Management of Race, Ethnicity, and Language Data in the Pediatric Emergency Department

Abstract

In this secondary analysis of a multicenter investigation, we describe several gaps in the collection and management of pediatric race, ethnicity, and language data. These findings highlight the ongoing need for reliable data management processes as a crucial step toward advancing pediatric health equity.

Introduction

Inequities within the United States healthcare system have been well documented. Children from historically and currently marginalized racial, ethnic, and language backgrounds receive different care and have worse clinical outcomes than their non-Hispanic White and English speaking peers.¹ For researchers and clinicians working to promote equity in child health, high-quality, self-reported race, ethnicity, and language (REaL) data are crucial for measuring and addressing disparities that arise due to structural racism and other biases.² However, administrative race and ethnicity data may be inaccurate, and data on language and language concordance is often unavailable.^3,4 In this study, we describe REaL data collection and classification processes across 34 academic pediatric emergency departments (EDs) from a large multi-site investigation.⁵

Methods

This is a secondary analysis of a multicenter study investigating the association between REaL and the management of febrile infants.⁵ Investigators from each site completed an online questionnaire querying their hospital’s processes for managing REaL data and obtained patient-level data through chart review (Appendix).⁵ Investigators completed the intake survey between March – November, 2021, but were asked specifically to identify practices that were in place during the time period under study (January 1, 2018 – December 31, 2019). Infant race and ethnicity were entered into the central study database as separate variables using U.S. Census Bureau categories. This included an option to select “other” and provide a free-text response, as well as options to select “patient declined to state” and “unknown”. More than one option could be selected within both race and ethnicity variables. Language was determined based on the language documented in the electronic health record. Investigators additionally assessed for documentation of professional interpreting during manual record review for each subject. We classified infants as receiving language-concordant care if they were identified as English speaking or if there was documentation of professional interpreting having occurred during the ED visit. We calculated descriptive statistics at the site and individual level and used Pearson’s Chi-square tests for comparisons between categorical variables.

Results

Of the 34 sites that participated in the primary study, 31 (91%) responded to the intake questionnaire to provide data on their hospital’s process for collecting and documenting race, ethnicity, and language data (Table 1). The majority (30/31, 97%) reported that their hospital collected race and ethnicity as two separate variables. Investigators from 23 (68%) hospitals reported that race and ethnicity were collected through a standard process of self-identification by the patient or caregiver. Twenty-five sites (81%) used a standardized method to identify interpreter need. Of these, most collected caregivers’ primary language (64%) or preferred language (52%), and 16% collected the language spoken at home (six sites reported using more than one approach). All sites had remote professional interpreting services; only 12 (38%) reported a standard process for documenting professional interpreting.

Table 1.

Details of language and interpreting services at participating pediatric emergency departments

	Site Responses (n = 31)
Process for identifying language1, No. (%)
Primary language	16 (50%)
Preferred language	13 (41%)
Language spoken at home	4 (13%)
No standard process	3 (9%)
Other identification process2	1 (3%)
Professional interpreting service availability1, No. (%)
In-person	25 (78%)
Remote - by phone	28 (88%)
Remote - by video	29 (91%)
Availability of in-person professional interpreting,1,3 No. (%)
Weekday daytime	24 (92%)
Weekday evenings	14 (54%)
Weekday overnight	7 (27%)
Weekend daytime	15 (58%)
Weekend evenings	11 (42%)
Weekend overnights	7 (27%)
Documentation of professional interpreting, No. (%)
There is a standardized format in the EHR	12 (38%)
Documentation of interpreting is at the discretion of the provider	25 (78%)
Other documentation process4	5 (16%)

¹Sum may be more than the total as participants could select more than one response

²Free text responses: also assess preferred written language

³Among 25 sites who responded that in-person interpreters were available in their emergency department

⁴Free text responses: not routinely documented (1), documented by the interpreter (1), available administratively (3)

Of the 4,178 infants in the dataset, 387 (9.3%) were missing race or ethnicity data (Table 2) and the proportion varied by site (Figure 1). Nearly one-third of infants with missing race or ethnicity data were documented to have declined to provide a response (29.6% of infants with missing race data, 27.3% of infants with missing ethnicity data). Most infants with missing data were missing only race (45.0%), the majority of whom had Hispanic ethnicity (64.9%). Similarly, the majority (67.2%) of infants listed as having “other” race had Hispanic ethnicity. All infants documented as having “other” ethnicity had free-text responses considered “Hispanic ethnicity” by federal reporting systems.

