Exploring Spatial Patterns of Disease with Geographic Information Systems

1
Exploring Spatial Patterns of Disease with
Geographic Information Systems

• Gerard Rushton
• Department of Geography and College of Public
  Health
• The University of Iowa
• Information and Health at Iowa Breaking
  Boundaries Building Bridges
• Iowa City, Iowa. October 29, 2004

2
Public Health functions that require information
processing capabilities that recognize location
and geography explicitly

• conduct surveillance to identify areas of need
• match resources geographically to areas of need
• measure the association between environmental
  exposures and disease rates

3
Organization

• Short vignettesapplications of Geographic
  Information Science in public health

4
Rothenburgs 1836 Map of Cholera
FRB/1999/09
5

Use of GIS Techniques in Spatial Epidemiology is
not New
Broad Street Pump
Dr. Snows map of deaths in London in a two week
period in 1855 from Cholera
FRB/1999/09
6
Location of Infant Births and Deaths 1996-98
7
Des Moines, Iowa Infant Mortality Rate ZipCode
Areas, 1989-1993
8
Des Moines, Iowa Infant Mortality Rate Census
Tracts, 1989-1993
9
Des Moines, Iowa Infant Mortality Rate Census
Block Groups, 1989-1993
10
The variability of any disease rate depends on
the size of the areas mapped.

• For this example of infant mortality in Des
  Moines, Iowa, 1989-1992
• for zip codes the imr rate is from 0 to 20 deaths
  per thousand
• for census tracts the rate is from 0 to 36
• for block groups the rate is from 0 to 72

11
Des Moines, Iowa Births and Infant Mortality
Rates All Races, 1996-98
12
Infant Mortality Rates at Three Different Spatial
Scales and Their Approximate Counterparts Using
Available Census Administrative Areas
Des Moines, Iowa 1989 - 1992
Spatial Filters 1.2 miles 0.8 miles 0.4 miles
13
We Can See the Geography Much Better by draping
the computed map on a USGS 124,000 TIF Image
23
14
Significant Development in GIS and Population
Health

• The question of selecting the best level of
  geography to attach to the health event is
  usually NOT the appropriate question
• 2. Controlling the geography and examining
  different spatial scales is usually important
• 3. Geocoding events permits analyses and
  interventions at different geographic scales

15
A Key Issue of GIS in Public Health--How to
define regions for public health
intervention?Where to intervene to reduce the
infant mortality rate in Des Moines, Iowa Where
are the high rate areas?
16
Des Moines Infant Mortality Project Area Seven
Census Tracts
I-235
Appendix A-1
17
Infant Mortality Rates for Census Tracts in Des
Moines, Iowa, 1991-1993
18
Infant Mortality Rate, 1993-94
Rates based on address-matched births and deaths
19
Comparison of Methods for Defining High IMR Areas
in Des Moines, Iowa 1991-1993
Healthy Start
Des Moines Region
1346
20
Significant Development in GIS and Population
Health
Using political or administrative areas to
define regions for public health interventions is
not the best way to define regions. GIS brings
spatial flexibility to the task of defining the
most appropriate regions for intervention
activities.
21
Susan S. Devesa1 Dan J. Grauman1 William J.
Blot2 Gene A. Pennello3 Robert N. Hoover1 Joseph
F. Fraumeni, Jr.1 1Division of Cancer
Epidemiology and Genetics National Cancer
Institute 2International Epidemiology Institute,
Ltd. 3Center for Devices and Radiological
Health Food and Drug Administration
Published December 1999
22
Associated Web site components
The Atlas
23
Web-based linked microplot maps. (Wang et al.
2002)
Source http//graphics.gmu.edu/xwang/cancer4/
24
Significant Development in GIS and Population
Health

• Web-based interactive mapping and linked
  microplots are significant developments.

25
Is there a geographic basis for cancer control
and prevention activities?

• Regional variations in the cancer burden include
• Rates of mortality?
• Rates of incidence?
• Rates of survival?
• Rates of stage at first diagnosis?
• Rates of screening?
• Objectives should be set to meet the needs of
  local, regional areas.
• Actions should reduce local cancer burdens.

26
The toolkit described in the CDC report clearly
identified the county as the unit for
evaluating cancer burden

• In the U.S. the county traditionally has been a
  reporting unit for morbidity and mortality data.
• But the county is rarely an area that
  characterizes people who belong to a health
  service area.
• The county is also an area with large differences
  in population which makes it unsuitable for
  computing and displaying disease rates.

27
Cases of Breast Cancer in Iowa 1993-1997
9,870 cases
9,738 included
131 excluded
1 excluded because
because of
of no zip code
particular morphology
418 Unstaged
9,320 Staged
Early 90.1
Late 9.9
28
Illustration of the raw data number of breast
cancer cases in the six-year period south east
Iowa.
29
Percent of Breast Cancer Cancer Patients with
Late Stage Tumors at Time of First Diagnosis.
Iowa, 1993-1997
30
Illustration 15 Mile Spatial Filter on a 4 Mile
Rectangular Grid

• The four colored dots represent four grid points
  spaced 4 miles apart.
• The four colored circles represent spatial
  filters of radius 15 miles.

31
Percent Late-stage Breast Cancer 1993-1997
32
Compare areas of high rates of lumpectomy/radiatio
n with locations of radiation treatment
facilitiesChoices per 1,000 cases of localized
breast cancer, 15 mile filter
33
Rates of Breast Conserving Strategiessmoothed
maps at three different scales 10 Mile
Filter 12.5 Mile Filter
15 Mile Filter
Choice rate per thousand women with localized
breast cancer--alternative choice was mastectomy.
34
Significant Development in GIS and Population
Health

• GIS could be used to make explicit the implicit
  questions about geographical variations in the
  CDC Framework for Comprehensive Cancer Control
  model.

