Study Design

In this observational study, we take advantage of a natural experiment in which retail clinics entered some markets and did not enter others. We conducted an ED-level analysis with a fixed-effects approach to determine whether growth of retail clinics in ED catchment areas is associated with a reduction in ED visits for low-acuity conditions.

We used data from the Healthcare Cost and Utilization Project State Emergency Department Databases from 2006 through 2012 for the 23 states that contributed data for each of those years. The contributing states included Arizona, California, Connecticut, Florida, Georgia, Hawaii, Indiana, Iowa, Kansas, Maryland, Massachusetts, Minnesota, Missouri, Nebraska, New Jersey, New York, Ohio, South Carolina, South Dakota, Tennessee, Utah, Vermont, and Wisconsin. The State Emergency Department Databases include the following relevant encounter-level data for all ED visits for patients who were treated and released (ie, visits that resulted in an inpatient admission were excluded): quarter of visit (January to March, April to June, July to September, and October to December), primary expected payer, diagnoses, patient age, patient zip code of residence, and income quartile in that zip code.

We combined State Emergency Department Databases data with 2006 to 2012 data from Merchant Medicine, a research and consulting firm specializing in the field of walk-in medicine, which include the dates of opening and closing and geocoded addresses of all retail clinics in the United States. Other data sources included the American Hospital Association Annual Survey of Hospitals for hospital characteristics and the Area Health Resource File for county-level characteristics.

The unit of analysis was the ED quarter. There were 2,504 EDs in the State Emergency Department Databases with data in any year from 2006 through 2012. We excluded EDs that either did not have reliable reporting that allowed the separation of ED visits from inpatient admissions through the ED or could not be matched to the American Hospital Association (N=2) ; were located in a rehabilitation hospital (N=90); had missing quarters or years of data from 2006 through 2012 (N=334); had large fluctuations in the number of ED visits from 2006 through 2012, which may indicate institutional changes such as mergers (N=32); or had no visits for low-acuity conditions in 2006 (N=3). This resulted in a final sample size of 2,043 EDs.

Retail clinic penetration was the primary independent variable of interest and was defined as the proportion of an ED’s catchment area that overlapped with any retail clinic’s catchment area. It was defined in 3 steps. First, to define ED catchment areas, we determined the residential zip code of patients who visited the ED for a low-acuity condition in 2006. We defined the ED’s catchment area as the collection of zip codes that accounted for the most, up to 75% of all, visits for these conditions in 2006.

Second, we defined retail clinic catchment areas by using a geographic information system program (ArcGIS; Esri, Redlands, CA) to identify the geographic area within a 10-minute driving distance around any retail clinic. Then, we used ArcGIS to determine the percentage of square miles of all zip codes that overlapped with the radius defined by this 10-minute driving distance.

Third, for each zip code in the ED catchment area, we multiplied the percentage of the zip code area that overlapped with any retail clinic catchment area by a weight that was the percentage of all ED visits for low-acuity conditions that came from that zip code in 2006. We summed this value across all zip codes in the ED’s catchment area to calculate the ED’s retail clinic penetration, which was calculated for each quarter from 2007 through 2012 and expressed as a percentage on a scale of 0 to 100.

We chose the 75% cutoff to exclude zip codes in which ED utilization was low and areas that were geographically distant from a hospital, and we chose the 10-minute drive time because it was consistent with the period used in a previous study. We conducted sensitivity analyses based on a 90% cutoff and 15- and 30-minute drive times.

Outcome Measures

The primary dependent variable was the rate (out of total treat-and-release ED visits) of treat-and-release ED visits for 11 low-acuity conditions (allergic rhinitis, bronchitis, conjunctivitis, other eye conditions, influenza, otitis externa, otitis media, pharyngitis, upper respiratory infections/sinusitis, urinary tract infections, and viral infections), which in previous work each accounted for more than 2% of retail clinic visits (see Table E1 [available online at http://www.annemergmed.com ] for specific International Classification of Diseases, Ninth Revision, Clinical Modification codes). We used all listed diagnoses on the ED claims to identify these conditions. The count of ED visits for low-acuity conditions was measured in each quarter from 2007 through 2012.

