Authors: Rama A. Salhi, Melissa A. Meeker, Kori S. Zachrison, Margaret E. Samuels-Kalow
Categories: Article, Health equity, Electronic health record data
Source: Annals of emergency medicine
Authors: Rama A. Salhi, Melissa A. Meeker, Kori S. Zachrison, Margaret E. Samuels-Kalow
Demographic data are critical in identifying and addressing disparities but are challenged by data classification issues, particularly for Hispanic or Latino patients. Ethnicity data are typically collected through (1) binary Hispanic or Latino response and/or (2) country-of-origin checklist; however, there is no consensus on which populations are represented by the term Hispanic or Latino. Our objective was to examine the agreement between the commonly collected binary ethnicity variable and country-of-origin-based definitions.
We conducted a cross-sectional study among patients in a regional health care system (January 1, 2021 to November 16, 2023). The primary outcome was agreement between the binary Hispanic or Latino ethnicity and country of origin. Given the variation in countries represented by the term Hispanic or Latino, we used multiple definitions including from the US Office of Management and Budget.
Among the 2,919,810 patients identified, 83.1% had completed responses to the binary Hispanic or Latino ethnicity question and 75.1% had completed responses to the country-of-origin ethnicity variable. Using the binary variable, 241,391 were documented as Hispanic or Latino and of these, 169,731 (70%) had countries of origin identified in the Office of Management and Budget definition. An expanded definition additionally including Brazil, Haiti, Belize, and Guyana had increased agreement (n=176,048; 73%).
Our findings highlight the limitations of using only the binary Hispanic or Latino ethnicity variable, specifically in that it may lead to underestimation. Efforts to improve data quality and nuance, particularly in the emergency department, are critical as inaccurate assessment of disparities may lead to misdirection of interventions, and, ultimately, missed opportunities to decrease disease burden.
Demographic data collection is critical to identifying and addressing disparities in exposures, clinical care, and disease outcomes. In the United States, ethnicity often refers to Hispanic, Latino, Latina, or Latinx people (hereafter “Hispanic or Latino”) and is intended to serve as a pan-ethnic descriptor. Organizations often employ inconsistent administrative definitions, combining the terms Hispanic and Latino despite distinct definitions and connotations.^1^ It is also important to note that these categories are not rooted in biology and are mutable over time and context. Currently, the term Hispanic often refers to people from Spanish-speaking countries (including, for example, Spain, Mexico, and Cuba while excluding Brazil and Belize), whereas the term Latino refers to those of Latin American descent (including Brazil and Belize, while excluding Spain). As the countries included in each definition may vary and are often not reflective of patient self-identification, this problematizes their aggregation into one variable.^2^
Prior work has shown that the accuracy of the Hispanic or Latino binary variable in the electronic health record compared with self-report can be as low as 60% to 70%.^3–5^ Despite the mainstream interchangeable use of the terms Hispanic or Latino, there is no current consensus on the exact populations that each represents. For example, although Haiti is often not included in the term Hispanic, it can be included in the term Latino.^6^ Drawing a sharp distinction between the two ignores the complicated geopolitical history of the region and may marginalize some patients.^7^ Despite these complexities, health systems collect information on Hispanic or Latino ethnicity in 2 (1) a binary yes/no question separate from race, though they are frequently aggregated and/or (2) through a country-of-origin checklist.^8^
As Hispanic or Latino patients experience ongoing gaps in care and health outcomes, accurate estimates based on reliable, high-quality data are critical to local and national efforts to address health disparities. Recent evidence underscores that disaggregation by country may help identify disparities in clinical measures such as hypertension, diabetes, and obesity.^9^ Given the limitations and complexities outlined, our objective was to examine the agreement between the commonly collected binary ethnicity variable and country-of-origin-based definitions.
We conducted a cross-sectional study of patients with at least one visit to any of 5 hospitals in a regional health care system (urban and suburban catchment) between January 1, 2021, and November 16, 2023. The Massachusetts General Brigham Institutional Review Board reviewed the study and deemed it exempt.
We included all patients presenting with emergency department (ED) or outpatient visits to our health system. As the medical record used in this hospital system overwrites any changes to demographic data, we pulled ethnicity information that was available at the patient’s most recent visit.
It is important to note several details with respect to how these data are collected in the local health system’s electronic health record. The question of Hispanic or Latino ethnicity as a binary yes/no is asked separately from country of origin. Online registration asks and records the Hispanic or Latino ethnicity question as “Hispanic,” whereas verbal registration typically asks the question as “Hispanic or Latino.” Most patients are initially registered through a verbal process; however, they have access to edit their demographic data through the online portal. We coded responses yes, no, declined (aggregate of “declined” and “prefer not to say”), or missing (absence of any of the previously mentioned entries).
Country of origin includes 146 countries or ethnic origins (and includes “Hispanic or Latino” in this list), plus additional categories for “declined,” “unavailable,” “doesn’t know,” and “other” for a total of 150 response options. The electronic health record collects data in noun form, allowing multiple selections. Per current guidance, we did not present our primary results in noun form; however, a full list as they appear in the electronic health record is shown in Appendix E1 (available at http://www.annemergmed.com).^10^
Other variables of interest included visit count (less than 6 visits versus 6+ visits), age (less than 18, 18 to 64, and 65+ years), primary language, recency of last visit (by year of visit), and insurance status. As online check-in processes for telemedicine visits often trigger an online review of demographic data in a standardized fashion, in contrast to in-person reviews, which may be subject to variability in registration practices, we also examined the modality of the most recent visit (telemedicine versus in-person).^11^
The primary outcome was agreement between documentation of the binary ethnicity variable and countries included in Hispanic or Latino ethnicity. As the countries included in Hispanic or Latino ethnicity (1) vary between sources and (2) are not completely interchangeable between Hispanic and Latino, we utilized recently updated federal guidelines from the Office of Management and Budget (OMB) as our primary definition.^8^ As a secondary definition, we included those utilized by the OMB plus Brazil, Haiti, Belize, and Guyana. These countries were selected for inclusion as they (1) may be included in the term Latino while being excluded from the term Hispanic, and (2) have patients represented in our patient population.^1^ Finally, we examined an expanded definition including all Latin American and Caribbean countries and ethnicities.