35
http//toxmap.nlm.nih.gov
36
http//toxmap.nlm.nih.gov/toxmap/home/welcome.do
37
(No Transcript)
38
(No Transcript)
39
(No Transcript)
40
http//toxmap.nlm.nih.gov
41
http//www.sciencedaily.com/releases/2004/10/04101
2082648.htm
42
http//www.arb.ca.gov/ch/chapis1/chapis1.htm
43
(No Transcript)
44
(No Transcript)
45
http//www.pca.state.mn.us/air/at-cep.html
46
Significant Development in GIS and Public Health

• National geospatial databases of environmental
  pollutants are being created by linking pollutant
  attributes to geospatial framework data from the
  national spatial data infrastructure.

47
COMPREHENSIVE ASSESSMENT OF RURAL HEALTH IN IOWA
Presentation and Demonstration of the CARHI GIS
Environmental Health Surveillance System
January 26, 2004 Iowa Department of Public
Health Des Moines, IA
Department of Geography Center for Health Effects
of Environmental Contamination (CHEEC) Department
of Occupational and Environmental Health
Department of Family Medicine
48
Carroll County, Iowa About 24 miles x 24
miles Major Urban Areas Carroll
Manning Coon Rapids
Carroll
Manning
Coon Rapids
Urban Places in Carroll County
Source Carroll Geographic Information System
49
Possible Sources of Environmental
Contamination This list was compiled by the
Center for Health Effects of Environmental
Contamination (CHEEC)
50
202 ft
Orthophotos 2 ft resolution
51
Orthophotos 2 ft resolution
2600 ft
52
Parcels with Livestock Building Permits in
Carroll County
Livestock Building Permits
53

AFO with manure management plan
Parcel with livestock permit
Parcel with NO livestock permit
54
Map of Animal Feedlot Operations (AFOs) in
Carroll County. There are 153 operations in
this dataset, of which 55 have manure management
plans (Source Iowa Geological SurveyIGSB) The
blue squares are parcels with livestock building
permits (Source County Records, Carroll County
GIS) There are 36 records common to both these
datasets, most of which are AFOs with manure
management plans. Note Parcel Identification
Numbers were added to each AFO in the IGSB
dataset
Legend

AFOs (Source IGSB)
Parcels with Livestock Building Permits
Roads
55
Example of data point from IGSB file and
corresponding parcel in the livestock database.
The livestock database gives name, permit number
and value of property. The AFO database gives
additional information about the type of
livestock building and manure storage plan. The
Identify Results window displays the
information from the AFO file for this particular
facility

AFO
Parcel polygon
The features are displayed on a 2 ft orthophotomap
56
Example of rural addresses in the Carroll County
E911 address file. Note USPS Standardization A
few addresses do not have a unit number (e.g. the
record following 32264, 330th street)
57
The distance between the E911 address point and
the improved point in this case is approximately
955 feet The orthophoto displayed is at the 2
ft resolution
955 feet
58
(No Transcript)
59
Address point map containing the locations of
rural addresses in Carroll County, IA
This map shows the extent of the address point
file as of August 2003. This file contains the
locations of 100 of the addresses in the E-W
direction and 2/3rds of the addresses in the N-S
direction. This file is now complete for all
rural addresses (January 2004)
Carroll
Manning
Legend
Coon Rapids
Locations of addresses
Roads
60
Attribute table of the E911 address point file
received from Carroll County
Attribute table of the improved address point
file created at the UI
The key field, Unitid, can be used to link the
two tables and hence associate an address with
each of the lat-long coordinates and vice versa.
The reason for doing is this is that it takes the
person less time to write down a single number vs
an entire street address. The numbers used are
arbitrary. The field Codeinfo classifies the
degree of certainty in definitively identifying
the location of the house
61
(No Transcript)
62
Roles of the GIS in the Health Surveillance System

• To georeference all types of data (geocode)
  environmental pollutants, rural and urban
  addresses, other framework data such as roads,
  property lines, other features.
• To provide access to typical GIS functions such
  as show coverage, make a buffer, count cases
  within a buffer, make a map
• To provide access to spatial analysis tools to
  link exposure data to the health encounter data.
• To develop and run simulations of possible
  environmental exposure effects to estimate the
  power of the analysis to detect health effects of
  environmental contamination.
• To provide visualizations of the associations
  between measures of health and potential
  environmental exposure.

63
When two sets of data are geo-coded spatial
relationships can be computed

• Visualizations of spatial choice
• Comparisons of chosen with potential places for
  interaction
• Imputing changes in spatial choices through time
• Finding distances to closest, second-closest,
  etc. places with given resources

64
(No Transcript)
65
(No Transcript)
66
(No Transcript)
67

• Sample spatial choice data can be used to create
  regions for more rational analysis of population
  health and relationships with health care
  resources
• The following map is based on a national random
  sample (17.1 million Medicare records).
• The primary health care areas project

68
Iowa Primary Care Service Areasproportion of
primary care patient visits made inside the local
service area
Sioux Center
Hampton
Maquoketa
Carroll
proportion
Manning
Miles
Source data files from pcsa.hrsa.gov See
Goodman et al. 2003
69
(No Transcript)
70
HRSA PCSA Datafile as accessed February, 2003.
71
(No Transcript)
72
Significant Development in GIS and Population
Health

• Computing geographic accessibility information
  from geo-coded data is a significant development
  in GIS for population health

73
Using geographic information science to link
information on health, population, and the
environment is an informatics challenge with
rewarding career opportunities.