Table 2.

Details of infant race and ethnicity

	Hispanic (n = 1082)	Non-Hispanic (n = 2856)	Other ethnicity (n = 27)	Missing ethnicity (n = 213)
Infant race by ethnicity
American Indian or Alaska Native	4 (0.37%)	7 (0.25%)	1 (3.7%)1	0 (0%)
Asian	5 (0.46%)	198 (6.93%)	0 (0%)	5 (2.35%)
Black	31 (2.87%)	649 (22.72%)	0 (0%)	11 (5.16%)
Multiracial2	9 (0.83%)2	38 (1.3%)3	0 (0%)	2 (0.94%)4
Native Hawaiian or Pacific Islander	4 (0.37%)	18 (0.63%)	0 (0%)	0 (0%)
Other, no details provided	299 (27.63%)	108 (3.78%)	5 (18.52%)5	34 (15.96%)
Other, additional details provided	55 (5.08%)6	8 (0.28%)7	17 (62.96%)8	1 (0.47%)9
White	562 (51.94%)	1769 (61.94%)	4 (14.81%)10	60 (28.17%)
Missing race	113 (10.44%)	61 (2.14%)	0 (0%)	100 (46.95%)

¹Other ethnicity: “Mexican” (1)

²Other race: “Multiracial” (5); Selection of multiple races: Black + White (3), American Indian or Alaska Native + Native Hawaiian or Pacific Islander (1), Black + other race, no details provided (1)

³Other race: “Multiracial” (13); Selection of multiple races: American Indian or Alaska Native + White (1), Asian + White (4), Asian + Black (1), Black + White (18), Native Hawaiian or Pacific Islander + White (1)

⁴One identified by selection of “other race” with “multiracial” indicated in free text, one identified by selection of AIAN + NHPI race

⁵Other ethnicity: “Dominican” (1), “Mexican” (4)

⁶Other race: “Hispanic or Latino” (45), “Mexican” (3), “Latin American” (4), “Guamanian or Chamorro” (1), “Honduran” (1), “Salvadorian” (1)

⁷Other race: “Muslim” (2), “Indian” (3), “Pakistani” (1), “Romanian” (1), Middle Eastern (1)

⁸Other race and ethnicity: “Latin American” (race and ethnicity, 5), “Mexican” (race and ethnicity, 11), “Thai” race and “Honduran” ethnicity (1)

⁹Other race “Muslim” (1)

¹⁰Other ethnicity: “Dominican” (1), “Mexican” (1), “Puerto Rican” (1), “South American” (1)

Figure 1.

This figure shows the proportion of subjects at each site with missing race and ethnicity data. Race and ethnicity were considered missing if ethnicity was missing or if race was missing with non-Hispanic or other ethnicity. Error bars represent 95% confidence intervals. Shape and shading indicate whether or not the site lead reported having a standardized process for collecting self-identified race and ethnicity data.

Table 3 shows the complete race, ethnicity, and language data of included infants. Of the 4,154 infants with available language data (99.5% of sample), caregivers of 490 (11.8%) used a language other than English (LOE), 79.8% of whom used Spanish. The caregivers of most infants (94.2%) were documented to have received language-concordant care during their ED visit. This included 3,664 caregivers who were identified as using English and 251 families with LOE who were documented to have received interpretation. Two-hundred and thirty-nine caregivers with LOE (48.8%) had no documentation of having received interpretation. There was no difference in the documentation of having received interpretation between those who spoke Spanish and those who used another LOE (49.1% vs 54.0%, p = 0.69).

Table 3.