ED visits for low-acuity conditions were excluded if they also included codes for conditions that cannot be treated in a retail clinic and require treatment in an urgent care facility (allergic rhinitis, bronchitis, conjunctivitis, other eye condition, influenza, otitis externa and impacted cerumen, otitis media and eustachian tube diagnoses, pharyngitis, upper respiratory infection/sinusitis, urinary tract infection, and viral infection) or ED (septicemia, acute myocardial infarction, congestive heart failure, acute cerebrovascular disease, facture of the neck of femur, and acute unspecified renal failure). The conditions specific to urgent care centers were chosen in accordance with previous work in this area. The conditions specific to the ED were chosen because they were among the 10 most frequent principal diagnoses encountered in the ED that resulted in an admission to the same hospital, excluding conditions that frequently have low-acuity stages (eg, pneumonia, urinary tract infection), which might be treated without admitting the patient.

Hospital characteristics included in the analyses were census region, urban or rural location (defined with the 2003 Urban Influence Codes), teaching status, bed size, and ownership. We characterized the ED catchment areas with county-level data from the Area Health Resource File on unemployment, percentage of residents in poverty, percentage of residents younger than 65 years and without health insurance, and population size. Using a county-to–zip code crosswalk, we merged the Area Health Resource File data with all zip codes in an ED’s catchment area (as defined above). To calculate ED-specific measures, we calculated the mean across all zip codes in the catchment area, weighted by the percentage of discharges for low-acuity conditions from that zip code out of all ED visits for the same conditions in the catchment area in 2006. Because the Area Health Resource File is available only by year, each quarter in a given calendar year was assigned the same value.

If urgent care centers entered the same markets as retail clinics, they could confound our results. Therefore, we used data from the American Hospital Association to identify whether each hospital in the State Emergency Department Databases had an urgent care center and controlled for the number of hospital-based urgent care centers in an ED’s catchment area.

Primary Data Analysis

We estimated Poisson regression models in which the outcome was the rate of ED visits for low-acuity conditions out of all ED visits. We ran a combined model across all payers. Because of known variation in ED utilization by insurance status, we also ran separate models by primary expected payer. Our models included hospital and year fixed effects (ie, separate intercepts for each hospital and year). By estimating a unique intercept for each ED, the ED fixed effect accounted for any unobserved time-invariant ED or ED catchment characteristics that were correlated with both retail clinic penetration and low-acuity ED visits.

We included additional covariates that we hypothesized might have changed over time and would therefore not be accounted for in the hospital fixed effects: hospital ownership (private for profit, private nonprofit, and public), the percentage of hospital discharges that were Medicaid or uninsured (excluded from models by payer), patient age, and income quartile in the patient’s zip code of residence, as well as catchment area characteristics of population size, unemployment, poverty, lack of health insurance, and number of hospital-based urgent care centers in a given quarter.

We multiplied the parameter estimate for retail clinic penetration by 10 and exponentiated this value to obtain the primary association of interest. The results can be interpreted as the rate ratio (RR) associated with increasing retail clinic penetration by 10 percentage points per quarter, or 40 percentage points in a given year, which is roughly equivalent to the average retail clinic penetration among EDs that had some growth in retail clinic penetration from 2007 through 2012. In other words, the RR can be interpreted as the change in the rate of low-acuity ED visits associated with an ED having no retail clinic penetration to having approximately the average penetration rate within 2012.

As stated above, if urgent care centers entered the same markets as retail clinics, they could confound our results. Unfortunately, there are no reliable data to identify all urgent care centers. In the American Hospital Association data, we identified only hospital-based urgent care centers, which represent a relatively small proportion of all urgent care centers. To address the potential influence of urgent care centers, we conducted a falsification test to assess the extent to which retail clinic penetration is associated with ED utilization for conditions that are treated in urgent care clinics but not retail clinics. These conditions are noted earlier and also listed in Table E2 (available online at http://www.annemergmed.com ). Because urgent care centers also can care for patients with low-acuity conditions, any correlation between retail clinic penetration and urgent care center penetration would bias the effect of retail clinic penetration. To the extent that urgent care center penetration is correlated with retail clinic penetration, we would observe an “effect” of retail clinics on ED utilization for conditions that are treated in urgent care clinics but not in retail clinics. More detailed information on methods are available in Appendix 1 .

