Abstract
Rationale
Patients who identify as from racial or ethnic minority groups who have sepsis or acute respiratory failure (ARF) experience worse outcomes relative to nonminority patients, but processes of care accounting for disparities are not well-characterized.
Objectives
Determine whether reductions in intensive care unit (ICU) admission during hospital-wide capacity strain occur preferentially among patients who identify with racial or ethnic minority groups.
Methods
This retrospective cohort among 27 hospitals across the Philadelphia metropolitan area and Northern California between 2013 and 2018 included adult patients with sepsis and/or ARF who did not require life support at the time of hospital admission. An updated model of hospital-wide capacity strain was developed that permitted determination of relationships between patient race, ethnicity, ICU admission, and strain.
Results
After adjustment for demographics, disease severity, and study hospital, patients who identified as Asian or Pacific Islander had the highest adjusted ICU admission odds relative to patients who identified as White in both the sepsis and ARF populations (odds ratio, 1.09; P = 0.006 and 1.26; P < 0.001). ICU admission was also elevated for patients with ARF who identified as Hispanic (odds ratio, 1.11; P = 0.020). Capacity strain did not modify differences in ICU admission for patients who identified with a minority group in either disease population (all interactions, P > 0.05).
Conclusions
Systematic differences in ICU admission patterns were observed for patients that identified as Asian, Pacific Islander, and Hispanic. However, ICU admission was not restricted from these groups, and capacity strain did not preferentially reduce ICU admission from patients identifying with minority groups. Further characterization of provider decision-making can help contextualize these findings as the result of disparate decision-making or a mechanism of equitable care.
Keywords: healthcare disparities, intensive care units, sepsis, respiratory insufficiency, delivery of health care
Patients who identify as racial minorities disproportionately die from sepsis and acute respiratory failure (ARF), two of the most common syndromes patients present with at acute care hospitals (1–7). Prior efforts to understand contributors to these disparities have focused on specific characteristic differences between hospitals, including teaching hospital status, size, rural location, and safety net status (2). Although these studies enrich our understanding of disparities in outcomes, hospital characteristics are difficult to change and, rather than being causal contributors, may be markers of underlying hospital cultures, resource availability between hospitals, or quality of care delivery (8).
By contrast, processes of care that vary among racial and ethnic groups within hospitals may reveal more readily modifiable mechanisms underlying disparities. Our prior work has shown that capacity strain, defined as limits in a care unit’s ability to deliver high-quality care because of resource limitation (9–11), is an important systems-level phenomenon with nuanced associations with processes and outcomes of care for patients with sepsis and acute respiratory failure (12–20). Because capacity strain is a dynamic phenomenon that is quantifiable, perceptible to clinicians, and associated with differences in healthcare decision-making (12, 21), it may contribute to disparities by acting as a cognitive stressor that alters decision-making (22, 23), thus potentially predisposing to racially biased care (24–26). In addition, provider admission decisions relate to healthcare delivery, quality, and outcome (27, 28); therefore, different patterns of admission decisions between patients from minority and nonminority groups are important to characterize.
For certain high-acuity conditions, the decision to admit to intensive care units (ICUs) is discretionary, offering the opportunity to discern bias in care. In this study, we explore race- and ethnicity-based differences in the likelihood of ICU versus ward admission for patients with sepsis and ARF who did not receive life support in the emergency department. Building from prior work showing that such patients are more commonly admitted to wards as hospital capacity strain increases (17, 18), we hypothesized that strain-induced decisions to admit patients to ICUs would differ on the basis of patient identification with minority compared with nonminority racial or ethnic groups. To test these hypotheses, we refined a previously developed hospital-wide capacity strain index (17), enabling us to determine relationships between patient race, ethnicity, probability of ICU admission, and how these relationships are modified relative to increasing strain. Whereas differences in ICU admission rates across patient groups may not represent disparities because of unmeasured confounding by patient preference, severity of illness, or confounding by indication, we follow a conceptual model in which disparate outcomes of care may result from the differential receipt of care processes (e.g., ICU admission) as a consequence of strain (9, 20).