Complete details of infant race, ethnicity, and language

Race and Ethnicity	Language	No.
American Indian and Alaskan Native
Hispanic	English	4
Non-Hispanic	English	6
	Other Language1	1
Other Ethnicity2	Spanish	1
Asian
Hispanic	English	5
Non-Hispanic	English	153
	Other Language3	45
Missing Ethnicity	English	4
	Other Language4	1
Black
Hispanic	English	27
	Spanish	4
Non-Hispanic	English	640
	Other Language5	9
Missing Ethnicity	English	10
	Other Language6	1
Multiracial
Hispanic	English	8
	Spanish	1
Non-Hispanic	English	38
Missing Ethnicity	English	2
Native Hawaiian and Pacific Islander
Hispanic	English	2
	Spanish	2
Non-Hispanic	English	18
Other race, no details provided
Hispanic	English	187
	Spanish	112
Non-Hispanic	English	93
	Spanish	6
	Other Language7	9
Other Ethnicity8	English	3
Other Ethnicity9	Spanish	2
Missing Ethnicity	English	28
	Spanish	5
	Other Language10	1
Other race, additional details provided
Hispanic	English11	27
	Spanish12	28
Non-Hispanic	English13	5
	Other Language14	3
Other Ethnicity15	English	12
Other Ethnicity16	Spanish	5
Missing Ethnicity	Other Language17	1
White
Hispanic	English	392
	Spanish	168
	Other Language18	2
Non-Hispanic	English	1758
	Other Language19	8
	Spanish	3
Other Ethnicity20	English	2
Other Ethnicity21	Spanish	2
Missing Ethnicity	English	57
	Spanish	1
	Other Language22	2
Missing race
Hispanic	English	64
	Spanish	48
	Other Language23	1
Non-Hispanic	English	58
	Spanish	1
	Other Language24	2
Missing Ethnicity	English	97
	Spanish	2
	Other Language25	1

¹Language: Arabic (1)

²Ethnicity: Mexican (1)

³Language: Mandarin (8), Burmese (6), Karen (6), Vietnamese (4), Hmong (4), Nepali (3), not specified (3), Cantonese (2), Japanese (2), Khmer (1), Farsi (1), Hindi (1), Pashto (1), Persian (1), Tagalog (1), Korean (1)

⁴Language: Vietnamese (1)

⁵Language: Somali (3), Swahili (2), French (2), Bambara (1), Hebrew (1), not specified (1)

⁶Language: not specified (1)

⁷Language: Arabic (4), Amharic (1), Armenian (1), Pashto (1), Portuguese (1), Romanian (1)

⁸Ethnicity: Mexican (1)

⁹Ethnicity: Dominican (1), Mexican (1)

¹⁰Language: Arabic (1)

¹¹Race: Hispanic/Latino (21), Latin American (3), Mexican (3)

¹²Race: Hispanic/Latino (24), Guamanian or Chamorro (1), Honduran (1), Latin American (1), Salvadorian (1)

¹³Race: Indian (2), Muslim (2), Middle Eastern (1)

¹⁴Race and language: Pakistani and Arabic (1), Indian and Bengali (1), Romanian and Romanian (1)

¹⁵Race and ethnicity: Mexican (7), Latin American (4), Thai and Honduran (1)

¹⁶Race and ethnicity: Mexican (4), Latin American (1)

¹⁷Race and language: Muslim and Arabic (1)

¹⁸Language: Romanian (1), not specified (1)

¹⁹Language: Arabic (5), Dari (1), Romanian (1), Uzbek (1)

²⁰Ethnicity: Dominican (1), Puerto Rican (1)

²¹Ethnicity: Mexican (1), South American (1)

²²Language: Arabic (1), Russian (1)

²³Language: Portuguese (1)

²⁴Language: Arabic (1), Farsi (1)

²⁵Language: Arabic (1)

Discussion

In this large sample of academic pediatric EDs, we identified several gaps in pediatric REaL data. Although self-identification is considered the gold standard when collecting race and ethnicity data,³ one-third of institutions had no standardized process for doing so. In the absence of standardized self-identification processes, race and ethnicity data may instead be collected based on staff assessment, which risks inaccuracy and misclassification.^6-8 Further, nearly one in ten infants in our sample were missing race or ethnicity data. Although this was sometimes due to caregiver choice, the varied nature of missing data between institutions suggests institutional practices play a role. Importantly, many infants with missing race data were Hispanic. Individuals who are Hispanic may not identify with a unique race category, and thus may decline to provide a response to questions assessing race separately from ethnicity.⁹ Alternatively, individuals who are Hispanic may feel compelled to select a race category with which they do not identify, such as White or “other” race. This has important implications for researchers: Hispanic infants will be systematically and inaccurately categorized as White, “other” race, or “missing” (and potentially excluded from analysis) if ethnicity is not considered during early stages of data preparation.¹ In light of this, the Office of Management and Budget of the United States has proposed changing to the use of a single variable that includes options for race and ethnicity categories together.¹⁰

