a b s t r a c t

Objectives: racial disparities in Emergency medical care are abundant, and processes aimed to increase through- put, such as a rapid triage fast-track (FT) systems, may exacerbate these inequities. A FT strategy may be more susceptible to implicit bias as subjective information is obtained quickly. We aim to determine whether a FT model was associated with greater disparities between Black and White emergency department (ED) patients. Methods: Triage-related outcomes were compared across race using a cohort selected from encounters in an ED that uses a FT model. White and Black patient encounters were exact-matched on potential confounders includ- ing sex; presence of Abnormal vital signs; ED arrival time; Insurance type; age category; and chief complaint. The primary triage-related outcome was use of the FT area (versus the main ED), and the secondary outcomes were wait time and assigned encounter acuity.

Results: Encounters for 5151 black patients were exact-matched with 7179 encounters for white patients. Weights were applied to address differential numbers of encounters from each group. Within this matched co- hort, Black patients were more likely to be triaged to FT than White patients (odds ratio = 1.28, 95% CI: 1.12; 1.46) and less likely to be given a high acuity score (odds ratio = 0.73, 95% CI: 0.66, 0.81). Among the high- acuity patients, Black patients were 40% more likely to be triaged to the FT area.

Conclusions: These results suggest that, after controlling for potential confounders, racial disparities may have been exacerbated in a FT ED triage process. In a FT model utilizing physicians and midlevel providers, this may create tiered Levels of care between Black and White patients – an unacceptable side-effect of an effort to increase ED throughput.

Published by Elsevier Inc.

1. Introduction

Structural racism and Racial bias are present in nearly all facets of modern American society, including in how patients of color receive care in the emergency department (ED). These racial disparities can be illustrated in many ways: from the disparity in the rate of missed ap- pendicitis diagnoses between White and Black [1] children [2] to the nearly doubled rate of Opioid prescriptions prescribed to White patients compared to their minority counterparts [3]. Some have argued that these disparities can be explained by factors outside of the control of the emergency physician and instead are a reflection of socioeconomic factors (such as insurance status) [4]. However, research consistently

* Corresponding author.

E-mail address: [email protected] (S. Boley).

finds disparate outcomes by race even after accounting for socioeco- nomic factors [5-7].

The triage process is especially prone to unintentional racial bias due to its reliance on obtaining predominately subjective information. Typ- ically, when a patient arrives in the ED, they are evaluated briefly bya nurse using a specific protocol, vital signs are obtained, and patients are assigned an acuity score. Although several Triage systems have been implemented in EDs worldwide, the Emergency Severity Index [8] is recommended by the emergency nurses Association and the American College of Emergency Physicians [9]. The ESI utilizes 5 Acuity levels (with ESI of 1 being most acute, and ESI of 5 the least acute), and relies on subjective determinations by the triage provider, including judgements of whether the patient is suffering from an imme- diate life threat and of their expected ED resource requirement. The only objective data considered in this algorithm are vital signs – which, if markedly abnormal, can move the patient one ESI level [8].

https://doi.org/10.1016/j.ajem.2022.07.030 0735-6757/Published by Elsevier Inc.

The acuity score influences what type of bed a patient is ultimately evaluated in (such as a designated high-acuity or stabilization bed), how quickly the patient is roomed, and how long a patient waits for evaluation and definitive management [10]. Using large databases (pri- marily the National Hospital Ambulatory Medical Care Survey [11]), previous authors have shown that Black patients wait longer in the waiting room than White patients [12-15]. Using the same database, others have shown that visits by Black patients tend to be classified as lower acuity [16,17]. Schrader et al. used a retrospective matched cohort design to study this phenomenon and confirmed that Black patients had longer wait times and their presentations were deemed to be less acute [18].

Due to the need to increase throughput as a result of ever-increasing annual ED visits, various strategies have been employed to more effi- ciently see patients [19]. One strategy is the utilization of a “rapid- triage” model, which has been shown to increase ED throughput mea- sures [20]. This model adds an additional determination in which pa- tients are triaged to a dedicated fast-track (FT) or lower-Acuity area instead of a bed in the main ED. This evaluation may be even more sub- jective than the traditional triage process and is often done without obtaining a complete set of vital signs. Protocols for these processes vary by institution and have been well reported previously [20-26]. It has recently been suggested that this rapid-triage model may be espe- cially prone to racial bias [27]. Addressing Racial inequities in our health system is crucial, and thus it is important to understand how interven- tions designed to increase throughput may affect whether care is equi- table.

This study aims to examine patterns of structural racism in the care of patients within the ED. Specifically, we aim to answer two questions:

1) Do triage outcomes differ for Black patients relative to White patients in the context of a rapid-triage protocol? 2) Are racial differences in triage outcomes similar across categories of chief complaint?

1. Methods
   1. Study design and setting

This is a retrospective matched cohort study of ED visits at a single institution in St. Paul, Minnesota, USA. The hospital is a 556-bed tertiary care center, the largest hospital in the city, and is both an accredited stroke center and a level 3 trauma center. The hospital serves as a refer- ral center for patients throughout Minnesota, the eastern Dakotas, and western Wisconsin and is located in a neighborhood with varied racial, ethnic, and socioeconomic demographics, resulting in a highly diverse patient population. The hospital is situated between the West 7th neighborhood (21.9% individuals of color and 30.2% with a Household income of <$35,000), the downtown neighborhood (31.5% individuals of color and 32.5% with a household income of <$35,000), and the Summit-University neighborhood (49.6% individuals of color and 37.5% with a household income of <$35,000) [28].