Study Design

In this observational study, we take advantage of a natural experiment in which retail clinics entered some markets and did not enter others. We conducted an ED-level analysis with a fixed-effects approach to determine whether growth of retail clinics in ED catchment areas is associated with a reduction in ED visits for low-acuity conditions.

We used data from the Healthcare Cost and Utilization Project State Emergency Department Databases from 2006 through 2012 for the 23 states that contributed data for each of those years. The contributing states included Arizona, California, Connecticut, Florida, Georgia, Hawaii, Indiana, Iowa, Kansas, Maryland, Massachusetts, Minnesota, Missouri, Nebraska, New Jersey, New York, Ohio, South Carolina, South Dakota, Tennessee, Utah, Vermont, and Wisconsin. The State Emergency Department Databases include the following relevant encounter-level data for all ED visits for patients who were treated and released (ie, visits that resulted in an inpatient admission were excluded): quarter of visit (January to March, April to June, July to September, and October to December), primary expected payer, diagnoses, patient age, patient zip code of residence, and income quartile in that zip code.

We combined State Emergency Department Databases data with 2006 to 2012 data from Merchant Medicine, a research and consulting firm specializing in the field of walk-in medicine, which include the dates of opening and closing and geocoded addresses of all retail clinics in the United States. Other data sources included the American Hospital Association Annual Survey of Hospitals for hospital characteristics and the Area Health Resource File for county-level characteristics.

The unit of analysis was the ED quarter. There were 2,504 EDs in the State Emergency Department Databases with data in any year from 2006 through 2012. We excluded EDs that either did not have reliable reporting that allowed the separation of ED visits from inpatient admissions through the ED or could not be matched to the American Hospital Association (N=2) ; were located in a rehabilitation hospital (N=90); had missing quarters or years of data from 2006 through 2012 (N=334); had large fluctuations in the number of ED visits from 2006 through 2012, which may indicate institutional changes such as mergers (N=32); or had no visits for low-acuity conditions in 2006 (N=3). This resulted in a final sample size of 2,043 EDs.

Retail clinic penetration was the primary independent variable of interest and was defined as the proportion of an ED’s catchment area that overlapped with any retail clinic’s catchment area. It was defined in 3 steps. First, to define ED catchment areas, we determined the residential zip code of patients who visited the ED for a low-acuity condition in 2006. We defined the ED’s catchment area as the collection of zip codes that accounted for the most, up to 75% of all, visits for these conditions in 2006.

Second, we defined retail clinic catchment areas by using a geographic information system program (ArcGIS; Esri, Redlands, CA) to identify the geographic area within a 10-minute driving distance around any retail clinic. Then, we used ArcGIS to determine the percentage of square miles of all zip codes that overlapped with the radius defined by this 10-minute driving distance.

Third, for each zip code in the ED catchment area, we multiplied the percentage of the zip code area that overlapped with any retail clinic catchment area by a weight that was the percentage of all ED visits for low-acuity conditions that came from that zip code in 2006. We summed this value across all zip codes in the ED’s catchment area to calculate the ED’s retail clinic penetration, which was calculated for each quarter from 2007 through 2012 and expressed as a percentage on a scale of 0 to 100.

We chose the 75% cutoff to exclude zip codes in which ED utilization was low and areas that were geographically distant from a hospital, and we chose the 10-minute drive time because it was consistent with the period used in a previous study. We conducted sensitivity analyses based on a 90% cutoff and 15- and 30-minute drive times.

Outcome Measures

The primary dependent variable was the rate (out of total treat-and-release ED visits) of treat-and-release ED visits for 11 low-acuity conditions (allergic rhinitis, bronchitis, conjunctivitis, other eye conditions, influenza, otitis externa, otitis media, pharyngitis, upper respiratory infections/sinusitis, urinary tract infections, and viral infections), which in previous work each accounted for more than 2% of retail clinic visits (see Table E1 [available online at http://www.annemergmed.com ] for specific International Classification of Diseases, Ninth Revision, Clinical Modification codes). We used all listed diagnoses on the ED claims to identify these conditions. The count of ED visits for low-acuity conditions was measured in each quarter from 2007 through 2012.