The infants in this sample demonstrated a significant amount of diversity by race and ethnicity at a granular level. As with this dataset, health equity researchers must balance the need for statistical power against the importance of recognizing and identifying differences facing smaller subgroups.¹¹ The diversity within race and ethnicity categories in this sample is a reminder that the categorization of race and ethnicity reflects social groupings rather than a shared genetic ancestry. In addition, the granular data reveal a wide variety of languages used by caregivers across race and ethnicity backgrounds, underscoring the importance of considering language separately from race and ethnicity.

Although most institutions in this study had a standardized process for assessing caregiver language, there was variability in how this was done. Notably, many institutions assessed language by asking about “primary language” or “language spoken at home,” which does not correlate sufficiently with interpreter need, performs less well for identifying patients at risk for disparities, and fails to acknowledge that language use, comfort and proficiency are context-dependent.^9,12,13 Institutions should instead assess either language for care or language preference in the healthcare setting, which explicitly acknowledge that language needs and preferences for healthcare may be very different than language use in other contexts.^13,14 Only half of caregivers with LOE were documented to have received professional interpreting, which is similar to previous studies showing interpreting occurs in clinical settings approximately half of the time.^15-17 This likely indicates significant gaps in language-concordant communication. However, even when there is documentation of interpreting, this does not necessarily indicate that all key information was interpreted throughout the visit.¹⁸ Further, provider language proficiency was not available in the dataset. Researchers investigating language concordance must use rigorous measurements of interpreting, such as direct observation or patient-level interpreter billing data, and devise reliable methods to capture provider language proficiency.^18,19

This secondary analysis is subject to limitations. Institutional practices for data collection were assessed by report of site investigators and may not reflect actual administrative data collection practices. Further, as awareness of and commitment to equity increases, practices may have already improved. Additionally, the practices in this sample of institutions may not be generalizable. However, this is a sample of large academic pediatric EDs that demonstrates significant gaps in standard self-report of REaL data and these issues may be even more pronounced in other settings where research and education are not as central to the culture. We were not able to assess for provider language proficiency, which may lead to an overestimation of the proportion of families who use LOE who received language discordant care. However, taken together, these findings highlight the ongoing need for standardized and reliable REaL data management processes as a crucial step toward advancing pediatric health equity.

Abbreviations:

REaL

race, ethnicity, and language

ED

emergency department

LOE

languages other than English

Appendix 1. Contributing Authors of the Pediatric Emergency Medicine Collaborative Research Committee Febrile Infants and Health Disparities Study Group

The following members of the Pediatric Emergency Medicine Collaborative Research Committee Febrile Infants and Health Disparities Study Group contributed to Investigation, Project administration, and Writing – Reviewing and Editing: Muhammad Waseem, MBBS, (Department of Pediatrics and Emergency Medicine, Lincoln Medical Center, Bronx, NY), Nidhi Singh, MD (Division of Pediatric Emergency Medicine, Department of Pediatrics, Baylor College of Medicine, Houston, TX), Rebecca S. Green, MD (Department of Pediatrics, Children’s Hospital of Philadelphia, Philadelphia, PA), Amy D. Thompson, MD (Department of Pediatrics, Nemours Children’s Hospital of Delaware, Wilmington, DE), Kathleen Jackson, MD (University of South Carolina School of Medicine Greenville, Greenville, SC), Nabila C. Kalari, MD (St. Christopher’s Hospital for Children, Philadelphia, PA), Samantha Lucrezia, MD (Johns Hopkins All Children’s Hospital, St. Petersburg, FL), Andrew Krack, MD, MS (Children’s Hospital Colorado, Aurora, CO), Jacqueline B. Corboy, MD, MS (Ann & Robert H. Lurie Children’s Hospital of Chicago), Myto Duong, MD (Southern Illinois University, Carbondale, IL), Bolanle Akinsola, MD (Emory University School of Medicine, Atlanta, GA), Jessica Kelly, MD (Children’s Hospital of Philadelphia, Philadelphia, PA), Laura F. Sartori, MD, MPH (Children’s Hospital of Philadelphia, Philadelphia, PA), Nirupama Kannikeswaran, MBBS (Children’s Hospital of Michigan, Detroit), Larissa L. Truschel, MD, MPH (Duke University School of Medicine), Jessica L. Chow, MD, MPH, MS (University of California, Los Angeles, CA), Jamie Chu, MD (Texas Children’s Pediatrics, Houston, TX), Leslie Dingeldein, MD (Rainbow Babies and Children’s Hospital, Cleveland, Ohio), Carly Theiler, MD (University of Iowa, Iowa City), Sonali Bhaldokar, MD (Yale University, New Haven, CT), Natalie J. Tedford, MD (University of Utah, Salt Lake City), Ahmed Lababidi, MD (AdventHealth, Orlando, Fl)