In early 2019, after a series of value stream mapping projects [29] were undertaken to improve throughput and decrease the number of patients who leave without being seen, the ED transitioned to a split- flow triage process in which patients are deemed to be appropriate for the main ED or assigned to a FT bed. The triage process proceeds as usual, with a nurse checking in the patient, determining a chief com- plaint, and obtaining a set of vital signs. Using the ESI protocol, the pa- tient is then given a score of 1 to 5, ranging from most to least acute. Immediately following this determination, the nurse determines whether the patient is appropriate for FT based on whether the patient satisfies the following requirements:

1. The patient is able to sit in a recliner.
2. The patient is ambulatory and able to speak.
3. The patient’s ESI score is 3, 4, or 5 (lowest acuity).
4. The patient is determined to be not critical based on the triage nurse’s determination.

Additional guidance is given to the triage nurse to avoid FT place- ment for pregnant patients, patients over 60 years of age, patients with altered mental status, and patients with abnormal vitals. However, these are not strict criteria. Patients identified for FT are then seen in a separate 5-bed area of the ED set up with recliners where patients can be seen, evaluated, and treated rapidly. Patients can receive intravenous fluids, medications, and receive laboratory and radiology tests in the FT area, and care is provided by a physician assistant (PA) or nurse practi- tioner (NP), nurse, and ED technician. These providers see patients inde- pendently but involve an ED physician from the main department depending on patient complexity – typically patients with high acuity ESI scores of 1,2, or 3, those who are likely to be admitted, and any pa- tient that requires advanced imaging. If a patient is deemed not eligible for FT they wait until an ED bed becomes available, where care is typi- cally provided by an ED physician (or PA in a minority of cases). The split-flow triage process operates from 9 am to 11 pm.

This study used a matched cohort design to compare triage process

outcomes by racial groups. Exact, full matching was used to adjust for several potential confounding patient-level and encounter-level charac- teristics. In addition to comparisons among the overall matched cohort, all outcomes were examined within subgroups based on four specific chief complaints: abdominal pain, chest pain, shortness of breath, and headache.

• 1. Study sample and data collection

All study data come from the electronic health record . ED en- counters were extracted from a one-year period starting after the full implementation of the rapid triage system (2/1/2019-1/31/2020). En- counters that would not be eligible for the rapid-triage process were ex- cluded, namely patients who arrived via ambulance or police, patients who arrived when the rapid-triage process was not operating (11 pm to 9 am), and patients who Left without being seen. In accordance with the Minnesota Health Records Act, encounters for patients without consent on file for the use of their EHR data for research were excluded. Additionally, encounters for patients aged <18 or whose age was miss- ing were excluded. Encounters were excluded if data were missing for key measures needed for the study (i.e., triage disposition, ESI, race/eth- nicity, wait time). After applying the exclusion criteria, we determined that the numbers of encounters for race/ethnicity groups other than White non-Hispanic (White-NH) and Black non-Hispanic (Black-NH) were too low to include in the analysis. Consequently, the comparison presented here is between White-NH and Black-NH patients. We fur- ther excluded White-NH encounters in which the patient needed a translator so that the comparison population included only patients least likely to experience bias or barriers.

The Allina Health Institutional Review Board (IRB) determined that IRB approval was not required for this study.

• 1. Measurements

The primary outcome was whether a patient was assigned to the FT area or the regular ED. The secondary outcomes were wait time (i.e., the time from ED arrival to rooming time, in minutes) and ESI score dichot- omized into high-acuity (ESI 1-3) and low-acuity (ESI 4-5). Dichoto- mous ESI score was treated as a separate outcome rather than a confounding variable given that, as described above, ESI assessment is highly subjective and potentially misleading if treated as an indepen- dent variable. This is because both ESI and FT eligibility are determined at the same time by the triage provider, therefore are both prone to the

same potential biases and should be treated as separate outcomes of the preexisting implicit bias rather than as confounders of each other.

Our primary comparison was between White-NH or Black-NH pa- tients. Race and ethnicity (i.e., Hispanic or non-Hispanic) are collected as separate measures in the EHR based on patient self-reporting. Addi- tionally, patients can report multiple races.

Other extracted measures include age (categorized as 18-39, 40-49, 50-59, 60-69, 70-79, and 80+ years), sex, Preferred language, ED pre- sentation time (9 am-1:59 pm, 2 pm-5:59 pm, 6 pm-11 pm), chief com- plaint (grouped into 49 categories, see Appendix 1), insurance type (private, Medicaid, Medicare, other), and discharge disposition (dis- charge or admit/transfer). Vital signs were also used in the matching process. Each vital sign was classified as normal or abnormal according to the following criteria to define normal ranges: respiratory rate (12-18 breaths/min), heart rate (60-100 beats/min), blood pressure (systolic 90-180/diastolic 40-100), temperature (36-38C), and oxygen saturation (>=90%). A dichotomous composite variable of all vital signs was used to divide the patients into those who had any abnormal vital signs and those who had all normal vital signs (to match the dichotomi- zation typically used in the acuity classification process).

• 1. Matched design

Encounters for White-NH patients were matched with encounters for Black-NH patients using exact full matching [30-32]. Encounters from these two Racial/ethnic groups were exactly matched on the fol- lowing characteristics: sex; vital statistics (i.e., any abnormal versus all normal); ED arrival time; insurance type; age category; and chief com- plaint. All unique combinations of the six matching variables create unique matched groups, or strata. To be included in the analysis, a stra- tum must contain at least one Black-NH encounter and at least one White-NH encounter; however, the number of White-NH and Black- NH encounters may differ depending on how many of each met the criteria. Weights were generated [32,33] to account for any differences in the number of encounters from each race/ethnic group within each unique stratum. Specifically, the White-NH encounters were weighted to be proportional to the number of Black-NH encounters within each stratum. Exact matching was chosen (as opposed to propensity score matching) to guarantee a balanced design within the subgroup analyses examining specific chief complaints.

