cart
{ Simplified access to client-specific clinic info
{ Reduced paper use
y Administrative interface Administrative interface
{ CMS based Administration
User groups
Data Types
9
Features:
{ Reduced paper use
User groups
Data Types
10
Features:
{ Reduced paper use
User groups
Data Types
11
Technologies leveraged:
y Microsoft NET Framework 3 5 y Microsoft.NET Framework 3.5
y SQL Server 2008
y Google Map JavaScript API V3
y TelerikRadControls (ASP.NET AJAX)
y jQuery
y HTML
y XML
Architecture Context Diagram
Browser
Map
SQL Server Web App
Mail Server
12
Architecture Flow Diagram
Browser
Map Data
Interactive Map Application
SQL
Server
Request
Data Associate
with Polygon
Browser
1
2
Criteria
Records, sets
& coordinates
4
3
5
6
Web
Server
2
Why Google Mapping?
y Compatibility y Compatibility
{ No 3
rd
party installation
y Cost
{ Free
{ Supports SSL Protocol
13
Why Google Mapping (cont.)?
y Mobile support y Mobile support
{ Google Android
{ iPhone and iPad
y Future component integration
{ Google Public Data Explorer
{ Fastest product development p p
y Large support
{ Google
{ Users & developers community
Development & Technical: Challenges
Polygon loading time
y Problem: Polygon load with each page submittal y Problem: Polygon load with each page submittal
{ Contains all Texas census tracts (4404) & ZIP Codes (2884)
(Portals expected to eventually cover state)
y Solution:
{ Detect upper left and lower right screen coordinates
{ Display / calculate only those polygons in viewing area
14
Consolidation and Standardization of Data
y Problem: Multiple databases y Problem: Multiple databases
{ 12 in original application
{ Consolidate single database
y Solution:
{ Store data definition and field type in database
{ Develop data definition and format for future data p
Interactivity, Usability, Presentation
y Problem: Users expected Windows like functionality y Problem: Users expected Windows-like functionality
in a Web environment, no forms
y Solution:
{ Add report button in hover state
{ Checkbox hover effect
{ (Future) Drag and drop polygons g p p yg
15
Current Implementation Challenges
On-the-Fly Calculations
y Problem: y Problem:
{ Rate and radius calculations increase response time
y Solution:
{ (in BETA) ASP.NET Data caching method
{ (Future) Save frequently used reports
{ (Future) Stand alone app: Calculate and store pp
Future
y Expand statewide
y 2010 Census Data y 2010 Census Data
y Mobile Devices Navigator access
y International applications
y Google Data Explorer
16
Future: Google Data Explorer
Discussion
Questions and comments?
Visit the portal at: http://www.slehc.org/
17
Contact Information (Last Page)
Linda Highfield, Ph.D
Director of Research Director of Research
3100 Main St., Suite 865, Houston, TX 77002
lhighfield@sleh.com
Jutas Arthasarnprasit
Director of Technology Director of Technology
3411 Mt. Vernon, Houston, TX 77007
713-528-2134
jutas@hexagroup.com
1
Spatial Patterns of
Water Insecurity:
Lessons from Accra, Ghana
Jus t i n St ol er , MS MPH PhDc ( San Di ego St at e
Uni ver s i t y)
John R. Weeks , PhD ( San Di ego St at e Uni ver i t y)
Gnt her Fi nk, PhD ( Har var d Sc hool of Publ i c Heal t h)
Al l an G. Hi l l , PhD ( Har var d Sc hool of Publ i c
Heal t h/Sout hampt on, UK)
This research was supported by grant number R01 HD054906
from the National Institute of Child Health and Human
Acknowledgments
from the National Institute of Child Health and Human
Development: Health, Poverty, and Place in Accra, Ghana
(John R. Weeks, Allan G. Hill)
Ghana Water Company Ltd.
GIS Office:
Richard Appiah Otoo, David Nunoo
Council for Scientific and Industrial Research,
Water Research Institute:
Joseph A. Ampofo, Anthony Karikari
2
The water and sanitation problem
y A billion people lack access to an improved
t water source.
y 1.6 million annual diarrheal deaths are related to
unsafe water, sanitation, and hygiene, mostly
among children under 5.
y About half the people in the developing world
suffer from one or more of the main diseases suffer from one or more of the main diseases
associated with inadequate water and sanitation
services.
Why do taps run dry in Accra?
y Overpopulation
{ Infrastructure cant keep up
y Water production limitations
{ Two water treatment plants; more needed
y Rationing of public drinking water
{ Influenced by both geography and income
Response: privatization
3 7/21/2011
4
Accras drinking water infrastructure
y 80% coverage geographically, but 45% with
h d ti house or yard connection
{ Connection disparities: 90% in high-income areas vs.
16% in low-income areas
y Ghana Water Company Ltd (GWCL):
mandate to reduce non-revenue water (NRW)
{ Focus on revenue-generating customers
Inequality of drainage infrastructure
vs.
5
Standpipes rarely flow daily
6 7/21/2011
7 7/21/2011
8
Water storage is often a necessity
Safe water storage is largely absent
y Goal: prevent water
contamination due to contamination due to
unhygienic storage
and handling
{ Keep hands and vectors out!
y Narrow dispensers for
filling; spouts/spigots for
di i dispensing
Do sachets obviate
safe storage needs? 7/21/2011
9
Sample Research Questions and Hypotheses
Who chooses sachets?
y Urban slum residents enduring worse living conditions are y Urban slum residents enduring worse living conditions are
more likely to consume sachets than fellow residents.
Is sachet-consumption neighborhood dependent?
y Residents are more likely to choose sachets if their
neighborhood experiences a higher degree of water rationing.
Might sachets offer a health advantage?
y Children under 5 in sachet using households are less likely to y Children under 5 in sachet-using households are less likely to
experience diarrhea in the previous two weeks.
Data sources and scales of analysis
WHSA HAWS
Vernacular
Neighborhoods
10
Non-geographic survey data
Percentages of households using piped drinking water vs.
sachet/other water as the primary source of drinking water by
DHS survey year for Ghana's ten administrative Regions.
Percentages of households using sachets as the primary source
of drinking water (by enumeration area cluster) within the Accra
Metropolitan Area, from the 2003 and 2008 Ghana DHS. 7/21/2011
11
Variation in mean days per week of running water in the Accra-
Tema Metropolitan Area by water district as recorded by Ghana
Water Company Ltd. in July 2009.
Neighborhood distribution of ATMA sachet filling businesses
registered with the Sachet Water Producers Association in 2010,
shown with GWCL rationing scheme and pipe infrastructure. 7/21/2011
12
Drinking water Quality Factor interpolated for Central Accra,
shown with GWCL pipe diameter and elevation.
Study design: quantitative
y The Housing and Welfare Study (HAWS) 2009-10
{ 2 095 women in 1 276 HHs in 37 EAs { 2,095 women in 1,276 HHs in 37 EAs
{ 813 children in 547 HHs in 37 EAs
Enumeration Area (EA)
Household Household Subject
SES
Measures
Rationing,
Land cover
y Analysis: hierarchical logistic regression
Indiv Indiv Individual
Age,
Ethnicity
13
Study design: qualitative
y Cluster sample drawn for water quality follow-up
(n = 32) (n 32)
{ Source vs. stored water in household
{ Total/fecal coliforms, residual chlorine
y In-depth interviews (n = 5)
{ Water acquisition, storage, use, health
Who uses sachets? Results from HAWS
Days/wk of water rationing **
Elevation
#
% Vegetated land cover
% Soil land cover **
EA
Type of dwelling ***
Source of lighting
Type of toilet access
Type of cooking fuel
#
Type of bathing facility ***
Method of solid waste disposal ***
Method of liquid waste disposal ***
Girls age < 5 in HH
Number of rooms in dwelling ***
Possessions index (0-10)
Daily expenditures (in GHC):
Phone ***
Transportation
Drinking water ***
Other water *
Bathroom
HH
Girls age 5 in HH
Boys age < 5 in HH *
Girls age 5-18 in HH ***
Boys age 5-18 in HH ***
Sex of HH head *
Ethnicity *
Food
Cooking Fuel ***
Monthly electricity expenditures
% of daily exp. for drinking water ***
Age *
*** p < 0.001; ** p < 0.01; * p < 0.05;
#
p < 0.10
Ind 7/21/2011
14
Morans I = 0.16, Z = 3.04
Mean days per week of GWCL water rationing (water shut off) in
37 Enumeration Areas surveyed by the 2009 Housing and
Welfare Study of Accra, Ghana.
Morans I = 0.11, Z = 2.25
Percent of households reporting sachets as the primary source of
drinking water in 37 Enumeration Areas surveyed by the 2009
Housing and Welfare Study of Accra, Ghana. 7/21/2011
15
Is sachet use neighborhood-dependent?
y Sachet consumption is associated with:
{ More days/week of water rationing { More days/week of water rationing
{ Living in worst types of housing
{ Having the least access to sanitation services
{ Having no children 5-18 in the household
{ Higher self-reported overall health
y Only about a third of neighborhood variance can y g
be accounted for; significant variation remains.
Where you live matters for sachet use
Are sachets associated with a health benefit?
y Reported diarrhea is associated with:
{ Non use of sachets as primary water source (!) { Non-use of sachets as primary water source (!)
{ Lower self-reported overall health of mother
{ Having lower household bathroom expenses
y 77% of neighborhood variation can be accounted
for, but only 25% of HH variation.
Child illness may be more influenced by
HH and indiv. factors than by neighborhood
16
Highlights: water quality assessment
y 56% store drinking water, usually in a plastic
t i d f d container, and for one day
y Stored water had consistently higher coliform
levels than the source (piped) samples
y 90% sachet samples tested were clean
Findings consistent with general literature
Highlights: in-depth interviews
y The task of fetching water is viewed as a greater
b d th i f it burden than paying for it
y Drinking water is not considered a major
expense compared to other HH expenses
{ Higher-priced sachet water may be overrated as far as
perpetuating poverty
l d il f h f h y People do not necessarily fetch water from the
closest source; social networks and
transportation logistics are important
17
A Fourth World Contradiction?
y The sachet industry is filling important service
b t ith h i t l t gaps, but with heavy environmental cost
y Sachet policy options: recycling? taxes? bans?
{ Potentially exacerbating the marginalization of the
urban poor due to rationing
y Peri-urban fringe: currently suburbs, but
i d b h b l poised to be the next urban slums?
A holistic understanding of water insecurity

1

Modeling Habitats of Mosquitoes Using Remote Sensing and GIS: Improving
Public Health through Vector Prevention
Brach Lupher, Lillian Reitz, Stacey Lyle, Yuxia Huang, Richard Smith, and Kevin Concannon
GISC Department, Department of Computing Sciences
Texas A&M University Corpus Christi

Abstract
Mosquitoes are the most common epidemiological vector of diseases, such as the West
Nile Virus, and are responsible for spreading maladies to approximately one-fifth of the human
population every year. Farm lands are usually vast areas that offer numerous potential breeding
grounds for mosquito populations to propagate, therefore creating a greater risk to public health.
An effective way to control mosquito populations is to reduce their larval (or breeding) habitats.
This paper reviews current methods used to identify potential locations of mosquito larva on
farm lands though remote sensing and Geographic Information Systems (GIS) with a focus on:
(1) using hydrological models to calculate the wetness index of low-lying areas, a main
environmental condition for mosquito larval habitats, derived from Digital Elevation Model
(DEM) data; (2) Using remote sensing and GIS to create spectrally distinct classifications of
environmental conditions relevant to mosquito larval habitats. This paper also reviews some case
studies on identifying potential mosquito larval habitat and comments on limitations of the
current methods. The research on remote sensing and GIS in identifying potential mosquito
larval habitats will be helpful for public health officials in: (1) confining vector control initiatives
to specific areas; and (2) promoting better methods for the prevention of future vector outbreaks
to the general public. The comprehensive and critical review provided in this paper will help in
identifying significant issues and directing future research in the fields of remote sensing and
GIS in both public health and agricultural management.

1 Introduction
Mosquitoes are the most common epidemiological vector of diseases, such as the West
Nile Virus Malaria, Dengue Fever, and Eastern Equine Encephalitis Virus. Disease spread by
mosquitoes not only infect one-fifth of the human population every year, but also pose health
threats to pets and livestock (Lawler & Lanzaro, 2005). For example, livestock that has been
affected by disease may become less useful, disabled, or die, which may cause economic strains
on the rancher or farmer to replace the animal. Real estate property values can also be affected
by overwhelming mosquito populations especially if there is a high risk of contracting a vector
borne disease in the area (Lawler & Lanzaro, 2005).
To create accurate and cost effective mosquito population models it is important to
understand the mosquito life cycle. Mosquitoes begin as eggs and develop into larva, then pupa
before reaching the adult stage of life. The first three stages of a mosquitos life cycle are spent
in an aquatic environment. The abundance of a mosquito population in a specific area has long
been tied to local climate factors like temperature and precipitation (Hay et al., 1998). The
temperature of the water affects mosquito development. Water that is too cold or warm will slow
or halt the mosquito life cycle (Shaman et al., 2006). Stagnant or standing water is important to
the complete development of mosquitoes; if the water is moving too fast the mosquito will not
survive to adulthood (Lawler & Lanzaro, 2005). The amount of water needed, and the specific
surface wetness and temperature conditions that are ideal for mosquitoes varies across the

2

species such that some species may require exceedingly wet conditions and some species may
thrive in dryer conditions (Shaman & Day, 2005). When the mosquitoes develop into the adult
stage, it is the female mosquito that bites animal or human hosts and hence is the vector of
disease. Ponds, puddles, tree holes, buckets, and depressions in the ground such as a hoof prints
can provide enough water for mosquito larva to develop. Irrigation systems on farm lands can
also provide ideal habitats for mosquito larva development (Lawler & Lanzaro, 2005).
The most common way of controlling mosquitoes, especially in third world countries, is
to use pesticides such as sprays or bednets against the adult mosquitoes (Li et al., 2009). It can be
extremely costly to continuously spray problem areas with pesticide that target primarily the
adult mosquitoes. Ground surveys of an area have been implemented in the past to observe
mosquito habitats and populations. However, ground surveys are time consuming, potentially
expensive, and difficult to conduct in areas with rough terrain. Covering large amounts of land is
also a problem with ground surveys. Since the 1970s, studies have been conducted that allow
for mosquito larval population locations to be predicted or modeled effectively and efficiently
through the use of hydrology models, remote sensing, and Geographic Information Systems
(GIS) (Hay et al., 1998). Determining mosquito larval habitat locations and then focusing
population reduction efforts directly to the water stages of the mosquito life cycle has been
shown to be effective and cost efficient through many independent studies. Studies have also
proven that elimination of potential mosquito habitats, such as reducing stagnant water on farms
and ranches, can greatly reduce the number of adult mosquitoes in these areas (Lawler &
Lanzaro, 2005).
This paper focuses on reviewing current methods used to identify potential locations of
mosquito larva on farm lands through several different methods in economic and efficient
manners. First, the paper discusses different methods used to identify potential mosquito larval
habitat locations, such as using hydrological runoff models for habitat prediction, using remotely
sensed aerial and satellite imagery to classify potential habitats, and then using various GIS
analysis techniques to refine the classifications made from the remote images to perform
statistical analyses and to reduce error and overestimation. Case studies using these different
techniques will then be discussed and reviewed in detail.