Appendix 2

A. Questionnaire items related to race, ethnicity, and language data

• Do you have the ability to access patient race and ethnicity in your EHR?

  • Yes
  • No

• Does your hospital collect race and ethnicity as one category or two separate categories? (i.e., is there a separate category for race and a separate category for ethnicity? This is the case for most institutions)

  • One single race/ethnicity category

  • Separate category for race and separate category for ethnicity

  Suggestion: If you are uncertain about the answers for the below questions, we recommend speaking with nursing leadership or a manager from patient registration

• How is patient race and ethnicity determined at your institution?

  • Self identification by the patient and/or parent or guardian
  • Determined by registration staff
  • No standard procedure
  • Other (free text)
• Is patient language routinely documented as a demographic in your hospital EHR?

  • Yes
  • No
• How is patient language preference/need identified in your ED? Select all that apply

  • Patient primary language (i.e., “What is your primary language?”)
  • Language spoken at home
  • Patient preferred language (i.e., “What language do you prefer for talking to doctors?”)
  • Patient English proficiency (i.e., “How well do you speak English?”)
  • No standardized approach
  • Other (free text)
• What professional interpretation is available in your ED? (select all that apply)

  • In person interpreters
  • Phone interpretation
  • Video interpretation
  • Other (free text)
  • None

• What languages are available for in-person interpretation in your ED? (free text)

  Approximately what hours are in-person interpreters available in your ED? (select all that apply)

  • Weekday business hours (e.g., Mon-Fri, 9a-5p)

  • Weekend days (e.g., Sat-Sun, 9a-5p)

  • Weekday evenings (e.g., Mon-Fri, 5–10p)

  • Weekend evenings (e.g., Sat-Sun, 5–10p)

  • Weekday overnight

  • Weekend overnight

• How is interpreter use documented in your EHR?

  • At provider discretion in the free text of the note (i.e., the provider must choose to add “professional interpretation provided by…”)
  • There is a standardized option in the EHR (such as a checkbox, dot phrase, or template) that the provider can activate when interpretation is used

  Other (free text)

B. Race, ethnicity, and language data collection within the primary study

• What is the patient’s race? Select all that apply.

  • American Indian/Alaska Native
  • Asian
  • Black/African American
  • Native Hawaiian/Other Pacific Islander
  • White
  • Patient declined to state
  • Other (free text)
  • Unknown/unable to determine
• What is the patient’s ethnicity? Select all that apply.

  • Hispanic/Latino
  • Non-Hispanic/Latino
  • Patient declined to state
  • Other (free text)
  • Unknown/unable to determine
• What is documented as the patient’s language?

  • English
  • Language other than English

    • Please select from the below options:

      • Spanish
      • Arabic
      • Amharic
      • Cantonese
      • Haitian/Creole
      • Hmong
      • Karen
      • Mandarin
      • Oromo
      • Portugues
      • Vietnamese
      • Other (free text)
      • Unknown/unable to determine
  • Unknown/unable to determine
• Did any member of the healthcare team document that an interpreter was used?

  • No interpretation documented
  • Yes – type not specified
  • Yes – in person professional interpreter
  • Yes – remote professional interpreter (such as by phone/video)
  • Yes – a family member or friend of the patient/caregiver
  • Yes – a bilingual staff member
  • Other (free text)