Background

Emergency departments (EDs) play a critical role in the United States health care system, delivering 11% of ambulatory visits and a quarter of acute care visits, and serving as the portal for half of hospital admissions. Modern EDs in the United States developed from the “emergency room” of acute care hospitals, and until recently most EDs were hospital based. Freestanding EDs were introduced in the 1970s to provide access to emergency care in communities that could not support a hospital-based ED. They deliver emergency care in a facility that is physically separate from an acute care hospital. Although some freestanding EDs are owned or operated by hospitals (“satellite”), others are independently owned and operated by physician groups and other entrepreneurs.

Importance

Recently, there has been significant growth of freestanding EDs, with a concentration in several states, notably, Texas. In 2007, Sullivan et al identified 80 freestanding EDs nationwide, whereas in 2009 the California HealthCare Foundation identified 222 freestanding EDs. Reports from the popular press reflect the rapid and accelerating growth of freestanding EDs. As the number of freestanding EDs increases, policymakers, health professional organizations, and payers are discussing how to react, including by making changes to policy and reimbursement. Yet current data on the number, geographic distribution, and analysis of the population and health services in areas in which freestanding EDs choose to locate have not been published, to our knowledge.

Goals of This Investigation

One question that recurs is whether freestanding EDs improve access to emergency care. In this article, we aim to describe the location and distribution of freestanding EDs and identify what populations are served by them. We test whether ZIP codes in which freestanding EDs locate differ from nonfreestanding ED ZIP codes along population growth, socioeconomic, and health services factors. Specifically, do freestanding EDs locate in areas with demographic features of underserved populations or fewer health services? Our findings can inform ongoing policy discussions of whether freestanding EDs help improve access to emergency care and for which patients.

Study Design and Setting

We created a national inventory of freestanding EDs and then conducted ZIP code–level geographic analyses in the 3 states with the most freestanding EDs: Texas, Ohio, and Colorado.

Methods of Measurement

We identified operational freestanding EDs by conducting a systematic national inventory. First, we gathered lists of licensed freestanding EDs from state departments of health and other state agencies. Second, we conducted Internet searches for each state, using Google to search for “freestanding” or “satellite” and “Emergency Department” or “ED” by state. The list of freestanding EDs in the inventory is current as of March 31, 2015. We collected data on facility characteristics (name and address), hospital ownership (satellite versus independent), and for-profit status (versus nonprofit). We used the inventory to calculate the number of freestanding EDs in each state.

We linked the freestanding ED inventory with ZIP code–level demographic, insurance, and health services data, using the 5-digit ZIP code corresponding to the freestanding ED’s location. We obtained demographic data, including population counts, annual growth rate, sex, age, race, education, and employment data from the 2013 ESRI Demographics files compiled by the Center for Geographic Analysis at Harvard University. Use of health services and medical facility data were also compiled by the Center for Geographic Analysis, using 2013 North American Industry Classification System (NAICS) business counts data. We used the 2013 American Hospital Association data to calculate the number of hospital EDs per ZIP code. We obtained ZIP code–level percentages of uninsured, privately insured, and Medicare and Medicaid patients from the 2013 American Community Survey.

Primary Data Analysis

We focused our geographic analysis on the 3 states with the highest number of freestanding EDs: Texas, Ohio, and Colorado. In each state, we flagged ZIP codes that contained at least 1 freestanding ED and ZIP codes that contained none. We found that ZIP codes with freestanding EDs had significantly higher total population and smaller land area compared with ZIP codes without freestanding EDs. To control for possible confounding because of differences in land area and population, in the 3 states we matched 187 freestanding EDs to 1,048 nonfreestanding ED ZIP codes on land area and population. We used a one-to-many coarsened exact matching algorithm that uses cut points to temporarily coarsen each variable, creates strata of land and population values, and matches observations within each stratum. The cut points were defined manually according to examination of the joint distributions of land and population in freestanding and nonfreestanding ED groups. Each freestanding ED ZIP code was allowed to be matched to more than 1 nonfreestanding ED ZIP code within the same population-land stratum. This one-to-many match produced different sample sizes, and we used coarsened exact matching weights to correct for sample size difference when computing confidence intervals. We used the cem function in Stata MP 13.1 to perform this analysis (StataCorp, College Station, TX). Matching produced groups that were similar with respect to the matching variables ( Table E1 , available online at http://www.annemergmed.com ). As a robustness check, we also performed a one-to-one match on land area and population, as well as a one-to-many match on population and 2010 to 2013 annual compound growth rate. Results of the one-to-one land area and population match were similar to the one-to-many match. Population and growth rates in freestanding ED and nonfreestanding ED samples were similar after we matched on these variables ( Tables E2 and E3 , available online at http://www.annemergmed.com ).