2 Methods for Larval Habitats of Mosquitoes

2.1 Using a Hydrological model to calculate the wetness index of low-lying areas
Originally, hydrology models and GIS developed independently of each other. Hydrology
models were more capable of representing models and analysis, but were unable to handle the
massive amounts of data processing and data management that a GIS could (Sui & Maggio,
1999). Starting in the late 1980s, an integration of hydrology models and GIS were researched.
There are four major ways of combining a hydrology model and a GIS; embedding a GIS into an
existing hydrological model, embedding a hydrological model into an existing GIS, loose
coupling, and tight coupling (Sui & Maggio, 1999). Each approach provides different advantages
and disadvantages and no one approach can be elected as the most efficient or effective, however
the combination of hydrological modeling and GIS have allowed for both platforms to expand.
When first modeling hydrological processes, especially in response to climate changes,
many land surface models viewed a soil column as the fundamental hydrologic unit (Stieglitz
et al., 1997). Stieglitz et al. argues that using the soil column as the fundamental hydrologic unit
is not the best way to represent soil moisture of an area and goes on to suggest a new model

3

based off of TOPMODEL (1997). TOPMODEL (a TOPography based hydrological MODEL)
created by K.J. Beven and M.J. Kirkby is a physically based model conceptual model that
focuses on contributing area variables, topography, and distance from saturation zones (Beven &
Kirkby, 1979; Franchini et al., 1996). Quinn et al. describe TOPMODEL as working through
the use of simple distribution empirical functions (1995). TOPMODEL is described as a set of
conceptual tools, not just a static model for one purpose, which is simple to use for many
different hydrological simulations (Beven, 1997). Stieglitz et al. create a model that incorporates
TOPMODEL equations to more effectively take into account the topography of an area and
presents the watershed as the fundamental hydrologic unit (1997). The study performed by
Stieglitz et al. provided a crucial platform to future hydrology models.
Current hydrological models focus on the movement of surface water over the land and
can be used to predict where the water might pool and create potential mosquito breeding
habitats. Mosquitoes breeding is critically influenced by the abundance of a suitable habitat for
the female mosquito to deposit her eggs (Shaman et al., 2002). When modeling water it is
important to understand the flow rate and temperature of the water because these factors can
directly influences mosquito egg, larva, and pupa development (Shaman et al., 2006).
Understanding how water moves on land after precipitation or flooding is crucial to create an
accurate prediction model of possible mosquito larval habitats. Precipitation is a large factor that
influences the amount water that pools in an area, but other factors that also affect surface
wetness of an area. The surface wetness of an area is a good indicator of whether an area is
indeed suitable for mosquito oviposition. The greater amount of water pools, ponds, and slow
moving streams, the greater chance of certain mosquito species using these locations for
breeding (Shaman et al., 2006). Typically a valley will have higher moisture content than higher
elevation regions (Shaman et al., 2002; Stieglitz et al., 1997). The topography of an area,
therefore, greatly influences the movement of moisture through a specific area.
Digital Elevation Models (DEM), referred to as topographic skeletons by Lawrence
Band, accurately model the topographic features of the land at a fine scale (Band, 1986). The use
of DEM data was not originally viable to early hydrology models, but the combination of the
GIS processing powers with hydrology models allowed for DEMs to be used to represent the
topographic features of an area (Sui & Maggio, 1999). The majority of current hydrological
models used to predict mosquito larval habitats use DEM data to model the land. It is extremely
important to have accurate representations of the land to accurate model the movement and flow
of water.
When modeling water movement and surface wetness it is important to not only
incorporate precipitation data, but to also incorporate data such as (but not limited to) :
temperature, prior wetness, soil type, land use, vegetation, topographic, evapotranspiration rate,
and humidity data (Shaman & Day, 2005; Shaman et al., 2002). Although most hydrology
models are created for a specific area or problem, dynamic hydrology models allow for
adjustments to be done to individual models to make the models affective over a wide range of
regions (Shaman et al., 2002). As a result, each model will be slightly different as to what
specific data is wanted or needed for the model to run accurately.

2.2 Using remote sensing and GIS to create spectrally distinct classifications of
environmental conditions relevant to mosquito larval habitats.
According to Cracknell and Hayes, remote sensing can be defined as the observation of,
or gathering information about, a target by a device separated from it by some distance (1991).

4

Of the many types of sensors available, terrestrial and satellite-multispectral sensors "measure
radiation from various regions in the electromagnetic spectrum (EMS), emitted, reflected, or
scattered from natural objects" (Hay et al., 1998). Because natural objects emit, reflect, or scatter
radiation differently throughout the EMS, they can be characterized with a 'spectral fingerprint'
(Hay et al., 1998).
False-color composite imagery is a type of remotely sensed imagery captured by a
satellite multi-spectral sensor producing several images, each corresponding to separate spectral
bands in the EMS (Cracknell & Hayes, 1991). Within this type of imagery, a wealth of
information exists that can be extracted from differences in reflectivity, as a function of
wavelength, of different objects on the ground (Cracknell & Hayes, 1991). Data from multiple
spectral bands can then be combined for further visual interpretation and analysis (Cracknell &
Hayes, 1991).
The interpretation of remote sensing imagery has been proven to be a useful tool in the
prediction of mosquito larval habitat locations since its first investigative use by the National
Aeronautics and Space Administration (NASA) in identifying larval habitats of the salt marsh
mosquito Aedes sollicitans from color infrared (CIR) aerial photography in 1971 (Hay et al.,
1998). Remote sensing imagery has been used to determine landscape features and climatic
factors (e.g. pond dynamics, vegetation coverage, turbidity (Lacaux et al., 2007)) associated with
the risk of vector-borne diseases and other factors affecting mosquito abundance and breeding
sites (Mushinzimana et al., 2006). The use of false-color composite imagery has been used to
identify typical mosquito larval habitats represented as mosaics of several dark pixels, or water,
adjacent to red pixels, or vegetation (Zou et al., 2006). In addition, Lacaux et al. mention turbid
bodies of water appear as a cyan color in false-color composite imagery (2007).
One of the most important pieces of environmental data used in remote sensing
classification are spectral vegetation indices (SVIs) which are a type of environmental index
derived from false-color composite imagery (Hay et al., 1998) used to identify vegetative areas
with a particular chlorophyllian activity (Lacaux et al., 2007). SVIs exploit the fact that healthy
vegetation has a low reflectance in the visible red spectrum because photosynthetic pigments in
plant tissues absorb such light and reflect strongly in near-infrared (NIR) as the structure of
mesophyll tissue reflects radiation at these wavelengths. The most commonly used SVI is the
Normalized Difference Vegetation Index (NDVI) (Hay et al., 1998), an indicator for density and
vigor of green vegetation and a proxy for rainfall (Lacaux et al., 2007), is a bounded ratio
equation of values between -1 and +1 where bare ground, active photosynthesized vegetation,
and water can be detected based on particular ranges of radiation wavelength values (Hay et al.,
1998). Water bodies can be identified using various image-derived environmental indices (e.g.
NDVI), flooding indices (FI) (Zou et al., 2006), spectral band ratios such as combinations of near
infrared (NIR) and red wavelengths or red and green wavelengths (Lacaux et al., 2007).
The classification process, in general, is performed either by a manual visual
interpretation of shades of color from false-color composite imagery or digitally with the help of
a computer (Cracknell & Hayes, 1991). Environmental indices are layered on top of remote
sensing imagery to contribute toward classifying areas of interest in the imagery (e.g. ponds) into
spectrally distinct, classified areas (Zou et al., 2006) using one of two major types of
classification: unsupervised or supervised. The classification method used by Mushinzimana et
al. involved performing a supervised classification to identify land-use and land-cover types
(2006), while the classification Lacaux, Tourre, Vignolles, Ndione & Lafaye used involved also
taking into account water turbidity because water bodies used for domestic and cattle needs

5

become muddy and sediment-filled and their associated radiometric responses behave like bared
soils (2007). To further simplify the process of interpreting natural features, Zou et al. reduced
the number of classifications by combining one or more spectrally-similar classified areas
together such as grassland and shrub areas (2006).
In addition to using hydrological modeling and remote sensing classification methods to
facilitate the identification of potential mosquito larval habitats, the use of GIS analysis has also
been well studied, such as spatial analysis of larval habitat locations (Zou et al., 2006) and
geostatistics of mosquito populations (Trawinski & Mackay 2008). Trawinski and Mackay used
geostatistical methods to quantify and model spatial dependence in mosquito populations and
make predictions for unsampled locations (2008). The purpose is to show that mosquito
abundance is spatially correlated and that spatial dependence differs between different mosquito
species. Similarly, Li et al. quantified spatial patterns of larval habitats using statistical methods
such as dispersion pattern analysis (a measure of how habitats are dispersed around a geographic
center), and compact analysis (the quantification of shape and range of habitats) (2009).

3 Case Studies

3.1 Using a Hydrological model to calculate the wetness index of low-lying areas
Shaman, Stieglitz, Stark, Le Blancq, and Cane recognized that the majority of public
health agencies do not have abundant resources available to do complete sampling and
monitoring of mosquito populations over a large area (2002). In their paper, Using a Dynamic
Hydrology Model to Predict Mosquito Abundances in Flood and Swamp Water, the authors
address the public health agencies deficiencies by creating a detailed dynamic hydrology model
to predict flood and swamp water mosquito abundances (Shaman et al., 2002). An accurate
prediction model can allow for public health agencies to more efficiently predict mosquito
populations over large areas with less cost than other mosquito prediction options. The
subsequent model (a TOPMODEL variation) created in this study takes topographic variability
into consideration and addresses the influence topography has over soil wetness heterogeneity
and water runoff with respect to the possible mosquito habitats of the area (Shaman et al., 2002).
The goal of this study and model was to allow for public agencies to be able to use the
predictions to launch vector control before the mosquitoes reach the adult stage of life. The study
focused on two main study areas: 1) seven individual sites near Pequest River in Jersey, and 2) a
single site in the Great Swamp National Wildlife Refuge in New Jersey (Shaman et al., 2002).
Mosquitoes were collected from the two study areas (Pequest River and Great Swamp National
Wildlife Refuge) for thirteen years and fifteen years, respectively. It is important to note that the
Great Swamp National Wildlife Refuge site was not influenced by the use of vector control in
the area. Mosquito collection was accomplished with New Jersey light traps and each trap
location was recorded with a GPS receiver. The mosquitoes were separated into individual
species and counted after collection. Meteorological data was collected from the National
Climate Data Center (NCDC) in Allentown Pennsylvania and solar radiation data was provided
by the North-east Regional Climate Center (NRCC) (Shaman et al., 2002). DEM data of the
study sites were downloaded from the U.S. Geological Survey (USGS) website.
Two components, physical and empirical, were considered while creating the model
(Shaman et al., 2002). First, the physical component was addressed by using the dynamic
hydrology model to recognize surface wetness areas that could conceivably support mosquito
larvae (Shaman et al., 2002). Using the hydrology model is considered to be a physical

6

component because it uses physical variables from the study site like topography, soil type, and
humidity, to determine surface wetness heterogeneity. Second, the surface wetness prediction
created by the model was then empirically related to the abundance of the mosquitoes (Shaman
et al., 2002). As a result a logistic regression model was created to link surface wetness to
anticipate or projected mosquito species populations. Logistic regression models are often used
in analysis when modeling relationships between two or more variables, an outcome variable and
independent variables (Hosmer & Lemeshow, 2000). The final model is able capture saturated
areas expanding and decreasing inside of a spatial frame work (Shaman et al., 2002).
As discussed previously in this paper, surface wetness is not solely determined by
precipitation, but is also influenced by many other factors. The dynamic hydrology model
developed for this study included precipitation, temperature, meteorological data (evaporation,
humidity, surface pressure, wind speed, long-wave radiation, and solar radiation) soil properties,
vegetation, prior wetness condition data along with the topography of the area to determine
surface wetness distribution (Shaman et al., 2002). Each of the variables listed above has
influences the hydrology models results. Shaman, Stieglitz, Stark, La Blancq, and Cane, state
that the model created can produce not only a daily time series, but an hourly time series of
variables such as: mean water table depth (WTD), percent surface saturation, and total surface
run-off (2002). By determining the WTD, possible mosquito habitat locations can be determined.
Time series of daily mean catchment WTD were generated and used to create indices of local
wetness (ILWs), which were then used for statistical analysis to estimate the association between
surface wetness and mosquito species prevalence (Shaman et al., 2002).
The authors concluded that the study suggests the model may not be the best way of predicting
predicting mosquito abundance, but the model did create a probabilistic forecast for mosquito
populations (Shaman et al., 2002). This case study is an example of how hydrologic models can
aid in the prediction of mosquito larval population for the use of public health agencies to
combat vector-borne disease.

2.2 Using remote sensing and GIS to create spectrally distinct classifications of
environmental conditions relevant to mosquito larval habitats.
The Powder River Basin of north central Wyoming, an area with many small water
bodies resulting from significant coalbed methane gas extraction, have created aquatic habitats
that have potential support for mosquito larval development of the mosquito species, Culex
tarsalis, a primary vector species in the state of Wyoming. The objective of this study was to
assess potential larval habitats in an effort to establish a basis for predicting risk exposure to
West Nile Virus. Authors Zou, Miller, and Schmidtmann used remotely sensed imagery to
identify mosquito larval habitats in order to promote cost-effective vector control operations in
nearby regions (2006).
The authors used Landsat TM data as base imagery for larval habitat assessment because
it offered spectral data for bands 4 and 5 of the EMS, or infrared and mid-infrared respectively,
which was deemed suitable for vegetation and water content analysis (Zou et al., 2006).
Mosquito larval habitats, recognized as mosaics of dark pixels (water) adjacent to red pixels
(vegetation), were identified from false-color composite Landsat TM images (Zou et al., 2006).
Areas of interest (AOI) were delineated by selecting pixels in known larval-positive sites (e.g.
ponds) (Zou et al., 2006).

7

In addition to six bands of Landsat TM data, four variables (e.g. NDVI and FI) were
derived and stacked with the original Landsat TM images because it was believed that these
variables would contribute to the image classification process (Zou et al., 2006).
An unsupervised classification using the Iterative Self-Organizing Data Analysis Technique
(ISODATA) method was used to generate four classes of AOIs extracted from the sites testing
positive for Cx. tarsalis larvae during field sampling operations (Zou et al., 2006). The authors
compressed five classes into three: water, dense vegetation, and grasslands and shrubs (Zou et
al., 2006).
The authors identified larval habitats of Cx. tarsalis in the classified image as dense
vegetative areas immediately adjacent to small water bodies and then extracted these potential
larval habitats from the classified image to meet this criterion (Zou et al., 2006). Final (refined)
habitats were generated using rule-based modeling according to knowledge from experts and the
authors field experiences (Zou et al., 2006). In August 2004, a field study was undertaken to
identify Cx. tarsalis larval habitats from training sites using sampling coordinates recorded using
a geographic positioning system (GPS), which were subsequently transformed into a GIS data
layer (Zou et al., 2006). Samples of mosquito larvae were collected along pond edges of the
training sites and sent to a nearby USDA laboratory where they were sorted by species (Zou et
al., 2006). Training sites were classified as positive if Cx. tarsalis larvae were found during
sampling (Zou et al., 2006). Positive sites were subsequently overlaid with the Landsat TM
image and used to extract appropriate spectral signatures (Zou et al., 2006).
The classified results were compared against digital orthophoto quarter-quadrangles
(DOQQs) of the area for accuracy assessment purposes (Zou et al., 2006). Small ponds in the
DOQQ were digitized by hand to serve as a proxy for ground truth (Zou et al., 2006). Spectral
reflectance curves of the five classes classified by the ISODATA algorithm were condensed into
three distinct classes: water, dense vegetation, and grassland and shrubs. Based upon field
observations, the authors considered dense vegetation areas immediately adjacent to small water
bodies as primary larval habitats of Cx. tarsalis and therefore used a high pass 30-m filter to
identify these pixels from the imagery (Zou et al., 2006).
The authors deemed it necessary to refine areas of potential Cx. tarsalis larval habitats using GIS
analysis techniques to solve the problem of overestimation, or improper classification and field
identification caused by differences in scale of various remote sensing imagery datasets used
(Zou et al., 2006). Based on an error matrix of classification results, it was observed that small
water bodies (less than two acres) had a greater chance of misclassification due to limitations of
imagery pixel size (Zou et al., 2006). These misclassified water bodies were refined by
performing a secondary classification accomplished by setting a threshold of exceedance in the
panchromatic band (which has a higher spatial resolution of 15m) followed by implementing a
rule set which filtered out a subset of dense vegetation pixels deemed to be areas of mixed water
and riparian vegetation (Zou et al., 2006). The authors stated that successful identification of
potential larval habitats was dependent on separating Cx. tarsalis larval habitats (based on spatial
proximity, size of water and dense vegetation areas) from other water and vegetation features,
and concluded that their classifier could efficiently create a spatial distribution of Cx. tarsalis
larval habitats at a large spatial scale deemed suitable for regional-scale vector control
management (Zou et al., 2006).