Methods
Data Source
We conducted this study using granular electronic health record data collected from 9.2 million patients admitted to 27 acute care hospitals across Penn Medicine and KPNC (Kaiser Permanente Northern California) from 2013 to 2018. This dataset has been described in prior studies (17–20) and contains detailed clinical care information from racially diverse populations from the city of Philadelphia, greater Philadelphia metropolitan area, and Northern California. The study protocol was approved with a waiver of informed consent by the institutional review boards of KPNC and the University of Pennsylvania.
Study Population
Patients included in this retrospective cohort were 18 years or older, admitted to a study hospital through the emergency department to a medical or medical–surgical ward or ICU, and met clinical criteria for sepsis or ARF. To focus on high-acuity patients whose admission decision may be discretionary (17, 20), we excluded patients with a LAPS2 (Laboratory-based Acute Physiology Score) less than 100 (who rarely go to ICU) (29–31) and those who received mechanical ventilation or vasopressor support in the emergency department (who almost always go to ICU). Patients were also excluded if on hospice or on orders for “comfort measures only” at the time of emergency department presentation.
Patients with sepsis were identified on the basis of modified Sepsis-3 criteria (17, 32–35) by having a suspected diagnosis of infection and evidence of organ dysfunction at the time of admission decision. Suspected infection was satisfied by having at least one antimicrobial medication order and at least one microbiologic culture order. For patients with suspected ARF, patients had evidence of hypoxemia while receiving supplemental oxygen, received supplemental oxygen for at least two hours, or had evidence of clinically significant hypercarbic and/or hypoxic respiratory failure (17, 19, 34, 35).
Study Variables
Demographic variables included patient race, ethnicity, age, gender, and insurance status. Patient race and ethnicity were self-identified, except when patients or surrogates were unable to self-identify. In these situations, race and ethnicity were determined by hospital staff from any available source as part of usual care. We operationalized race and ethnicity for this study as non-Hispanic White, non-Hispanic Black, non-Hispanic Asian and Pacific Islander (AAPI), non-Hispanic multiracial, and Hispanic (36). Gender was recorded binarily as woman or man, on the basis of reporting at study hospitals. Insurance was categorized as either private insurance (which also included patients who had Medicare with private secondary insurance), Medicare (which included patients with Medicare only or with dual Medicare and Medicaid eligibility), Medicaid, or unknown. To assess acute and chronic disease severity, we used two validated scores. We used LAPS2, a predictor of inpatient mortality on the basis of physiological variables and laboratory tests (30), because of its superior model discrimination and calibration for patients of color compared with other commonly used acute disease severity scores (34). We also used the COPS2 (Comorbidity Point Score 2), which predicts mortality using 12 months of a patient’s comorbid disease history (29–31). Both scores were calculated at the time of each patient’s hospital admission.
Strain index creation
Strain metrics included in a multivariable composite strain index in our prior work were retained for this study (17). Each metric represents a unique measure of strain in one of three domains: occupancy, census disease acuity, and census turnover. In total, we calculated 22 strain metrics that quantified unique components of these domains within different hospital areas, as experienced by each patient during the time of emergency department decision-making. Strain metric coefficients were calculated within each study hospital and standardized relative to unit-, hospital-, and year-specific averages.
Strain indices were created on the basis of our methods described in prior studies (17, 19) and modified to not incorporate patient race or ethnicity during strain index creation. Because strain may exert effects on clinical care processes and outcomes differently for patients of different racial and ethnic identities, controlling for race at the stage of strain index modeling could mask true differences. The new strain indices were built from a logistic regression model that used emergency department disposition decisions (e.g., ICU admission relative to ward) as the dependent variable. Covariates included patient age, gender, insurance type, COPS2 and LAPS2 at presentation, and each of the 22 strain metrics, each assigned on the basis of time of presentation to the emergency department. To account for center-level effects, strain indices were stratified by admitting hospital. In addition, models were stratified by presenting disease (sepsis or ARF). β coefficients corresponding to each strain metric were extracted from the regression model, and the sum of the products of the coefficients and the standardized strain measure created an individual patient’s strain index.