• 1. Analysis

Descriptive statistics were used to document the study sample before and after matching, using weighted analysis for the post-matching sam- ple. Weighted logistic regression was used to model dichotomous out- comes of FT (versus non-FT) and high ESI acuity (versus low), and comparisons were summarized as odds ratios (ORs) with 95% confidence intervals (95% CIs). Weighted linear regression was used to analyze the continuous measure of wait time, and comparisons were summarized as differences in average wait time with 95% CIs. All models were run on the matched sample as well as within categories of major chief com- plaints: abdominal pain, chest pain, shortness of breath, and headache. To assess whether the association between Patient race/ethnicity and FT is modified by acuity level, additional models for FT included an inter- action term for acuity and race, with results reported stratified by ESI acuity level. The unit of analysis was patient encounters, so individual pa- tients may be represented multiple times; our analysis did not account for within-patient correlation. Stata 17 was used for all analyses [34].

1. Results
   1. Study sample

There were 50,484 ED encounters during the 12-month study pe- riod. After the initial exclusion criteria were applied, 23,052 encounters

were eligible for rapid triage and had complete data. A majority of en- counters were excluded because the patient arrived by ambulance/ police or arrived outside of the rapid-triage program operation hours (Fig. 1). After narrowing the sample further to encounters only with Black-NH and White-NH patients, there were 19,651 encounters (6313 for Black-NH patients and 13,338 for White-NH patients). The distribution of several patient characteristics was substantially different between White-NH patients and Black-NH patients (Table 1) prior to matching. After applying exact matching criteria were applied, 5151 (82%) Black-NH encounters were matched to 7179 White-NH English- speaking encounters (Fig. 1). The final sample includes 12,330 encoun- ters for 9704 unique patients.

Black-NH patient encounters that did not have matching encounters with White-NH patients were more likely to be male, older, and more likely to have insurance through Medicare relative to the Black-NH encounters that did have matches (Appendix 2). After matching and applying weights for White-NH comparisons, the distribution of demo- graphics was similar between the two study groups. In both groups, the participants were 70% female, with an average age of 37 years, and 57.9% had insurance through Medicaid (Table 1). Some differences remained for unmatched variables such as marital status and blood pressure.

• 1. Triage outcomes

Table 2 summarizes the comparison of Black-NH and White-NH pa- tients for all triage outcomes among the entire matched cohort and within subgroups based on chief complaint. Black-NH patients were sig- nificantly more likely to be triaged to the FT workflow area compared to matched White-NH patients (22.6% versus 18.5%, OR = 1.28, 95% CI: 1.12, 1.46). Within all four chief complaint subgroups, the odds of a triage disposition of FT for Black-NH patient encounters were higher relative to White-NH patients, however this association was only statis- tically significant for the headache group: abdominal pain OR = 1.50 (95% CI: 0.84, 2.70), chest pain OR = 1.76 (95% CI: 0.77, 4.02), shortness

of breath OR = 1.56 (95% CI: 0.62, 3.94), and headache OR = 2.10 (95% CI: 1.01, 4.39).

Black-NH patients were significantly less likely to be triaged as high acuity than matched White-NH patients (59.8% versus 67.0%, OR = 0.73, 95% CI: 0.66, 0.81). Trends were similar for all four complaint cat- egories, with a significant difference for shortness of breath OR = 0.45 (95% CI: 0.22, 0.94), and non-significant differences for chest pain OR

= 0.63 (95% CI: 0.37, 1.07), abdominal pain OR = 0.53 (95% CI: 0.24,

1.15), and headache OR = 0.73 (95% CI: 0.43, 1.24). In brief, among pa- tients with shortness of breath, Black-NH patients were about twice as likely as White-NH patients to be triaged as low acuity.

The mean wait time from ED arrival to being assigned an ED bed was

3.47 min shorter, and was statistically significant (95% CI-6.56, -0.37) for Black-NH patients (mean: 54.0; 95% CI: 52.3, 55.7) than White-NH patients (mean: 57.5; 95% CI: 54.9, 60.1). Within the chief complaint categories, the average wait times ranged from a statistically significant 18 min shorter for Black-NH patients with chest pain (95% CI: -28.9,

-8.7) to a non-statistically significant 8 min longer for Black-NH patients with headache (95% CI: -5.6, 21.8).

Table 3 shows the results of the ED FT model with interaction of race and ESI-level, specifically comparing Black-NH and White-NH patients stratified by low- and high-acuity. Overall, among patients who were categorized as low-acuity, there was no difference between Black-NH and White-NH patients with regard to FT status. However, among pa- tients categorized as high-acuity, Black-NH patients were about 40% more likely than White-NH to be fast-tracked with a statistically signif- icant association (OR = 1.40, 95% CI: 1.05, 1.87). Of note, the high-Acuity group was primarily made up of patients with an ESI score of level 3, with no level 1 patients, and a single level 2 patient. A similar pattern was observed among patients with abdominal and chest pain, with larger non-statistically significant effect sizes within the high-acuity

Fig. 1. Sample inclusion.

groups (ORs of 2.26 (95% CI: 0.91, 5.60) and 6.57 (95% CI: 0.90, 48.14)).

Within the abdominal pain category, a much lower OR of 0.29 was seen in the low-acuity group. In contrast, the direction of the effects changed among patients with headache, suggesting that chief complaint plays a role in the effect modification of acuity on ED FT status.

1. Discussion

Optimizing the triage process in the ED has become more important as ED volumes increase. Competing priorities influence how the process is undertaken, including the desire to increase throughput while ensur- ing that the sickest patients are being recognized and cared for rapidly. Triage nurses often rely on subjective determinations of a patient’s acu- ity, including an overall gestalt of a patient’s illness severity. Unfortu- nately, such subjective determinations are prone to the influences of implicit bias, and they may play a large role in exacerbating health disparities in our healthcare system.