8

4 Conclusion
As stated before, hydrological models can be used to aid in the prediction of mosquito
habitats and mosquito populations. Hydrological models have a few advantages over other
methods of mosquito larval habitat prediction methods. Proxy data such as NDVI is not used in
hydrological models (Shaman et al., 2002). Hydrological modeling is also not hindered or
limited by weather conditions that might slow a process that uses aerial imagery to predict
mosquito larval habitats (Shaman et al., 2006). Dynamic hydrological models are also easily
integrated with local and global climate models, allowing for real time predictions to be made
(Shaman et al., 2002; Shaman & Day, 2005). Limitations of hydrological models may occur
when weather patterns across a study area are monitored from a weather station that is not
located nearby or when the scale at which mosquito populations can be monitored does not
match the scale at which weather conditions can be monitored (Shaman et al. 2002).
The interpretation and classification of remote sensing imagery has been shown to be a
fairly reliable method of detecting potential mosquito larval habitats since its first purported use
by NASA in 1971 (Hay et al., 1998). Some of the benefits of unsupervised classification
methods include the advantage of being easy to apply by both image processing and statistical
software packages (Xie et al. 2008; Hay et al. 1998) and the ability to automatically convert raw
image data into useful information so long as higher classification accuracy is achieved (Xie et
al., 2008). However, a disadvantage exists for unsupervised classification methods in that the
classification process must be repeated if new data (samples) are added (Xie et al., 2008). On the
contrary, the addition of new data (samples) has no impact on the classification process of
supervised classification methods and the classification process therefore does not have to be
repeated (Xie et al., 2008). A particular disadvantage of both types of classification is that
different vegetation types may possess similar spectral reflectance values which degrade the
accuracy of classification results (Xie et al. 2008; Zou et al. 2006). However, the accuracy of
results from a classification process can be improved if image enhancement is performed
emphasizing or sharpening particular image features prior to the classification process (e.g.
image pre-processing) (Xie et al., 2008). Despite limitations caused by low-resolution spatial
imagery and meteorological variables (Hay et al., 1998) such as cloud coverage, prior research
has shown that this method is useful for relatively large-scale operations whether it be used to
create disease transmission prediction maps (Lacaux et al., 2007) or for use by public health
officials for vector control operations (Zou et al., 2006).
In addition, GIS analysis is important for examining spatial autocorrelation of larval
habitat locations and mosquito population dynamics. According to Trawinski & Mackay, a
better understanding of vector population structures in space is vital to improving knowledge of
population dynamics and ability to control these populations (Trawinski & Mackay 2008;
Srividya 2002; Ryan et al. 2004; Diuk-Wasser et al. 2006). It is important to have a detailed
knowledge of the spatial distribution patterns in mosquito larval habitat distributions so that
vector control efforts can be focused to primary mosquito breeding sites which would in return
make the GIS analysis method more cost effective and widespread in practice (Li et al., 2009).