Statistical Analyses
Continuous variables were analyzed using ANOVA if normally distributed and Kruskal-Wallis if nonnormally distributed. Categorical variables were compared using Pearson’s chi-squared test. All analyses were performed using Stata 15.1; significance was considered at P < 0.05.
To test for the association of race with ICU admission, we constructed two logistic regression models, one for each disease cohort, with emergency department disposition (ICU vs. ward) as the dependent variable. Variables for both models included the created strain index together with patient-level factors of race, ethnicity, age, gender, insurance, COPS2, LAPS2, and study hospital as a fixed effect. Study hospital was included as a fixed effect to account for the possibility that hospital-specific ICU admission patterns may confound relationships between patient characteristics and ICU admission (20).
To test our primary hypothesis that strain would differentially affect ICU admission across racial and ethnic groups, we examined statistical interactions between patient race, ethnicity, and the created SIs in the sepsis and ARF models. We used the Wald test for the significance of the interaction term and quantified adjusted ICU admission probability at several percentiles of strain intensity.
Sensitivity analyses adjusting for code status
Because do-not-resuscitate (DNR) orders that are present in the emergency department may influence admission decisions, and because certain minority groups may be less likely to have DNR orders (12), we considered code status as a possible confounder in sensitivity analyses. DNR status was defined as having a DNR order that was placed in the emergency department or up to 3 hours after hospital admission. Patients who did not have a DNR order placed in the emergency department were designated full code. Code status was binarily coded as either full code or DNR. In the sensitivity analysis, code status was added as a covariate to final models for ICU admission that included race, ethnicity, age, gender, insurance, COPS2, LAPS2, study hospital, and strain index as additional covariates. As described earlier, patients receiving only comfort measures at emergency department presentation were excluded.
Results
Study Population Clinical Characteristics
We identified 102,362 eligible patients, of which 84,685 (82.7%) met criteria for sepsis and 42,008 (41.0%) met criteria for ARF. Criteria for both conditions were met by 24,331 (23.8%) patients. In each disease population, racial and ethnic groups significantly differed in baseline characteristics (P < 0.001 for all comparisons) (Table 1). Patients who identified as Black tended to be youngest, least likely to use private insurance, and most likely full code compared with all other racial and ethnic groups.
Table 1.
Characteristics of the study population stratified by race and ethnicity
| Sepsis Population |
ARF Population |
|||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| White | Black | Asian and PI | Hispanic | Multiracial | Total | White | Black | Asian and PI | Hispanic | Multiracial | Total | |