Our results suggest that, in this single institution that utilizes a split- flow rapid-triage ED system, Black patients are more likely to move through the rapid workflow area of the ED compared to their White counterparts, despite matching them in terms of objective patient en- counter characteristics. These same Black patients are also less likely to be characterized as having high-acuity presentations (based on ESI score). Given that these determinations – ESI score and determination of FT eligibility – occur at the same time they are considered to be com- mon effects of the same cause rather than confounding factors. To- gether, these triage outcomes undoubtedly have downstream effects

on patient care. Previous studies have shown that ESI score is directly correlated to the resource utilization [35], admission likelihood [36], and specialty consultation utilization [36].

The shorter average wait time for Black-NH patients compared to White-NH patients found in our study likely results from the Black-NH patients being more likely to be given FT status with increased FT bed availability and pressure to room them in the FT area quickly. Given that FT beds tend to be staffed by PAs in our institution, these results may lead to different tiers of care for Black-NH patients versus White- NH patients. This unfortunate sequela of a system designed to improve Patient throughput is unacceptable given our aim to provide equitable care for all. This particular discrepancy in care could be partially ad- dressed by staffing the FT area with an emergency physician rather than a PA or NP. Nevertheless, overcoming the anchoring bias (i.e., relying too heavily on the first piece of information) associated with seeing patient in the FT area would likely continue to affect patient care, even with an emergency physician directing care, especially re- garding the ordering of laboratory tests and advanced imaging studies. Davey et al. first investigated whether a rapid-triage system that in- cludes primarily subjective determinations of acuity is prone to racial bias [27]. They retrospectively looked at nearly 3000 patients at a single institution who presented with either chest pain or abdominal pain. Their results indicated that non-White patients were twice as likely to be triaged to the low-acuity area than their White counterparts (24% vs 12%). Outside of the chief complaint inclusion criteria, Davey et al. did not adjust for potential confounders. In our matched cohort – in- cluding all chief complaint categories – we found nearly identical

Study sample before and after matching.

	Before matching				Matcheda
	White-NH (n = 13,338)	Black-NH (n = 6313)	p-value		White-NH (n = 7179)	Black-NH (n = 5151)	p-value
Sex
Female	59.6% (7943)	67.5% (4259)	<0.001		70.4% (4727)	70.4% (3628)	1.000
Male	40.5% (5395)	32.5% (2054)			29.6% (2452)	29.6% (1523)
Age, mean(SD)	53.54 (19.97)	37.88 (14.81)	<0.001		37.44 (36.98)	36.95 (13.25)	0.113
18-29	29.6% (3946)	63.1% (3982)	<0.001		66.6% (3198)	66.6% (3432)	1.000
30-49	13.3% (1768)	15.5% (980)			14.0% (1003)	14.0% (720)
50-59	16.0% (2140)	11.0% (697)			9.9% (1044)	9.9% (511)
60-69	16.9% (2249)	6.8% (432)			6.3% (982)	6.3% (325)
70-79	13.0% (1732)	2.5% (155)			2.4% (667)	2.4% (123)
80+	11.3% (1503)	1.1% (67)			0.8% (285)	0.8% (40)
Preferred language
English	99.9% (13,320)	93.1% (5878)	<0.001	99.9% (7169)		93.4% (4810)	<0.001
Spanish	0% (2)	0% (0)		0.04% (2)		0% (0)
Somali	0% (0)	3.9% (249)		0% (0)		3.9% (201)
Other	0.1% (16)	3% (186)		0.04% (8)		2.7% (140)
Interpreter needed, % yes	0% (0)	5.5% (345)	<0.001	0% (0)		5.1% (264)	<0.001
Marital Status
Married/ partner	43.4% (5783)	21.1% (1332)	<0.001	28.9% (2946)		21.5% (1106)	<0.001
Single	56.4% (7517)	78.4% (4950)		70.8% (4214)		78.0% (4019)
Missing	0.3% (38)	0.5% (31)		0.2% (19)		0.5% (26)
Insurance category
Private	40.2% (5359)	16.7% (1055)	<0.001	18.6% (3027)		18.6% (957)	1.000
Medicaid	18.5% (2467)	54.8% (3461)		57.9% (1942)		57.9% (2981)
Medicare	37.2% (4962)	12.8% (808)		11.6% (1864)		11.6% (595)
Other	4.1% (550)	15.7% (989)		12.0% (346)		12.0% (618)
Chief complaint – top 10 Categories
Body pain/injury	14.3% (1144)	19.4% (717)	<0.001	20.5% (854)		20.5% (680)	1.000
Abdominal pain	15.3% (1223)	16.7% (617)		18.1% (949)		18.1% (600)
Chest pain	13.4% (1074)	10.4% (384)		10.3% (809)		10.3% (343)
GI problem	8.9% (713)	10.7% (393)		10.9% (475)		10.9% (362)
Weak/dizzy	10.5% (837)	5.9% (218)		5.5% (460)		5.5% (183)
SOB	10.2% (815)	6.4% (236)		5.5% (448)		5.5% (184)
Trauma	8.2% (659)	6.7% (248)		6.7% (286)		6.7% (223)
Asthma/COPD/cough	5.0% (397)	11.4% (422)		11.3% (265)		11.3% (376)
Back/Neck pain	6.1% (487)	8.6% (319)		7.9% (314)		7.9% (263)
Cardiac symptom	8.3% (662)	3.7% (137)		3.2% (325)		3.2% (106)
Vital statistics composite measure: normal if all individual vital measures are normal, otherwise abnormal
Normal	68.1% (9083)	70.2% (4433)	0.003	74.3% (5365)		74.3% (3827)	1.000
Abnormal	31.9% (4255)	29.8% (1880)		25.7% (1814)		25.7% (1324)
Vital Statistics, individual (continuous and dichotomous) measures
Respiration, mean(SD)	17.17 (2.40)	17.03 (2.27)	<0.001	17.08 (3.58)		16.97 (2.07)	0.265
Normal, 12-18/min	88.0% (11,741)	90.1% (5689)	<0.001	91.0% (6506)		91.2% (4699)	0.729
Abnormal	12.0% (1597)	9.9% (624)		9.0% (673)		8.8% (452)
Heart Rate, mean(SD)	79.94 (15.09)	82.85 (14.03)	<0.001	83.45 (15.2)		82.63 (12.41)	0.015
Normal, 60-100/min	84.7% (11,301)	86.0% (5427)	0.023	86.3% (6255)		87.8% (4521)	0.074
Abnormal	15.3% (2037)	14.0% (886)		13.7% (924)		12.2% (630)
Systolic BP, mean(SD)	133.35 (21.06)	132.61 (20.41)	0.021	129.25 (19.83)		131.48 (18.08)	<0.001
Normal, 90-180 mmHg	97.3% (12,975)	97.3% (6143)	0.908	99.0% (7060)		97.8% (5038)	<0.001
Abnormal	2.7% (363)	2.7% (170)		1.0% (119)		2.2% (113)
Diastolic BP, mean(SD)	76.74 (13.39)	79.77 (14.04)	<0.001	77.39 (13.87)		79.09 (12.45)	<0.001
Normal, 40-100 mmHg	95.6% (12,750)	92.5% (5839)	<0.001	96.2% (6918)		93.9% (4834)	<0.001
Abnormal	4.4% (588)	7.5% (474)		3.8% (261)		6.2% (317)
Temperature, mean(SD)	98.1 (0.85)	98.28 (0.66)	<0.001	98.18 (0.88)		98.28 (0.6)	<0.001
Normal, 36-38 ?C	97.3% (12,980)	98.0% (6184)	0.007	98.0% (7044)		98.2% (5057)	0.498
Abnormal	2.7% (358)	2.0% (129)		2.0% (135)		1.8% (94)
Oxygen saturation, mean(SD)	96.87 (2.96)	97.79 (3.00)	<0.001	97.46 (2.28)		97.88 (2.58)	<0.001
Normal, >=90%	98.5% (13,136)	99.4% (6272)	<0.001	99.4% (7109)		99.5% (5124)	0.618
Abnormal	1.5% (202)	0.7% (41)		0.6% (70)		0.5% (27)

