Patient Sex, Racial and Ethnic Disparities in Emergency Department Triage: A Multi-Site Retrospective Study

Abstract

Objectives:

There is limited evidence on sex, racial, and ethnic disparities in Emergency Department (ED) triage across diverse settings. We evaluated differences in the assignment of Emergency Severity Index (ESI) by patient sex and race/ethnicity, accounting for age, clinical factors, and ED operating conditions.

Methods:

We conducted a multi-site retrospective study of adult patients presenting to high-volume EDs from January 2019-February 2020. Patient-level data were obtained and analyzed from three EDs (academic, metropolitan community, and rural community) affiliated with a large health system in the Southeastern United States. For the study outcome, ESI levels were grouped into three categories: 1–2 (highest acuity), 3, and 4–5 (lowest acuity). Multinomial logistic regression was used to compare ESI categories by patient race/ethnicity and sex jointly (referent = White males), adjusted for patient age, insurance status, ED arrival mode, chief complaint category, comorbidity score, time of day, day of week, and average ED wait time.

Results:

We identified 186,840 eligible ED visits with 56,417 from the academic ED, 69,698 from the metropolitan community ED, and 60,725 from the rural community ED. Patient cohorts between EDs varied by patient age, race/ethnicity, and insurance status. The majority of patients were assigned ESI 3 in the academic and metropolitan community EDs (61% and 62%, respectively) whereas 47% were assigned ESI 3 in the rural community ED. In adjusted analyses, White females were less likely to be assigned ESI 1–2 compared to White males although both groups were roughly comparable in the assignment of ESI 4–5. Non-White and Hispanic females were generally least likely to be assigned ESI 1–2 in all EDs. Interactions between ED wait time and race/ethnicity-sex were not statistically significant.

Conclusions:

This retrospective study of adult ED patients revealed sex and race/ethnicity-based differences in ESI assignment, after accounting for age, clinical factors, and ED operating conditions. These disparities persisted across three different large EDs, highlighting the need for ongoing research to address inequities in ED triage decision-making and associated patient-centered outcomes.

1. Introduction

With nearly 150 million emergency department (ED) visits each year, of which over 20 million result in a hospital admission, emergency care plays a significant role in health care delivery and population health.¹ Disparities in emergency care by sex, race, and ethnicity are well documented.^2–5 These disparities are likely due to a complex interplay of factors, including explicit and implicit biases, sociocultural differences, and systemic sexism and racism. With an increasingly diverse patient population, the ED is an ideal setting in which to understand, address, and achieve equity in health care.

In 2021, the American College of Emergency Physicians (ACEP) convened a multi-disciplinary work group on the development of quality measures to address disparities in ED care.⁶ The workgroup suggested measurement of disparities should focus on the processes of care delivery. There is growing evidence that women and racial and ethnic minorities are undertriaged in the ED compared to male and White patients.^7–11 With some inherent subjectivity, the triage nurse’s initial assessment and assignment of acuity level are prone to inequitable decisions.¹² Prior research has also identified significant sex-race interactions in ED triage although the results vary by ED setting and populations.^{11, 13} Further, inequities in triage could be exacerbated during suboptimal ED operating conditions (e.g., crowding) because cognitive stress can cause clinical decision makers to rely on personal biases.¹⁴ Investigation of the factors that influence the ED triage disparities can inform the development of measures and interventions to monitor and promote equitable emergency care delivery.⁶

There is limited research on the role of ED conditions and settings in sex, racial, and ethnic disparities in ED triage. Therefore, with retrospective data from three diverse EDs, we estimated sex, racial, and ethnic differences in the assignment of Emergency Severity Index (ESI),¹⁵ adjusting for patient age, clinical factors, and ED operating conditions. Female and minority race and ethnicity patients were hypothesized to have lower acuity ESIs assigned compared to male and White counterparts. Further, we tested whether ED-wide wait times modified any observed differences in ESI assignment by patient sex, race, and ethnicity. Results were stratified and compared by ED site to describe how ED triage disparities vary across different health care setting.