| n (%) | 51,749 (61.1) | 10,586 (12.5) | 8,957 (10.6) | 8,444 (10.0) | 4,949 (5.8) | 84,685 | 25,840 (61.5) | 6,425 (15.3) | 3,982 (9.5) | 3,422 (8.1) | 2,339 (5.6) | 42,008 |
| Both sepsis and ARF, n (%) | 15,450 (29.9) | 3,017 (28.5) | 2,485 (27.7) | 1,959 (23.2) | 1,420 (28.7) | 24,331 (28.7) | 15,450 (59.8) | 3,017 (47.0) | 2,485 (62.4) | 1,959 (57.2) | 1,420 (60.7) | 24,331 (57.9) |
| Age, median (IQR) | 77 (67–86) | 69 (58–80) | 77 (66–85) | 73 (59–83) | 79 (69–86) | 76 (65–85) | 76 (67–85) | 67 (57–77) | 76 (66–85) | 72 (60–82) | 77 (68–85) | 75 (64–84) |
| Women, n (%) | 24,835 (48.0) | 5,332 (50.4) | 3,933 (43.9) | 3,838 (45.5) | 2,537 (51.3) | 40,475 (47.8) | 13,288 (51.4) | 3,471 (54.0) | 1,724 (43.3) | 1,623 (47.4) | 1,304 (55.8) | 21,410 (51.0) |
| Insurance | ||||||||||||
| Private | 41,670 (80.5) | 6,014 (56.8) | 7,626 (85.1) | 6,780 (80.3) | 4,568 (92.3) | 66,658 (78.7) | 19,658 (76.1) | 3,110 (48.4) | 3,240 (81.4) | 2,601 (76.0) | 2,128 (91.0) | 30,737 (73.2) |
| Medicare | 6,050 (11.7) | 2,176 (20.6) | 148 (1.7) | 464 (5.5) | 113 (2.3) | 8,951 (10.6) | 3,657 (14.2) | 1,637 (25.5) | 68 (1.7) | 220 (6.4) | 54 (2.3) | 5,636 (13.4) |
| Medicaid | 985 (1.9) | 1,072 (10.1) | 201 (2.2) | 345 (4.1) | 119 (2.4) | 2,722 (3.2) | 566 (2.2) | 711 (11.1) | 102 (2.6) | 182 (5.3) | 75 (3.2) | 1,636 (3.9) |
| Unknown | 3,044 (5.9) | 1,324 (12.5) | 982 (11.0) | 855 (10.1) | 149 (3.0) | 6,354 (7.5) | 1,959 (7.6) | 967 (15.1) | 572 (14.4) | 419 (12.2) | 82 (3.5) | 3,999 (9.5) |
| LAPS2, median (IQR) | 124 (111–143) | 125 (111–145) | 125 (112–145) | 124 (111–142) | 126 (111–145) | 124 (111–143) | 127 (112–146) | 123 (110–143) | 129 (113–151) | 125 (112–146) | 129 (113–149) | 126 (112–146) |
| COPS2, median (IQR) | 108 (67–148) | 116 (64–165) | 103 (57–144) | 100 (56–142) | 117 (78–153) | 108 (65–149) | 117 (79–152) | 117 (68–159) | 111 (67–149) | 110 (68–147) | 123 (93–157) | 116 (78–153) |
| Code status | ||||||||||||
| DNR | 19,205 (37.1) | 1,785 (16.9) | 2,560 (28.6) | 2,069 (24.5) | 1,721 (34.8) | 2,7340 (32.3) | 9,443 (36.5) | 894 (13.9) | 1,132 (28.4) | 863 (25.2) | 758 (32.4) | 13,090 (31.2) |
| Full code | 32,544 (62.9) | 8,801 (83.1) | 6,397 (71.4) | 6,375 (75.5) | 3,228 (65.2) | 57,345 (67.7) | 16,397 (63.5) | 5,531 (86.1) | 2,850 (71.6) | 2,559 (74.8) | 1,581 (67.6) | 28,918 (68.8) |
| ICU admission, n (%) | 9,261 (17.9) | 2,830 (26.7) | 1,821 (20.3) | 1,713 (20.3) | 910 (18.4) | 16,535 (19.5) | 6,519 (25.2) | 2,346 (36.5) | 1,291 (32.4) | 1,054 (30.8) | 642 (27.4) | 11,852 (28.2) |
Definition of abbreviations: ARF = acute respiratory failure; COPS2 = Comorbidity Point Score 2; DNR = do not resuscitate; ICU = intensive care unit; IQR = interquartile range; LAPS2 = Laboratory-based Acute Physiology Score 2; PI = Pacific Islander.
Descriptive statistics of the study populations. Continuous variables were analyzed by analysis of variance if normally distributed and Kruskal-Wallis if nonnormally distributed. Categorical variables were compared by Pearson’s chi-squared test. P < 0.001 for all comparisons.
