by Jess Behrens
© 2005-2019 Jess Behrens, All Rights Reserved
Included below is a brief description of several of the GIS projects I've worked on over the years.
Use of the Gravity Model to Adjust & Analyze Case Distribution
This project looked at the rates of Diabetic Retinopathy (DR) diagnosis by Chicago zip code, specifically if the rates of DR diagnosis are low in areas where the risk of diabetes is high. Using the gravity model (FCA 2 Step) to adjust for the distance of each zip code from the primary location for each of the participating health centers, I was able to show that there is a significantly lower rate of DR diagnosis in zip codes where residents are at a much higher risk of having diabetes. Conversely, there is a much higher rate of DR diagnosis in areas where the risk of having diabetes is significantly lower. The final product of the analysis was the number of DR cases above the expected number (based on national averages) for both total general hospital visits and total DR cases.
These two areas also conform to a strong socio-economic split. with the low Diabetic Retinopathy diagnosis areas also being much more socio-economically challenged, and vice versa for the high DR areas.
I am not able to reproduce the images or tables involved I produced for this project here due to the publication process & rules. However, I have applied for a provisional patent, using different but related data. The Patent Receipt Number is:
62/740,508 Novel Method for Adjusting Disease Case Ratios, or Other Essential Fact Ratios, by Zip Code or other Geographic Unit, using the Gravity Model
Liquor Licenses & Gunshot Wounds
A unique project that melded traditional statistical programming & GIS, while with Northwestern University's Medical School, I worked on a project looking at the relationship between gunshot wounds & packaged goods liquor stores. The method involved using the gravity model as described in the FCA 2 Step Method for estimating the probable relationship between a gunshot wound and all of the liquor licenses within the study area.
These gravity estimates were then aggregated by census tract and used as an independent variable in an iterative geographically weighted regression (GWR) analysis. GWR derives from multi-variate Ordinary Least Squares (OLS) regression but focuses on the spatial non-stationarity of the independent variables (measured using the Koenker statistic). Thus, significant, traditional OLS work must first be done prior to GWR as well as during each step of breaking the study areas into regions. Logistic regression was then used to identify the Odds Ratio for liquor licenses within each region.
To ensure that the relationship was not spurious, the process was done repeatedly using multiple business and location types. Schools, churches, grocery stores, etc. were included in the gravity calculations used in this second stage of the analysis. Two comparisons were made:
The first where the liquor licenses were included with all other business types
A second where only the other business types were included (liquor licenses left out)
While risk remained with the liquor licenses included, the risk disappeared when the liquor licenses were dropped.
Of note is the fact that risk was unequal across the study area, and the liquor licenses actually decreased the overall risk of gunshot wounds in some areas. Obviously, the focus was on those areas where the risk significantly increased, but noting this contrary fact is vitally important. The relationship between gunshot wounds and liquor licenses is a complex one that involves multiple social, economic, & spatial dynamics.
Due to publication rules, I can't reproduce the maps and tables here. If you're interested in reading about the study, it has been published in the American Journal of Surgery.
Crandall M, Kucybala K, Behrens J, Schwulst S, Esposito T. Geographic association of liquor licenses and gunshot wounds in Chicago American Journal of Surgery 2015;210(1): 99-105.
Emergency Room Wait Times
In this unfinished manuscript (ERPaper.pdf), I use national rates to estimate the number of emergency room visits by US Census Block groups. I then use tools from economics (price elasticity, time cost, etc.), the gravity model, & ascendency theory (from ecology) to model real hospital wait times in the Chicago area. The work speaks for itself, even though I'm pretty sure I used incorrect symbology in the equations. Regardless, the math works out just as it is described.
While with Northwestern Medical School, I collaborated on a research project that resulted in a patent (Serial No. 62/287,164). This idea is mine & goes back to a conversation I had with a fellow teaching assistant in the Dept. of Geography-Geology at the University of Nebraska-Omaha (UNO) in December on 1997. The discussion was around the limits to which a geographer could interpolate, or impute, data that had already been aggregated. At that time, I used data from a project at UNO and applied something called 'pycnophylactic interpolation' by way of ESRI's scripting language Avenue. When I went on to study Ecology at Colorado State University, I learned more about the field of geo-statistics, and even attempted a limited version of this work using data from a different project.
The goal of the work, and what we successfully accomplished, was to combine a limited public health dataset with US Census Data to develop a fine resolution, geo-spatially referenced raster showing probable case location. The process involved a number of steps, including Monte Carlo simulations, a krig & a gaussian geo-statistical simulation. Due to publication and patent rules, I can't reproduce images or the math here. However, if you'd like to read about the work, it has been published in the Journal of the American Medical Informatics Association as well as the proceedings from the 2016 American Medical Informatics Association Conference.
Small Area Estimates to Predict Where Medicaid Patients Reside, Journal of the American Medical Informatics Association Conference, 2016.
Behrens, J., Wen, X., Goel, S., Zhou, J., Fu, L., & Kho, A.N. Using Monte Carlo/Gaussian Based Small Area Estimates to Predict Where Medicaid Patients Reside AMIA Annu Symp Proc, 2016: 305-309.