Document Details
These maps were created based on cancer data aggregated for ...
- Click : 0
- Download : 0
- File Size : 9.49 K
- Document Type :
Memo
:
These maps were created based on cancer data aggregated for different geographic units and smoothed to estimate the rates of cancer incidence, late stage cancer, and cancer mortality as a continuous surface
ICCCC Cancer Maps: A Guide
to Interpretation
These maps were created based
on cancer data aggregated for selected geographic units and smoothed
to estimate the rates of cancer incidence (newly diagnosed cancers),
late stage cancer (cancers diagnosed in an advanced stage), and cancer
mortality (cancer deaths) as a continuous surface, providing local detail
of the cancer burden in Iowa communities. These maps are posted
on the following website: <a href=http://www.uiowa.edu/%7Egishlth/ICCCCMaps/http://www.uiowa.edu/~gishlth/ICCCCMaps/
The objective of these maps
is to compare the number of cancer cases or deaths we observe to the
number we would expect to see if local rates of cancer incidence and
mortality were typical of the state as a whole. Local rates that
are higher than what we would expect are shown in red, while rates lower
than what we would expect are shown in blue, with darker shades of color
indicating the most extreme rates. Areas experiencing rates close
to what we would expect are not represented by a color. On the
black and white maps, darker shades represent higher than expected rates,
while lighter shades represent lower than expected rates.
The background used for these
maps is the 1:100,000 scale topographic map from the US Geological Survey,
which shows information such as the locations of cities and towns, and
physical landmarks such as rivers and lakes. The major highways
in Iowa, and names of some Iowa cities, are also included to assist
in interpretation.
Incidence and Late-Stage
Maps
The incidence and late-stage
maps were created based on ZIP Code level data received from the Iowa
Cancer Registry for the years 1998-2003.
For these maps, incidence and
late stage rates were adjusted in order to take into account the age
and sex of populations in each ZIP Code. If this process, called
standardization, had not been done, the highest rates would most
likely be found in places with larger populations of older people, as
older people are more likely to be affected by cancer, and therefore
would reflect the age of the population instead of differences in the
cancer burden due to other factors. The process we used is called
indirect age-sex standardization and results in a measure called
a standardized incidence ratio (SIR).
To make the incidence rate
map, the expected number of cancer cases for each ZIP Code was
computed by multiplying the statewide rates of cancer for each age-sex
group (for example, the rate experienced by males ages 50-54) by the
numbers of people in the same age-sex group in the ZIP Code area, and
then summing up the total number of expected cases for all age groups
together. For the late-stage map, the expected number of late
stage cases for each ZIP Code was computed by multiplying the statewide
rate of late stage cancer (the proportion of all cancers diagnosed in
the late stages) for each age-sex group to the number of cancer cases
in each age-sex group in each ZIP Code, and summing up the total number
of late stage cases we would expect for all age groups together.
The observed numbers of cancer cases (the data we received from
the Cancer Registry for each ZIP Code), and the number of expected cases
(which we calculated), were then placed at a single point location within
each ZIP Code area that represented the geographic center of that ZIP
Code's population.
Then, in order to create a
continuous surface, a method called adaptive filter kernel density
estimation was used. Density estimation methods attempt to
estimate the density of a phenomenon, such as cancer cases per population
(a cancer rate), for a particular location based on the distribution
of that phenomenon in the surrounding area. This process assumes
that places closer in space are more likely to be similar than places
far from each other - this is known as the "first law of geography."
Kernel density estimation is
accomplished by placing a uniform grid of points over an area and estimating
the density at each grid point by pulling in observations from the surrounding
area using a circular window centered on the grid point. In adaptive
filter kernel density estimation, the size of this window will vary
for each grid point because it will expand until a certain number of
cases are pulled in to ensure that the rate calculated is stable -
in other words, the rate is not subject to very much variation and is
likely to represent the true rate experienced in a local area.
Grid points located in urban areas will have smaller windows than those
in rural areas because the window does not need to expand very far before
it pulls in enough cases, due to the higher populations, and more cases,
in urban areas.
To construct our incidence
maps, a grid of points was placed over the state of Iowa, and a standardized
incidence ratio (SIR) was calculated for each grid point by pulling
in observed and expected cases from the surrounding area. A SIR
is the number of observed cases divided by the number of expected cases.
A value of 1 means that we observed exactly the number of cases we would
expect. A value greater than one means that we observe more cases
than we would expect, while a value less than one means that we observe
fewer cases than we would expect - given the statewide cancer rates.
The further a rate is from 1, the more it differs from what we would
expect. For each grid point, we pulled in observed and expected
cases until we had pulled in at least 50 expected cases in order to
ensure a stable cancer rate. We then smoothed the data between
the grid points in order to create a continuous surface.
Mortality Maps
The mortality maps were created
based on city and county level data received from the Iowa Department
of Public Health for the years 1999-2003.
