Appendix B. Methodological and Technical Notes

A. County Definitions

Data from a number of different sources were used in the generation of this report. We used the Federal Information Processing Standard (FIPS) codes to link county definitions across datasets. For some data, modifications mere made to county definitions in order to achieve comparability. The standard definitions used were those in the Area Resource File (ARF). In part this definition was used to enable the utilization of the ARF's contiguity matrix when spatially smoothing death rates.

Modifications were limited to collapsing data from various independent cities in several states ( Virignia, Missouri), into surrounding counties in addition to reclassification for several U.S. counties that no longer exist. For details regarding the county definitions, refer to the Area Resource File User Documentation.

B. Data Sources

1.) Mortality Data

Death  certificate data for the years 1985-1997 were obtained through the National Vital Statistics System maintained by the National Center for Health Statistics. For each cause-of-death used in this analysis, the underlying cause of death listed on the death certificate was utilized. All causes of death were coded to according to the International Classification of Disease–9^th Revision (ICD-9).  For each decedent, underlying cause of death, age, race/ethnicity, gender, and county of residence at the time of death were abstracted from computerized death certificate data.

Rates for infant mortality were taken directly from the Area Resource File (see below under Medical Care Resources).

2.) Hospitalization Data

Inpatient hospitalization for the year 2000 were obtained for 10 Appalachian States through the Health Care Cost and Utilization Project (HCUP), sponsor by the Agency for healthcare Research and Quality (AHRQ). This study used the HCUP State Inpatient Datasets (SID), which consist of individual data files from data organizations in 29 participating states. They are composed of annual, state-specific files that share a common structure and many common data elements, which represent the universe of inpatient discharge abstracts in participating states. The data are provided for all patients regardless of the payer, including persons covered by Medicare, Medicaid, private insurance, and the uninsured. No personal information is provided, nor is it possible to distinguish hospitalizations the may have been incurred by the same individual. SID data for 6 of the ten Appalachian states used in this analysis (New York, Maryland, West Virginia, Kentucky, North Carolina, South Carolina, were obtained from the HCUP Central Distributor. Comparable data for the remaining 4 state ( Georgia , Tennessee, Pennsylvania, and Virginia) were obtained through hospital organizations in each respective state.

Data elements retained for analysis in this study included principal diagnoses at discharge, age, gender, and either the county for ZIP code area of residence. For some states ZIP codes were used to identify the patients area of residence as opposed to  the county. In these cases ZIP codes were collapsed into their respective county. The file used to collapse ZIP codes into counties was provided by the Appalachian Regional Commission. Because ZIP code areas often cross county boundaries, the county that comprised the largest portion of the ZIP code area was used to define the county of residence.

3. Socioeconomic and Population Data

Demographic and socioeconomic data for all U.S. counties were obtained from the 1990 and 2000 decennial censuses. Population count data for all U.S. counties were obtained through the Bureau of Census for the years 1985-1997. These intercensal estimates were calculated by the Bureau of Census through extrapolation of linear trends in population growth and inter-county migration patterns between census years.

4. Behavioral Risk Factor Data

Data on behavioral risks was obtained from the Behavioral Risk Factor Surveillance System for the years 1993-1997. The BRFSS is a telephone survey administered each year by state health departments across the country and is designed to "collect uniform, state-specific data on preventive health practices and risk behaviors that are linked to chronic disease, injuries, and preventable infectious disease" . Although the BRFSS began in 1984, county identifiers were not universally provided until 1993. Therefore, sub-state areas cannot be analyzed using data prior to 1993. Beginning the 1998, county of residence for respondents is not publicly released for counties which have fewer than 50 respondents. The target population for the BRFSS survey is the adult (aged 18 and older) non-institutionalized population of each state.  A two-stage random-digit telephone sampling protocol is used to contact households, and a third stage randomly selects a single eligible adult from within each sampled household.  The median response rate for participating states was 70.4% in 1992.  The format and content of specific questions in the BRFSS surveys replicate questions asked in the National Health Interview Survey and the National Health and Nutrition Examination Surveys whenever possible.

5.) Medical Care Resource Data

Medical care resource data were obtained from the Area Resource File (February 2001 release). The Area Resource File (ARF) represents a compilation of health-related data abstracted from multiple data sources by the National Center for Health Workforce Information & Analysis, Bureau of Health Professions, Health Resources and Services Administration, and the Department of Health and Human Services .

A. Spatial Smoothing of Mortality Rates

1.) Spatial Smoothing Methods

For each county, deaths (numerators) and population counts (demonimators) for ten-year age groups (e.g. 35-44, 45-54, etc.) were summed for the eight-year study period 1990-1997. County numerators and denominators were then combined with death count numerators and population count denominators for all neighbor counties using a distance-weighted approach. Neighbors were defined solely on the basis of contiguity. Distances were calculated from the geographic centroid of each county. These distances were then used to weight the contrubution of each neighbor to the overall death count numerators and population count denominators. The schematic diagram below provides a graphic example of the process.

This process produced spatially 'smoothed' age-specific death rates which were then age-adjusted to the 2000 United States standard population for each age group used in this study (35 to 64 and 65 and older). The process of spatially smoothing death rates helps to minimize local variability that may result from small populations. The spatial moving average used in this analysis was adapted from a method used in two recent publications published by the Centers for Disease Control and Prevention and West Virginia University: Heart Disease in Women: An Atlas of Geographic Disparities and Heart Disease in Men: An Atlas of Geographic Disparities. For this study we have added the distance-weighted approach to partially deal with issues involving counties of varying sizes and configurations.

A constraint was also applied to maximize the likelihood calculating stable county death rates. For each demographic sub-group, a death rate was not calculated for any county for which the total number of deaths in that county plus the neighboring counties was fewer than 20 for the period 1990-1997.

A. Calculation of Hospitalization Rates

For each county, hospitalizations (numerators) and population counts (demonimators) for ten-year age groups (e.g. 35-44, 45-54, etc.) for the year 2000 were used to generate age-specific hospitalization rates. Age-specific death rates were then age-adjusted to the 2000 United States standard population for each age group used in this study (35 to 64 and 65 and older).

A spatial smoother of the type used for mortality rates could not be used for this analysis due to the lack of complete geographic coverage for hospitalization data.

A constraint was also applied to maximize the likelihood calculating stable county hospitalization rates. For each demographic sub-group, a hospitalization rate was not calculated for any county for which the total number of hospitalizations was fewer than 20 for the year 2000.

E. Mapping Procedures

All maps were generated in a manner which depicts less favorable conditions in dark shades and more favorable conditions in lighter shades. For the maps of death and hospitalization rates, the distribution of rates are represented as quartiles with the addition of low and high outliers where they are present. Outliers are counties that had statistically unusually low or high rates relative to the majority of counties. Low outliers were identified as counties with death rates lower than the 25^th percentile minus 1.5 times the interquartile range (25^th-75^th percentile) and high outliers were identified as counties with death rates higher then the 75^th percentile plus 1.5 times the interquartile range.