2. Methods

2.1. Ethical Approval & Reporting

This study was reviewed and approved by the University of North Carolina at Chapel Hill Institutional Review Board, in accordance with human subjects research regulations. It was determined that the risk involved was no more than minimal and was approved by expedited review. The STrengthening the Reporting of OBservational studies in Epidemiology (STROBE) guidelines¹⁶ were followed to ensure all relevant information was included in this manuscript.

2.2. Study Design & Population

We conducted a retrospective study with electronic health record (EHR) data on all adult (≥18 years old) patients presenting to the ED from January 2019-February 2020 (prior to the COVID-19 pandemic) within a large health care system in the Southeastern United States. We selected three high-volume EDs (>60,000 visits annually) affiliated with different hospital types: academic, community serving a metropolitan area, and community serving mostly rural areas. We excluded ED visits with missing or implausible values for ESI, sex, race, ethnicity, payer type, ED arrival mode, and ED timestamps. Patient sex not female or male and patient race or ethnicity documented as “Prefer not to answer” or “Unknown” were also excluded. ED visits with a chief complaint related to obstetrics/gynecology were excluded because these complaints are unique to female patients and complicate the interpretation of female-male comparisons.

2.3. Outcome & Exposures

The primary outcome was ESI assignment by the ED triage nurse ranging from 1 (highest acuity) to 5 (lowest acuity). ESI levels are typically determined by the immediate or potential need for life-saving intervention (ESI 1–2) and the anticipated need for resources, such as lab tests, medications, and specialty consultation (ESI 3–5). For this study, we classified ESI’s into three categories (1–2, 3, and 4–5) because of small numbers in the highest and lowest levels. Patient sex, race, and ethnicity were the main exposures of interest. Patient demographics are collected during the ED registration process and documented into a standard EHR system across all EDs. ED registration staff complete a training on how to collect demographic information. Patients of either female or male sex were eligible for this study. Patients are asked to self-report their race and ethnicity from fixed options in the EHR. Because of small numbers, American Indian or Alaska Native, Asian, Native Hawaiian or other Pacific Islander, and mixed races were collapsed into an “Other” group. Ethnicity is documented in the EHR as Hispanic or Latino, Not Hispanic or Latino, Prefer not to answer, or Unknown. In this study, known race and ethnicity information was combined into four race/ethnicity groups (non-Hispanic White or Caucasian, non-Hispanic Black or African American, non-Hispanic Other, and Hispanic or Latino). We decided a priori to evaluate joint exposures with interactions between sex and race/ethnicity.

2.4. Covariates

Patient age was categorized into 5 groups: 18–34, 35–49, 50–64, 65–79, ≥80 years old. Primary and secondary payer types were combined into four insurance status groups: Commercial or other, Medicaid (any), Medicare (only or plus Comm/Other), and Self-Pay. Arrival mode to the ED was classified as EMS or other means (e.g., private vehicle, public transportation, law enforcement). The chief complaint was categorized into 11 groups: Cardiovascular, Ear/Noise/Throat, Gastrointestinal, Environmental/Ophthalmologic/General, Genitourinary, Mental Health/Psychiatric/Substance Misuse, Neurologic, Orthopedic, Respiratory, Skin, and Trauma. Patient comorbidities were summarized into a weighted Elixhauser comorbidity score.¹⁷ Visit day of week - was included to characterize ED conditions. Visit time of day was categorized into six 4-hour blocks (12–3:59AM, 4–7:59AM, 8–11:59AM, 12–3:59PM, 4–7:59PM, 8–11:59PM). Overall ED wait time for each visit was operationalized as the hourly average time from triage to provider seen (in 15-min increments).