A procedure similar to the
one described above was used to create the prostate cancer mortality
map. For the mortality data, we received the county of residence
for each person who died of cancer, as well as the city they lived in,
or whether they lived outside city limits, in the "rest of the county."
We used population data from the US Census to calculate mortality rates.
This information is only available for some cities. For the cities
for which we could find population data, we aggregated all cancer deaths
associated with that city. The rest of the deaths were aggregated
by the county of the person's residence, and were placed at the geographic
center - the centroid - of the county. Population data for
these cases was calculated by subtracting the populations of the cities
for which we had information from the county populations.
We then used the same process
of adaptive filter kernel density estimation to pull in observed and
expected cancer deaths for each grid point in order to calculate standardized
cancer mortality rates. In this case, the circular window expanded
until it pulled in at least 37 expected deaths.
What do these maps tell
us?
Maps help us to visualize information
and to understand how phenomena, such as the burden of cancer in Iowa,
vary geographically. It is often the case that one will encounter
maps presenting cancer rates based on geographic units such as counties
or census tracts. Unfortunately, in these types of maps, large
geographic areas with small, rural populations often dominate the map
and mislead the map reader by drawing his/her eye. Because rural
areas have lower populations, it is often the case that rates mapped
for rural areas are unreliable because not enough population exists
in order to calculate a stable rate. Unfortunately, these rates
are also very high or very low as well as being unreliable, which causes
the map reader to become focused on these areas, ignoring other, smaller
areas that may have rates that are both stable and high.
In order to avoid these problems,
we created our maps as a smooth surface. These "smoothed"
maps are made in order to provide the most accurate estimation of cancer
incidence and mortality rates in local areas as possible, while avoiding
the problem of unstable rates and large areas with small populations
dominating the map. However, while the method of adaptive filter
kernel density estimation helps us to smooth rates over space and make
them more reliable, the map reader must keep in mind that rates calculated
in rural areas are based on larger areas than those calculated in urban
areas - in other words, the resolution is different across
the map.
Overall, these maps can help
to guide our cancer prevention and control efforts by offering us information
about what places within our state are more effected than others by
certain cancers. If a community is experiencing a rate that is
higher than expected, that community may begin to question what is causing
this elevated rate, and consider actions that can be taken to address
their local cancer burden.
Contact Dr. Gerry Rushton (<a href=mailto:gerard-rushton@uiowa.edugerard-rushton@uiowa.edu) if any questions.
Created by Kirsten Beyer, 10/2006.
These maps were created based on cancer data aggregated for different geographic units and smoothed to estimate the rates of cancer incidence, late stage cancer, and cancer mortality as a continuous surface
ICCCC Cancer Maps: A Guide
to Interpretation
These maps were created based
on cancer data aggregated for selected geographic units and smoothed
to estimate the rates of cancer incidence (newly diagnosed cancers),
late stage cancer (cancers diagnosed in an advanced stage), and cancer
mortality (cancer deaths) as a continuous surface, providing local detail
of the cancer burden in Iowa communities. These maps are posted
on the following website: <a href=http://www.uiowa.edu/%7Egishlth/ICCCCMaps/http://www.uiowa.edu/~gishlth/ICCCCMaps/
The objective of these maps
is to compare the number of cancer cases or deaths we observe to the
number we would expect to see if local rates of cancer incidence and
mortality were typical of the state as a whole. Local rates that
are higher than what we would expect are shown in red, while rates lower
than what we would expect are shown in blue, with darker shades of color
indicating the most extreme rates. Areas experiencing rates close
to what we would expect are not represented by a color. On the
black and white maps, darker shades represent higher than expected rates,
while lighter shades represent lower than expected rates.
The background used for these
maps is the 1:100,000 scale topographic map from the US Geological Survey,
which shows information such as the locations of cities and towns, and
physical landmarks such as rivers and lakes. The major highways
in Iowa, and names of some Iowa cities, are also included to assist
in interpretation.
Incidence and Late-Stage
Maps
The incidence and late-stage
maps were created based on ZIP Code level data received from the Iowa
Cancer Registry for the years 1998-2003.
For these maps, incidence and
late stage rates were adjusted in order to take into account the age
and sex of populations in each ZIP Code. If this process, called
standardization, had not been done, the highest rates would most
likely be found in places with larger populations of older people, as
older people are more likely to be affected by cancer, and therefore
would reflect the age of the population instead of differences in the
cancer burden due to other factors. The process we used is called
indirect age-sex standardization and results in a measure called
a standardized incidence ratio (SIR).
