Today HealthLandscape is releasing Geospatial Brief #2 “Where are “Hot Spots” of Medicare Spending and Preventable Hospitalizations and “Cold Spots” of Preventive Care. Brief #2 focuses on using advanced geospatial methods to identify priority areas for preventive care. This blog briefly describes different ways of defining hot spots and cold spots. We argue that it doesn’t necessarily matter how these terms are defined, as there are multiple methods for using these techniques to identify priority regions.
The term “hot-spotting” has become popular in the healthcare realm thanks to the work of Dr. Jeffrey Brenner in Camden, NJ, who identified a very small group of super-utilizers, which made up a disproportionate share of hospitalizations, ER visits, and healthcare costs. These “super-utilizers” were often concentrated in small geographic areas, identified as hot spots, such as apartment buildings or city blocks, which were poor, under-resourced areas. Thus, by focusing on these hot-spots and providing coordinated care and social services, Brenner and his team were able to improve health, reduce unnecessary hospital visits, and lower costs (Gawande, 2011). Brenner’s hot-spotting approach has been very successful and is being replicated in regions throughout the U.S., with the Robert Wood Johnson Foundation dedicating substantial funding for hot-spotting programs (RWJ, 2012).
Dr. John Westfall took an alternate view, wondering if the problem wasn’t really hot spots, but rather cold spots – “communities in which the social determinants of health, support, and access to primary care have broken down.” (Westfall, 2013). Westfall explains that identifying cold spots and working to improve the conditions in these areas could have a larger overall impact on improving population health. However, Westfall argues that addressing the needs of cold-spot communities is much more complex than dealing with a small group of super-utilizers, and requires a broader, communities of solution approach (Griswold et al., 2013).
Hot Spots and Cold Spots in Geospatial Analysis
While general awareness of the terms hot spots and cold spots have increased, these terms have different meanings in the field of geospatial analysis, where hot spots are defined as clusters of high values and cold spots as clusters of low values. These clusters are compared to random geographic patterns to test if they are statistically significant. Several methods exist for exploring hot spots and cold spots, including the Local Moran’s I (Anselin et al., 2006). For example, if you map Diabetes prevalence (Medicare) for counties in the U.S. (see map below), you would find clusters of high values throughout the southeast and Appalachia, and clusters of low values in the upper Midwest and throughout the Western part of the U.S. To test for statistically significant Diabetes hot spots, you would have to use advanced geospatial methods to determine if the clusters of high values are significantly different from random geographic patterns.
Diabetes Prevalence by County (Medicare)
Source: CMS Geographic Variation, 2013; HealthLandscape Medicare Data Portal
Making Sense of Hot Spots and Cold Spots
There are similarities in how hot spots and cold spots described above are defined. For example, if you had census tract data for 30-day readmission rates from a local hospital, you could map these data and visually identify census tracts with high rates (i.e., hot spots using Brenner’s definition). Next, you could use geospatial methods (such as a Local Moran’s I) to determine if the clusters of high readmission rates are significantly different from a random pattern of hospital readmissions (i.e., hot spots using geospatial definition). Similarly, you could map census tract education data to visually identify census tracts with low levels of education (i.e., cold spots using Westfall’s definition), and then use Local Moran’s I to determine if clusters of low education census tracts are statistically different from random patterns of education levels (cold spots using geospatial definition).
While there are many different ways to define hot spots and cold spots, there are all useful for identifying priority areas for place-based interventions. The key issue for future research is how we use the results of hot-spot and cold-spot analyses to target interventions.
Anselin, Luc, Ibnu Syabri and Youngihn Kho (2006). GeoDa: An Introduction to Spatial Data Analysis. Geographical Analysis 38 (1), 5-22.
Centers for Medicare and Medicaid (CMS), 2013. Geographic Variation Public Use File
Accessed at https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/Medicare-Geographic-Variation/
Dartmouth Atlas of Health Care, 2013. Data Downloads.
Gawande, Atul. (2011). “The Hot Spotters: Can we lower medical costs by giving the neediest of patients better care?” The New Yorker. January 24, 2011.
Griswold, Kim S., Sarah E. Lesko, and John M. Westfall (Folsom Group). (2013). Communities of Solution: Partnerships for Population Health. Journal of the American Board of Family Medicine 26(3): 232-238.
HealthLandscape Medicare Data Portal
Robert Wood Johnson Foundation, (2012). “Expanding “Hot Spotting” to New Communities: What We’re Learning about Coordinating Health Care for High-Utilizers.”
Accessed at http://www.rwjf.org/en/library/research/2012/01/expanding–hot-spotting–to-new-communities.htm
Topmiller, Michael. (2016). “Do Regions with More Preventive Care have Lower Spending and Fewer Hospitalizations?” HealthLandscape Geospatial Research Brief #1.
Accessed at http://www.healthlandscape.org/Geospatial-Analysis.cfm
Westfall, John M. (2013). Cold Spotting: Linking Primary Care and Public Health to Create Communities of Solution. Journal of the American Board of Family Medicine, 26(3): 239-240.