9

References
Band, L.E. (1986). Topographic partition of watersheds with digital elevation models. WATER
RESOURCES RESEARCH, 22(1), doi: 10.1029/WR022i001p00015
Beven, K. (1997). Topmodel: a critique. Hydrological Processes, 11, 1069-1085.
Beven, K.J., & Kirkby, M.J. (1979). A physically based, variable contributing area model of basin
hydrology. Hydrological Sciences Journal, 24(1), doi: 10.1080/02626667909491834
Cracknell, A.P., & Hayes, L.W.B. (1991). Introduction to remote sensing. Bristol, PA: Taylor & Francis
Inc.
Franchini, M., Wendling, J., Obled, C., & Todini, E. (1996). Physical interpretation and sensitivity
analysis of the topmodel. Journal of Hydrology, 175, 293-338
Hay, S.I., Snow, R.W., & Rogers, D.J. (1998). From predicting mosquito habitat to malaria seasons using
remotely sensed data: practice, problems and perspectives. Parasitology Today, 14(8), 306-313.
Hosmer, D.W., & Lemeshow, S. (2000). Applied logistic regression, second edition. Danvers, MA: John
Wiley & Sons, Inc.
Lacaux, J.P., Toure, Y.M., Vignolles, C., Ndione, J.A., & Lafaye, M. (2007). Classification of ponds from
high-spatial resolution remote sensing: application to rift valley fever epidemics in
senegal. Remote Sensing of Environment, 106, 66-74.
Lawler, S., & Lanzaro, G. (2005). Managing mosquitoes on the farm. Manuscript submitted for
publication, Division of Agriculture and Natural Resources, University of California, Oakland,
California.
Li, L., Bian, L., Yakob, L., Zhou, G., & Yan, G. (2009). Temporal and spatial stability of anopheles
gambiae larval habitat distribution in western kenya highlands. International Journal of Health
Geographics, 8(70).
Mushinzimana, E., Munga, S., Minakawa, N., Li, L., Feng, C., Bian L., Kitron U., Schmidt C., Beck L.,
Zhou G., Githeko A.K., Yan G. (2006). Landscape determinants and remote sensing of
anopheline mosquito larval habitats in the western kenya highlands. Malaria Journal, 5(13).
Quinn, P.F., Beven, K.J., & Lamb, R. (1995). The ln(a/tanp) index: how to calculate it and how to use it
within the topmodel framework. Hydrological Processes, 9, 161-182.
Shaman, J., & Day, J.F. (2005). Achieving operational hydrologic monitoring of mosquito borne disease.
Emerging Infectious Diseases, 11(9), 1343-1350.
Shaman, J., Stieglitz, M., Stark, C., Le Blanq, S., & Cane, M. (2002). Using a dynamic hydrology model
to predict mosquito abundances in flood and swamp water. Emerging Infectious Diseases, 8(1), 6-
13.
Shaman, J., Spiegelman, M., Cane, M., & Stieglitz, M. (2006). A hydrologically driven model of swamp
water mosquito population dynamics. Ecological Modeling, 194, 395-404.
Stieglitz, M., Rind, D., Famiglietti, J., & Rosenzweig, C. (1997). An efficient approach to modeling the
topographic control of surface hydrology for regional and global climate modeling. Journal of
Climate, 10, 118-137.
Sui, D.Z., & Maggio, R.C. (1999). Integrating gis with hydrological modeling: practices, problems, and
prospects. Computers, Environment and Urban Systems, 23, 33-51.
Trawinski, P.R., & Mackay, D.S. (2008). Spatial autocorrelation of west nile virus vector abundance in a
seasonally wet suburban environment. J Geogr Syst, 11, 67-87.
Xie, Y., Sha, Z., & Yu, M. (2008). Remote sensing imagery in vegetation mapping: a review. J Plant
Ecol, 1(1), 9-23.
Zou, L., Miller, S.N., & Schmidtmann, E.T. (2006). Mosquito larval habitat mapping using remote
sensing and gis: implications of coalbed methane development and west nile virus. J. Med.
Entomol., 43(5), 1034-1041. 7/15/2011
1
Predicting Larval Habitats of
Mosquitoes on Farm Lands Using
Remote Sensing and GIS in South Texas
UNDERGRADUATE STUDENTS:
THOMAS LUPHER
LI LLI AN REI TZ
FACULTY MENTORS :
DR. S TACEY L YL E
DR. L UCY HUANG
MR. RI CHARD S MI TH MR. RI CHARD S MI TH
DR. KEVI N CONCANNON
Texas A&M University Corpus Christi
2
Conrad Blucher Institute of Surveying and Science
GISC Programs
y Bachelors of Science, Geographic Information Science
{ Geomatics
{ Geographic Information Systems
y Masters of Science, Geospatial Surveying Engineering
{ Geographic Information Systems
y Post-Baccalaureate Certificates
Post Baccalaureate Certificates
{ Geomatics
{ Geographic Information Systems 7/15/2011
3
Grant Sponsor
USDA U-Search Program
Project Summary
y Five TAMUCC students chosen to conduct research in one
of three study areas y
1. Predicting mosquito larval habitats
2. Integrating RTK-GPS to create automated tractors
3. Auditing farm production using geospatial and remote sensing
technologies
y Research guided by GISC faculty members
| Dr Stacey Lyle
| Dr. Stacey Lyle
| Dr. Lucy Huang
| Mr. Richard Smith
| Dr. Kevin Concannon (Dept. of English) 7/15/2011
4
Predicting Mosquito Larval Habitats: Project Goals
y First Stage (Dec. 2010- June 2011):
{ Acquire a sufficient amount of mosquito sample data from South { Acquire a sufficient amount of mosquito sample data from South
Texas counties
{ Gain an understanding of current methods of predicting mosquito
larval habitats
{ Write a review paper on current mosquito larval prediction methods
y Second Stage (July 2011-July 2012): g y y
{ Use gained knowledge to develop a mosquito larval habitat
prediction model to be used in agricultural areas throughout South
Texas
Tasks for Stage One
y Define study areas in South Texas
y Acquire GIS boundary data for farms/ranches y Acquire GIS boundary data for farms/ranches
y Contact public departments for sample data
{ Daily mosquito counts
{ Arbovirus detection
{ Mosquito trap locations
{ Trapping/Prevention methods used pp g/
y Read/Review existing academic studies focusing on
mosquito larval habitat prediction models or
mosquito population prediction models
5
Mosquito Life Cycle
y of the mosquito life
cycle is aquatic
{ Egg Larva Pupa
Adult
y Typical habitat:
{ Stagnant water Stag a t wate
{ Warmer temperature
{ High vegetation coverage
http://www.osceola.org/mosquitocontrol/129-6426-0/mosquito_life_cycle.cfm
Mosquito Larval Habitat Prediction Methods
1. Hydrology models
2. Aerial and Satellite image classification
{ Using GIS spatial analysis techniques to refine
classification
6
Hydrology Models
y Hydrology run-off models are used to simulate surface wetness
{ Track the movement of water through soil and atmosphere
{ Can be used to simulate surface pooling
y Physical vs. Conceptual
{ Physical models reflect field measurements as parameters
{ Conceptual models uses simplified schemes to emulate total flow
y TOPMODEL
O h b d h d l O { TOPography based hydrology MODEL
{ Combination of physical and conceptual model
{ Simple model that can be updated or used for a variety of studies
{ Calculated topographic indices
ln(a/tan) a=upslope contributing area and tan=local slope angle
Connecting Hydrology Models to Mosquito Habitat Prediction
y Assimilation of hydrology models and climate models
{ Real-time Meteorological data g
{ Hourly and Daily data including:
Precipitation, air temperature, humidity, surface pressure, wind speed,
and radiation data
{ Allows for surface water variation to be tracked at a rate that can
affect mosquito larval development
y Functions within an algorithm can be used to represent g p
mosquito life cycle
{ Aquatic development
{ Mortality rate
If then 7/15/2011
7
Case Study
y Using a Dynamic Hydrology Model to Predict Mosquito Abundances in
Flood and Swamp Water
J l E i I f i Di { Journal: Emerging Infectious Diseases, 2002
{ Authors: J. Shaman, M. Stieglitz, C. Stark, S. Le Blancq, M. Cane
y Data used included:
{ Vegetation, soil type, meteorological, and DEM data
y Logistic regression model created included 10-day lag
{ Lag representative of average mosquito larval development time
R lt y Results:
{ Regression analysis confirmed a correlation between surface wetness and
mosquito abundance
{ The model created provided a probabilistic forecast of mass increase in
mosquito species population
Digital Elevation Model
(DEM) of study area
Land surface wetness at a
single point in time
(Shaman, Stieglitz, Stark, Le Blancq, & Cane, 2002) 7/15/2011
8
Advantages Advantages Disadvantages Disadvantages
Hydrology Models
y Does not use proxy
data
y Not affected by cloud
cover
y Can incorporate local
y Scales
y Distances between
study area and
meteorological station
y Mosquito population of y Can incorporate local
and global climate
models
y Mosquito population of
study area assessed
only through adult
mosquitoes
Method 2: Classification of
Remote Sensing Imagery
9
Typical Larval Habitat Locations
y Mosquito larvae thrive in areas of standing water and
vegetation vegetation
y Identify and classify these areas from RS imagery
Source: L. Zou, S. Miller, E. Schmidtmann, 2006
Spectral Reflectance Curves
10
Data Sources
y Remote Sensing Imagery:
{ Landsat 5 Thematic Mapper (TM) (30m) { Landsat 5 Thematic Mapper (TM) (30m)
{ Landsat 7 ETM+ (30m)
{ Ikonos (1m)
{ Color Infrared Digital Orthophoto Quarter Quadrangles
(DOQQs) Aerial Imagery (1m)
{ Panchromatic Aerial Imagery
y Spatial Data:
{ Digital Elevation Model (DEM) (30m)
{ National Land Cover (NLCD) (30m)
{ Major Hydrography Features
Method 2: Classification of Remote Sensing Imagery
y Two Types of Classification
{ Unsupervised { Unsupervised
{ Supervised
y Environmental Variables derived from RS Imagery
{ e.g. Vegetation indices, flooding index
y Pixels from RS imagery classified into spectrally-distinct
l h t ti l it l l h bit t
Source: L. Zou, S. Miller, E. Schmidtmann, 2006
classes where potential mosquito larval habitat areas
identified as water surrounded by dense vegetation
y Potential larval habitat locations refined using GIS
analysis methods and can be assessed using DOQQs
11
Case Study
Journal: Mosquito Larval Habitat Mapping Using Remote Sensing and
GIS: Implications of Coalbed Methane Development and West Nile Virus
A h L Z S Mill E S h id { Authors: L. Zou, S. Miller, E. Schmidtmann
{ Source: Journal of Medical Entomology, 43(5):1034-1041. 2006.
Study Area: Northern Wyoming
Problem: Substantial increase of water bodies and potential
mosquito larval habitats (primarily Culex tarsalis) as a result of
i d lb d h i i increased coalbed methane extraction operations
Solution: Map these potential larval habitat locations using
RS classification and GIS analysis techniques to help local
vector control authorities prioritize their operations
Case Study: (Simplified) Classification Workflow
Landsat
TM Image
C. tarsalis
positive site data
Derive:
Environmental
variables
Stack Variables:
on Landsat TM image
Unsupervised
and Supervised
Classifications
Rule Set:
Ponds < 10 acres,
Slope < 5,
Not connected to
major streams
New 3-class
Landsat TM Image
Assessment:
Compared with
DOQQ i
GIS Layers:
DEM,
Land cover,
Water features
Combine GIS layers
and Apply rule set
Refined:
C. tarsalis
larval habitats
Landsat TM Image
DOQQ imagery 7/15/2011
12
Case Study: Results
y Their classification process identified potential larval
habitat locations to an accuracy of 70% habitat locations to an accuracy of 70%
y Results showed a 75% increase in potential larval
habitats from 1999 to 2004 in the study area,
primarily due to a large increase of small ponds as a
result of coalbed methane extraction
Case Study: Results
75% increase
Classified larval habitats of C. tarsalis in 1999 and 2004 inside Landsat TM
coverage areas (Sheridan, Johnson, and Campbell counties, Wyoming).
Source: L. Zou, S. Miller, E. Schmidtmann, 2006 7/15/2011
13
Conclusion
y As the most common epidemiological vector of
disease it is important to: disease it is important to:
{ Understand the mosquito life cycle and their habitats
{ Accurately predict possible mosquito larval habitats
{ Recognize early signs of possible mosquito-borne disease
outbreaks
H d l d l I l ifi ti t h i y Hydrology models, Image classification techniques,
and GIS analysis are effective and efficient ways of
predicting mosquito populations
Future Goals
y Use knowledge gained from extensive research to:
{ Better understand the mosquito habitats on farms in South { Better understand the mosquito habitats on farms in South
Texas
{ Create our own mosquito habitat prediction model
{ Validate our model with real-time data
14
Questions
Lillian Reitz lreitz@islander.tamucc.edu
Thomas Lupher tlupher@islander.tamucc.edu
PROTECTI NG THE ENVI RONMENT THROUGH
GI S
PRESENTED BY XAVI ER V. DAVI S
GI S SPECI ALI ST/ FOG PROGRAM
COORDI NATOR
PEACHTREE CI TY WATER AND SEWERAGE
AUTHORI TY
PCWASA FOG Program
About Peachtree City, Georgia
Founded in 1959.
Has over 35,000 residents
as of 2006.
Also known to be a golf
cart community.
Considered one of the best
places to retire.
In July 2009, Ranked 8
th
on its list of the 100 Best
Places to Live in the
United States.
About Peachtree City Water and Sewerage
Authority (PCWASA)
Publicly Owned and Operated
Utility since 1996. (Prior to
1996, Privately Owned by
Georgia Utilities)
Responsible for collection and
treatment of public,
commercial, and industrial
wastewater within the City of
Peachtree City.
Potable Water Supply
Infrastructure owned and
operated by Fayette County
Water System (not operated by
PCWASA).
Purpose of PCWASA FOG Program
Assist in preventing FOG (fats, oils, and grease)
blockages and Sanitary Sewer Overflows (SSOs) from
occurring in PCWASA Wastewater Collection System
Main purpose is to manage commercial and
residential generated FOG.
Overview of PCWASA FOG Program
Consist of FOG Program Coordinator who
administers and directs FOG program.
Ensure FOG generators are in compliance with
Sewer Use Ordinance.
Educate general public on commercial and
residential FOG problems.
What is FOG?
Derived from living cells of animal and vegetable
matter.
Generated by restaurants, cafeterias, residential
homeowners, etc.
Used as an aid in food preparation.
Produced during food preparation.
Why is FOG a problem?
Causes blockages in sanitary sewer which could later
result in SSOs.
SSOs in return are hazardous to the environment.
Blockages can result in foul odors and drain back-
ups in residential and commercial establishments.
SSOs can lead to property damage, increased
cleanup costs and civil penalties.
Picture of FOG in Sanitary Sewer
Pipe
Source: http://griffinfog.com/
How GIS enhances PCWASA FOG Program
Used as a problem analysis tool to identify possible
root causes of FOG blockages and SSOs.
Assists PCWASA GIS Specialist in geocoding grease
trap locations.
Aids PCWASA FOG Program Coordinator in grease
trap inspection planning and routing.
GIS Maps used to identify possible
root causes of FOG Blockages and
SSOs
FOG Backup Identification Map
SSO Identification Map
How GIS assists PCWASA GIS
Specialist in geocoding Grease Trap
Locations
What is a Grease Trap?
A plumbing device
designed to intercept
greases and solids before
they enter the sanitary
system.
Inside Grease Trap
Usually installed in facilities
that have small amounts of
FOG.
Located in delicatessens,
smoothie shops, etc.
Typically ranges from 40 to
100 lb.
Pictures of Outside Grease Trap
(Before & After)
Outside Grease Trap
Normally located
underground
Typical sizes are 1000, 1500
and 2000 gallons
Usually installed at
restaurants that produce
heavy FOG (fast food, ethnic,
seafood, steakhouse, etc.)
Also installed at grocery
store and school locations
Basic Grease Interceptor Diagram
What is Geocoding?
A GIS operation that converts street addresses into
spatial data.
This spatial data is displayed as features on a map.
Street address information is referenced in GIS from
the street data layer.
In order to geocode, an address locator is required.
What is an Address Locator?
A dataset in GIS which stores address attributes.
Translates non-spatial descriptions of places into
spatial data (i.e. street addresses.)
Contains snapshot of the reference data used in
geocoding to match addresses.
Create Address
Locator
Geocode Addresses
Create Address Locator
Address Locator
Style US Streets
Address Locator
Reference Data
GIS Roads Layer
Field Map
Left Odd Street Numbers
Right Even Street Numbers
Road Name Data
Output Address Locator
Create file name and file
location for address locator.
Click OK
Geocode Address
Input Table Grease
Trap Locations
Spreadsheet
Input Address
Locator Street
Address Locator
created previously.
Output Feature Class
Geocode grease
trap locations on
map.
Click OK
Geocoded Grease Trap Locations Map
How GIS aids PCWASA FOG
Program in Grease Trap Inspection
Planning and Routing
Grease Trap Inspection Planning
Based on one or more of the following:
Grease Trap Location
Grease Trap Cleaning Date
Grease Trap Inspection Date
GIS Tools Used in Grease Trap Inspection Planning
and Routing
Using ArcReader Map
Find Tool
My Places Tool
Find Route Tool
To find grease
trap location,
click Find icon
Type in
Restaurant Name
and click Find
button
Then click Add
to My Places
Under Tools,
click My Places
Right-click on
Restaurant
Location
Select Add to
Route as Stop
To Find Route,
click Tools, then
click Online
Services
Click Find
Route
Select View
Directions to
obtain directions
Select Create
Printable
Version to print
directions
Map Routing Information for Grease Trap
Inspections
PCWASA FOG Program - Fayette
County (GA) Environmental Health
Department Connection
PCWASA FOG Program - Fayette County (GA)
Environmental Health Department Connection
The PCWASA FOG Program receives every month
information from local Environmental Health Department
of new Food Preparation Commercial Facilities new to
Peachtree City.
Local Environmental Health Department informs these
new facilities to contact PCWASA for grease trap
installation approval.
With information from both local Environmental Health
Department and new Food Preparation Commercial
Facilities, the PCWASA FOG Program is able to properly
inspect and approve new grease trap installations.
Conclusion
In this presentation, you learned how GIS is a useful
tool in the following areas:
Problem analysis to identify possible root causes of FOG
blockages and SSOs.
Assistance for PCWASA GIS Specialist to geocode grease trap
locations.
Used as an aid to assist PCWASA FOG Program Coordinator in
grease trap inspection planning and routing.
FOG Information
For more information on FOG, please check the
following web sites:
www.georgiafog.com(Georgia FOG Alliance)
www.wef.com(Water Environment Federation)
Questions?
Thank You!
Contact Information
Xavier V. Davis
Peachtree City Water & Sewerage Authority
1127 Highway 74 South
Peachtree City, GA 30269
Office: 770-487-7993
Mobile: 678-618-7648
Fax: 770-631-5380
E-Mail: XavierD@pcwasa.org
7/7/2011
1
More than a dot More than a dot
h h on a map: the on a map: the
geography of geography of
Brownfields Brownfields and and
health health health health
University of Illinois University of Illinois Urbana UrbanaChampaign Champaign
Marilyn O Ruiz Marilyn O Ruiz moruiz@illinois.edu moruiz@illinois.edu and William Brown and William Brown
University of Illinois University of Illinois Chicago Chicago
Serap Serap Erdal Erdal, Sean , Sean Crudgington Crudgington and Margaret and Margaret Sietsema Sietsema
Presented at the 3rd
URISA GIS for Public
Health conference.
29 June 2011
Atlanta, GA
Background Background
zz Focus on the health effects of Focus on the health effects of brownfields brownfields on on
neighborhoods and changes in health over time with neighborhoods and changes in health over time with
brownfield brownfield redevelopment redevelopment
zz Project Project title title Best Management Practices and Benefits Best Management Practices and Benefits
of Sustainable Redevelopment of Brownfield of Sustainable Redevelopment of Brownfield Sites Sites
zz Funded by a USEPA Funded by a USEPA Brownfields Brownfields Training, Research Training, Research yy g, g,
and Technical Assistance Grant and Technical Assistance Grant
zz Project lead by The Institute for Environmental Project lead by The Institute for Environmental
Science and Policy (IESP) at the University of Illinois at Science and Policy (IESP) at the University of Illinois at
Chicago Chicago
7/7/2011
2
What is a Brownfield? What is a Brownfield?
Version 1 Version 1 official official
zz DEFINITION OF BROWNFIELD SITE DEFINITION OF BROWNFIELD SITE Section 101 of the Section 101 of the zz DEFINITION OF BROWNFIELD SITE DEFINITION OF BROWNFIELD SITE Section 101 of the Section 101 of the
Comprehensive Environmental Response, Compensation, Comprehensive Environmental Response, Compensation,
and Liability Act [commonly known as Superfund] of 1980 and Liability Act [commonly known as Superfund] of 1980
(42 U.S.C. 9601) is amended by adding at the end the (42 U.S.C. 9601) is amended by adding at the end the
following: following:
zz EPA definition: With certain legal exclusions and EPA definition: With certain legal exclusions and
additions, the term ` additions, the term `brownfield brownfield site means real property, site means real property, additions, the term additions, the term brownfield brownfield site means real property, site means real property,
the expansion, redevelopment, or reuse of which may be the expansion, redevelopment, or reuse of which may be
complicated by the presence or potential presence of a complicated by the presence or potential presence of a
hazardous substance, pollutant, or contaminant. hazardous substance, pollutant, or contaminant.
http://www.esaphaseone.com/epadefinitionofabrownfieldsite/
What is a Brownfield? What is a Brownfield?
V 2 V 2 more descriptive more descriptive
zz Brownfields Brownfields are abandoned idle or underused commercial are abandoned idle or underused commercial zz Brownfields Brownfields are abandoned, idle or underused commercial are abandoned, idle or underused commercial
or industrial properties, where the expansion or or industrial properties, where the expansion or
redevelopment is hindered by real or perceived redevelopment is hindered by real or perceived
contamination. contamination.
11
zz Brownfields Brownfields vary in size, location, age, and past use vary in size, location, age, and past use they they
can be anything from a five can be anything from a fivehundred acre automobile hundred acre automobile
assembly plant to a small, abandoned corner gas station. assembly plant to a small, abandoned corner gas station.
11
zz A A brownfield brownfield is largely what a community identifies as a is largely what a community identifies as a
brownfield. brownfield.
22
1. From: http://dnr.wi.gov/org/aw/rr/rbrownfields/index.htm#bf
2. Paraphrased from meeting with brownfield staff at ATSDR, Chicago, Nov. 2010
7/7/2011
3
Estimated 400,000 Estimated 400,000 brownfields brownfields nationwide nationwide
here is a sample of the distribution here is a sample of the distribution
From: US Council of
Mayors report
Recycling Americas
Land: National
Report on
Brownfields
Redevelopment,
2008
Objectives Objectives
zz Study area: Cook County, IL (home of Study area: Cook County, IL (home of zz Study area: Cook County, IL (home of Study area: Cook County, IL (home of
Chicago), population approximately 5.2 million in Chicago), population approximately 5.2 million in
2010 2010
zz Is the health of neighborhoods near Is the health of neighborhoods near brownfields brownfields improved after improved after
brownfield brownfield remediation? remediation?