2.5. Statistical Analysis

Descriptive characteristics of patient demographics and clinical factors and ED conditions were summarized by ED: academic, metropolitan community, and rural community. Ordinal shift graphs were created comparing unadjusted ESI distributions between non-Hispanic White and non-Hispanic Black patients and male and female patients. Multinomial logistic regression models computed odds ratios (ORs) and 95% confidence intervals (CIs) for the associations between ESI groups (1–2 and 4–5 with 3 as the baseline) and patient sex and race-ethnicity, adjusting for all covariates. Model building involved testing the statistical significance of several interactions. Statistical significance was defined as p<0.05 throughout. First, race/ethnicity*sex*ED wait time product terms were tested using the likelihood ratio test and retained if statistically significant for any ED-specific model. Second, if these three-way product terms were not retained, models were fit with race/ethnicity*sex interactions. If statistically significant for any ED-specific model, all race/ethnicity*sex contrasts were presented with White males as the common referent. Lastly, if no statistically significant interactions, then final models included only main effect terms. We report the adjusted ORs and their associated 95% CIs for sex and race-ethnicity comparisons. Estimated ORs and 95% CIs were interpreted individually for each comparison. Since we did not conduct any joint or global tests, we did not adjust for multiple comparisons. Because large sample sizes can find statistical significance for small differences, our interpretation of results focused on the magnitude and precision of estimated associations. Data processing and statistical analyses were performed with R software.

3. Results

We identified 186,840 eligible ED visits with 56,417 from the academic ED, 69,698 from the metropolitan community ED, and 60,725 from the rural community ED (Figure 1). The patients seen at the metropolitan community ED were on average older than those at the academic and rural community EDs (Table 1). The academic ED had the highest proportion of Hispanic or Latino patients (12%). The majority of patients seen at the rural community ED were non-Hispanic Black (64%). The proportions of self-pay and Medicaid patients were highest in the academic and rural community EDs, suggesting lower socioeconomic status compared to metropolitan community ED patients. The three most common chief complaint categories were cardiovascular, gastrointestinal, and orthopedic for all EDs. However, based on the mean weighted Elixhauser scores, the metropolitan ED patients had the most comorbidities whereas the rural community ED patients had the least. The median ED wait (i.e., triage to provider seen) times were highest in the academic and metropolitan community (26 and 25 minutes, respectively) compared to rural community (21 minutes) (Table 2).

Figure 1.

Flow Diagram by ED

Table 1.

Patient demographic and clinical characteristics by ED

		Academic ED N=56,417		Metropolitan Community ED N=69,698		Rural Community ED N=60,725
		No.	%	No.	%	No.	%
Age group
	18–34	15257	27%	15541	22%	17604	29%
	35–49	12956	23%	12733	18%	13519	22%
	50–64	14272	25%	15417	22%	14127	23%
	65–79	9533	17%	15444	22%	10718	18%
	80+	4399	8%	10563	15%	4757	8%
Female sex		30437	54%	39848	57%	34603	57%
Race-ethnicity
	Non-Hispanic Black	16692	30%	21640	31%	38790	64%
	Non-Hispanic Other	2601	5%	2928	4%	1901	3%
	Hispanic or Latino	6860	12%	3490	5%	1288	2%
	Non-Hispanic White	30264	54%	41640	60%	18746	31%
Insurance status
	Commercial or Other	23153	41%	36444	52%	18403	30%
	Medicaid (any)	13558	24%	9721	14%	20092	33%
	Medicare (only)	7521	13%	14125	20%	8664	14%
	Self-Pay	12185	22%	9408	13%	13566	22%
EMS arrival		15611	28%	22136	32%	16452	27%
Chief complaint category
	Cardiovascular	7311	13%	14201	20%	8830	15%
	ENT	2635	5%	2674	4%	3533	6%
	Gastrointestinal	9590	17%	11823	17%	8954	15%
	General/Ophth/Environ	6314	11%	5540	8%	4757	8%
	Genitourinary	2351	4%	3246	5%	3101	5%
	MH/Psych/Subst	4015	7%	1541	2%	2564	4%
	Neurologic	5085	9%	6783	10%	5651	9%
	Orthopedic	6149	11%	7763	11%	8107	13%
	Respiratory	4580	8%	7103	10%	7064	12%
	Skin	4198	7%	3629	5%	3428	6%
	Trauma	4189	7%	5395	8%	4736	8%
Elixhauser comorbidity score (weighted)
	mean, SD	2.83	7.08	3.91	7.57	1.94	5.50

Table 2.