ICU Admission Rates
The crude proportion of patients with sepsis admitted to ICUs was highest among patients who identified as Black and lowest among patients who identified as White (Black patients, 26.7%; AAPI patients, 20.3%; Hispanic patients, 20.3%; multiracial patients, 19.5%; White patients, 17.9%; P < 0.001) (Table 1). In fully adjusted models, patients who identified as AAPI had the highest independent odds of ICU admission (odds ratio [OR], 1.09; 95% confidence interval [CI], 1.03–1.16; P = 0.006) (Table 2). ICU admission was similar and nonsignificant between patients who identified as White and those who identified as Black (OR, 1.04; 95% CI, 0.97–1.10), Hispanic (OR, 1.00; 95% CI, 0.94–1.07), and multiracial (OR, 1.02; 95% CI, 0.94–1.11).
Table 2.
Unadjusted and adjusted odds of intensive care unit admission
| Sepsis Population |
ARF Population |
||||||
|---|---|---|---|---|---|---|---|
| Model | Race or Ethnicity | OR | 95% CI | P Value | OR | 95% CI | P Value |
| Model 1 (unadjusted) | White (ref) | 1 | – | – | 1 | – | – |
| Black | 1.67 | (1.59–1.76) | <0.001 | 1.70 | (1.61–1.81) | <0.001 | |
| Asian and PI | 1.17 | (1.11–1.24) | <0.001 | 1.42 | (1.32–1.53) | <0.001 | |
| Hispanic | 1.17 | (1.10–1.24) | <0.001 | 1.32 | (1.22–1.43) | <0.001 | |
| Multiracial | 1.03 | (0.96–1.11) | 0.389 | 1.12 | (1.02–1.23) | 0.018 | |
| Model 2 (model 1 + demographic variables) | White (ref) | 1 | – | – | 1 | – | – |
| Black | 1.34 | (1.27–1.41) | <0.001 | 1.45 | (1.37–1.55) | <0.001 | |
| Asian and PI | 1.10 | (1.04–1.17) | 0.001 | 1.31 | (1.22–1.41) | <0.001 | |
| Hispanic | 0.98 | (0.93–1.04) | 0.594 | 1.13 | (1.04–1.22) | 0.003 | |
| Multiracial | 1.07 | (0.99–1.16) | 0.082 | 1.11 | (1.01–1.22) | 0.033 | |
| Model 3 (model 2 + acuity scores) | White (ref) | 1 | – | – | 1 | – | – |
| Black | 1.30 | (1.23–1.37) | <0.001 | 1.49 | (1.40–1.58) | <0.001 | |
| Asian and PI | 1.08 | (1.02–1.15) | 0.008 | 1.28 | (1.18–1.38) | <0.001 | |
| Hispanic | 0.99 | (0.93–1.05) | 0.781 | 1.13 | (1.04–1.23) | 0.003 | |
| Multiracial | 1.05 | (0.97–1.13) | 0.27 | 1.08 | (0.98–1.20) | 0.112 | |
| Model 4 (model 3 + strain index) | White (ref) | 1 | – | – | 1 | – | – |
| Black | 1.28 | (1.21–1.35) | <0.001 | 1.40 | (1.31–1.49) | <0.001 | |
| Asian and PI | 1.06 | (0.99–1.12) | 0.08 | 1.21 | (1.12–1.31) | <0.001 | |
| Hispanic | 0.97 | (0.92–1.04) | 0.424 | 1.10 | (1.01–1.20) | 0.022 | |
| Multiracial | 1.03 | (0.95–1.12) | 0.464 | 1.06 | (0.95–1.17) | 0.292 | |
| Model 5 (model 4 + facility fixed effects) | White (ref) | 1 | – | – | 1 | – | – |
| Black | 1.04 | (0.97–1.10) | 0.260 | 1.07 | (0.99–1.16) | 0.074 | |
| Asian and PI | 1.09 | (1.03–1.16) | 0.006 | 1.26 | (1.16–1.37) | <0.001 | |
| Hispanic | 1.00 | (0.94–1.07) | 0.933 | 1.11 | (1.02–1.21) | 0.02 | |
| Multiracial | 1.02 | (0.94–1.11) | 0.584 | 1.04 | (0.94–1.16) | 0.424 | |
Definition of abbreviations: ARF = acute respiratory failure; CI = confidence interval; OR = odds ratio; PI = Pacific Islander; ref = reference group.