zz Are health indicators in neighborhoods near Are health indicators in neighborhoods near brownfields brownfields different different
from those farther from from those farther from brownfields brownfields? ?
zz How do past land use and site contamination vary by site type and How do past land use and site contamination vary by site type and
how are these associated with neighborhood characteristics? how are these associated with neighborhood characteristics?
zz Characterize neighborhoods with historic and current census data, Characterize neighborhoods with historic and current census data,
mortality data from vital records, cancer registry and other health mortality data from vital records, cancer registry and other health
data as available. data as available.
7/7/2011
4
What and where are What and where are brownfields brownfields? ?
geographic data perspective geographic data perspective
Need an unbiased and well structured database to carry out Need an unbiased and well structured database to carry out Need an unbiased and well structured database to carry out Need an unbiased and well structured database to carry out
analysis of health effects of analysis of health effects of brownfields brownfields on neighborhoods on neighborhoods
and change over time with cleanup. and change over time with cleanup.
zz Spatially representative of study region Spatially representative of study region
zz Representative of all types of Representative of all types of brownfields brownfields
zz Have enough points to carry out empirical analysis and Have enough points to carry out empirical analysis and
consistent information about each site consistent information about each site consistent information about each site consistent information about each site
zz Need to have the same data on all sites related to timing of Need to have the same data on all sites related to timing of
site contamination and cleanup, the nature of the past use site contamination and cleanup, the nature of the past use
and the contaminants present and the contaminants present
Search for the Search for the brownfield brownfield data data
zz What data are available What data are available
options considered options considered
zz What methods have other What methods have other
researchers used to represent researchers used to represent
brownfields brownfields for analysis? for analysis?
zz What we decided to do What we decided to do
7/7/2011
5
US EPA US EPA Brownfields Brownfields program program
http://www.epa.gov/brownfields/index.html
The US EPA Assessment, Cleanup and The US EPA Assessment, Cleanup and
Redevelopment Exchange System Redevelopment Exchange System ACRES ACRES
includes data on includes data on brownfield brownfield projects funded with projects funded with
EPA grants EPA grants EPA grants EPA grants
7/7/2011
6
U.S. Environmental Protection Agency
Geospatial data access
http://www.epa.gov/enviro/geo_data.html
ACRES sites ACRES sites
in Cook County in Cook County
N = 50 in Cook
and surrounding
counties. N = 42
in Cook alone.
Compare with Compare with
Chicagos
estimate of 500
brownfields in
2008 survey
7/7/2011
7
Is there a Is there a
Brownfield Brownfield
l i f l i f
From BROWNFIELD.shp
Date: unknown
Agency: unknown
Purpose: unknown
layer in one of layer in one of
the local the local
municipal GIS municipal GIS
databases? databases?
N = 74
http://www.epa.state.il.us/land/database.html
7/7/2011
8
Site
Remediation
Program (SRP)
d t i l d data includes
thousands of
records for sites
funded state
wide.
What data have been used to represent What data have been used to represent
brownfields brownfields in other studies? in other studies?
zz Vacant and underused parcels > 1 acre from Baltimore Vacant and underused parcels > 1 acre from Baltimore zz Vacant and underused parcels, > 1 acre, from Baltimore Vacant and underused parcels, > 1 acre, from Baltimore
City Planning Dept records. N=480. City Planning Dept records. N=480. Litt Litt & Burke 2002 & Burke 2002
zz Projects undertaken through the Charlotte Brownfield Projects undertaken through the Charlotte Brownfield
program. N=76. Chilton, Schwartz & Godwin 2009. program. N=76. Chilton, Schwartz & Godwin 2009.
zz Massachusetts Department of Environmental Protection Massachusetts Department of Environmental Protection
parcels identified as parcels identified as brownfields brownfields in Worcester, MA. N=485. in Worcester, MA. N=485. pp
Brill 2009. Brill 2009.
zz USEPA ACRES data for Lowell, MA. N=30. Personal USEPA ACRES data for Lowell, MA. N=30. Personal
communication. communication.
7/7/2011
9
Our Approach Our Approach
zz Focus on SRP sites from Illinois EPA Focus on SRP sites from Illinois EPA
O l id i h di d O l id i h di d h i d h i d zz Only consider sites that were remediated Only consider sites that were remediated have received at have received at
least on No Further Remediation letter least on No Further Remediation letter
zz Select sites to include a representative sample of sites in Cook Select sites to include a representative sample of sites in Cook
County. All areas, include slightly more larger sites. Stratified County. All areas, include slightly more larger sites. Stratified
random approach. random approach.
zz Select 90 sites to give a large enough number for empirical Select 90 sites to give a large enough number for empirical
analysis but small enough to allow for a more detailed analysis but small enough to allow for a more detailed
characterization of HISTORY and CHEMICAL contamination of characterization of HISTORY and CHEMICAL contamination of
each site. each site.
zz Dry cleaners were about 1/3 of all sites, so 60 sites were selected Dry cleaners were about 1/3 of all sites, so 60 sites were selected
with stratified random sampling design, without dry cleaners with stratified random sampling design, without dry cleaners
and then dry cleaners selected randomly from that subset. and then dry cleaners selected randomly from that subset.
Map of SRP brownfield sites
In Cook County there are 1454 unique IEPA
IDs (from database downloaded on March
10, 2010) with at least one NFR letter
1656 NFR records were listed for those
sites
Map of stratifying divisions. Brownfield
data for 4 of the 8 divisions were first
examined closely for review:
Div 1=NNW Chicago;
Div 3=W Chicago;
Div 7= Southside; and
Div 8=East Chicago/Lakeside.
7/7/2011
10
FOIA request to
Illinois EPA for all
records on the 93
selected sites
93 sites
Use records to
characterize history
and contamination
Link information to Link information to
parcels in Cook
County parcel
database.
1,412,949 parcels
Site Records Site Records
zz PDF files with all letters and reports on PDF files with all letters and reports on pp
the referenced site. the referenced site.
zz Not searchable. Voluminous. Exact Not searchable. Voluminous. Exact
content is site specific. Some on content is site specific. Some on
microfiche. microfiche.
zz Developed data scheme for Developed data scheme for zz Developed data scheme for Developed data scheme for
zz Historic use Historic use
zz Level and type and of contamination Level and type and of contamination
7/7/2011
11
Among the
documents are
the primary site
evaluation
reports
The site legal
description and
historic use
sections are
most useful for
geography and
history data.
GOAL of the
property owner
No Further
Remediation
notice is
recorded with
the property the property
title.
7/7/2011
12
Our data Our data Historic Use Historic Use
PIN10 the property appraisers number for each parcel
DB_ID LPC# for all parcels associated with the original site, even if they are not
part of the SRP.
NAMESITE the name given by the SRP program
ParcelAddress street address of the parcel
ParcelCity City of the site
Area_Acr Acres in the parcel as measured by the GIS
Site_Type SRP = SRP site; SRB_OBA = SRP site that is in the OBA funding
program; SRP_Aux = Part of the original site but not in the SRP program
Use_1 Dirty use prior to redevelopment.
Use 1SYr Year when Use 1 started Use_1SYr Year when Use 1 started
Use_1EYr Year when Use 1 ended
Use_2 Current use
Use_2SYr Year when Use 2 started
Use_3 Other important use other than Use 1
Use_3SYr Year when Use 3 started
Use_3EYr Year when Use 3 ended
ParcelNFRDate date of final NFR letter for this parcel
Establish the geography of the Establish the geography of the
site from the parcel file site from the parcel file
Galewood Railroad
Yard
7/7/2011
13
3 examples of 3 examples of brownfield brownfield sites sites
zz Walts automotive Walts automotive gas station with leaking underground gas station with leaking underground
storage tank storage tank
zz Former Former Flexweld Flexweld site site plant making steel pipes plant making steel pipes
zz Barrie Park Barrie Park manufactured gas plant manufactured gas plant
Example 1 Example 1 Walts Automotive Walts Automotive
59 pdf records
Gas Station 1940 1973
Automotive Repair 1974 1990
Current use Bank
7/7/2011
14
Walts is a LUST site Walts is a LUST site owner is selling, needs owner is selling, needs
to prove that soil is clean to get NFR on title to prove that soil is clean to get NFR on title
Example 2 Example 2 Former Former Flexweld Flexweld Site Site
Village of Bartlett g
redeveloping area
into new Town
Center.
Flexweld is a manufacturing
l f fl bl l plant of flexible steel piping.
Volatile organic compounds
(VOC) of concern. Also PCBs,
lead and arsenic and others.
Center of property near original
buildings Plant was operational
on the site from about 1970 to
1999.
7/7/2011
15
Historic air photos
(below) 193841 and
other records indicate
other use prior to
Flexweld in 1970, but this
C t f it
earlier use is not clarified
in the IEPA documents
Center of site was
more contaminated
location of former
plant (above)
Example 3 Example 3 Barrie Park Barrie Park
Former Manufactured
Gas Plant from late
1800s to 1930. Plant
razed and property
transferred to Village of
Oak Park. Became a
park in 1965 park in 1965.
7/7/2011
16
Our original sites included 2
residential parcels near the park
Barrie Park
93 sites 93 sites uses found uses found
Some sites
have more
than one use
7/7/2011
17
Observations Observations
zz The location and characteristics of SRP The location and characteristics of SRP brownfield brownfield site site
does not always represent the original contaminated use does not always represent the original contaminated use does not always represent the original contaminated use does not always represent the original contaminated use
(i.e. the (i.e. the brownfield brownfield issue) issue)
zz The size, shape and intensity of past use varies The size, shape and intensity of past use varies
zz Large original parcels subdivided into smaller lots Large original parcels subdivided into smaller lots
zz Parcels combined to form larger lot Parcels combined to form larger lot
zz Shape of parcels can be compact or dispersed Shape of parcels can be compact or dispersed
zz The point locations provided in the Illinois EPA database The point locations provided in the Illinois EPA database
are generally very good are generally very good they tend to be located at the they tend to be located at the
center of the area of concern center of the area of concern
zz Characterization of the neighborhood around the site is Characterization of the neighborhood around the site is
affected by its size and shape affected by its size and shape
Neighborhood effect differences Neighborhood effect differences
300 mbuffer 300 mbuffer
around point
and around
parcelbased
area
7/7/2011
18
Future analysis Future analysis
zz Multi Multilevel approach. level approach. zz Multi Multi level approach. level approach.
zz Site characteristics Site characteristics
zz Size and shape of site Size and shape of site
zz Level and type of contamination Level and type of contamination
zz Neighborhood characteristics Neighborhood characteristics
zz Vital records mortality at address level Vital records mortality at address level
zz Census Censusbased housing and socio based housing and socioeconomic economic zz Census Censusbased housing and socio based housing and socioeconomic economic
zz Other possible data Other possible data blood lead levels, crime, cancer blood lead levels, crime, cancer
zz Other contaminated facilities / sites Other contaminated facilities / sites
zz Changes over time and differences between places Changes over time and differences between places
Challenges Challenges
zz Contamination levels and types is very difficult Contamination levels and types is very difficult some is some is
off offsite site off off site site
zz Dates of clean up not always clear Dates of clean up not always clear
zz Not all history is available in FOIA records Not all history is available in FOIA records how much how much
more time to take on this? more time to take on this?
zz Vital Records research requests suspended due to lack of Vital Records research requests suspended due to lack of q p q p
personnel at Illinois Department of Public Health personnel at Illinois Department of Public Health
7/14/2011
1
Physical and Social Environmental
Impacts on Obesity and Risk Factors
in the U.S.
AKI HI KO MI CHI MI , PH. D.
GEOGRAPHI C I NFORMATI ON SCI ENCE CENTER OF EXCELLENCE
SOUTH DAKOTA STATE UNI VERSI TY
Purpose
y To investigate the influences of physical and social
characteristics of obesity in the conterminous U.S. characteristics of obesity in the conterminous U.S.
Metro vs. nonmetro counties
Risk factors: lower levels of physical activity, fruit/vegetable
consumption
Project I
Mapping spatial patterns of obesity & risk factors
Project II
Head-banging Smoothing
Project II
Supermarket accessibility
Project III
Recreational/outdoor amenities
Head banging Smoothing
Before After
7/14/2011
2
Conceptual Model of Obesity
Environment
Spatial
Pattern
Individuals
Age/Sex
Race/
Ethnicity
Natural
Social Income
y
Education
Behavior
Physical
Activity
Food
Consumption
Physical
Activity
Food
Consumption
Economic
y p
Health Outcomes
Obesity
y p
Obesity
Biology/
Genetics
Data
Project I Smoothed Maps
y 2000-2006 BRFSS (Behavioral Risk Factor
Surveillance System) Surveillance System)
State-based telephone surveys conducted by the CDC
Self-report on weight, height, food intakes, etc.
Sample size: 1,802,676 civilian non-institutionalized adults 18+
years living in the conterminous U.S.
% obese (BMI 30)
% participating leisure time physical activity (PA) in past month
% participating leisure time physical activity (PA) in past month
% consuming fruit & vegetables (F/V) 5 or more times a day
http://www.cdc.gov/BRFSS/
7/14/2011
3
Methods
y Weighted Head-banging algorithm (Mungiole et al. 1999)
Weighted median derived from neighboring counties Weighted median derived from neighboring counties
Results
Obesity Raw Data Obesity Smoothed Data
% Obese (BMI >30)
% Obese (BMI >30)
% Obese (BMI >30)
0.000 - 0.190
0.191 - 0.218
0.219 - 0.240
0.241 - 0.258
0.259 - 0.280
0.281 - 0.313
0.314 - 0.663
STATES
0.088 - 0.211
0.212 - 0.225
0.226 - 0.237
0.238 - 0.246
0.247 - 0.255
0.256 - 0.271
0.272 - 0.445
STATES
Prevalence, %
Metro: 21.3 [21.2, 21.5]
Nonmetro: 24.6 [24.4, 24.8]
Source: Michimi and Wimberly. (2010). Spatial patterns of obesity and associated risk
factors in the conterminous U.S. American Journal of Preventive Medicine, 39(2), e1-e12.
7/14/2011
4
%Obese (BMI >30)
0 088 0 211
Pearsons Correlation
Coefficient
Obesity & PA r = - 0.62 *
Obesity & F/V r = - 0.37 *
0.088 - 0.211
0.212 - 0.225
0.226 - 0.237
0.238 - 0.246
0.247 - 0.255
0.256 - 0.271
0.272 - 0.445
STATES
% Obesity - Smoothed
* p < .0001
% Physical Activity
0.793 - 0.912
0.770 - 0.792
0.751 - 0.769
0.734 - 0.750
0.713 - 0.733
0.692 - 0.712
0.578 - 0.691
%Fruit/Veg 5+ Day
0.262 - 0.352
0.237 - 0.261
0.223 - 0.236
0.214 - 0.222
0.203 - 0.213
0.187 - 0.202
0.136 - 0.186 (blank - no data) % F/V - Smoothed
% PA - Smoothed
Source: Michimi and Wimberly. (2010). Spatial patterns of obesity and associated risk
factors in the conterminous U.S. American Journal of Preventive Medicine, 39(2), e1-e12.
Project II
Mapping Supermarket Accessibility
y Purpose
Assess supermarket accessibility in the conterminous U S Assess supermarket accessibility in the conterminous U.S.
Population-weighted mean distance to supermarket at the county
level
Urban vs. rural food deserts
7/14/2011
5
Project II
Mapping Supermarket Accessibility
Hypotheses
1.The population-weighted mean distances to supermarket are
positively related to obesity prevalence and negatively related to
fruit and vegetable consumption at the national level
2.The associations of obesity and fruit and vegetable consumption
with distance to supermarket vary between metro and nonmetro
p y
areas, reflecting the unique characteristics of urban vs. rural food
deserts
Data
Project II - Mapping Supermarket Accessibility
y Obesity and F/V consumption 5+ per day from the
BRFSS BRFSS
y Population-weighted ZIP Code Tabulation Area
(ZCTA) centroids
y U.S. Census Bureau ZIP Code Business Pattern
Supermarkets/Supercenters (SM)
Size of SM by the number of employees
y p y
1-19 : Small
20-49 : Medium
50 and more : Large
y Calculating pop-weighted mean distance to SM
7/14/2011
6
Sioux Falls
Pierre
Rapid City
Sioux Falls
Pierre
Rapid City
(a) (b)
ZCTA Centroids Containing Small SMSCs
ZCTA Centroids Containing Medium SMSCs
ZCTA Centroids Containing Large MSMCs ZCTA
ZCTA
ZCTA Centroids Containing Small SMSCs
ZCTA Centroids Containing Medium SMSCs
ZCTA Centroids Containing Large MSMCs
131 - 145
County
Large/Medium SMSCs
87 - 102
102 - 116
116 - 131
44 - 58
58 - 73
73 - 87
0 - 15
15 - 29
29 - 44
(c) (d)
Sioux Falls
Pierre
Rapid City
0 - 635
636 - 783
784 - 947
948 - 1144
1145 - 1510
1511 - 7214 ZCTA Centroids Containing Large/Medium SMSCs
Sioux Falls
Pierre
Rapid City
2.5 - 6.0
6.0 - 10.9
10.9 - 14.0
14.0 - 20.8
20.8 - 33.7
33.7 - 43.6
43.6 - 95.1
ZCTA Centroids Containing Large/Medium SMSCs
Source: Michimi and Wimberly. (2010). Associations of supermarket accessibility and obesity and fruit and vegetable consumption in the conterminous United
States. International Journal of Health Geographics, 9, 49.
Mean Distance (km)
0.03 - 4
4 - 5.7
5.7 - 7.1
7.1 - 8.5
8.5 - 10
10 - 14.4
14.4 - 99.9
Metro Counties
STATES
Large SMs only
Large, medium, or small SMs Large or medium SMs
Source: Michimi and Wimberly. (2010). Associations of supermarket accessibility and obesity and fruit and vegetable consumption in the conterminous United
States. International Journal of Health Geographics, 9, 49.
7/14/2011
7
Methods
y Generalized linear model for logistic regression:
Obesity Obesity
F/V consumption
Distances to SM as covariates
Other demographic variables to control (e.g. age, sex, SES)
M t d t d l d t l
Metro and nonmetro modeled separately
Results
Metro Nonmetro
OR 95% CI OR 95% CI
Obesity
Distance to large SM 1.24 [1.19, 1.29]** 1.00 [0.94, 1.06]
F/V
Distance to large SM 0.95 [0.93, 0.99]* 0.99 [0.88, 1.04]
* p < 0.05, ** p < 0.01
Results were similar for different sizes of SM
7/14/2011
8
Project III
Mapping Recreational/Outdoor Amenities
y Purpose
Examine the association of outdoor amenities with obesity and Examine the association of outdoor amenities with obesity and
physical activity (PA) in nonmetro areas of the U.S.
Areas with greater recreational values = increased PA
3 Outdoor Activity Potential (OAP) indices
Recreational opportunity
Outdoor amenity
Outdoor amenity
Nonmetro recreation county
Mt. Hood, Oregon
Project III
Mapping Recreational/Outdoor Amenities
Hypothesis
yThe natural amenities and outdoor recreational opportunities are
key drivers for the prevalence of obesity and PA in nonmetro areas
of the U.S.
Salmon River, Idaho
Rocky Mountain National Park
7/14/2011
9
Data: 3 OAP Indices
Project III - Mapping Recreational/Outdoor Amenities
1. Recreational Opportunity Index
1997 National Outdoor Recreation Supply Information System 1997 National Outdoor Recreation Supply Information System
(NORSIS) by USDA Forest Service
24 variables related to outdoor PA (e.g. numbers of facilities
for walking, biking, hiking, swimming, etc.)
Factor analysis to produce 8 factors and factor scores
http://www.srs.fs.usda.gov/
Recreational Opportunity Index
Standardized z scores: continuous
7/14/2011
10
Data: 3 OAP Indices
2 Outdoor Amenity Index 2. Outdoor Amenity Index
8 variables related to rural development/pop change (McGranaham 2010)
1. January sunlight hours
2. January mean temperature
3. July mean temperature
4. July relative humidity
kWh/m
2
/day, 19601990
F, 19712000 normals
F, 19712000 normals
%, ambient and saturation
5. Topographic relief
6. Water area
7. Forest cover
8. Tourism
meter, Max Min, DEM
mi
2
2000 Census
%, 2001 NLCD
L.Q. 2000 Census CBP
Outdoor Amenity Index
Standardized z scores: continuous
7/14/2011
11
Data: 3 OAP Indices
2 Outdoor Amenity Index 2. Outdoor Amenity Index
3. Nonmetro Recreation County Index
2004 County Typology Code by USDA Economic Research
Service
http://www.ers.usda.gov/
Nonmetro Recreation County Index
1 = Yes, 0 = No: binary
Share of earnings in recreation
Share of seasonal use housing units
Per capita receipts from hotels
7/14/2011
12
Methods
y Generalized linear model for logistic regression
Obesity Obesity
PA
Each OAP index modeled separately
Other demographic variables to control (e.g. age, sex, SES)
Recreation Opportunity Outdoor Amenity Nonmetro Recreation
Results
OR 95% CI OR 95% CI OR 95% CI OR 95% CI OR 95% CI OR 95% CI
Obesity
Recreation Opportunity 0.94 [0.92, 0.95]**
Outdoor Amenity 0.91 [0.90, 0.93]**
Recreation County 0.84 [0.80, 0.88]**
PA
Recreation Opportunity 1.09 [1.07, 1.12]**
Outdoor Amenity 1.08 [1.07, 1.10]**
** p < 0.01
Recreation County 1.23 [1.17, 1.29]**
7/14/2011
13
Overall Project Summaries
Project I Project II Project III
% Distance to SM 3 OAP indices
Obesity Obesity F/V Obesity PA Obesity Obesity F/V Obesity PA
Metro Lower Risk Risk
Nonmetro Higher Not Not Risk Risk
y Clusters of obesity
Rural South - links to demography, SES
y Supermarket accessibility
Di t t SM h d i t b it F/V i t
Distance to SM had no impacts on obesity or F/V in nonmetro
Car ownership, travel behavior, food preference, etc.
y Outdoor amenity/recreation opportunities
Lack of places for recreation/PA in nonmetro
Areas with extreme climate, environmental hazards - indoor facilities
Thank you - Questions?
Acknowledgements
Funding support
This project was supported by National Research Initiative Grant from the USDA National
Research Initiative, Influences of Physical and Social Landscapes on the Health of Rural
Communities (2007-04544).
Coauthors and collaborators
Michael C. Wimberly, PhD
1
Bonny L. Specker, PhD
2
Howard E. Wey, PhD
2,3
Natalie W Thiex PhD MPH
2
P j t b it
Natalie W. Thiex, PhD, MPH
Lacy McCormack, MPH, RD
2
Institutions
1
Geographic Information Science Center of Excellence, South Dakota State University
2
Ethel Austin Martin Program in Human Nutrition, South Dakota State University
3
College of Nursing, South Dakota State University
Project website
http://globalmonitoring.sdstate.edu/projects/obesity
Presentation Outline:
y Introduction
y System Architecture
y Framework of Measures
y Library of Geospatial Data
y Tool Development
y Challenges & Future
Directions
Childhood Obesity Epidemic
y Rapid growth of obesity,
widespread throughout the
US