^a Data for the matched sample show weighted proportions and means and actual n’s.

differences in those triaged to the lower acuity workflow. They suggest that relying on largely subjective data can exacerbate racial disparities. Schrader et al. used a matched cohort design similar to that in our study at a single institution to investigate triage acuity scores and wait times between races [18]. They matched patients based on age, sex, chief complaint, insurance status, time of presentation, and presence of abnormal vital signs. They investigated over 19,000 encounters and were able to match 4210 Black patients with White patients. Looking

at mean ESI scores, they found that Black patients were scored as consis- tently less acute (in the range of 0.1-0.3 difference) than White patients for all chief complaints they investigated with the exception of syncope. Although we chose to look at the more clinically significant grouped “high” or “low” acuity scores rather than the raw numbers, we also found evidence of a bias toward lower acuity determination for Black patient encounters. They further found that Black patients waited on av- erage 10.9 min longer compared to their matched White counterparts

Table 2

Comparison of triage outcomes of Black and White patients eligible for rapid-triage for overall cohort and among four chief-complaints.

	Black-NH	White-NH	Comparisona	p-value
Total
ED Fast track	22.6% [21.4, 23.7]	18.5% [16.8, 20.3]	1.28 [1.12, 1.46]	<0.001
High acuity	59.8% [58.4, 61.1]	67.0% [65.0, 69.1]	0.73 [0.66, 0.81]	<0.001
Mean wait time, minutes	54.0 [52.3, 55.7]	57.5 [54.9, 60.1]	-3.47 [-6.56, -0.37]	0.028
Abdominal Pain
ED Fast track	4.8% [3.1, 6.5]	3.3% [1.1, 5.4]	1.50 [0.84, 2.70]	0.303
High acuity	94.7% [92.9, 96.5]	97.1% [95.2, 99.0]	0.53 [0.24, 1.15]	0.107
Mean wait time, minutes	69.3 [63.3, 75.3]	78.8 [70.0, 87.7]	-9.52 [-20.07, -0.03]	0.028
Chest Pain ED Fast track	7.9% [5.0, 10.7]	4.6% [2.4, 6.8]	1.76 [0.77, 4.02]	0.082
High acuity	88.9% [85.6, 92.2]	92.7% [90.0, 95.5]	0.63 [0.37, 1.07]	0.086
Mean wait time, minutes	58.1 [50.9, 65.2]	76.9 [69.7, 84.0]	-18.82 [-28.93, -8.72]	<0.001
Shortness of Breath ED Fast track	6.5% [3.0, 10.1]	4.3% [1.4, 7.2]	1.56 [0.62, 3.94]	0.343
High acuity	87.0% [82.1, 91.8]	93.6% [90.1, 97.2]	0.45 [0.22, 0.94]	0.034
Mean wait time, minutes	60.4 [50.0, 70.7]	59.5 [49.1, 69.8]	0.91 [-13.7, 15.52]	0.903
Headache ED Fast track	12.4% [7.7, 17.2]	6.3% [2.8, 9.8]	2.10 [1.01, 4.39]	0.048
High acuity	79.5% [73.6, 85.3]	84.2% [78.9, 89.5]	0.73 [0.43, 1.24]	0.238
Mean wait time, minutes	64.4 [54.7, 74.1]	56.4 [46.7, 66.0]	8.07 [-5.62, 21.76]	0.247

^a Comparison of proportions are odds ratios and comparisons of means are differences.