Presented are the odds of intensive care unit admission for indicated minority groups relative to White patients. Models were adjusted as indicated using logistic regression. Demographic variables refer to patient age, gender, and insurance status. Acuity scores refer to Laboratory Acute Physiology Score and Comorbidity Point Score.
Patients with ARF were more commonly admitted to the ICU overall, with similar patterns of admission between patients of different racial and ethnic groups. Specifically, crude overall proportions of ICU admission were highest for patients with ARF who identified as Black (36.5%), followed by patients who identified as AAPI (32.4%), Hispanic (30.8%), multiracial (27.4%), and White (25.2%) (all P < 0.001) (Table 1). In fully adjusted models, patients who identified as AAPI had the highest odds of ICU admission (OR, 1.26; 95% CI, 1.16–1.37; P < 0.001), followed by patients who identified as Hispanic (OR, 1.11; 95% CI, 1.02–1.21; P = 0.02) (Table 2).
Race and Ethnicity-specific ICU Admission and Capacity Strain Intensity
Capacity strain was inversely associated with ICU admission across all races, ethnicities, and disease populations (Figure 1). In both the sepsis and ARF populations, strain-induced reductions in ICU admission probability were proportionally similar among patients from minority and nonminority groups, with no evidence of effect modification (P > 0.05 for all interaction term ORs) (Table 3 and Figure 2).
Figure 1.
Race- and ethnicity-specific capacity strain associations with intensive care unit (ICU) admission. Probabilities represent the independent likelihood of ICU admission within the indicated racial and ethnic group on the basis of fully adjusted models of ICU admission. Models were built by logistic regression and stratified by disease population at each decile of hospital-wide capacity strain. (A) Race- and ethnicity-specific associations between capacity strain and ICU admission in patients with sepsis. (B) Race- and ethnicity-specific associations between capacity strain and ICU admission in patients with ARF. ARF = acute respiratory failure; PI = Pacific Islander; SI = strain index.
Table 3.
Interaction effects between patient race, ethnicity, and hospital capacity strain
| Race | Disease | Race or Ethnicity OR | 95% CI | P Value | Strain OR | 95% CI | P Value | iOR | 95% CI | P Value |
|---|---|---|---|---|---|---|---|---|---|---|
| Black | Sepsis | 1.03 | (0.97–1.10) | 0.376 | 0.37 | (0.35–0.39) | <0.001 | 0.97 | (0.84–1.12) | 0.709 |
| ARF | 1.06 | (0.98–1.15) | 0.141 | 0.38 | (0.35–0.40) | <0.001 | 0.94 | (0.82–1.07) | 0.358 | |
| Asian and PI | Sepsis | 1.09 | (1.02–1.16) | 0.015 | 0.37 | (0.35–0.40) | <0.001 | 1.01 | (0.85–1.19) | 0.944 |
| ARF | 1.30 | (1.19–1.42) | <0.001 | 0.37 | (0.35–0.40) | <0.001 | 1.05 | (0.88–1.26) | 0.563 | |
| Hispanic | Sepsis | 0.97 | (0.91–1.03) | 0.339 | 0.37 | (0.35–0.39) | <0.001 | 0.96 | (0.82–1.13) | 0.634 |
| ARF | 1.09 | (0.99–1.19) | 0.078 | 0.37 | (0.35–0.40) | <0.001 | 1.04 | (0.88–1.23) | 0.618 | |
| Multiracial | Sepsis | 1.03 | (0.95–1.12) | 0.484 | 0.37 | (0.35–0.39) | <0.001 | 1.15 | (0.93–1.43) | 0.200 |
| ARF | 1.04 | (0.93–1.17) | 0.447 | 0.37 | (0.35–0.40) | <0.001 | 0.95 | (0.75–1.19) | 0.631 |
Definition of abbreviations: ARF = acute respiratory failure; CI = confidence interval; iOR = interaction term odds ratio; OR = odds ratio; PI = Pacific Islander.