y Obesity heightens the risk
for Type 2 diabetes,
Coronary heart disease, &
various cancers and impacts
well-being & health

y Early intervention is crucial
Why Build Healthy Kids GIS?
y Causes of childhood obesity are multiple
combination of factors

y GIS is underutilized in childhood obesity research
& prevention yet useful for examining the place-
variant determinants & risk factors of obesity

y Simply a need for a centralized childhood obesity
GIS information hub that could extend resources
to underserved communities & build their
capacity

GIS enables the exploration of relationships
among multiple risk factors of obesity

Objectives:
{ Identify core measures
{ Identify relevant geospatial data
{ Develop initial GIS prototypes at the
national, state, local map scales
System Architecture & Data Flow
y 3 avenues for loading data: batch
data load (GIS database), end-user
entry/load (digitize or upload their own
data), & web-services technology
(connect to data stored externally).

y Prototype Physical architecture Details:
y Geodatabase: managed by ArcSDE
through the ArcGIS desktop application
& the database is stored & managed in
Oracle through Oracle client.
y Information is hosted through a Web-
server using an Internet Information
Services (ISS) application, Geocortex
Essential Components (for web-based
mapping), & the GIS server via ArcGIS
Server manager.

Web Server
Geocortex Essential
.Net Web ADF
GIS Server
Browser
Request
Map Definition
Geocortex Essential
Manager
ArcGIS Server
Manager
ArcGIS Desktop
Layer Info
Database
ArcSDE
Layer Info Viewer
Response
GeoDatabase
Oracle
y Geographic data mostly
available for the
Environmental
measures, SES & Public
assistance measures

y Cognitive/Social
outcomes & pregnancy
measures could not be
operationalized in GIS

y Data geographically
aggregated to a large
map unit
y
Examples: screen time
aggregated to state
level; adult obesity
mostly aggregated to
state map unit,
morbidity rates
aggregated to county
map unit

GIS Reality
GIS Data Library: Active Living, Built Environment
y Destinations
(points) &
Active Living
category more
often at the
county & state
level

*Walkscore
-Cont. Built Environment: Food & Safety
y Variety of
food data
layers - some
provided as
locational
points (food
retailers) &
polygons
(grocers per
1000
persons)
-Cont. Socioeconomic Environment
y ESRI estimates for
2009 (based on US
Census) anticipate
updates

y Demographics
presented in block
groups, counties, &
zip codes

y % & Density
measures facilitate
cross- comparisons

Multi-Scale Comparison of Tools
Drawing Tools. Example: propose changes to the Built Environment and create a 1-mile buffer around proposed change.
Useful for selecting at-risk populations
Useful for saving work, sharing it with others, and downloading info.
Tools Available for the National, State, and Local Healthy Kids GIS Prototypes
Example: Find out descriptive information, such as average income in a census block group. Measure the distance from a school to a grocery store.
Tools Available for Childhood Obesity Analysis
National
State (Arkansas)
Local (Alachua)
Many tools were more suitable & useful for
analysis at the local level
GIS Tool Example: Access to Healthy Food

Improve Access to
Healthy Affordable
Food:
Find food desert
(red)
Propose a grocery
store location using
Markup
Examine how this
grocery store would
serve at-risk
populations within
walking distance
from proposed
grocer
Reporting Tool: Highlight School Characteristics
y Enables cross-comparisons across a region
y Example info provided: SES, participation levels
in the free/reduced school lunch program, HH
income, adult education level, adult obesity rates,
adult physical inactivity rates, etc.
Click tool & Select Query:
+ Additional Parameters (if desired)
=
*Only portion of report shown.
Challenges
y Most of the core measures were not suitable
for national level analysis (Env., Individual,
Structural domains) map unit lends itself to
local level analysis

y Create a user-driven system that is easy-to-
use for a broad range of users requires
feedback

y Personal health information restrictions

y Challenges with presenting an enormous
amount of GIS data: need intuitive
organization & search tools
{ Metadata documentation critical

How to build system tools for
users with different data?
Need data standards
this is important for
maintaining
functionality of tools.
Future Directions
y Further work in a state or local
prototype

y Extend the work to create a more
generalized health GIS data
library combine the relevant
geospatial resources for a variety
of health issues & possibly
support Health Impact
Assessments

y Add more geospatial data &
advance the tools & make
continual improvements
1
Spatial Model to Evacuate
Special Needs Population to
Appropriate Shelters
NATI ONAL CENTER FOR BI ODEFENSE COMMUNI CATI ONS
JACKSON STATE UNI VERSI TY
JACKSON, MS
Spatial Model to Evacuate Special Needs
Population to Appropriate Shelters
y Evacuation route planning is a vital component to
prepare for both natural and man-made disasters prepare for both natural and man made disasters
y Developing thorough emergency evacuation plans
plays a vital role in ensuring public safety
y Planning identifies routes to minimize the time to y Planning identifies routes to minimize the time to
evacuate the effected vulnerable population to a set
of destinations (Shelters)
2
y Risk situations can be identified, evaluated by
integrating geospatial data.
g g g p
y Integrated information system supports emergency
responders to plan alternative traffic directions.
y Such systems helps to develop route plan for safe
movement of vulnerable population to safer areas
y Emergency Response System should be equipped
with tools to overcome resource shortfalls
Pre-Disaster Activities
y Mitigation: measures to reduce the likelihood of damage or Mitigation: measures to reduce the likelihood of damage or
to lessen its impacts.
y Preparedness: emergency plans that provide for a decision-
making structure
Post-Disaster Activities
y Response: mobilization of first
d i i f responders, provision of emergency
support services (evacuation if necessary).
y Recovery: reestablishment of normal operations and return
of evacuees to affected areas.
3
Application of GIS technologies
Information tools developed using Geospatial technologies:
y Can be used through out the phases of emergency response
system.
y Determine the incident locations and analyze the range of
emergencies.
y Allows predicting an evacuation route from incident point to
closest shelters closest shelters.
y Suggest directions to follow the route (specifics to a
disaster).
y Provide scenario based rescue strategies to decision makers.
Status of Emergency response tools in MS
y The state of MS lacks centralized accessible source of
data to the emergency responders on shelter data to the emergency responders on shelter
locations and vulnerable populations with medical
needs.
y The Mississippi Department of Health lists the rural
character of the state and its disproportionate
number of disabled people.
y Information about the citizens particularly who
needs medical attention and transportation is of
paramount importance in early preparedness.
4
Objective of the study
y To provide a centralized accessible source of data for
emergency responders. emergency responders.
y To develop network-based information tool that can
Determine the location of incident in the state
Create a scenario based impact buffer region
Identify population centers, especially with medical needs
Determine the closest shelters from the travel time based
road network built for the state. road network built for the state.
Generate evacuation routes leading to the shelters from the
population centers.
y To integrate and disseminate the generated
information on DISCOVER MS
DISSEMINATION :DISCOVER MS
5
Methodology: GIS-Based Evacuation Tool
Major Components and Data Sources
Navigable Data Model Geospatial Database
Travel time based
Road Network layer using
Demographic profile
Red Cross Shelter Locations
Network Analyst
Red Cross Shelter Locations
MSDH Hospital Locations
Travel-Time based network dataset Generation of Service Areas
6
Automate the process using Model Builder
GIS-Based Evacuation Tool
INCIDENT LOCATION AND HOSPITALS IN THE IMPACT ZONE
7
GIS-Based Evacuation Model
GIS-Based Evacuation Model
8
Future Work
2
nd
Closest
Shelter:700
3rd Closest
Shelter: 100
Hospital Capacity
(1000)
Closest
Shelter: 200
2