(although this was in an ED that did not utilize a FT system). Our results contrast with this in that wait time was actually shorter for Black-NH patients for reasons suggested previously.

The present study has several important limitations, most notably the retrospective nature of the analysis. The bias-prone nature of retro- spective studies has been well described [37], and every effort was made to minimize such bias in our data. Unfortunately, the accuracy of retrospective data cannot be determined; i.e., we cannot determine whether the data were inaccurately input into the EHR. Further, certain variables are especially prone to potential bias – none more than the race and ethnicity variables. Such a variable is prone to bias at several points, including: 1) whether the patient chooses to divulge their race accurately, 2) whether the correct race is input into the EHR, and most concerning, 3) whether the EHR can accommodate a full range of race and/or ethnicity. Unfortunately, a prospective randomized design inves- tigating these outcomes is impossible given the inability to randomize encounters by patient race. Further, obtaining a large enough sample size for interpretation is only feasible in a retrospective design.

The EHR utilized in the institution studied here has a single category for “Black or African American” which is selected for patients who may self-identify as Black, as well as those patients who are more recent

immigrants from the African continent. These groups of patients have profoundly different cultural experiences and backgrounds and likely suffer from implicit bias in different ways. Unfortunately, it is difficult to differentiate such populations (as demographic information such as “preferred language” or “country of origin” is inconsistently obtained), a limitation that should be addressed immediately so as to allow for a richer and more accurate picture of patient demographics in our health system. A major limitation to obtaining more granular data is the EHR vendors’ lack of incentive to do so, as the demographic information they collect is guided by what is required by the Health information technology for Economic and Clinical Health Act (HITECH). Expanding beyond what was laid out in this act would require significant expense [38].

Another limitation is possible confounding due to imbalance of un- matched factors. For example, after matching imbalance remained for preferred language and need for interpreters, however this difference was expected given our choice of comparison group of English- speaking non-Hispanic white patients. Other imbalances remained for some vital signs (i.e., blood pressure), however, we exact matched for abnormal vitals for any measure, which is the main criteria used for triaging patients.

Table 3

Comparison of ED fast track outcome of Black and White patients stratified by low and high ESI acuity level for overall cohort and among four chief-complaints.

	Black-NH	White-NH	Odds ratio	p-value
Total
ED fast track among low acuity	51.1% [48.9, 53.2]	51.2% [48.8, 53.6]	0.99 [0.87, 1.13]	0.934
ED fast track among high acuity	3.4% [2.8, 4.1]	2.5% [1.9, 3.0]	1.40 [1.05, 1.87]	0.024
Abdominal Pain ED fast track among low acuity	43.8% [26.6, 60.9]	73.2% [52.3, 94.0]	0.29 [0.08, 1.02]	0.053
ED fast track among high acuity	2.6% [1.3, 4.0]	1.2% [0.3, 2.1]	2.26 [0.91, 5.60]	0.079
Chest Pain ED fast track among low acuity	52.6% [36.8, 68.5]	59.1% [39.8, 78.4]	0.77 [0.28, 2.14]	0.616
ED fast track among high acuity	2.3% [0.6, 4.0]	0.4% [-0.3, 1.0]	6.57 [0.90, 48.14]	0.064
Shortness of Breath ED fast track among low acuity	50% [30.0, 70.0]	50% [21.4, 78.7]	1.0 [0.25, 4.04]	0.999
ED fast track among high acuity	no estimate	1.2% [-0.4, 2.8]	no estimate
Headache ED fast track among low acuity	55.3% [39.5, 71.1]	27.0% [10.9, 43.1]	3.34 [1.19, 9.44]	0.023
ED fast track among high acuity	1.4% [-0.5, 3.2]	2.4% [0, 4.9]	0.55 [0.10, 3.09]	0.497

This study is also limited by the lack of a true measure of acuity of presentation. As noted, ESI is a measure derived from almost entirely subjective information and so could not be directly utilized here, as it is just as prone to being influenced by implicit bias as the primary out- come we are studying.

Our study was further limited by the focus on a single institution, which has a patient population with a specific demographic make-up and used a specific approach to implementing a FT model. This signifi- cantly limited the generalizability of the present results. We had initially planned to include other affiliated hospitals in the health system to in- crease overall generalizability but found that the split-flow model was utilized inconsistently elsewhere, making it very difficult to determine when patients were eligible for the accelerated pathway and what time of day it was active. We ultimately decided that the reliability and accuracy we could attain by investigating a single institution with consistent utilization outweighed the goal of increasing the generaliz- ability of the results. We look forward to repeating the analysis with affiliated institutions around the metropolitan area as the FT process is implemented and standardized across the health system.

An additional limitation that may affect interpretation of the results is due to the exclusion of patients who came in by EMS. The decision to exclude these patients was because they typically bypass the triage pro- cess in our institution and are brought directly to a main ED room. That being said, the decision to come in by EMS may be affected by race and thus would confound our results. This decision may also be motivated by a lack of transportation, differences in medical knowledge, socioeco- nomic factors, et cetera.

Finally, our study is limited due to being unable to determine whether the system was being implemented consistently across all pa- tients and across all triage providers. There may have been a period of time where providers were “getting used” to the new protocols which may have affected which patients were or were not triaged to the fast track area. Further, we did not examine or control for which triage pro- viders were working during any period of time in the ED so cannot determine the effect of any individual on the final results.

1. Conclusion

In summary, a rapid-triage split-flow model appears to lead to under-triaging Black patients in the ED as compared to matched White counterparts. This, in turn, may lead to disparities in the level of care they receive – further exacerbating health disparities between the races. Efforts should be made both to rely on less subjective informa- tion in the triage process (recognizing the profound effect of implicit bias) and to explore ways to lessen implicit bias among ED staff.