To make the incidence rate
map, the expected number of cancer cases for each ZIP Code was
computed by multiplying the statewide rates of cancer for each age-sex
group (for example, the rate experienced by males ages 50-54) by the
numbers of people in the same age-sex group in the ZIP Code area, and
then summing up the total number of expected cases for all age groups
together. For the late-stage map, the expected number of late
stage cases for each ZIP Code was computed by multiplying the statewide
rate of late stage cancer (the proportion of all cancers diagnosed in
the late stages) for each age-sex group to the number of cancer cases
in each age-sex group in each ZIP Code, and summing up the total number
of late stage cases we would expect for all age groups together.
The observed numbers of cancer cases (the data we received from
the Cancer Registry for each ZIP Code), and the number of expected cases
(which we calculated), were then placed at a single point location within
each ZIP Code area that represented the geographic center of that ZIP
Code's population.
Then, in order to create a
continuous surface, a method called adaptive filter kernel density
estimation was used. Density estimation methods attempt to
estimate the density of a phenomenon, such as cancer cases per population
(a cancer rate), for a particular location based on the distribution
of that phenomenon in the surrounding area. This process assumes
that places closer in space are more likely to be similar than places
far from each other - this is known as the "first law of geography."
Kernel density estimation is
accomplished by placing a uniform grid of points over an area and estimating
the density at each grid point by pulling in observations from the surrounding
area using a circular window centered on the grid point. In adaptive
filter kernel density estimation, the size of this window will vary
for each grid point because it will expand until a certain number of
cases are pulled in to ensure that the rate calculated is stable -
in other words, the rate is not subject to very much variation and is
likely to represent the true rate experienced in a local area.
Grid points located in urban areas will have smaller windows than those
in rural areas because the window does not need to expand very far before
it pulls in enough cases, due to the higher populations, and more cases,
in urban areas.
To construct our incidence
maps, a grid of points was placed over the state of Iowa, and a standardized
incidence ratio (SIR) was calculated for each grid point by pulling
in observed and expected cases from the surrounding area. A SIR
is the number of observed cases divided by the number of expected cases.
A value of 1 means that we observed exactly the number of cases we would
expect. A value greater than one means that we observe more cases
than we would expect, while a value less than one means that we observe
fewer cases than we would expect - given the statewide cancer rates.
The further a rate is from 1, the more it differs from what we would
expect. For each grid point, we pulled in observed and expected
cases until we had pulled in at least 50 expected cases in order to
ensure a stable cancer rate. We then smoothed the data between
the grid points in order to create a continuous surface.
Mortality Maps
The mortality maps were created
based on city and county level data received from the Iowa Department
of Public Health for the years 1999-2003.
A procedure similar to the
one described above was used to create the prostate cancer mortality
map. For the mortality data, we received the county of residence
for each person who died of cancer, as well as the city they lived in,
or whether they lived outside city limits, in the "rest of the county."
We used population data from the US Census to calculate mortality rates.
This information is only available for some cities. For the cities
for which we could find population data, we aggregated all cancer deaths
associated with that city. The rest of the deaths were aggregated
by the county of the person's residence, and were placed at the geographic
center - the centroid - of the county. Population data for
these cases was calculated by subtracting the populations of the cities
for which we had information from the county populations.
We then used the same process
of adaptive filter kernel density estimation to pull in observed and
expected cancer deaths for each grid point in order to calculate standardized
cancer mortality rates. In this case, the circular window expanded
until it pulled in at least 37 expected deaths.
What do these maps tell
us?
Maps help us to visualize information
and to understand how phenomena, such as the burden of cancer in Iowa,
vary geographically. It is often the case that one will encounter
maps presenting cancer rates based on geographic units such as counties
or census tracts. Unfortunately, in these types of maps, large
geographic areas with small, rural populations often dominate the map
and mislead the map reader by drawing his/her eye. Because rural
areas have lower populations, it is often the case that rates mapped
for rural areas are unreliable because not enough population exists
in order to calculate a stable rate. Unfortunately, these rates
are also very high or very low as well as being unreliable, which causes
the map reader to become focused on these areas, ignoring other, smaller
areas that may have rates that are both stable and high.
In order to avoid these problems,
we created our maps as a smooth surface. These "smoothed"
maps are made in order to provide the most accurate estimation of cancer
incidence and mortality rates in local areas as possible, while avoiding
the problem of unstable rates and large areas with small populations
dominating the map. However, while the method of adaptive filter
kernel density estimation helps us to smooth rates over space and make
them more reliable, the map reader must keep in mind that rates calculated
in rural areas are based on larger areas than those calculated in urban
areas - in other words, the resolution is different across
the map.
Overall, these maps can help
to guide our cancer prevention and control efforts by offering us information
about what places within our state are more effected than others by
certain cancers. If a community is experiencing a rate that is
higher than expected, that community may begin to question what is causing
this elevated rate, and consider actions that can be taken to address
their local cancer burden.
Contact Dr. Gerry Rushton (<a href=mailto:gerard-rushton@uiowa.edugerard-rushton@uiowa.edu) if any questions.
Created by Kirsten Beyer, 10/2006.