Racial Differences in Triage Assessment at Rural vs Urban Maine Emergency Departments

Data continues to accumulate demonstrating that those belonging to racialized groups face implicit bias in the emergency care delivery system across many indices, including triage assessment. The Emergency Severity Index (ESI) was developed and widely implemented across the US to improve the objectivity of triage assessment and prioritization of care delivery; however, research continues to support the presence of subjective bias in triage assessment. We sought to assess the relationship between perceived race and/or need for translator and assigned ESI score and whether this was impacted by hospital geography. We performed retrospective EMR-based review of patients presenting to urban and rural emergency departments of a health system in Maine with one of the top ten most common chief complaints (CC) across a 5-year period, excluding psychiatric CCs. We used multivariable regression to analyze the relationships between perceived race, need for translator, and gender with ESI score, wait time, and hallway bed assignments. We found that patients perceived as non-white were more likely to receive lower acuity ESI scores and have longer wait times as compared to patients perceived as white. Patients perceived as female were more likely to receive lower acuity scores and wait longer to be seen than patients perceived as male. The need for an interpreter was associated with increased wait times but not significantly associated with ESI score. After stratification by hospital geography, evidence of subjective bias was limited to urban emergency departments and was not evident in rural emergency departments. Further investigation of subjective bias in emergency departments in Maine, particularly in urban settings, is warranted.


Introduction
Emergency departments (EDs) have long served as critically important medical entry points and safety nets for many individuals, particularly those who belong to socially or economically disadvantaged groups.Many patients rely on EDs not only for emergent care but also for access to routine and preventive care services 1,2 .While all individuals, regardless of race, culture, or gender, should ideally receive fair, high-quality healthcare, many experience significant structural barriers and inequitable treatment at EDs nationwide [1][2][3][4][5][6][7][8][9][10] .Previously, studies have demonstrated significant disparities in treatment for socially disadvantaged groups across a wide range of acute conditions 1- 3,7,9-11 .For example, non-white patients experience significantly longer wait times for chest pain and acute coronary events, increased testing for substance use, decreased admission rates for traumatic injury or stroke, and less opioid administration pain management than white patients [7][8][9]11,12 . Otherstudies have demonstrated that need for interpreter increases ED length of stay (LOS), which is in turn associated with decreased patient satisfaction 5,13 .One potential explanation for these disparities is bias.In the ED, providers are often under immense pressure to make quick, critical decisions in high stakes situations where cognitive resources are strained.Prior work has demonstrated that this type of work environment can lead to increased reliance on cognitive shortcuts like bias and stereotypes [14][15][16] .Additionally, ED providers often face gaps in knowledge of patients' medical histories, lack long-standing/continuity of patientprovider relationships, and endure numerous job-related distractions, all of which can translate into medical decisions based on individuals' stereotypes or group-level judgements instead of best medical practice.Biased care can begin as early as the triage process where cognitive shortcuts may affect who receives care first and can introduce discriminatory decisions at the first touch point 17,18 .
In an effort to standardize triage and decrease bias, EDs widely implemented the Emergency Severity Index (ESI) 19 .The ESI takes into account presenting vital signs (VS), mental status, severity of pain, and projected resource utilization to generate a score ranging from 1 (most acute) to 5 (least acute).This score is used to prioritize the order in which patients are seen by medical providers in the emergency department (ED).Accumulating evidence suggests that patient race affects ESI score assignment [20][21][22][23][24][25] .These racial disparities often intersect with disparities experienced by people who live in rural areas 24 .Studies have shown that rural status confers additional disadvantage for many health care access measures.For example, Black/African American individuals residing in rural localities have lower rates of basic health screening tests, are less likely to have a primary care provider, and experience higher transportation burden when accessing medical or dental care then rural white or urban Black individuals [26][27][28] .Maine is considered the most rural state in America, with 40% of its population living in one of 11 rural counties 29 .The rural counties in Maine are designated as primary care health professional shortage areas (HPSAa), often contributing to increased ED utilization for non-emergent services.The potential impact of hospital geography (e.g.urban vs. rural) on racial disparities in the triage process has not previously been assessed, either in Maine or elsewhere nationally.In our health system, demographic characteristics such as race and ethnicity are often based upon the registered nurse's (RNs) perception rather than direct patient query.Understanding how person-perception processes, specifically racial and gender biases, affects providers' perceptions of patients and ESI assignment remains relatively under-researched.We hypothesized that patients who were perceived as non-white would be more likely to receive lower acuity ESI scores, independent of objective measures of illness severity (heart rate, respiratory rate, blood pressure, and oxygen saturation) and that this effect would be moderated by the geographic location of the hospital (urban vs. rural).
We recognize that gender identity, race, and ethnicity are socially-constructed, rather than biological, categories.The health disparities experienced by these social groups are the result of intersecting structural and individual factors, including the experiences of social devaluation via stigmatization, stereotyping, and bias 30 .It is therefore important to note that in studies that examine social group differences, race, ethnicity, and gender identity are proxies for social devaluation mechanisms, like racism and sexism.In the case of the current study, we examined differences in outcomes by perceived race and gender as a proxy for racial and gender bias in triage providers' perceptions of patients.