Presented are adjusted odds ratios of intensive care unit admission for patient race, ethnicity, the capacity strain index, and the interaction between patient race, ethnicity, and capacity strain. The race- and ethnicity-specific OR uses the odds of intensive care unit admission of White patients as the reference group. Models were built by logistic regression and fully adjusted for age, gender, insurance status, Comorbid Point Score, Laboaratory Acute Physiology Score, and study hospital.
Figure 2.
Interaction effects between patient race, ethnicity, and capacity strain. Presented are fully adjusted probabilities of intensive care unit (ICU) admission for patients from the indicated racial and ethnic groups. The strain index is presented in deciles that correspond to increasing intensities of strain. Models were built by logistic regression, stratified by disease population at each decile of hospital-wide capacity strain, and adjusted for age, gender, insurance status, Comorbid Point Score, Laboratory Acute Physiology Score, and study hospital. (A) Race- and ethnicity-specific associations between capacity strain and ICU admission in patients with sepsis. (B) Race- and ethnicity-specific associations between capacity strain and ICU admission in patients with acute respiratory failure. SI = strain index.
ICU Admission and Code Status
Patients who identified as from minority groups were less likely to have DNR code status at the time of hospital admission (Table 1). Adjusting for code status in models for ICU admission tended to reduce the odds of ICU admission for patients who identified from minority groups relative to patients who identified as White; however, the odds of ICU admission remained significant for patients with ARF who identified as AAPI (OR, 1.19; 95% CI, 1.10–1.30; P < 0.001) (Figure E1 in the online supplement).
Discussion
In this study, we demonstrate that patients with ARF and sepsis who are from certain minority racial and ethnic groups that might plausibly be managed in either the general ward or the ICU were more likely to receive initial ICU care than patients from nonminority groups. In fully adjusted models, patients who identified as AAPI were 9% and 26% more likely than patients who identified as White to be admitted to the ICU rather than the ward when presenting for sepsis and ARF, respectively. Patients who identified as Hispanic were also 11% more likely than patients who identified as White to be admitted to the ICU when presenting for ARF. For patients that identified as Black or multiracial, ICU admission was similar to White patients and well explained by between-hospital practice patterns. We found no evidence that capacity strain modified ICU admission decisions for patients from minority compared with nonminority groups.
Our study provides evidence that with respect to ICU admission decisions, systematic differences exist between patients on the basis of the racial and ethnic group from which they identify. However, we found that patients from certain minority groups were more likely, rather than less likely, to receive ICU admission. Though some might believe our data implies equity in the delivery of critical care resources, we believe that our findings do not provide sufficient evidence of equitable ICU resource allocation. Importantly, systematic biases, whether implicit or explicit, may be major determinants of ICU admission decisions (25, 26, 37, 38). Future studies should use qualitative methods to understand the content of providers’ decision-making for ICU resource delivery and clarify the extent to which equitable healthcare delivery is a goal in the provision of ICU care.
In addition, our choice to define the study population based in part on a disease severity score could have biased our results. In prior work, we have shown that available acuity measures are racially biased, such that patients who identify as Black die less commonly than predicted on the basis of available scores (34). Though the LAPS2 performs with less racial bias than alternatives such as the Sequential Organ Failure Assessment score, subtle bias could nonetheless be introduced as a result of the differential performance of the LAPS2 among patients who identify as White or Black. Therefore, our approach may have identified a population of Black patients who were less severely ill than other groups and who might have been more likely to receive general ward admission. Our study, therefore, highlights the need for improved disease severity scores that more effectively predict mortality regardless of patient racial and ethnic background.