ED visits for low-acuity conditions were excluded if they also included codes for conditions that cannot be treated in a retail clinic and require treatment in an urgent care facility (allergic rhinitis, bronchitis, conjunctivitis, other eye condition, influenza, otitis externa and impacted cerumen, otitis media and eustachian tube diagnoses, pharyngitis, upper respiratory infection/sinusitis, urinary tract infection, and viral infection) or ED (septicemia, acute myocardial infarction, congestive heart failure, acute cerebrovascular disease, facture of the neck of femur, and acute unspecified renal failure). The conditions specific to urgent care centers were chosen in accordance with previous work in this area. The conditions specific to the ED were chosen because they were among the 10 most frequent principal diagnoses encountered in the ED that resulted in an admission to the same hospital, excluding conditions that frequently have low-acuity stages (eg, pneumonia, urinary tract infection), which might be treated without admitting the patient.

Hospital characteristics included in the analyses were census region, urban or rural location (defined with the 2003 Urban Influence Codes), teaching status, bed size, and ownership. We characterized the ED catchment areas with county-level data from the Area Health Resource File on unemployment, percentage of residents in poverty, percentage of residents younger than 65 years and without health insurance, and population size. Using a county-to–zip code crosswalk, we merged the Area Health Resource File data with all zip codes in an ED’s catchment area (as defined above). To calculate ED-specific measures, we calculated the mean across all zip codes in the catchment area, weighted by the percentage of discharges for low-acuity conditions from that zip code out of all ED visits for the same conditions in the catchment area in 2006. Because the Area Health Resource File is available only by year, each quarter in a given calendar year was assigned the same value.

If urgent care centers entered the same markets as retail clinics, they could confound our results. Therefore, we used data from the American Hospital Association to identify whether each hospital in the State Emergency Department Databases had an urgent care center and controlled for the number of hospital-based urgent care centers in an ED’s catchment area.

Primary Data Analysis

We estimated Poisson regression models in which the outcome was the rate of ED visits for low-acuity conditions out of all ED visits. We ran a combined model across all payers. Because of known variation in ED utilization by insurance status, we also ran separate models by primary expected payer. Our models included hospital and year fixed effects (ie, separate intercepts for each hospital and year). By estimating a unique intercept for each ED, the ED fixed effect accounted for any unobserved time-invariant ED or ED catchment characteristics that were correlated with both retail clinic penetration and low-acuity ED visits.

We included additional covariates that we hypothesized might have changed over time and would therefore not be accounted for in the hospital fixed effects: hospital ownership (private for profit, private nonprofit, and public), the percentage of hospital discharges that were Medicaid or uninsured (excluded from models by payer), patient age, and income quartile in the patient’s zip code of residence, as well as catchment area characteristics of population size, unemployment, poverty, lack of health insurance, and number of hospital-based urgent care centers in a given quarter.

We multiplied the parameter estimate for retail clinic penetration by 10 and exponentiated this value to obtain the primary association of interest. The results can be interpreted as the rate ratio (RR) associated with increasing retail clinic penetration by 10 percentage points per quarter, or 40 percentage points in a given year, which is roughly equivalent to the average retail clinic penetration among EDs that had some growth in retail clinic penetration from 2007 through 2012. In other words, the RR can be interpreted as the change in the rate of low-acuity ED visits associated with an ED having no retail clinic penetration to having approximately the average penetration rate within 2012.

As stated above, if urgent care centers entered the same markets as retail clinics, they could confound our results. Unfortunately, there are no reliable data to identify all urgent care centers. In the American Hospital Association data, we identified only hospital-based urgent care centers, which represent a relatively small proportion of all urgent care centers. To address the potential influence of urgent care centers, we conducted a falsification test to assess the extent to which retail clinic penetration is associated with ED utilization for conditions that are treated in urgent care clinics but not retail clinics. These conditions are noted earlier and also listed in Table E2 (available online at http://www.annemergmed.com ). Because urgent care centers also can care for patients with low-acuity conditions, any correlation between retail clinic penetration and urgent care center penetration would bias the effect of retail clinic penetration. To the extent that urgent care center penetration is correlated with retail clinic penetration, we would observe an “effect” of retail clinics on ED utilization for conditions that are treated in urgent care clinics but not in retail clinics. More detailed information on methods are available in Appendix 1 .