Study Design and Data
We performed a retrospective analysis of de-identified data extracted electronically from the MaineHealth electronic medical record (EMR).The study was reviewed and exempted by the MaineHealth (MH) Institutional Review Board (IRB).All encounters from adult patients (≥ 18 years old) presenting to MH EDs between Jan 1, 2016 -Dec 31, 2020 were eligible for inclusion.Encounters were categorized by chief complaint to control for the impact of different illness scripts and symptoms on triage scores.Patients with one of the ten most frequent categories (excluding psychiatric complaints) were included in our analysis (N=301,050).These chief complaints included: abdominal pain, back pain, shortness of breath, chest pain, dental pain, dizziness, fall, fatigue, headache, and urinary symptoms.Patients presenting with psychiatric chief complaints were excluded.

Study Outcomes
The primary outcome was ESI score, assigned by the triage RN on presentation to the ED.To characterize how patient characteristics and initial ESI score impact a patient's ED experience, we included the following secondary outcomes: 1) time from presentation to the ED to first medical provider evaluation, and 2) whether the patient was placed in a hallway bed.

Perceived patient characteristics
Perceived patient characteristics (race, need for an interpreter, and gender) were defined based on documentation from triage providers.Patients were defined as white if perceived race was reported as "white" and perceived ethnicity was reported as "non-Hispanic."All other patients were categorized as Black, American Indian/Native American/Alaska Native/Native Hawaiian/Pacific Islander/Asian (AI/NA/AN/NH/PI/A), or multiracial/other.Perceived need for an interpreter was categorized as yes / no.Due to small numbers for individual languages, we could not analyze specific language categories.
Perceived gender was categorized as male, female, or transgender/nonbinary/another gender identity.

Statistical Analyses
Clinical and demographic variables were examined by perceived race category using univariate statistics, the Kruskall-Wallis rank sum test for continuous variables, and chi-square analysis for categorical variables.
Multivariate ordinal logistic regression was used to assess the associations between perceived patient characteristics and ESI score.Regression estimates were log-transformed to produce odds ratios for study outcomes.Multivariate linear regression was used for the secondary outcome of time to first provider evaluation and multivariate logistic regression was used for hallway bed assignment.All models were mutually adjusted for the primary exposures (race, need for an interpreter, and gender).Additional covariates included age, payor type, mode of arrival (selfpresentation vs. arrived in ambulance), triage vital signs, chief complaint category, and measures of ED busyness.Where available, the National Emergency Department Overcrowding Study (NEDOCS) score was used to reflect ED busyness 31 .Where not available, we used time of day and day of the week as a proxy for ED busyness, given known patterns of ED use 6,32,33 .We used multiplicative interaction terms to assess effect modification between race and rurality.In secondary analyses, results of the analysis of ESI score were stratified by chief complaint.
Demographic and clinical characteristics are summarized as frequencies and percentages, medians and interquartile ranges, or means and 95% confidence intervals (CIs), as appropriate for the data.Results of regression analyses are presented as odds ratios (OR) for logistic regressions and beta coefficients for linear regression, with associated 95% CIs.We accepted statistical significance at p < 0.05 using two-tailed tests.Analyses were completed using R version 3.6.2statistical software.and more likely to be seen in a rural (vs.urban) ED (p<0.001).White patients were also less frequently Medicaid beneficiaries (p<0.001) and more frequently participated in Medicare (p<0.001).There were small but statistically significant differences in illness severity (as measured by triage vital signs) across racial categories, as displayed in Table 1.

