Data sources
We gathered person-level data on VHA-users from national VHA enrollment files for the year 2013, and linked it to county-level data on: 1) non-VHA provider availability from the Area Health Resource File (AHRF) maintained by the Health Resources and Service Administration (HRSA); 2) median household income and total adult population over age 18 from the American Community Survey (ACS) fielded by the US Census Bureau; 3) rurality based on Urban Influence Codes (UIC) created by the Economic Research Service of the US Department of Agriculture; and 4) health status measures from the Robert Wood Johnson County Health Rankings [
6‐
8]. We compiled data for 3107 counties in the contiguous US. VHA-users and counties in Alaska and Hawaii were excluded because of the relatively small numbers of Veterans and unique geography in these states.
A VHA-user was defined as any Veteran who accessed any VHA care in 2013 in inpatient or outpatient medical, mental health, or substance use treatment settings. We determined whether individual VHA-users were eligible for non-VHA care under the Choice Act, based on their estimated driving distance to the nearest VHA care site that provided any form of inpatient or outpatient care (i.e. < 40 miles or > 40 miles). We used Federal Information Processing Standards (FIPS) county codes to link each VHA user to data on county of residence.
Data on non-VHA health care providers were obtained from the 2013 Area Health Resource File, a publicly-available data set provided by the US Health Resources and Services Administration (HRSA) [
6]. Counties were classified as health professional shortage designations for primary care and mental health care using criteria created by HRSA, based on provider-to-population ratios [
9]. Primary care provider (PCP) shortage areas were defined as < 1 PCP per 3500 persons. Mental health care shortage areas were defined as either: 1) < 1 psychiatrist / 20,000 persons and < 1 PCP / 6000 persons; OR 2) < 1 PCP / 9000 persons; OR 3) < 1 psychiatrist / 30,000 persons.
We also determined the availability of various non-VHA providers of specialized care in each county, including psychiatrists, cardiologists, pulmonologists, neurologists, and physical medicine and rehabilitation (PM&R) specialists. Only non-federal physicians involved in patient care were counted. We also determined the number of community health centers and community mental health centers for each county. Community health centers were limited to HRSA-grantees and community mental health centers were limited to certified Medicare providers. All physician and facility measurements were from 2013. Availability of specialty providers and facilities in each county were categorized as any or none.
County-level rurality was classified using Urban Influence Codes (UIC) created by the US Department of Agriculture’s Economic Research Service [
8]. Following a commonly-applied framework, we collapsed the 12 codes into a four level measure of rurality: 1) metropolitan / (i.e. counties with population clusters > 50,000 people, UIC 1–2); 2) non-metropolitan / –adjacent to metropolitan areas (UIC 3–7); 3) nonmetropolitan - micropolitan (i.e. not adjacent to metropolitan counties but with town/urban cluster of 10,000–50,000 people, UIC 8); and 4) nonmetropolitan - remote (i.e. the remainder, UIC 9–12). For simplicity, we refer to these groups as metropolitan, rural – adjacent, rural-micropolitan, and rural-remote [
8].
We used median household income as a county-level proxy for socioeconomic status. Median household income values were single-year, model-based estimates from the 2009–2013 Small Area Income and Poverty Estimates (SAIPE) provided by the US Census Bureau [
10]. Based on inspection of county-level distributions of median yearly household income, we categorized counties as median household income < $30,000 (i.e. roughly bottom decile of counties), $30–$40,000; $40,000–$50,000; $50,000–$60,000; and > $60,000 (i.e. top decile of counties).
County-level health status measures were drawn from the County Health Rankings & Roadmaps program, a collaboration between the Robert Wood Johnson Foundation and the University of Wisconsin Population Health Institute [
7]. As a proxy for population health status, we used age-adjusted years of potential life lost (YPLL) per 100,000 people, aggregated over three years (2010–2012). The reference age was 75. For example, a person who dies at age 45 would contribute 75–45 = 30 YPLL. The county-level YPLL is a sum of individual YPLL over all premature deaths in a county. Rates per 100,000 people were given after adjusting for differences in the age distribution over counties. YPLL values were missing for 134 counties, which were excluded from analyses of health status. County-level health status was categorized as Very Poor, Poor, Good, and Very Good according to quartiles of YPLL.
Analysis
We began by examining the distributions of the overall US adult and VHA-user populations according to characteristics of counties of residence, including non-VHA provider availability, rurality, household income, and health status. We further stratified the VHA-user population according to eligibility for purchased care under the Choice Act based on driving distance to the nearest VHA care site (i.e. < 40 miles or > 40 miles). We repeated analyses for the subset of all VHA-users living in rural counties. We used Chi-square tests to compare distributions across county categories.
Because non-VHA providers will be more impacted by reforms to purchase care for Veterans in areas where VHA has recently delivered care for larger portions of the overall population, we also calculated the density of VHA-users in the total adult population (i.e. VHA-users / 1000 adults), according to county rurality and distance to the nearest VHA care site. Counties were categorized according to their distance to the nearest VHA care site by estimating the driving distance from the population-weighted centroid for each county to the nearest VHA care site. Population-weighted centroid coordinates were determined based on 2013 Census data, using the MABLE/Geocorr tool from the Missouri Census Data Center [
11]. Coordinates of VHA facilities were collected from the Department of Veterans Affairs National Center for Veterans Analysis & Statistics, furnished by ESRI (
http://www.va.gov/vetdata/maps.asp). Driving distances were estimated using ArcOnline [
12]. Other analyses were completed using SAS software v9.2 (Cary, NC). All analyses were approved by the Institutional Review Board at the University of Iowa.