Operating conditions and assigned ESI levels by ED

		Academic ED N=56,417		Metropolitan Community ED N=69,698		Rural Community ED N=60,725
		No.	%	No.	%	No.	%
Time of day
	12:00AM-3:59AM	4753	8%	5523	8%	4915	8%
	4:00AM-7:59AM	3892	7%	4823	7%	5051	8%
	8:00AM-11:59AM	11465	20%	14521	21%	14025	23%
	12:00PM-3:59PM	14141	25%	17679	25%	14621	24%
	4:00PM-7:59PM	12578	22%	15557	22%	12664	21%
	8:00PM-11:59PM	9588	17%	11595	17%	9449	16%
Day of week
	Monday	8932	16%	10588	15%	9516	16%
	Tuesday	8371	15%	10214	15%	9097	15%
	Wednesday	8166	14%	9933	14%	8785	14%
	Thursday	8031	14%	9855	14%	8582	14%
	Friday	8257	15%	10058	14%	8589	14%
	Saturday	7327	13%	9515	14%	7919	13%
	Sunday	7333	13%	9535	14%	8237	14%
Average ED wait time (min)
	median, IQR	26.4	10.9–58.5	24.7	10.2–55.4	22.4	9.5–52.4
Emergency Severity Index (ESI)
	1 (most urgent)	1140	2%	1259	2%	585	1%
	2	12134	22%	13851	20%	14707	24%
	3	34572	61%	43404	62%	28323	47%
	4	8008	14%	10526	15%	15872	26%
	5 (least urgent)	563	1%	658	1%	1238	2%

The majority of patients in the academic and metropolitan community EDs had ESI level of 3 (61% and 62%, respectively). Patients seen in the rural community ED were more likely to be assigned an ESI level of 4 (26%) compared to the academic and metropolitan community EDs (14% and 15%, respectively). Unadjusted comparisons of ESI levels by race showed Black patients were less likely to be assigned ESI 1 or 2 (highest) and more likely to be assigned ESI 4 or 5 (lowest) compared to White patients across all EDs (Figure 2). Also across all EDs, female patients were less likely to be assigned ESI 1 or 2 compared to male patients although assignment of the lowest ESI levels was similar between sexes.

Figure 2.

Ordinal shift in ESIs (unadjusted) between race and sex by ED

Interaction terms for race/ethnicity*sex*ED wait times were not statistically significant in any ED (academic: p=0.685, metropolitan community: p=0.529, rural community: p=0.345). Interactions for race/ethnicity*sex were statistically significant for academic ED (p=0.030) and metropolitan community (p<0.001) but not for rural community (p=0.086). Therefore, multinomial regression models were fit with main exposures of Black female, Black male, Hispanic female, Hispanic male, Other female, Other male, White female, and White male (referent). Figure 3 displays ED-specific ORs and 95% CIs for ESI 1–3 versus ESI 3 and ESI 4–5 versus ESI 3 by race/ethnicity-sex groups adjusting for patient age, insurance status, EMS arrival mode, chief complaint category, weighted Elixhauser score, time of day, day of week, and ED wait time. In general, Black and Hispanic females were least likely to be assigned ESI 1 or 2. Sex differences were observed among all race-ethnicity groups. In all EDs, Black females and males were most likely to be assigned ESI 4 or 5. Sex differences in assignment of ESI 4 or 5 appeared minimal except for Black female and males in the metropolitan community ED.

Figure 3.

Adjusted Multinomial Logistic Regression by ED

The complete results of the unadjusted and adjusted regression models are provided in Supplemental Tables 1 and 2, respectively. Parameter estimates changed with covariate adjustment although the direction and magnitude varied by race/ethnicity-sex group and ED. As expected, age, arrival by EMS, and more comorbidities according to the Elixhauser score were strongly associated with ESI assignment. After adjusting for age, patients with Medicare (versus those with commercial insurance) were less likely to be assigned the lowest acuity ESIs in all three EDs. Further, self-pay patients were assigned lower ESIs than insured patients. Statistically significant associations between longer ED wait times and assignment of the highest acuity ESIs were observed in all three EDs although the magnitudes were small.