Association of Perceived Race, Need for an Interpreter, and Gender with ESI Score
The associations of race, need for an interpreter, and gender with ESI score are presented in

Effect modification by rurality
Analysis of effect modification by rurality found that differences in ESI scores between Black and white patients, and between multiracial/other and white patients, were only significant in urban hospitals (p=0.001,p=0.006 for Black-rural interaction term and multiracial/other-rural interaction term, respectively).There were no differences in ESI score by race in rural locations, as depicted in Figure 1.

Chief complaint subgroup analysis
We performed a subgroup analysis by chief complaint to assess for the presence of racebased differences in ESI assignment for each chief complaint separately (Table 3).Differences in ESI score assignment between perceived racial groups were present across multiple chief complaints, including shortness of breath, headache, fall, back pain, dizziness, and fatigue, with Black patients consistently more likely to receive lower acuity (higher) ESI scores for each of these complaints.
Those in the AI/NA/AN/NH/PI/A group were also significantly more likely to receive lower acuity (higher) ESI scores within the dizziness chief complaint, as were patients in the multiracial/other group for the shortness of breath complaint in comparison to white patients.We did not observe race-based differences in ESI score assignment for the chief complaint categories of abdominal pain, chest pain, urinary symptoms, or dental pain.

Association of Perceived Race, Need for an interpreter, and Gender with Wait Time and Hallway bed assignment
Linear regression analysis revealed that, on average, Black patients waited approximately 16 minutes longer (95% CI: 9.8-23.0minutes, p<0.001) than did white patients to be seen by a medical provider.Similarly, those in the multiracial/other group waited 11 minutes longer (95% CI: 3.0-20.0minutes, p=0.008) than their white counterparts.Those who required an interpreter experienced longer waits (23 minutes, 95% CI: 12.0-34.0minutes, p<0.001) compared to those who did not need an interpreter.Conversely, male patients were seen more quickly than female patients (-2.3 minutes, 95% CI: -3.8--0.73minutes, p=0.004).Additional wait time findings are summarized in Table 4.
The odds of being assigned to a hallway bed, rather than an individual treatment space, were approximately 2% less for those in the multiracial/other group, as compared to their white counterparts (OR 0.98, 95% CI: 0.96-1.00,p=0.038).Other race-based differences in hallway assignment were not noted, as seen in Table 5.Those who required an interpreter also had significantly lower odds of being placed in a hallway space (OR 0.97, 95% CI: 0.94-0.99,p=0.019).
Full details of these findings are summarized in Table 5.