Funding

Funding for this study came from the United Hospital Foundation.

Declaration of Competing Interest

None.

Acknowledgements

The authors would like to acknowledge Martin Cozza for editing assistance on the final manuscript.

Appendix 1

Chief complaint list. Abdominal pain

Altered mental status/dementia Assault

Asthma/COPD/cough Back/neck pain

Body pain/injury Breast problem Cardiac symptom Chest pain

Colon or rectal problem Dental problem Endocrine problem

feeding tube/vascular access problem Flank/kidney problem

Foreign body Gastrointestinal problem

Head/eyes/ears/nose/throat (HEENT) problem Head/face injury

Headache/migraine Hematology problem Infection other Laceration Mammalian bite

Medication effect/allergy Mental Health Musculoskeletal pain/injury Neurological problem/other OB/Gyn problem occupational medicine Other abdominal problem Other/nonspecific Outpatient result referral Post-operative problem Primary care need Seizure/convulsion Shortness of breath

Skin problem Substance use problem Throat problem

Toxicology or overdose problem Trauma

Urology/male genitourinary problem

Urinary tract infection/sexually transmitted infection Weak/dizzy

Appendix 2

Comparison of Black-NH encounters included and excluded from final sample.

CRediT authorship contribution statement

Sean Boley: Writing – original draft, Supervision, Methodology, Investigation, Funding acquisition, Conceptualization. Abbey Sidebottom: Writing – original draft, Visualization, Project administra- tion, Methodology, Investigation. Marc Vacquier: Writing – review & editing, Visualization, Methodology, Formal analysis. David Watson: Writing – review & editing, Visualization, Methodology, Formal analysis. Jeremy Olsen: Writing – review & editing, Funding acquisition, Concep- tualization. Kelsey Echols: Writing – review & editing, Funding acquisi-

Age, mean (SD)	37.88 (14.81)	36.95 (14.50)	41.99 (15.45)
18-29	63.1% (3982)	66.6% (3432)	47.3% (550)
30-49	15.5% (980)	14.0% (720)	22.4% (260)

tion, Conceptualization. Sara Friedman: Writing – review & editing,

Black encounters eligible for

Black encounters with matches included in final

Black encounters excluded due to no matching white

p-values?

<0.001

	study (n = 6313)	analysis (n = 5151)	encounters (n = 1162)
Sex
Female	67.5% (4259)	70.4% (3628)	54.3% (631)	<0.001
Male	32.5% (2054)	29.6% (1523)	45.7% (531)

Funding acquisition, Conceptualization.

(continued on next page)

Appendix 2 (continued)

Black encounters eligible for study (n = 6313)

Black encounters with matches included in final analysis (n = 5151)

Black encounters excluded due to no matching white encounters (n = 1162)

p-values?

1. Zhang X, Carabello M, Hill T, Bell SA, Stephenson R, Mahajan P. Trends of racial/eth- nic differences in emergency department care outcomes among adults in the United States from 2005 to 2016. Front Med. 2020;7:300.
2. Schrader CD, Lewis LM. Racial disparity in emergency department triage. J Emerg Med. 2013;44:511-8.
3. Morley C, Unwin M, Peterson GM, Stankovich J, Kinsman L. Emergency department crowding: a systematic review of causes, consequences and solutions. PLoS One. 2018;13:e0203316.

50-59 11.0% (697) 9.9% (511) 16.0% (186)

60-69 6.8% (432) 6.3% (325) 9.2% (107)

70-79 2.5% (155) 2.4% (123) 2.8% (32)

80+ 1.1% (67) 0.8% (40) 2.3% (27)

Preferred language

English	93.1% (5878)	93.4% (4810)	91.9% (1068)	0.071
Spanish	0% (0)	0% (0)	0% (0)
Somali	3.9% (249)	3.9% (201)	4.1% (48)
Other Interpreter needed,	3% (186) 5.5% (345)	2.7% (140) 5.1% (264)	4.0% (46) 7.0% (81)	0.012
% yes

Marital Status Married/ partner 21.1% (1332)		21.5% (1106)	19.5% (226)
Single	78.4% (4950)	78.0% (4019)	80.1% (931)
missing	0.5% (31)	0.5% (26)	0.4% (5)

0.290

1. Murrell KL, Offerman SR, Kauffman MB. Applying lean: implementation of a rapid triage and treatment system. West J Emerg Med. 2011;12:184-91.
2. Breen LM, Trepp Jr R, Gavin N. Lean process improvement in the emergency depart- ment. Emerg Med Clin North Am. 2020;38:633-46.
3. Vashi AA, Sheikhi FH, Nashton LA, Ellman J, Rajagopal P, Asch SM. Applying lean principles to reduce wait times in a VA emergency department. Mil Med. 2019; 184:e169-78.
4. Eller A. Rapid assessment and disposition: applying LEAN in the emergency depart- ment. J Healthc Qual. 2009;31:17-22.
5. Peng LS, Rasid MF, Salim WI. Using modified triage system to improve emergency department efficacy: a successful lean implementation. Int J Healthc Manag. 2021; 14:419-23.
6. Garrett JS, Berry C, Wong H, Qin H, Kline JA. The effect of vertical split-flow patient management on emergency department throughput and efficiency. Am J Emerg Med. 2018;36:1581-4.
7. Meislin HW, Coates SA, Cyr J, Valenzuela T. Fast track: urgent care within a teaching hospital emergency department: can it work? Ann Emerg Med. 1988;17:453-6.
8. Davey K, Olivieri P, Saul T, Atmar S, Rabrich JS, Egan DJ. Impact of patient race and primary language on ED triage in a system that relies on chief complaint and general

