About the Data
HOW DOES THE CALCULATOR PRODUCE ITS RESULTS?
The researchers studied the statistical relationship between county death rates or diabetes prevalence and local education and income levels across thousands of U.S. counties, using a statistical technique called generalized linear regression. When the slider tool is moved, a computer program uses the regression equation to estimate the proportional change in death rate or diabetes prevalence based on the proposed change in the education or income level. This factor is then applied to the selected county to arrive at an estimate of avertable deaths and avertable diabetes cases. The number of avertable deaths is obtained by subtracting the predicted number of deaths (if education or income levels were changed to the values shown on the slider) from the actual number of deaths in the county. Similarly, the number of avertable diabetes is obtained by subtracting the predicted number of diabetes (if education or income levels were changed to the values shown on the slider) from the actual number of diabetes in the county. Diabetes cost is the product of the estimated number of diabetes cases, subject to slider values of education and income, times an estimate of the per capita diabetes care cost. This cost is estimated by multiplying the national per capita diabetes cost times the ratio between county-level and national-level Medicare expenditures. Costs saved are calculated by subtracting the predicted diabetes costs from the actual diabetes costs. Estimates for states, or for the nation, are an aggregate—derived by pooling data from the counties they encompass.
WHAT WERE THE DATA SOURCES?
Researchers at Virginia Commonwealth University collected data for almost all counties in the United States—more than 3,000 counties in all. They obtained county education and income statistics directly from the U.S. Census Bureau and county death rates and estimates of the prevalence of diabetes directly from the U.S. Department of Health and Human Services. The county death rates were taken from 2008 to 2010 the most recent years for which analyzable data were available. Data were combined over three years to produce a three-year sample that was large enough for valid statistical analysis.
Estimates of diabetes prevalence are based on self-reports of respondents to a national survey (Behavioral Risk Factor Surveillance System Survey) in 2009; the question asks, “Have you ever been told by a doctor that you have diabetes?” The estimates of medical spending on diabetes rely on an American Diabetes Association estimate of the costs of diabetes care and a county variation factor derived from The Dartmouth Atlas of Health Care Medicare reimbursement.1
More details about assumptions and the limitations of these analytic methods appear in the Technical Report.
WHAT ARE THE LIMITATIONS?
The tool provides a “ballpark estimate” of avertable deaths, diabetes, and diabetes cost. It uses data to give the public a sense of magnitude about the importance of social determinants of health, but several factors affect its precision. The data are from 2008-2010 and may not reflect the current economy or current conditions in a county or state. Because the estimates are approximations based on a generalized linear regression equation, true values for any given county or state may vary from the values predicted by the equation. In some cases, the tool relies on “imputations” of data for rural or small counties with sparse populations. The tool examines cross-sectional associations (the correspondence between education or income levels and the death rates during the same time period), the health consequences of social determinants often occur many years later.
The most important caveat is that the tool does not imply “causality”: one cannot assume that all the deaths or diabetes that this tool designates as “avertable” can be prevented by education or income alone. While education and income would help, the avertable deaths or diabetes calculated by the tool reflect the total package of living conditions that exist in areas with higher education or income. People with more education or income tend to have better jobs and working conditions and greater access to medical care when they need it. They also tend to live in healthier neighborhoods with better access to nutritious foods, bonus casino, cleaner air and drinking water and less violent crime. Additionally, residents in such circumstances are less likely to smoke or be overweight or obese. All of these factors increase a person’s chances of living a healthier, longer life.
The bottom line: saving lives or reducing the prevalence of disease, such as diabetes, requires more than improving education and income alone—it requires addressing the “package” of healthier living conditions that exist in areas where education and income are higher.
1. American Diabetes Association. Economic costs of diabetes in the U.S. In 2007. Diabetes Care. 2008 Mar;31(3):596-615.