Discussion
Even though the ESI algorithm was implemented in an effort to limit triage subjectivity, research shows bias impacts acuity assignment.Previous research has documented that there are racial, geographic (rural), and gender-based disparities in multiple dimensions of ED care, including ESI assignment 10 .Here, we aimed to assess the degree to which ESI scores assigned at triage varied as a function of patients' perceived race, gender, or need for a language interpreter.
We observed that non-white patients were more likely to receive lower acuity (higher) ESI scores compared to white patients, female patients were more likely to receive lower acuity (higher) ESI scores compared to male patients, and transgender or gender non-conforming patients were more likely to receive lower acuity (higher) ESI scores compared to female patients.Our analysis of wait times reveled that, on average, Black patients and patients in the multiracial/other group waited significantly longer to been seen by a medical provider than white patients.Female patients also waited longer to see a medical provider than male patients, as did patients who needed an interpreter relative to those who did not need an interpreter.
Overall, our findings are consistent with previous studies showing racial bias in the triage process 4,20,25,34,35 .The significance of this discrepancy cannot be understated; all studies examining race and ESI assignment, including our data collected here, show that non-white patients receive discriminatory care which negatively impacts downstream health outcomes.In multiple studies, non-white patients received less acute ESI scores, waited longer to see a provider, were less likely to get tests ordered, less likely to be admitted, and have a higher mortality rate within the ED and inhospital 1,2,[4][5][6][7][8][9][10][11][12][13]27 .
We also examined whether there existed race- Additionally, many studies increased rates of ED utilization for routine healthcare services by race and ethnicity confounded by SES and health literacy. 3,36,37While we did not measure SES and health literacy in this study, we did look at illness severity and found a small but statistically significant difference in illness severity across race which could suggest a similar pattern of utilization here and may impact the ESI scoring.However, this small difference should arguably been mitigated by the need for a translator (for some of these patients) that is built into the ESI scoring system.While designed to be a place for emergency care, the paradigm ED medicine is shifting.With an increasing number of immigrants and long wait times for primary care providers, the ED must continue to serve as a critical first touchpoint and entryway for healthcare services in our communities.
When we examined the moderating effect of hospital rurality, we found that ESI scores differed between non-white and white patients in urban but not rural hospitals.This was an unexpected finding as non-white patients often experience barriers to basic medical and dental care in comparison to rural white and urban non-white patients [26][27][28]38 . We uspect this finding could be related to higher patient volumes at our urban locations as increased cognitive load is associated with increased risk of biased behavior.Additionally, two of our three urban locations are training environments for emergency medicine residents.While not assessed here or in prior studies, in the training environment there exists some emphasis on evaluating residents by the total number of patients seen potentially making residents more likely to avoid interpreter patients as these encounters invariably require more time.Finally, at our urban locations with long wait times, there is a process in place where laboratory tests and imaging studies are ordered by a triage doctor so while patients are "waiting to be seen by first provider" their initial work-up is in process and those results often impact care in addition to the ESI score (ie if someone has abnormal labs, a provider will see them more quickly while those with normal tests will wait longer).The "doctor in triage" model is not implemented in rural locations because wait times are generally shorter overall.
While not assessed here, initial bias in triage and less acute ESI scores may influence providers and affect downstream ED care including amount of tests ordered and admission rate.
Importantly, some studies show a higher mortality rate for non-white patients in the ED and hospital,. 3,10,24,25,39,40Whether there is a causal link between racial bias in the triage process and downstream indicators of health inequity requires further investigation.

Limitations
This study has several limitations that qualify our results.First, race, ethnicity, and gender were often identified by triage providers or the registration team; thus, it is possible that triage providers' perceptions of patients' social group and patients' self-identification are not the same.
Related to this, we were unable to separate ethnicity data for non-white participants as this data was rarely reported.Future research should examine both provider and patient reported perspectives, as well as congruence between patient and providers' denotation of race identification.Relatedly, differences in triage providers' assessment of patients as a function of how they identified of patients' social group membership served in the current study as a proxy for bias; because providers were most likely using phenotypical or observational characteristics to make their category judgements, and theses assessments are often rooted in perceptions of what is stereotypically typical of racial, ethnic, and gender group membership.Future studies should incorporate other operationalizations of bias, including measures of patients' subjective experiences.While our results do offer empirical support for racial and gender bias serving as an underlying mechanism driving disparities in ED care, our current ED triage format and retrospective study design do not allow for any distinction between implicit or explicit biases.Implicit and explicit bias are both activated differentially and can elicit differential outcomes; therefore, it would be necessary for future research to specifically assess implicit versus explicit bias in order to optimize the development of future interventions [41][42][43] .
Another limitation of the current study was that some racial and ethnic groups were underrepresented in our dataset, thus we condensed our categories for statistical power.Similarly, some individual languages were infrequently reported so we could not analyze these specific language categories.Additionally, the white patients in our dataset were generally older than nonwhite patients, and our results could be biased due to residual confounding by age.
Future data collection efforts and research should strive to improve our processes for collecting data related to patient's social identities, and in particular groups that are systematically underrepresented in health research, so we obtain higher quality data with which to understand the experiences of different racial and ethnic groups.Lastly, because of the exploratory nature of our study; findings should be replicated in future studies, and across contextually diverse settings.