4. Discussion

In this retrospective study of adult patients in three diverse and high-volume EDs, we observed differences in ESI assignment by patient sex and race/ethnicity after accounting for age, insurance status, comorbidities, chief complaint, and ED conditions. Notably, in the rural community ED in which almost two-thirds of patients were Black/African American, compared to White patients, Black/African American patients were less likely to be assigned ESIs 1–2 and more likely to be assigned ESIs 4–5. Statistically significant interactions between race/ethnicity and sex were found in the academic and metropolitan community EDs, where minority females tended to be triaged less urgently. Lastly, we did not find statistical evidence that longer ED wait times modified race/ethnicity-sex differences in ESI assignment as we had hypothesized.

Our findings that females and racial and ethnic minorities were undertriaged compared to male and White counterparts is consistent with the literature.^7–11 The determination of an ESI level of 2 takes into account pain and distress, which can be treated or perceived as less urgent in women and minorities.^{18, 19} Racial and ethnic minorities, but not females, were assessed at the time of triage to require fewer resources, i.e., ESI levels 4 and 5. While these results were adjusted for insurance status, health insurance information is collected during patient registration after initial medical examination and is unlikely to be known by the triage team. In our study, Medicaid or no insurance (“self-pay”) could be a proxy for low socioeconomic status, which is susceptible to stigma and bias. Future research on the influence of social determinants of health on ED care decisions and delivery is needed. Our finding that differences in ESI assignment by race/ethnicity vary by sex has important implications for research and quality improvement programs to address disparities in ED care. Existing and novel measures to detect disparities in quality by racial and ethnic groups may need to consider sex subgroups in certain ED settings.

We hypothesized that average wait times for the entire ED would modify ESI assignment differences by sex and race/ethnicity. Even though we did not detect statistically significant interactions between these disparities and longer ED wait times in 15-min increments, we cannot conclude there was no effect of the environmental conditions of ED. There may be other operational metrics, such as ED boarding or staffing shortages, that are better proxies for cognitive stressors. It is also possible that environmental conditions do not substantially influence disparities in initial triage decisions in the ED. Since ESI assignment occurs by the triage nurse early in the ED encounter, waiting times to be seen by a provider might not influence this step as much as later decision-making. In our study, ED wait times and ESI assignment were weakly associated across the three EDs. The ED conditions in this study were prior to the COVID-19 pandemic. With increased ED boarding times during the pandemic,^20,21 there could be different ED wait time interactions due to more extreme operating conditions. Because many EDs within the study health system modified triage processes in response to COVID-19, we only examined ED visits prior to COVID-19. Future research on disparities in decision-making should consider the pandemic effects on ED volumes and operations.

Further investigation into clinician-level determinants of ED triage disparities is needed. The ESI is known to have limited interrater reliability.²² Moreover, because of the subjective nature of ED triage, these decisions are prone to providers’ implicit and explicit biases.¹² While various forms of bias and stigma are introduced in the current Emergency Severity Index Handbook (5^th Edition), nursing educational programs need to continue to evolve and incorporate the latest evidence on how to mitigate biases that can lead to inaccurate decisions. Disparities research that examines patient subgroups and ED conditions such as ours can be used to create clinical case vignettes for trainings. To develop strategies that promote equitable decision-making in the ED, future research needs to look at clinician biases, sociocultural and economic contexts, and systemic and structural factors as well.

4.1. Limitations

This study has some notable limitations. Our findings are from a single health system within one state and may not be fully generalizable to other ED settings. However, we provide results from three large EDs with different patient demographics. Also, in this study population, we were only able to examine broad race-ethnicity categories due to smaller numbers of Asians, Native Americans, and other minorities. Because this study used data readily available in the EHR, we were not able to reliably include important sociocultural factors like gender identity and primary language. Also, we did not have data on vital signs at presentation and were not able to adjust for their influence on ESI assignment. There was no gold standard measure of ESI ascertained, so we were not able to determine the accuracy of ESI assignment. Lastly, we did not have data on triage nurses’ experience level or demographics or the use of provider in triage programs,²³ which may influence ESI assignment overall and by patient sex and race/ethnicity.

4.2. Conclusions

This multi-ED study found that ESI levels assigned to adult ED patients varied by sex and race/ethnicity, after accounting for age, clinical factors, and ED operating conditions. These disparities were consistently observed across three different, large EDs. Continuous quality improvement and future research are needed to monitor and evaluate disparities in ED triage decisions and identify strategies to promote equitable ED decision-making.