Insurance category				appearance. Am J Emerg Med. 2017;35:1013-5.
Private 16.7% (1055)	18.6% (957)	8.4% (98)	<0.001	[28] Saint paul profile (2014-2018). https://www.mncompass.org/profiles/city/saint-
Medicaid 54.8% (3461)	57.9% (2981)	41.3% (480)		paul; 2022.
Medicare 12.8% (808)	11.6% (595)	18.3% (213)		[29] Henrique DB, Rentes AF, Godinho Filho M, Esposto KF. A new value stream mapping
Other 15.7% (989)	12.0% (618)	31.9% (371)		approach for healthcare environments. Prod Plan Control. 2016;27:24-48. [30] Rosenbaum PR. A characterization of optimal designs for observational studies. J R I
Discharge Disposition				State Dent Soc. 1991;53:597-610.
Admit 13.3% (840)	13.1% (674)	14.3% (166)	0.512	[31] Hansen BB. Full matching in an observational study of coaching for the SAT. J Am
Transfer 0.7% (44)	0.7% (37)	0.6% (7)		Stat Assoc. 2004;99:609-18.
Discharge 86% (5429)	86.2% (4440)	85.1% (989)		[32] Stuart EA, Green KM. Using full matching to estimate causal effects in nonexperi-

* p-values are comparing values for encounters included versus those excluded after matching.

References

1. Mack K, Palfrey J. Capitalizing black and white: grammatical justice and equity. https://www.macfound.org/press/perspectives/capitalizing-black-and-white- grammatical-justice-and-equity; 2022.
2. Mahajan P, Basu T, Pai C-W, Singh H, Petersen N, Bellolio MF, et al. Factors associated with potentially missed diagnosis of appendicitis in the emergency department. JAMA Netw Open. 2020;3:e200612.
3. Heins JK, Heins A, Grammas M, Costello M, Huang K, Mishra S. Disparities in analge- sia and opioid prescribing practices for patients with musculoskeletal pain in the emergency department. J Emerg Nurs. 2006;32:219-24.
4. Jarman MP, Pollack Porter K, Curriero FC, Castillo RC. Factors mediating demographic determinants of injury mortality. Ann Epidemiol. 2019;34:58-64. e2.
5. Williams DR, Sternthal M. Understanding racial-Ethnic disparities in health: socio- logical contributions. J Health Soc Behav. 2010;51(Suppl):S15-27.
6. Dubay LC, Lebrun LA. Health, behavior, and Health care disparities: disentangling the effects of income and race in the United States. Int J Health Serv. 2012;42:607-25.
7. Kirby JB, Taliaferro G, Zuvekas SH. Explaining racial and ethnic disparities in health care. Med Care. 2006;44:I64-72.
8. Wuerz RC, Milne LW, Eitel DR, Travers D, Gilboy N. Reliability and validity of a new

five-level triage instrument. Acad Emerg Med. 2000;7:236-42.

1. Fernandes CMB, Tanabe P, Gilboy N, Johnson LA, McNair RS, Rosenau AM, et al. Five- level triage: a report from the ACEP/ENA five-level triage task force. J Emerg Nurs. 2005;31:39-50.
2. FitzGerald G, Jelinek GA, Scott D, Gerdtz MF. Emergency department triage revisited. Emerg Med J. 2010;27:86-92.
3. Nawar EW, Niska RW, Xu J. National hospital ambulatory medical care survey: 2005 emergency department summary. Adv Data. 2007:1-32.
4. Sonnenfeld N, Pitts SR, Schappert SM, Decker SL. Emergency department volume and racial and ethnic differences in Waiting times in the United States. Med Care. 2012;50:335-41.
5. Pines JM, Russell Localio A, Hollander JE. Racial disparities in emergency department length of stay for admitted patients in the United States. Acad Emerg Med. 2009;16: 403-10.
6. Okunseri C, Okunseri E, Chilmaza CA, Harunani S, Xiang Q, Szabo A. Racial and ethnic variations in waiting times for emergency department visits related to nontraumatic dental conditions in the United States. J Am Dent Assoc. 2013;144:828-36.
7. Alrwisan A, Eworuke E. Are discrepancies in waiting time for chest pain at emer- gency departments between African Americans and whites improving over time? J Emerg Med. 2016;50:349-55.
8. Lopez L, Wilper AP, Cervantes MC, Betancourt JR, Green AR. Racial and sex differ- ences in emergency department triage assessment and test ordering for chest pain, 1997-2006. Acad Emerg Med. 2010;17:801-8.

mental studies: examining the relationship between adolescent marijuana use and

adult outcomes. Dev Psychol. 2008;44:395-406.

1. Hirano K, Imbens GW, Ridder G. Efficient estimation of average treatment effects using the estimated propensity score. Econometrica. 2003;71:1161-89.
2. StataCorp. Stata statistical software: Release 17. College Station, TX: StataCorp LLC.; 2021..
3. Gilboy N, Tanabe T, Travers D, Rosenau AM. Emergency Severity Index : A triage tool for emergency department. Rockville, MD: Agency for Healthcare Research and Quality; 2011..
4. Elshove-Bolk J, Mencl F, van Rijswijck BTF, Simons MP, van Vugt AB. Validation of the Emergency Severity Index in self-referred patients in a European emergency department. Emerg Med J. 2007;24:170-4.
5. Kaji AH, Schriger D, Green S. Looking through the retrospectoscope: reducing bias in emergency medicine chart review studies. Ann Emerg Med. 2014;64:292-8.
6. Douglas MD, Dawes DE, Holden KB, Mack D. Missed policy opportunities to advance health equity by recording demographic data in electronic health records. Am J Pub- lic Health. 2015;105(Suppl. 3):S380-8.