Association of Race/Ethnicity With Emergency Department Destination of Emergency Medical Services Transport

Key Points Question Are racial/ethnic minorities who use emergency medical services transported to the same emergency department as white residents living in the same zip code? Findings In this cohort study of 864 750 Medicare enrollees from 4175 zip codes, the proportion of white patients transported to the reference (or most frequent) emergency department destination was high (61.3%), compared with the proportion of black patients (difference of −5.3%) and Hispanic patients (difference of −2.5%). Meaning This study suggests that emergency department destination is substantially different on the basis of the race/ethnicity of patients living in the same zip code.


eAppendix. Differences in Emergency Department Destination of Emergency Medical
Services Transport of Co-located Patients by Race/Ethnicity and Geography

Sampling Universe
Eligible enrollees: Our study sample was drawn from the universe of Medicare enrollees during 2006-2012 (N=45.6 million in 2006 to 53.6 million in 2012); a sample was drawn from the universe for each year separately. For this study we selected those aged 66 and older and have continuous Fee for Service (FFS) coverage; for enrollees from 2006 to 2010, we required continuous coverage for 3 years, for enrollees from 2011 we required continuous coverage for 2 years and for enrollees from 2012 we required continuous coverage for 1 year. eTable 1 gives the counts for the selected subgroups for 2006 and 2012, overall and by race/ethnicity.
Zip codes with race/ethnic diversity: Using the residence zip code (reported in the Medicare beneficiary files) of continuously FFS enrolled members for each year, we stratified all zip codes by racial/ethnic diversity; a zip code was categorized as diverse if it was the residence zip code of >10 white, >10 black and >10 Hispanic eligible enrollees. Appendix Table 2 gives the composition of zip codes by racial/ethnic diversity for the 2009 study-eligible sampling universe; these figures were similar for the other years. An additional requirement of the study zip codes was that there be at least 5 EMS transports (to ED) for each of the three race/ethnic groups (from each zip code) during the study period.
The above table identifies qualifying 3,953 zip codes in 2009 data; similar analysis for all years identified a total of 4,175 unique zip codes, which formed the eligible zip codes for this study.

Sample Size
From the universe of eligible Medicare enrollees from the selected zip codes with race/ethnic diversity we obtained stratified random samples separately for each year. Medicare utilization records were only obtained for the sample cohorts. Following the rolling cohort design of national surveys (AHRQ's Medicare Expenditure Panel Survey and CMS' Medicare Current Beneficiary Survey), we stratified the sample for each year into three cohorts with each cohort followed for 1 to 3 years. 1-3 eTable 3 identifies the counts and follow-up periods for the 9 distinct cohorts in the study. Cohort 1 consisted of 123,791 enrollees sampled from the eligible universe in 2006; we obtained utilization records for 1 year (i.e., 1 year of follow-up). Cohorts 2 and 3 were also obtained from the 2006 universe but with longer follow-up periods. Cohort 4 is introduced in 2007, based on the 2007 universe, and replaces the retired Cohort 1. Cohorts 3 to 7 have 3 follow-up years. Cohort 8 is followed for 2 years and Cohort 9 for 1 year. Enrollees may be re-sampled in a subsequent cohort if they are no longer in an actively followed cohort. While the total sample count is 1,048,960, after excluding those re-sampled in multiple rounds, the number of unique enrollees was 864,750; we have reported this figure as the underlying sample size for all the ED visits examined in this study. For more about re-sampling, particularly from Massachusetts, see section 3 below.

Sampling by co-location
The desired sample count from the universe of each year was obtained by random sampling of enrollees stratified by zip code and race/ethnicity. Following were the sampling criteria.
1) Our preference was to obtain equal number of the 3 race/ethnic groups from each zip code; however, zip codes with racial/ethnic diversity varied considerably in the number of enrollees that could be sampled. The number that can be sampled is given by the number of the smallest race/ethnic group in the zip code; for instance, if a zip code has 100 white enrollees, 20 black enrollees and 12 Hispanic enrollees, then we can sample at most 12 members of each group from the zip code. For better representation of the overall population we chose larger sample sizes from zip codes with larger number of enrollees that could be sampled.
2) The study data was designed as part of a larger study aimed at comparing ED use changes in Massachusetts vs. the remaining states in the country. Therefore, we over-sampled Massachusetts enrollees; approximately one-third of the total sample was from Massachusetts by design. In particular, for many zip codes, we were more likely to select all the black or Hispanic enrollees from the zip codes; consequently, these enrollees were likely to be re-enrolled in a later cohort. eTable 4 provides the sampling rates by zip code categories: those with >100 enrollees from each race/ethnic group (category 1); those with 26 to 100 enrollees from each race/ethnic group (category 2) and those with 11 to 25 enrollees from each race/ethnic group (category 3). It presents the sampling numbers in Massachusetts and rest of the country separately.

Sampling weights
Our sampling strategy amounts to stratification of all eligible Medicare enrollees in these zip codes at two levels, first, by zip code, and second, by race/ethnicity. That is, all eligible enrollees in each zip code are stratified into four groups by race/ethnicity: Hispanics, (non-Hispanic) blacks, (non-Hispanic) whites and Others (all the remaining groups combined). As sampling of enrollees was done randomly from each group, the sampling probability for each selected enrollee is given by the ratio between the sample size and the total number of eligible enrollees from the racial/ethnic group in the zip code. The sampling weight is the inverse of this ratio.
Application of sampling weights leads to estimates generalizable to approximately 7 million underlying study-eligible Medicare enrollee population each year. This cohort is detailed in eTable 2 for 2009; there were 6,934,344 enrollees in the study-eligible zip codes.

Estimation
We estimated linear probability models of the following form: where is a dichotomous indicator of transportation to the reference ED (and, in secondary analysis, to a safety-net ED) and ∆ denotes the difference between individual from the mean value at the zip code level.
is an indicator of of patient characteristics (age, sex, dual Medicaid coverage, principal diagnosis cohort, comorbidity indicators).
is an indicator of race/ethnicity and denotes dichotomous (fixed effects) indicators of each year of patient transport. We used least squares estimation and obtained standard errors robust to clustering at zip code level Note: a) Difference in distance is defined as the distance to the second most frequent destination ED minus the distance to the first most frequent ED. b) All patients from each zip code, regardless of race/ethnicity, were included in this estimation. Estimates of difference in mileage are based on a linear regression of mileage on an indicator of whether the destination ED was the first or second most frequent ED. No other covariates were included as our focus was on estimating the distance to ED of all patients located in different parts of the zip code, regardless of their clinical or sociodemographic characteristics. eTable 8. Concordance in Destination ED Between Whites and non-Whites, by Age, Sex and Medicaid Eligibility a