m
i
l
e
s
1 mile
Conclusion
y Currently NCBCs spatial model facilitates
To access centralized data for the state of MS from DISCOVER MS
To build travel time based road network data set for Mississippi
To analyze travel based service areas of the critical facilities
To determine incident location and identify the vulnerable
population centers and health facilities in their impact region.
To identify 4 closest shelter locations (time based)and generate
evacuation routes leading to these facilities.
To disseminate the results on DISCOVER MS To disseminate the results on DISCOVER MS.
y Future work: NCBC focuses on improvising the tool in
order to facilitate the emergency responders to
Determine the closest shelters based on their capacity and its type
(medical facilities/pet friendly)
9
Questions !!
1
Geographic
Opportunities in Medicine
Amy J. Bl at t , Ph. D. , GI SP
Guest Edi t or
Journal of Map & Geography Li brari es
( JMGL) :
Advances i n Geospat i al I nf ormat i on,
Col l ect i ons and Archi ves Col l ect i ons and Archi ves
Quest Di agnost i cs, West Nor r i t on, PA
E- mai l :
Amy. J. Bl at t @Quest Di agnost i cs. com
Agenda
y JMGL: History and background y g
y Special theme issues in JMGL
y Presentation focus: increase access to geographic
resources for medical geographers and public
health researchers
y Case study: Center for Geographic Analysis at Harvard
University
URISA 2011 GIS in Public Health Conference June 27-30, 2011 - Atlanta, Georgia - www.urisa.org URISA 2011 GIS in Public Health Conference June 27-30, 2011 - Atlanta, Georgia - www.urisa.org URISA 2011 GIS in Public Health Conference June 27-30, 2011 - Atlanta, Georgia - www.urisa.org
y
y Enabling map/geography libraries to distribute resources to
a wider audience 6/20/2011
2
History of JMGL
y Co-founded in 2002 by Mary Lynette Larsgaard (University of
California Santa Barbara) and Paige G. Andrew (The Pennsylvania ) g ( y
State University)
y Volume 1 published in 2004 by Haworth Press
y JMGL was biannual for the 1
st
6 volumes and is now published 3x per
year
y Haworth Press purchased by Taylor and Francis in 2009
{ Additional support for color illustrations
St dit i l/ bli h t
{ Stronger editorial/publisher team
{ Broader marketing group
y Co-edited by Paige G. Andrew and Kathy H. Weimer (Texas A&M
University Libraries) in 2010
JMGL Special Theme Issues
y Geospatial innovations at Department of Energy:
Volume 4, 2008
y Preservation of digital geospatial materials:
Volume 6, Issue 1, 2010
y Geographic opportunities in medicine:
Volume 7, Issues 1 and 3, 2011
y Crisis mapping: Volume 8, Issue 2, 2012 6/20/2011
3
Geographic Opportunities in Medicine
y How can map/geography librarians increase access to
geographic resources for medical geographers and public g g p g g p p
health researchers?
y Libraries as guardians of electronic geospatial information
and data storage
{ New York State Public Library
{ Washington State Uni ersit libraries
{ Washington State University libraries
{ Syracuse University
{ Harvard University
Role of the Map and Geographic Information Library in
Medical Geographic Research
y Author: Ellen K. Cromley, University of Connecticut School
of Medicine, Farmington, Connecticut , g ,
y Libraries provide geographic data on places, spatially-
referenced data on populations, spatially-referenced data
on health events
y Librarians can help with metadata standards development,
archiving digital spatial databases, and adopting
georeferences to support place-based information retrieval 6/20/2011
4
Yellow Fever Cases in Lower Manhattan, 1819
(National Library of Medicine, NIH)
Typhoid in Washington, D.C., 1906 1909
(Public Health Services Reports)
5
Enabling Geographic Research Across Disciplines
Building an institutional infrastructure for geographic
analysis at Harvard University y y
y Authors: Weihe Wendy Guan
a
; Bonnie Burns
b
; Julia L.
Finkelstein
c
; Jeffrey C. Blossom
a
{
a
Harvard University Center for Geographic Analysis,
Cambridge, Massachusetts
{
b
Harvard Map Collection, Cambridge, Massachusetts
{
c
Harvard School of Public Health, Boston,
Massachusetts
History of Geography at Harvard University
Harvard Map Collection (HMC) founded
Geography department dissolved
1818
1948
Geography department dissolved
Computer Graphics and Spatial Analysis Laboratory
dissolved
HMC adds GIS capabilities
HMC hosts Massachusetts Electronic Atlas (MEA)
d i l ib ( ) f d d
1948
1991
1992
1995
2001
Harvard Geospatial Library (HGL) founded
Professor and historian Peter Bol forms faculty
committee on spatial analysis
Center for Geographic Analysis (CGA) formed
2001
2003
2006 6/20/2011
6
Acquiring and Sharing Spatial Data
y Harvard Map Collection strengths in geospatial data
acquisition and archiving q g
Harvard Geospatial
Library depository
d t l f and catalogue of
electronic geospatial
data and metadata
7
y Center for Geographic Analysis manages university-wide
licenses of geospatial, analytical, and visualization software g p , y ,
and their capabilities
{ Consultation and help desk services
{ GIS training
{ Analytic services
{ Technology monitoring
{ Web map services
CGA Faculty
Steering
Committee
Harvard
College
Library
CGA Technical
Advisory
Committee
Harvard
University
Library
CGA HMC
y
HGL
Standing
Committee
y
HGL
GIS Data
Services
GIS
Training
and Course
Support
Geographic
Analysis
Services
Users (everyone with a Harvard ID) 6/20/2011
8
Applying GIS Methods to Public Health
Research at Harvard University
y Authors: Jeffrey C. Blossom
a;
Julia L. Finkelstein
b
; Weihe Wendy
Guan
a
; Bonnie Burns
c
{
a
Harvard University Center for
Geographic Analysis, Cambridge,
Massachusetts
{
b
Harvard School of Public
Health, Boston, Massachusetts
{
c
Harvard Map Collection,
Cambridge Massachusetts
Cambridge, Massachusetts
Workflow for Working with Geographic
Data at CGA
y Initial consultation
{ Project objectives
y Major projects categories
{ Geocoding and census
{ Project objectives
{ Data sources
{ Methodologies
{ Tools/software
{ Deliverables
y Fees: 1st 4 hours are
g
variable extraction
{ Map creation for print
publications
{ Dynamic web map
creation
(http://maps.cga.harvard.
du/gmaps_instruction/)
free, then $75 for each
additional hour
{ GIS analysis and
visualization (Python,
Java, .NET, PHP) 6/20/2011
9
CGA Project Specification Document
Examples of Public Health Research at CGA
y Nurses Health Study (Drs. Speizer and Willett)
{ 238 000 registered nurses { 238,000 registered nurses
{ Relationship between risk factors and non-
communicable diseases
Particulate matter exposure and cardiovascular
mortality
Northern latitudes and rheumatoid arthritis
Northern latitudes and rheumatoid arthritis 6/20/2011
10
Examples of Public Health Research at CGA
y AfricaMap framework for organizing and distributing
Africa data from a variety of disciplines y p
{ Supports collaborative research and teaching on Africa
{ Harvard School of Public Health
{ Harvard Medical School and affiliated hospitals
{ Harvard Humanitarian Initiative
{ Harvard Initiative for Global Health
{ Faculty of Arts and Sciences
Malaria
Distribution
in AfricaMap
11
Child Physical Activity and the Built
Environment
y Managing spatial data using an open-source framework:
Langley and Messina (Michigan State University)
Enabling Map/Geography Libraries to
Distribute Resources to a Wide Audience
g y g y
12
Expanding graduate
medical education medical education
using GIS:
Hayashi, Bazemore, and
McIntyre (The Robert
Graham Center for
Policy Studies, and
Georgetown
Georgetown
University, Providence
Hospital Family
Medicine Residency
Program)
GIS-based race-
prediction p
model to target
areas for
reducing health
disparities:
Vernon and Ting,
(WellPoint, Inc.,
parent company of
Anthem BlueCross
BlueShield) 6/20/2011
13
Further Considerations
y Provide education on geographic literacy
F t d t i t di i li i ti ti y Foster and promote interdisciplinary investigations
y Establish appropriate personnel and technological
infrastructures
y Perform annual stakeholders analysis
1
Web GIS Framework for Surveillance of Malaria: A
Case Study of Allahabad District, India y ,
Gupta, R.D.
1
, Siddiquie, Shekh Faisal
2
,Srivastava, Shweta
3
1- Professor & Coordinator, GIS Cell; Department of Civil Engineering; Motilal Nehru National
Institute of Technology, Allahabad- 211004, India
(e-mail: gupta.rdg@gmail.com, Tel: 0532-2271308 (O), 2541505 (R))
2, 3- M. Tech. (GIS & Remote Sensing); GIS Cell; MNNIT, Allahabad, India
Presented by:
Dr. R. D. Gupta
Professor & Coordinator, GIS Cell
About My Institute & GIS Cell
Motilal Nehru National Institute of Technology (MNNIT), Allahabad, India
Institute of National Importance, as declared by Govt. of India Institute of National Importance, as declared by Govt. of India
GIS Cell (having the status equivalent of a Department)
Running M. Tech. (GIS & Remote Sensing) PG programme since 2006
Offers Institute fellowship for Ph.D. programme in the field of Geoinformatics
2
About Me
Dr. R.D. Gupta
Professor (Geoinformatics)
Coordinator, GIS Cell
Department of Civil Engineering
l l h l f h l ll h b d d Motilal Nehru National Institute of Technology (MNNIT), Allahabad, India
Ph.D. (GIS & Remote Sensing) IIT Roorkee
M. Tech. (Remote Sensing & Photogrammetric Engg.) IIT Roorkee
B. Tech. (Civil Engineering) IIT Roorkke
23 years of experience in the field of Geoinformatics (Teaching & Research)
Fields of Interests: Open Source GIS, SDI, Web GIS, GIS based DSS
Mapping from High Resolution Satellite Data
Motivation for the Present Work
To demonstrate the usefulness of Open Source GIS software
as a viable alternative to proprietary GIS solutions in India
To work in the field of Health Sector as we all are
concerned and affected by this
To use open Source GIS for development and implementing
of Health GIS and Heath SDI
To convert the available Health data in the form of Health
GIS
To do a little bit for the society and to improve the quality
of human life
3
GIS can be used for mapping and analysis of:
The location of health and other services (hospitals, dispensaries
health centres, schools, etc.)
Role of GIS in Health Sector
Epidemiology (morbidity, mortality, drug susceptibility)
Population distribution and their movement (location of towns,
villages, hamlets and road networks)
Socio-economic pattern
The environmental parameter (physical features, land use, surface
water)
Meteorology (rainfall, temperature, humidity) gy ( , p , y)
For health sector, disease data sharing is important for the collaborative
preparation, response, and recovery stages of disease control.
Data heterogeneities integration interoperability and cartographical
Web GIS for Health Sector
Data heterogeneities, integration, interoperability, and cartographical
representation are still major challenges in the health geographic fields.
These challenges cause barriers in extensively sharing health data and
restrain the effectiveness in understanding and responding to disease
outbreaks.
To overcome these challenges in health sector, mapping and sharing of
spatiotemporal disease information in an interoperable framework based on
OGC specifications under GIS environment is the need of the hour.
Web GIS based frame ork can be sed to pro ide a real time and d namic
6
Web GIS based framework can be used to provide a real-time and dynamic
way to represent disease information on maps.
4
Objectives of the Present Work
To develop a geo-statistical model for intra-district malaria
analysis and to find the most susceptible community development
block for Malaria
(Hierarchy- District --- Tehsil --- Block ---- Village)
To develop a open source based Web GIS framework for disease
7
To develop a open source based Web GIS framework for disease
surveillance (in particular- Malaria)
Open Source Paradigm
5
OSS is currently one of the most debated phenomena in the software
industry.
Increasing trend in the OSS movement as a new paradigm for
software development.
Open Source Software (OSS)
software development.
9 OSS hailed as a substitute for proprietary software
OSS greeted with great fanfare in a variety of forums
9 OSS defined as computer software for which the human-readable
source code is available
But also under a copyright license which is free of cost
9 General public alter the programmes to develop/ improve it
More efficient way to develop software compared to an isolated
group of programmers on the payroll of a company
Free software movement launched in 1983 but in 1998, this term
was replaced by Open Source Software (OSS)
Biggest Advantages:
OSS
9 Users have free access to the source code, unlike the
traditional proprietary paradigm of software development
9 Users can change & improve the software, and can
redistribute it in modified or unmodified form
Biggest Handicaps:
9 Lack of awareness about the capabilities of OSS 9 Lack of awareness about the capabilities of OSS
9 Lack of proper training
9 Above all, the mindset
6
Definition is based on the Debian Free Software Guidelines, written &
adapted primarily by Bruce Perens (Founder of the OSI- Open Source Initiative)
Under the Open Source Definition, licenses must meet Ten conditions:
OSS
Open Source Definition (OSD)
1. Free Redistribution: can be freely given away
2. Source Code: must either be included or freely obtainable
3. Derived Works: redistribution of modifications must be allowed
4. Integrity of The Author's Source Code: licenses may require that
modifications are redistributed only as patches
5. No Discrimination against Persons or Groups
6. No Discrimination Against Fields of Endeavor: commercial users
cannot be excluded.
OSS
7. Distribution of License: The rights attached to the program must
apply to all to whom the program is redistributed without the need
for execution of an additional license by those parties
8 License Must Not Be Specific to a Product: the program can not be 8. License Must Not Be Specific to a Product: the program can not be
licensed as part of a larger distribution.
9. License Must Not Restrict Other Software: the license can not insist
that any other software to be distributed with this must also be open
source
10. License Must Be Technology-Neutral: No provision of the license
may be predicated on any individual technology or style of interface.
7
Open Source license is a copyright license that makes the source code available
under terms that allow for modification and redistribution without having to pay
the original author
Such licenses may have requirements to preserve the name of the authors and
the copyright statement within the code
OSS.
Open Source License
the copyright statement within the code
There are also shared source licenses which have some similarities with open
source, such as the Microsoft Reference License (MS-RL), but are not
compatible with the OSD
OS Licenses
Mozilla Public License (MPL) 1.1
Apache Software License 2.0
Eclipse Public License
Common Development and Distribution License
Common Public License 1.0
GNU General Public License (GPL)
Affero General Public License (AGPL)
OSS..
Open Source vs. Proprietary
Open Source Closed Source
Programmers freely provide improvements
for recognition
The software firms hire programmers and
work for compensation for recognition work for compensation
If program is popular, a very large number
of programmers may work on it
The firms resources limit the number of
programmers working on the code
Programmers compete with each other for
recognition
Programmers do not compete with each
other
User do not pay price; hence particularly
attractive to user
Users pay a price to offset the cost of R&D
that the software firm invests
Less time and efforts are spent by the
individual user/ programmer on the R&D,
as there are more programmers
More time and effort spent on the R&D by
the software firm, as there are fewer
programmers.
8
Open Source GIS Software
Open Source GIS Software
Open Source GIS based projects can offer:
9 rich functionality
9 Robustness
9 cooperation from contributing developers &
9 continuous improvement
Existence of several OS libraries and GIS suites
OS GIS software may not only meet but in some cases also surpass
the capabilities of proprietary software to produce robust spatial
solutions in a cost-effective and efficient manner.
Open Source GIS Software can be used for:
geospatial database creation
spatial analysis
geospatial web services
9
OS GIS Software
OS GIS software categorisation based on Language used:
(the list is indicative and not exhaustive):
OS GIS Software..
C/ C++ Based JAVA Based
DEEGREE
GeoServer
Open JUMP
Quantum GIS
ILWIS
GRASS
Open JUMP
uDIG
gvSIG
GRASS
PostgreSQL
MapServer
Open Source based Web GIS Framework
10
Conceptualization of Web GIS Framework
Aimed to develop and implement an efficient Web GIS framework Aimed to develop and implement an efficient Web GIS framework
This should supports integration of applications running on
different resource specific devices
9 Need to conceptualize a distributed architecture
9 Need to conceptualize an Interoperable architecture
Geospatial web services are an integral part of any SDI
A Geospatial Web Service is an application that can retrieve data
from GIS databases
Geospatial Web Services for SDI Development
It provides geographic information through a browser interface
Client
Geospatial
Web Services
Geographic
Data
Request
Response
Geospatial Web Service Architecture
11
Geospatial Web Capabilities can be realised using the client-server
architecture- either through thick client and thin client model.
In the present work, a thin client architecture has been used.
In thin client model, most of the processing work is done on demand in the
server and the client does not perform any task other than to display.
Client Server Architecture for Geospatial Web Services
Thin client GIS Architecture
Open GIS Software follows Open standards provided by Open Geospatial
Consortium (OGC) and World Wide Web Consortium (W3C)
Open Standards and protocols are used for transferring and sharing GIS
data across a heterogeneous global network
Th O W b GIS S i hi d b OGC i h i Fi
Open Source based Geospatial Web Services
The Open Web GIS Service architecture, proposed by OGC, is shown in Figure:
Web Registry
Services
(WRS)
Web Coverage
Services
(WCS)
Web Map
Services
(WMS)
Web Feature
Services
( S)
OGC Web Services
(WRS) (WCS) (WMS) (WFS)
Register Services
Get Features
Transaction WFS
Get Coverage Get Map
Get Feature Info
12
Web Map Services (WMS)
enables the creation of a network of distributed map servers from which clients can
build customised maps.
provides the image files and graphic languages like SVG and WebCGM.
can be invoked using a standard web browser by submitting requests in the form of
URLs.
produces maps of spatially referenced data dynamically from geographic information
Web Feature Services (WFS)
returns actual vector data.
provides an interface allowing requests for geographical features p g q g g p
across the web using platform-independent calls.
define interfaces for data access and manipulation operations on
geographic features using HTTP.
Framework of Open Source Based Web GIS Model: Option-I
Geospatial Health Database
(Using Quantum GIS/ Open Jump)
Storing of Geospatial health data in
shape file
Non-spatial data
Web Map
Services
Web Feature
Services
Wamp Server, ALOV, MySQL & PHP
Business Process Data (Security)
Utility Services
Web Coverage
Services
Client 1
.........
Client 2 Client N
13
Open SDI Framework
Salient Features:
For the creation of integrated geospatial health database, the present
framework uses Quantum GIS software
ALOV and Wamp Server have been integrated for imparting the geospatial
web capabilities in terms of WMS & WFS
The business process data are meant to give additional storage for security
purpose
MySQL is used for storing of security aspects and non-spatial data for
decision making
PHP: Hypertext Preprocessor language has been used for dynamic server PHP: Hypertext Preprocessor language has been used for dynamic server
side scripting in the framework
Geospatial Database
Storing of Geospatial
data
Non-spatial data
Geospatial Database
Storing of Geospatial
data
Non-spatial data
Modified Web GIS Framework
WMS
WFS
GeoServer, Apache
Tomcat, PostGIS, MySQL
& JSP
Utility
Service
WMS
WFS
GeoServer, Apache
& JSP
Utility
Service
WFS
Master Server
Slave Server Slave Server
Client 1
.........
Client 2 Client N
GeoServer, Apache
& JSP
WMS Utility
Service
Master Server
14
Distributed Architecture
A Master server is used for sharing of the geospatial data and
publishing on the web
Different servers can be taken to act as slave server and may
consist of the geospatial data of different domains consist of the geospatial data of different domains
There would be no effect on the system if any of the slave servers
is shut down.
For example:
IP address of the slave server is 172.31.10.67 and 172.31.10.85
IP address of the master server is 172.31.10.76. IP address of the master server is 172.31.10.76.
Distributed Web GIS services will be accessed by the client using:
http://172.31.10.76/webgis
Approach for Development of Open SDI Model
The SDI prototype development is based on:
Jacobsons method of Object Oriented Software Engineering (OOSE)
for incorporating strong user focus & time critical nature for incorporating strong user focus & time critical nature
In OOSE method, the software development process adopts a
sequence of steps including:
requirement specification
analysis and design
implementation and testing
complete module &
model observation model observation
The process is usually cyclic or incremental in nature
15
SDI Development
Successive iterations allows to take into consideration more informed view
of SDI requirements
Incremental development strategy allows the development of framework to
be tackled in smaller, more manageable portions
1. Requirement Specifications:
aims to specify the behaviour of model from the perspective of a user
Use Case Model is adopted to specify sequence of actions that need to be
performed
To specify multi-level users for security, i.e.,
SDI Development
To specify multi level users for security, i.e.,
9 Administrative user
9 General user
9 Developer
2. Analysis & Design
Translating user requirements into underlying algorithms and an interface
The preliminary investigations determine the feasibility of system to be
developed, & includes
9 Technical feasibility
9 Operational feasibility &
9 Economical feasibility
The design phase involves the creation of geospatial database.
Spatial data have been created in the form of shape files
16
3. Implementation & Testing:
Coding is done to fulfil the overall requirements and for testing
Coding is one of the challenging tasks for achieving the real success .
Testing phase is essential for increasing the QOS (Quality of Service) of
the software
SDI Development
4. Complete Module:
The prototype consists of three main modules, e.g.,
9 Module I for registration
9 Module II for Allahabad District Malaria Mapping & Surveillance
9 Module III for Utility Services
5 M d l Ob ti 5. Model Observation:
The Open JUMP, ALOV, Apache Tomcat, MySQL, PHP and Java Development Tool Kit
are used for overall development of the system.
Involves viewing all the features of developed prototype
If all the required features are satisfactorily implemented, then OK
ORELSE repeat all the stages of OOSE design
Case Study:
Some Snapshots
of
Prototype Framework Developed
17
GIS is a valuable tool to assist in health research and education
Health GIS can facilitate decision makers to plan the strategy from preparedness to
combat the emergence of diseases
Case Study: Allahabad District (India)
One vector borne disease i e Malaria has been selected for the case study One vector-borne disease, i.e., Malaria has been selected for the case study
The information in terms of the various aspects of Malaria included:
9 Susceptibility regions
9 Annual Blood Smear Examination Rate (ABER)
9 Slide Positive Rate (SPR)
9 Annual Parasite Index (API)
Thematic layers created include: map of India with State boundaries, Allahabad
district with block boundaries, Malaria information for different blocks of district with block boundaries, Malaria information for different blocks of
Allahabad district, etc.
Allahabad district (India) has been taken up as the study area to demonstrate the functionalities
of developed Health SDI framework
The district has an area of 7261 sq. km and lies between 24
0
47'00N to 25
0
47'00N and
81
0
09'00E to 82
0
21'00E with about 50 lacs. total population
Snapshot of Prototype Health SDI: Home Page
18
Snapshot of Health SDI: User Registration Page
Snapshot of Health SDI: Malaria Mapping & Surveillance Page
19
Snapshot of Health SDI: Malaria Mapping & Surveillance Page ..
Snapshot of Health SDI: Utilities Page
20
HEALTH SDI
Link to Health-SDI
http://localhost/hsdi
Concluding Remarks
The present research work is focused at adopting OGC standards
for creating, accessing, integrating and sharing the geospatial
health information on the web.
The developed framework adopts a modular and flexible structure,
d id ffi i t h i f th ti d d li and provides an efficient mechanism for the generation and delivery
of value-added spatial information.
The present web GIS framework extends the concept of geospatial
web services in the field of health sector.
The experience gained in using open source GIS software suggests
that various tools and software like Q- GIS, ALOV, Apache Tomcat/
Wamp Server, MySQL and JSP/ PHP are available for creation of
40
p , y
spatial datasets and implementation of geospatial web services
The widespread use of these open source resources in the
development of GIS based applications on the web instead of going
for costly proprietary solutions could benefit a vast user community
and should be encouraged.
21
THANK YOU
Dr. R. D. Gupta
Professor & Coordinator GIS Cell Professor & Coordinator, GIS Cell
MNNIT Allahabad, India
E-mail: gupta.rdg@gmail.com, rdg@mnnit.ac.in
1
GIS and Health:
Where Can We Go From Here?
Cl os i ng Pl enar y
El l en K. Cr oml ey, Ph. D.
Depar t ment of Communi t y Medi ci ne and
Heal t h Car e
Uni ver s i t y of Connect i cut School of Medi ci ne
Purpose
y Purpose
To provide a brief overview of milestones affecting the
evolution of public health GIS to today
To describe the local trap as a basis for considering
future challenges
To suggest ways to make health applications of GIS more
spatial and better able to uncover from the vast data
available the key configurations of factors that affect
health in particular places 7/7/2011
2
How Far Have We Come?
y Key Milestones
1970s Landsat
1980s GBF/Dime {HIV/AIDS} 1980s GBF/Dime {HIV/AIDS}
1990s TIGER/Line
files {Emerging Infectious Diseases}