Using matched ZIP code–level data, we computed group-level means of demographic, health insurance, and use and availability of health services variables. To compare characteristics of ZIP codes with and without freestanding EDs, we calculated differences in means across matched groups. We computed confidence intervals for these differences by using univariate regression with freestanding ED group dummy and cem weights to correct for sample size differences. We also performed a multivariable logistic regression analysis to assess the joint contribution of state, demographic, and health services variables to freestanding ED location ( Table E4 , available online at http://www.annemergmed.com ). Links to the datasets and the coarsened exact matching code are available online ( Figure E1 , available online at http://www.annemergmed.com ). All analyses were performed in Stata 13.1.

Study Design and Setting

We created a national inventory of freestanding EDs and then conducted ZIP code–level geographic analyses in the 3 states with the most freestanding EDs: Texas, Ohio, and Colorado.

Methods of Measurement

We identified operational freestanding EDs by conducting a systematic national inventory. First, we gathered lists of licensed freestanding EDs from state departments of health and other state agencies. Second, we conducted Internet searches for each state, using Google to search for “freestanding” or “satellite” and “Emergency Department” or “ED” by state. The list of freestanding EDs in the inventory is current as of March 31, 2015. We collected data on facility characteristics (name and address), hospital ownership (satellite versus independent), and for-profit status (versus nonprofit). We used the inventory to calculate the number of freestanding EDs in each state.

We linked the freestanding ED inventory with ZIP code–level demographic, insurance, and health services data, using the 5-digit ZIP code corresponding to the freestanding ED’s location. We obtained demographic data, including population counts, annual growth rate, sex, age, race, education, and employment data from the 2013 ESRI Demographics files compiled by the Center for Geographic Analysis at Harvard University. Use of health services and medical facility data were also compiled by the Center for Geographic Analysis, using 2013 North American Industry Classification System (NAICS) business counts data. We used the 2013 American Hospital Association data to calculate the number of hospital EDs per ZIP code. We obtained ZIP code–level percentages of uninsured, privately insured, and Medicare and Medicaid patients from the 2013 American Community Survey.

Primary Data Analysis

We focused our geographic analysis on the 3 states with the highest number of freestanding EDs: Texas, Ohio, and Colorado. In each state, we flagged ZIP codes that contained at least 1 freestanding ED and ZIP codes that contained none. We found that ZIP codes with freestanding EDs had significantly higher total population and smaller land area compared with ZIP codes without freestanding EDs. To control for possible confounding because of differences in land area and population, in the 3 states we matched 187 freestanding EDs to 1,048 nonfreestanding ED ZIP codes on land area and population. We used a one-to-many coarsened exact matching algorithm that uses cut points to temporarily coarsen each variable, creates strata of land and population values, and matches observations within each stratum. The cut points were defined manually according to examination of the joint distributions of land and population in freestanding and nonfreestanding ED groups. Each freestanding ED ZIP code was allowed to be matched to more than 1 nonfreestanding ED ZIP code within the same population-land stratum. This one-to-many match produced different sample sizes, and we used coarsened exact matching weights to correct for sample size difference when computing confidence intervals. We used the cem function in Stata MP 13.1 to perform this analysis (StataCorp, College Station, TX). Matching produced groups that were similar with respect to the matching variables ( Table E1 , available online at http://www.annemergmed.com ). As a robustness check, we also performed a one-to-one match on land area and population, as well as a one-to-many match on population and 2010 to 2013 annual compound growth rate. Results of the one-to-one land area and population match were similar to the one-to-many match. Population and growth rates in freestanding ED and nonfreestanding ED samples were similar after we matched on these variables ( Tables E2 and E3 , available online at http://www.annemergmed.com ).

Using matched ZIP code–level data, we computed group-level means of demographic, health insurance, and use and availability of health services variables. To compare characteristics of ZIP codes with and without freestanding EDs, we calculated differences in means across matched groups. We computed confidence intervals for these differences by using univariate regression with freestanding ED group dummy and cem weights to correct for sample size differences. We also performed a multivariable logistic regression analysis to assess the joint contribution of state, demographic, and health services variables to freestanding ED location ( Table E4 , available online at http://www.annemergmed.com ). Links to the datasets and the coarsened exact matching code are available online ( Figure E1 , available online at http://www.annemergmed.com ). All analyses were performed in Stata 13.1.