GeoMed
Sample data source: Case Studies in
Biometry, 1994, N. Lange, L. Ryan, L. Billard, D. Brillinger, L. Conquest,
and J. Greenhouse (eds.), John Wiley & Sons, Inc., New York (http://lib.stat.cmu.edu/datasets/csb/.index.html)
Chapter 1: Spatial Pattern Analyses to
Detect Rare Disease Clusters, L.A. Waller, B.W. Turnbull, L.C. Clark, and P.
Nasca
This data set consists of 592 Leukemia cases
(1978-1982) diagnosed in an 8 county region in upstate New York and reported to
the state Cancer Registry. The 8 counties were divided into 790 subregions for
which 1980 Census counts were available to estimate population at risk. There
are 5 columns of data for each of the 790 rows. The first column is a label
identifying the subregion. The second and third columns are x- and y-
coordinates, indicating the location of the centroid of the subregion. The
forth and fifth columns were swapped to meet the input data file format
required by GeoMed. The forth column represents the (fractional) number of
cases, and the fifth column represents the population-at-risk size for each
subregion.
Try the GeoMed statistical advisor
(double click on GeoMed.chm in the directory containing the GeoMed application). It will provide guidance
for selecting a method appropriate for your investigation and data from those
methods available in the GeoMed software.
These data were collected retrospectively in
order to investigate the spatial pattern of leukemia occurrence in the
five-year period 1978-1982. This is in
contrast to data collected and monitored in an ongoing fashion with the goal of
detecting clustering soon after it occurs (surveillance). These data are aggregated over the entire
study period, so we are interested in investigating spatial clustering. Keep in mind that this data set consists of case
and population-at-risk counts aggregated by subregion.
Work your way through the advisor to a
global, spatial method.
UMich students: due to technical difficulties, you may get a missing page error at the point at which you arrive at the method of choice and select it. Never fear: you can get the information from Method list in the contents box.
You can read up a bit on the method (look in the index
to find general information) and then fire up the GeoMed software
(GeoMed.exe - at the dialog box titled "New", simply hit okay).
Run
the global spatial method by selecting it from the Statistical Analysis
menu option. You will be prompted to
open a data file for analysis, open the file titled leukemia.txt. You will be presented with a data entry
dialog box. Try a cluster cutoff size =
6. Run 999 Monte Carlo
randomization runs.
Run the analysis by clicking the OK button
Check the value of the test statistic for
global clustering (R) against the values generated from Monte Carlo simulation
runs. Select the menu option Graphical display > MC Frequency
Distribution.
Check the map of the subregion coordinates
and position of local clusters. Select
the menu option Graphical display > Map.
Try using the Layers list. To view only the circular cluster buffers,
unselect the transparency check mark for the "cluster locations"
layer. To view only the subregion
coordinates, move the "Region coordinates" layer to the top and
unselect its transparency check mark.
Try using the Tools. Click the magnifying glass and
"left" click the map to zoom in, "right" click to zoom out. Move the scroll bars to pan around
Run another analysis with a new cluster
cutoff size = 8.
Return to
statistical advisor and find an appropriate (new) local spatial method
for the same data set, and run the method in GeoMed. Set the population radius
= 5,000.
Check the Monte Carlo frequency distribution
against the three highest test statistic values. Check the map. To view
only the location of the cluster with the largest test statistic, move it to
the top of the list followed by the Region coordinates, and unselect the
transparency check mark for the Region coordinates
Run another analysis with a population
radius = 10,000. It's hard to
see the locations of the clusters with the 2nd and 3rd
highest test statistics, with the originally presented Layers list order, (1st
largest, followed by 2nd and then 3rd largest, and lastly
the Region coordinates) unselect the last two check marks.
You are now interested in examining whether
or not disease cases tend to cluster around particular locations. There is a hazardous waste cite located at
the x,y coordinate (-0.14, -67.19).
Return to
statistical advisor and find an appropriate focused spatial method for the same
data set simply hypothesizing that risk
increases with closer proximity to the focus, we don't assume much else
regarding the disease pattern surrounding the putative exposure source. Run the method in GeoMed.
To see the focus location on the map, with
the focus as the top layer, unselect its check mark. Run 2 more analyses with coordinates
(-4.47, -67.65) and (11.98, -71.61)
Multiple Comparisons:
We've run multiple analyses for each of the
Besag & Newell, Turnbull , and Score analyses
Select the menu option Statistical
analysis > Multiple Comparisons…
Run adjustments for Besag & Newell,
Turnbull (all statistics), and Score
Return to
statistical advisor and find an appropriate focused spatial method for the same
data set, but assume you have a
priori knowledge of how risk varies with distance from the focus.
Use the original focus coordinate = (-0.14,
-67.19), with relative risk model parameters Beta = 10 and Phi = 2. Hit the Visualize button (and
move the dialog window out of the way as best you can). Run the analysis and close all function
scatter plots in order to see the results
Data set source: provided by Peter Diggle
from his web site - the incinerator data
This data was collected retrospectively to
investigate the possible association between the spatial distribution of cases
of cancer of the larynx and the location of the now-disused industrial
incinerator (354.5, 413.6). The
spatial locations of 'control' subjects (lung cancer cases) were also
collected and included in the data set.
Return to the statistical advisor to find an
appropriate method. Run the method with
the file diggle.txt. Try relative
risk model parameters Alpha = 20 and Beta = .5. A plot of the fitted raised incidence model
(with the data plotted as cases (1) and controls (0)) is presented as part of
the output. Have a look at the map of the data cases (red), controls (blue),
and the focus (black x).
Data set source: SpaceStat Data: Africa (http://www.spacestat.com/data.htm)
Data for 42 African countries (1966-1978)
was provided by the Conflict and Peace Data Bank. An ascii text file with a
single column, the total conflict index, will be used in the analysis. An
ArcView Shape file will be used to determine the boundaries of the African
countries. The associated .dbf file has a name column used to uniquely identify
all countries to be analyzed. These countries are ordered in correspondence
with the total conflict indices in the text file.
Return to the statistical advisor to find an
appropriate local spatial method for these retrospective data. The unit of analysis is not exactly a
disease rate, but it is similar enough to pretend that is what we want to
analyze. Run the local spatial method
in GeoMed.
You will first be prompted to open the ascii
text data file for analysis, open the file titled afcon.txt. You will next be prompted to open a
shape file with boundary information, open the file titled afcon.shp.
Run 999 Monte Carlo Randomization runs. The Map provides the region boundaries,
eventually it will provide more information.
Data Set source: SaTScan Data: New Mexico
(made available with the downloadable SaTScan software at http://www.dcpc.nci.nih.gov/BB/SaTScan.html)
Cases of brain cancer in 43 New Mexico
counties from 1973 to 1991 are listed. Census population counts are provided
for the years 1973, 1982, and 1991. All covariate information has been removed.
The following files are used during analysis:
Return to the statistical advisor. Here we have data with both spatial and
temporal references, so lets take full advantage of this information. Find an appropriate spatio-temporal method
and then run it.
You will first be prompted to open the data
file with study region centroid values, open the file titled nmCoord.txt. You will next be prompted to open a data
file with case data, open the file titled nmCase.txt. You will last be prompted to open a data
file with population data, open the file titled nmPop.txt.
Run 999 Monte Carlo Randomization runs. This could take awhile: if your patience is running out, hit the cancel button and the processing will stop, using as many Monte Carlo runs as have been run to that point.