Join CSTE   |   Career Center   |   Print Page   |   Contact Us   |   Report Abuse   |   Sign In
CSTE Features
Blog Home All Blogs
Search all posts for:   


View all (132) posts »

Exploring Geographic Rate Variation Using Indirect Spatial Estimation in Montana

Posted By Luke Baertlein, Friday, October 30, 2015
Updated: Friday, October 30, 2015

In the Montana Asthma Control Program we were faced with a problem of wanting to report on the geographic distribution of asthma burden while not having spatial data or data we could report at county-level aggregation. To get around this, we used an indirect spatial estimation method to explore and report on the spatial distribution of asthma morbidity. Indirect spatial estimation can be conducted when aggregate data are available for small areas, providing a means to approximate the spatial distribution of a metric when spatial data are unavailable. We analyzed emergency department (ED) visits for asthma and estimated the spatial variation in population-based rates using kernel density estimation (KDE) applied to ZIP-code area aggregated data.

Data for this project came from emergency department datasets for 2010 through 2013, accessed through the Montana Hospital Discharge Data System (MHDDS) and provided by the Montana Hospital Association. At the time of the analysis, our data use agreement did not allow reporting data aggregated below the region level, including at the county level. This was later changed, allowing us to compare our spatial estimates to county-aggregated estimates. MHDDS datasets are based on the Uniform Billing forms collected from participating hospitals and represent over 90 percent of hospitalizations and ED visits in the state. Asthma ED visits were defined as any visit with a primary diagnosis coded as ICD-9-CM 493. The ZIP code of residence was used to approximate location of residence.

The creation of the spatial map followed four key steps:

  1. Create a polygon map of rates: The ZIP-code area counts of asthma ED visits and population were used to estimate and map ZIP-code area rates of asthma ED visits per 100,000 persons per year.
  2. Convert from polygon to raster format: A cell grid was overlaid on the map of ZIP-code area rates and each cell was assigned the rate of the ZIP-code area containing it.
  3. Apply KDE to raster data: KDE with a 50km bandwidth was applied to the cell grid, producing a rate for each point equal to the average rate within a 50km radius of the point. A bandwidth that would cross ZIP-code boundaries from most points was used so that the rates were smoothed across ZIP-code areas.
  4. Test for regions with rates different from the statewide average: Significance testing was applied using the Getis-Ord GI* statistic, again with a 50km bandwidth, to test the difference of each point and its surrounding points from the statewide rate.
The map of the spatial estimates of asthma ED visit rates is shown in Figure 1. For comparison, the same data aggregated at the county level is shown in Figure 2.
Figure 1. Spatial estimates of the relative rate of asthma emergency department
visits, Montana, 2010-2013, Montana Hospital Discharge Data System
Figure 2. County estimates of the relative rate of asthma emergency department
visits, Montana, 2010-2013, Montana Hospital Discharge Data System

We found a trend in overlap of regions with rates higher than the state and American Indian reservations. Of the six regions with rates higher than the state, five overlapped with reservations. Only one of the six reservations did not have a rate detectably higher than average. While a racial disparity in asthma prevalence in Montana has been found in the statewide BRFSS, race is not recorded in the Montana hospital discharge data system. By examining the spatial distribution, we were able to point to a potential racial disparity in ED visit rates. However, the pattern is not as apparent in the county-aggregated map. This could be further examined by including more detailed geographic race distribution data, such as census data, in the analysis.

As with all methods, this one is not without limitations. There is a general bias in this method against detecting small areas with high rates surrounded by areas with low rates. These tend to be estimated to have a lower than actual rate due the lower rates of their surrounding points. For example, geographically small cities with high rates located in regions with low rates outside the city would likely be underestimated. There is also potential bias from the use of ZIP-code area aggregation. The spatial approximation assumes that the rate is constant within each ZIP-code area. Also, the spatial estimates are influenced by choice of bandwidth which is somewhat arbitrary. For this map a 50km bandwidth was chosen to ensure adequate smoothing over ZIP-code boundaries at a state-wide level. However, a smaller bandwidth may have been more appropriate for areas with smaller ZIP code areas, such as major cities. Finally, this method does not take the precision of the ZIP-code area rate estimates into account. Given these limitations, inferences about the true spatial distribution of asthma ED visit rates based on this method should be made with caution.

While the limitations of an indirect spatial estimation using the method outlined above may limit its use for scientific inference, it may be useful for public health planning and communication when other options are not available, especially when used in conjunction with political-area estimates, such as county asthma ED rate estimates in this case. A map of a spatial distribution rather than of a distribution by political boundaries, such as by counties in a state, can be a tool to communicate the geographic distribution of a disease that promotes consideration of environmental factors (in a broad sense, including the social and economic environment as well as the physical) while de-emphasizing the role of local political areas (e.g. counties).

Luke Baertlein, MPH is an epidemiologist for the Asthma Control Program at the Montana Department of Public Health and Human Services. For more information about environmental health issues related to asthma, visit CSTE’s Environmental Health / Occupational Health / Injury Steering Committee page.

This post has not been tagged.

Share |
Permalink | Comments (0)
Association Management Software Powered by YourMembership  ::  Legal