Conclusions
This study provides evidence of systematic differences in ESI assignment across perceived racial groups, reflecting possible biases rather than true differences in illness severity.More research is needed in the area of disparities between socially advantaged and disadvantaged groups and ED care, and particularly as it relates to social group identification and providers' perceptual processes.
Future research should also aim to identify the long-term health consequences of these disparities, as well as pragmatic interventions.

Figure 1 .
Figure 1.Estimated ESI scores by race, as modified by emergency department rurality.

Figure 1 .
Figure 1.Analysis of effect modification by geographic location (urban vs. rural emergency departments) revealed that differences in estimated ESI scores, controlling for covariates, for Black vs. white and multiracial/other vs. white patients were only significant in urban hospitals.Differences were not observed in rural hospitals.N = No (not a rural hospital) Y = Yes (rural hospital)

Table 2 .
Non-white patients were more likely to receive lower acuity (higher) ESI scores compared based differences in ESI assignment for each chief complaint.Differences in ESI score assignment between racial groups were present across multiple chief complaints, including shortness of breath, headache, fall, back pain, dizziness, and fatigue, with Black patients consistently more likely to receive lower acuity (higher) ESI scores for each of these complaints.AI/NA/AN/NH/PI/A patients were also significantly more likely to receive lower acuity (higher) ESI scores within the dizziness chief complaint, as were patients in the multiracial/other group for the shortness of breath complaint.Interestingly, in the category of chest pain which is a highly protocolized work-up pathway, there was no difference in any race categories, which suggests that protocols that can be applied to all patients no matter age, gender, and race might help decrease bias.Of note, 39% of white patients were >66 years of age in comparison to 26.5% of AI/NA/AN/NH/PI/A patients, 8.3% of Black patients, and 14.1% of multi-racial patients.Although we controlled for age in our models, we grouped age into categories with 18-40 years as the reference category and may have lost granularity in doing so leading to some residual confounding by age in our findings.

Table 1 .
Patient characteristics by racial group.
Notes: AI/NA/AN/NH/PI/A = American Indian/Native American/Alaska Native/Native Hawaiian/Pacific Islander/Asian; ESI = Emergency Severity Index; EMS = Emergency Medical Services; During COVID = Visit occurred during COVID-19 lockdown period; BP = blood pressure; HR = heart rate; bpm = beats per minute; SpO2 = oxygen saturation; ED = emergency department; Wait time = time from arrival to first seen by a medical provider; IQR = interquartile range.

Table 2 .
Adjusted odds ratios for assigned ESI score by patient characteristic.AI/NA/AN/NH/PI/A = American Indian/Native American/Alaska Native/Native Hawaiian/Pacific Islander/Asian; ESI = Emergency Severity Index, *indicates statistical significance at the 0.05 level Model is adjusted for age, payor type, mode of arrival (self-presentation vs. arrived in ambulance), triage vital signs, chief complaint category, and measures of ED busyness.

Table 3 .
Adjusted odds ratios for the association between patient race and assigned ESI score, by chief complaint.

Table 4 .
Coefficients for provider wait time by patient characteristic.AI/NA/AN/NH/PI/A = American Indian/Native American/Alaska Native/Native Hawaiian/Pacific Islander/Asian.Model is adjusted for age, payor type, mode of arrival (self-presentation vs. arrived in ambulance), triage vital signs, chief complaint category, and measures of ED busyness.