WorldWideWeb
GIS and health conferences
Desktop GIS software
GPS
Metadata standards
Google {Neighborhoods and health}
Shapefiles
p
2000s GIS and health textbooks {Homeland Security}
GIS and health journals {Reducing health disparities}
Spatial statistical software {Natural disasters}
Google Earth
/
Google Maps

{Global climate change}
Health mashups
Smart phones {Health insurance reform}
Where are We?
y Status of GIS and Health
Education of health professionals
Organization of GIS services in health agencies
Data
Approaches to analysis
Approaches to analysis
GIS systems
Dissemination of and public access to data, maps, and results 7/7/2011
3
The Local Trap
y The idea that the local scale is the only
meaningful scale in contextual studies of meaningful scale in contextual studies of
population health
Originated in the development and planning literature
More recently raised as an issue in studies of
neighborhood contextual factors and health
Serves as a point of departure for survey of current
challenges
The Local Trap in Time
y The idea that the present is the only meaningful
point in time in GIS point in time in GIS
Exposure occurs in time and space
Latency periods between exposure and health outcomes
are important
Health analysts need to be able to time-stamp and archive
data to recreate past patterns of environmental conditions
Increasing interest in health over the life course 7/7/2011
4
Education of Health Professionals
y Issues
GIS still not widely and deeply embedded in public health GIS still not widely and deeply embedded in public health
education
y Consequences
Local pockets of expertise
Need to engage in on-the-job training
Need to engage in on the job training
Impacts on GIS in public health agencies, health journal
editorial boards and review panels, IRB and health grant
review panels, and other health-related entities
Organization of GIS Services in Agencies
y Issues
Shortages of professionals within health agencies who Shortages of professionals within health agencies who
have GIS expertise
y Consequences
Difficult to build a sound GIS infrastructure within
agencies local pockets of expertise
Fewer models of how GIS can best be organized within
public health organizations
Outsourcing GIS 7/7/2011
5
Data
y Issues
Massive amounts of data Massive amounts of data
y Consequences
Effort to manage data leaves little time for analysis
Implications for data quality and confidentiality
Difficult to find the critical signals or patterns of interest
Approaches to Analyses
y Issues
Health studies that are geographic but not spatial Health studies that are geographic but not spatial
y Consequences
Cross-sectional studies of one place
Inconsistent results across localities and difficulties
identifying groups of similar localities
identifying groups of similar localities
Most analyses use global not local methods
Lack of attention to spatial sampling and spatial error 7/7/2011
6
Hardware and Software
y Issues
Keeping pace with technological change, especially the Keeping pace with technological change, especially the
challenges posed by cloud computing
y Consequences
Health data confidentiality in the cloud
Responding to changes in software and necessary
Responding to changes in software and necessary
hardware
Public Access and Participation
y Issues
So much health information is acquired, managed, and So much health information is acquired, managed, and
analyzed by private entities outside the view of public
health and the public and public agency dissemination has
changed
y Consequences
Despite technological advances, the public has less access
Despite technological advances, the public has less access
to data of interest in some cases
Growing concern about management and disclosure of
data from health insurance and medical care contact 7/7/2011
7
Where are We Going?
y Creating a spatially extensive configurational
approach to GIS and health approach to GIS and health
Linking groups of health professionals and researchers in
spatially extensive studies
Global and local statistics
Attention to spatial sampling and spatial error
Global Versus Local Statistics
Global Statistics Local Statistics
Summarize data for entire regions Summarize data for individual places
within entire regions
Single statistic Multiple statistics, one for each place
Uninteresting when mapped (aspatial) Interesting when mapped (spatial)
Provide misleading interpretations of
local relationships
Useful for exploratory data analysis,
confirmatory analysis, and building
more accurate global models
Adapted from A. Stewart Fotheringham, Chris Brunsdon, and Martin Charlton, 2002,
Geographically Weighted Regression: The Analysis of Spatially Varying Relationships. 7/7/2011
8
Global Spatial Statistics
y Methamphetamine Abuse Study
Population-based ecological approach using simultaneous
autoregressive spatial (SAR) models to correct for spatial
autocorrelation
Spatially extensive view of changing enforcement patterns
in local areas and drug use over time
Implications
Implications
--Displacement out of areas with greater law enforcement
on a larger urban-rural scale?
--Growth of markets in rural areas where producers can
evade law enforcement?
Local Spatial Statistics
y Connecticut CODES Project
Used to investigate patterns of motor vehicle collisions in
Connecticut
Defined places with high numbers of collisions by type of
collision
Calculated local proportions and local odds ratios to assess
importance of individual, environmental, and behavioral
factors 7/7/2011
9
Statewide Motor Vehicle Collisions
Collisions on federal and state roads, 1995-1996
Definition of Collision Places
y Creation of box-shaped kernels
Box-shaped kernel based on stopping distance to group
collisions along road segments
Identified places with high numbers of collisions by type of
collision
Eliminate overlapping places 7/7/2011
10
Fixed Object Collision Places
10 places with
highest frequency of highest frequency of
fixed object
collisions
Local Odds Ratios
Place
ID
Rain or
Snow
Dry Road Daylight Age
25-44
Male Too Fast
for Conditions
1 2.79
z=4 91
.29
z= 5 58
.45
z= 3 76
.73
z 1 56
.87
z 69
11.09
z=10 08 z=4.91 z=-5.58 z=-3.76 z=-1.56 z=-.69 z=10.08
2 2.83
z=1.30
.22
z=-1.63
1.27
z=.27
.00
z=.00
1.70
z=.66
4.71
z=1.78
3 4.80
z=3.75
.17
z=-4.46
.36
z=-2.52
1.34
z=.80
.56
z=-1.54
17.50
z=5.65
4 2.89
z=3.21
.19
z=-5.16
.30
z=-3.82
1.25
z=.73
.63
z=-1.53
8.62
z=6.14
5 3.06
z=3.33
.28
z=-3.74
.48
z=-2.27
1.32
z=.87
1.03
z=.09
3.11
z=3.17
6 1.05
z=.09
.54
z=-1.10
.23
z=-1.86
1.14
z=.27
1.14
z=.27
2.64
z=1.98
7 5.53
z=4.65
.16
z=-4.87
1.03
z=.08
1.24
z=.65
.36
z=-2.90
16.55
z=6.41
8 3.04
z=3.21
.32
z=-3.27
.35
z=-.292
1.20
z=.55
1.20
z=.55
10.50
z=5.81
9 2.33
z=2.25
.29
z=-3.28
.46
z=-2.02
.83
z=-.52
1.02
z=.06
3.32
z=2.85
10 .94
z=-.19
.78
z=-.81
.32
z=-3.59
.80
z=-.72
.92
z=-.25
3.86
z=3.81
State Odds
Ratio
1.76 0.50 0.40 1.13 1.10 6.38
7/7/2011
11
Linking Researchers in Spatially Extensive Studies
y Malaria Atlas Project www.map.ox.ac.uk
Initiated in 2006 to develop global map of malaria
Gathered prevalence reports based on strict inclusion criteria and
georeferenced them
Used geostatistical methods to make a continuous, age-standardized,
urban-corrected malaria prevalence map
Validated it and reported spatial error
Placed the final database in the public domain
Malaria Atlas Results
y Results
World, regional, and country maps
available for limits and endemicity
Downloadable mean endemicity
surface in GeoTIFF and ASCII
formats
12
Malaria Atlas Uncertainty
y Error
Uncertainty and uncertainty class
maps for the world, regions, and
countries
Bright green area shows area of
medium (higher) uncertainty
l d bl i f
Downloadable uncertainty surface
in GeoTIFF and ASCII formats
Breaking the Link Between Extent and Scale
y Developing open access resources that are
breaking the link of geographical extent and level g g g p
of detail
Large area no longer means limited detail
Health GIS applications can help us generalize key
relationships to the set of local placesthe spatial
d i h h l i f h l h
domainwhere they explain patterns of health 7/7/2011
13
An Invitation to
y Participate in regular efforts to increase the pool
of health researchers with GIS expertise of health researchers with GIS expertise
y Pay attention to spatial sampling
y Adopt global and local statistics and time-space
methods
y Georeference, time stamp, and archive reports,
maps, data and publications so they can be used in
health meta projects into the future
health meta-projects into the future
y Engage in current policy debates affecting
geospatial data on individual health
Visualization techniques of multidimensional health data
Alexandra Pinto*, Fernando Bao
, Victor Lobo
*Laboratory of Biomathematics, Faculty of Medicine of Lisbon,
ISEGI, New University of Lisbon,
CINAV - Portuguese Navy Research Center, Escola Naval

Data visualization is an emerging technology that is benefiting from the growing capacity
of computers and data mining techniques. Data visualization allows us to extract useful
information from data and is suitable for the analysis of large databases and multivariate
data. Visualization is a powerful way of analysis that helps to discover hidden patterns and
trends in the data. However, a chart can become visually difficult to understand if the
number of variables or groups to represent is high, or if the measurement scales are
different (Saary 2008).
A major problem of data visualization is the choice of an appropriate graphical
representation of information to be treated (Carmo 2003).
1. Introduction
7. References
1. ACS. (2009). Consumo de medicamentos ansiolticos, hipnticos e sedativos e antidepressivos no mercado do SNS, em ambulatrio.
http://www.acs.min-saude.pt/pns/acessibilidade-ao-medicamento/health-impact-assessment/. (Access 5th April 2010).
2. Bushardt, R. L., E. B. Massey, et al. (2008). Polypharmacy: Misleading, but manageable. Clinical Interventions in Aging 3(2): 383-389.
3. Carmo, M. B. (2003). Visualizao de Informao. Modelo Integrado para o Tratamento de Filtragem e Mltiplas Representaes.
Universidade de Lisboa.
4. Pinto, A., Rodrigues, T., Bao, F. e Lobo, V. (submitted). Medication and Polymedication in Portugal.
5. Saary, M. J. (2008). Radar plots: a useful way for presenting multivariate health care data. Journal of Clinical Epidemiology, 60, 311317.
6. Tufte, E. (2006). Beautiful Evidence. Graphics Press.
6. Conclusion
The table-graph allowed us to easily make transversal comparisons between and within
regions, in the same representation.
We used the graphs to draw some conclusions about polymedication, and we showed
that in this case the table-graph is an improvement, presenting data more clearly than the
multiple radar plots, common tables or line charts.
3. Radar plots
The radar plot is a two-dimensional graphical representation of multivariate data and is
widely used for three or more quantitative variables (Figure 1). Radar plots proved to be
an unhelpful tool when representing several variables. If we try to build a radar plot with
more than eight variables the graph will become messy, especially if we need to add labels
(Figure 2). In this case a new algorithm for labeling the axis should be implemented but it
probably wont solve the entire problem.
For the examples given below, we used MATLAB and Excel to compare Table-graph with
common table and line charts.
5. Other representations
Figure 1: Map of Portugal with radar
plots for each region of NUTS II for
some of the most common pathologies
or kinds of drugs
The aim of this study is to find an alternative representation of the classical radar plot
technique when the number of variables is high.
In our study we use medication data from the 4th National (Portuguese) Health Survey
(NHS), 2005/06, and we intend to represent pathologies or kinds of drugs most consumed
in Portugal.
2. Objective
0
5
10
15
20
High blood pressure
Joint pains
High cholesterol
Contraceptive pills
Sleeping pills
Headache or M.
Other pain
Other Cardiov. D.
Anxiety or N. Depression
Stomach problems
Diabetes
Antibiotics
Allergic symptoms
Menopause or O. H.
Asthma
Chronic bronchitis
Figure 2: Radar plot for the 18 pathologies mencioned in Portuguese NHS
We suggested another approach, the table-graph, cited by Tufte (Tufte 2006). This
graphical representation uses multivariate data and provides a visual image of each
element of the chart, without loss of information. Table-graph displays all values of each
variable connected with a line (horizontal, ascending or descending) like a common line
chart. We implemented Table-graph in MATLAB 7 (Figure 3).
4. Table-graph
Figure 3: Table-graph for the nine most
common pathologies or kinds of drugs
in Portugal, for each region of NUTS II
Norte Centro LVT Alentejo Algarve
2
4
6
8
10
12
14
16
18
20
Pathologies / Kinds of drugs
%
Other Cardiov. Disease
Other pain
Headache or Migraine
Sleeping pills
High cholesterol
Contraceptive pills
Joint pains
High blood pressure
Other drugs
Norte Centro LVT Alentejo Algarve Aores Madeira
0
5
10
15
20
Pathologies / Kinds of drugs
%
Pathologies / Regions
Kinds of drugs Norte Centro LVT Alentejo Algarve Aores Madeira
Other Cardiovascular Disease 6,3 7,4 6,5 6,2 4,7 5,9 4,0
Other pain 6,5 8,1 7,1 5,5 8,0 8,3 6,1
Headache or Migraine 7,8 5,5 8,2 5,8 6,7 5,0 2,2
Sleeping pills 8,5 8,1 8,6 7,6 7,1 7,3 3,3
High cholesterol 8,9 8,8 10,6 9,3 8,3 5,3 5,2
Contraceptive pills 9,4 9,1 7,6 5,7 8,1 7,5 6,2
Joint pains 10,2 9,6 10,1 10,9 7,0 8,0 6,4
High blood pressure 13,2 16,8 17,3 18,1 14,0 12,6 11,2
Other drugs 16,0 19,4 14,6 13,1 13,8 12,8 10,7
Pathologies /
Kinds of drugs
%
NUTS II
OCD
Norte
6.3
Centro
7.4
LVT
6.5
Alentejo
6.2
Algarve
4.7
Aores
5.9
Madeira
4
OP
Norte
6.5
Centro
8.1
LVT
7.1
Alentejo
5.5
Algarve
8
Aores
8.3
Madeira
6.1
HM
Norte
7.8
Centro
5.5
LVT
8.2
Alentejo
5.8
Algarve
6.7
Aores
5
Madeira
2.2
SP
Norte
8.5
Centro
8.1
LVT
8.6
Alentejo
7.6
Algarve
7.1
Aores
7.3
Madeira
3.3
HC
Norte
8.9
Centro
8.8
LVT
10.6
Alentejo
9.3
Algarve
8.3
Aores
5.3
Madeira
5.2
CP
Norte
9.4
Centro
9.1
LVT
7.6
Alentejo
5.7
Algarve
8.1
Aores
7.5
Madeira
6.2
JP
Norte
10.2
Centro
9.6
LVT
10.1
Alentejo
10.9
Algarve
7
Aores
8
Madeira
6.4
HBP
Norte
13.2
Centro
16.8
LVT
17.3
Alentejo
18.1
Algarve
14
Aores
12.6
Madeira
11.2
OD
Norte
16
Centro
19.4
LVT
14.6
Alentejo
13.1
Algarve
13.8
Aores
12.8
Madeira
10.7 Other drugs
High blood pressure
Joint pains
Contraceptive pills
High cholesterol
Sleeping pills
Headache or migraine
Other pain
Other cardiovascular disease
Figure 4: Table for the nine
most common pathologies or
kinds of drugs in Portugal for
each region of NUTS II
Figure 5: Matlab Line chart
(without value labels) for the
nine most common pathologies
or kinds of drugs in Portugal for
Figure 6: Excel Line chart
(with value labels) for the nine
most common pathologies or
kinds of drugs in Portugal for

GIS PUBLIC HEALTH PROCEEDINGS 2011

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

GIS PUBLIC HEALTH PROCEEDINGS 2011

Uploaded by

Copyright:

Available Formats

URISAs 2011 GIS in Public Health Conference PROCEEDINGS

contiguity matrices can be constructed, as well as distancebased or knearest neighbor adjacency

Food store and

s03 5637 11/29/2007 21:54 1271066 228945 2 1.44 63.8 321.62

files {Emerging Infectious Diseases}

*Laboratory of Biomathematics, Faculty of Medicine of Lisbon,

ISEGI, New University of Lisbon,

CINAV - Portuguese Navy Research Center, Escola Naval

You might also like