We calculated frequencies and proportions of the predefined country-of-origin definitions, the binary Hispanic or Latino ethnicity, and their agreement. We conducted all analyses using R version 4.3.1.^12^
We identified a total of 2,919,810 patients for inclusion in our analysis. Of these, 83.1% had completed responses to the binary Hispanic or Latino ethnicity question, 75.1% had completed responses to the country-of-origin ethnicity variable, and 65.1% (n=1,902,833) had both. A total of 241,391 (8.3%) patients were documented as Hispanic or Latino using the binary category (Table 1). Hispanic or Latino patients were also more likely to identify “other” as a racial category as compared to the overall population (47% versus 5%). When considering the individual countries of origin included in the OMB definition of Hispanic or Latino, 67% to 95% also answered “yes” to the binary Hispanic or Latino ethnicity question (Table 2). Patients from Brazil, Haiti, Belize, and Guyana were noted to have lower rates of identification as Hispanic or Latino (25%, 3%, 42%, and 11%, respectively). Appendix E2 (available at http://www.annemergmed.com) presents patient characteristics with agreement and disagreement between the binary and OMB definitions. Overall, 74.7% (n=1,422,233 of 1,902,833) were identified as non-Hispanic White. Of the remaining 480,600 patients, 9% had documented disagreement (versus 2% overall).
Of the 241,391 patients documented as Hispanic or Latino in the medical record, 169,731 (70%, Figure) identified a country of origin included in the OMB definition. When adding Brazil, Haiti, Belize, and Guyana to the OMB definition, this rose to 176,048 (73%). Alternatively, 92.9% of patients with a Hispanic or Latino country of origin, as defined by the OMB, were documented as Hispanic or Latino in the binary variable available in the medical record. When Brazil, Haiti, Belize, and Guyana were included, 77.7% were documented as Hispanic or Latino in the medical record.
An important limitation of our study is the absence of standardized collection processes or documentation of data source (eg, self-report, caregiver, and staff observation). Although data collection standards prioritize self-reporting, prior studies have shown that approximately 50% of hospitals report collecting race and ethnicity from staff observation (eg, based on the patient’s appearance or assumption based on surname).^4,11^ Additionally, although missingness rates in our data were noted to be quite low (1.1%), many people (n=494,685, 16.9%) were marked as having unavailable ethnicity documented in the binary variable. Individuals with unavailable/missing binary data frequently had unavailable/missing country-of-origin data as well (n=341,182/494,685, 69%; Figure), limiting further exploration of how this may influence population estimates. Additional qualitative work specific to addressing incomplete and inaccurate data is needed to understand the root causes of these patterns. Finally, 8% of patients identified as Hispanic or Latino, as compared to approximately 19% in the City of Boston, which may reflect selection bias of our sample. This underrepresentation may have attenuated our findings.
We observed variability in the overlap between patients’ documentation in the binary Hispanic or Latino ethnicity question and their identified country of origin. Of people who identified countries of origin included in the OMB definition of Hispanic or Latino, 92.9% also answered “yes” to the binary Hispanic or Latino ethnicity question. However, there was wide variability by individual country (67% to 95%). Interestingly, when considering Latin American countries of origin excluded from the OMB definition of Hispanic or Latino, many patients (3% to 25%) simultaneously identified as Hispanic or Latino, which may indicate potential incongruence between administrative definitions and lived experiences. This has been seen in other settings and highlights the limitations of attempting to use geopolitical borders interchangeably with ethnic identity.^2^ Alternatively, a proportion may instead reflect documentation errors or assumptions the recorder made.
Based on our findings, 5% to 21% of patients identified by country of origin were documented as not Hispanic or Latino in the binary variable. As of 2023, there were 65 million people in the United States identified as Hispanic or Latino. If interpreted as a potential underestimation, a variability of 5% would correspond to a difference of 3.25 million people. As community and individual identities evolve, ongoing community engagement is critical in maintaining data collection practices that appropriately reflect the communities they are intended to capture. This is particularly important in maintaining the health of communities, as inaccurate data estimates may cause us to underestimate disease burden and miss opportunities to intervene.^13^
We also observed instances where people identified as Hispanic or Latino while identifying a country not included in any of the definitions. Several factors may explain these differences. First, as country of origin may be interpreted as either heritage or birthplace, we saw many people (11%) who identify as Hispanic or Latino also simultaneously identify as American as their country of origin (Appendix E1). Second, patients with multiethnic identities may elect to only report a country of origin not included in the OMB definition while simultaneously identifying as Hispanic or Latino.
Our findings highlight the limitations of using the binary Hispanic or Latino ethnicity variable. As the ED is a unique and critical access point, often serving to provide care for medically marginalized populations, the inclusion of ED-specific efforts to improve data collection is vital to ensure representation. Several potential solutions can be considered for improvement of data (1) ongoing engagement with communities to improve the representativeness of Hispanic or Latino ethnicity data collection, (2) using multiple data variables for data quality and improvement efforts, and (3) longitudinal data collection/confirmation (eg, in different venues or on repeat ED visits).^14^ These efforts are of critical importance as inaccurate assessment of disparities may lead to misdirection of interventions, and, ultimately, missed opportunities to decrease disease burden.