Despite these potential limitations, we observe increased admission rates for patients who identify as AAPI or Hispanic. We hypothesize that language differences between patients and providers may be an important mechanism for these findings. Patients who have limited English proficiency or a preferred language other than English have been shown to receive less desirable communication in ICUs, disparate care, and worse outcomes compared with patients who primarily speak English (39–41). In the context of these prior studies, we hypothesize that differences in patient–provider spoken language might represent a barrier to developing a complete symptom inventory, presenting history, or characterizing goals of care. This might nudge admitting providers toward recommending ICU admission, in which the perceived abundance of resources may serve as a “safety net” for clinical deterioration. Although this decision-making framework may seem on the surface to preserve equity, it may not be patient-centered (42). Future studies designed to investigate between-group differences in patient preferences for ICU care will be necessary to understand whether our findings represent equitable ICU admission decision-making or are the result of inequitable care delivery.
Lastly, it is not clear that increased ICU care always results in improved outcomes. Our group has recently shown that ICU admission may increase in-hospital mortality and length of stay for patients with sepsis and decrease mortality and length of stay for patients with ARF (20). Therefore, preferential admission of minority patients to ICUs could be an important mechanism that drives disparate outcomes for patients with sepsis or maintains equity among patients with ARF. Future work might adapt causal effect techniques to determine whether patient-centered triage acts as a mechanism that influences outcomes.
Strengths and Limitations
A strength of our study is that it leverages a racially diverse, multiple-center sample and revises a previously developed model of capacity strain to investigate determinants of disparities in critical care. However, our study has some limitations. First, we emphasize that our analysis does not include clinician or patient characteristics (besides patient race and ethnicity) that could relate to triage decision-making and elicitation of patient goals of care. Among them, spoken language, preferred language, and concordance between patients or providers are unmeasured in our population. Because these and other relevant patient-level factors may impact our results, additional studies are needed to understand how language differences may impact ICU admission. Second, we were unable to determine situations in which patient race or ethnicity was indicated by individuals other than patients, surrogates, or family members. We expect the proportion of patients with demographics determined by healthcare providers to have been small, but we were unable to quantify this number in our dataset. Third, insurance is an incomplete assessment of individual socioeconomic status. Because patient experiences of poverty have been shown to associate with clinical outcomes for sepsis (43), additional studies are needed to consider how capacity strain influences care delivery for patients with poverty. Fourth, we only measure DNR status at the time of emergency department presentation. It is possible that patients who present with preexisting advanced care plans may have received even less ICU admission than those who had code status decisions documented at the time of emergency department presentation. Additional studies with a more comprehensive accounting of advanced care planning are necessary. Fifth, certain features of our population, such as the high degree of racial diversity, the relatively high proportion of patients with DNR orders, and the high prevalence of private insurance, may not be similar to other institutions and may limit generalizability to other health systems. Lastly, the possibility of unmeasured confounding may limit our ability to comment on causal associations between patient race, ethnicity, and ICU admission.
Conclusions
By quantifying hospital-wide capacity strain using a refined strain index, our study suggests that ICU admission was not restricted from patients who present for sepsis or ARF and who identify as from racial and ethnic minority groups. Patients who identified as AAPI or Hispanic were independently more likely to be selected for critical care management when presenting for ARF, whereas patients who identified as AAPI alone were most likely to be initially admitted to ICUs when presenting for sepsis. We found no evidence that strain modified decisions to admit patients to ICUs. Future studies are needed to evaluate the equity of outcomes and processes related to intensive care delivery in light of these complex relationships.
Footnotes
Supported by the American Thoracic Society Fellowship in Health Equity (C.F.C.); National Heart, Lung, and Blood Institute (NHLBI) (F32 HL160166-01 and T32 HL098054 [C.F.C.]; R01HL136719 [G.L.A., M.C., V.X.L., and S.D.H.]; and K23HL161353 [G.L.A.]).
Author Contributions: All authors were responsible for the conception, study design, and data acquisition; data analysis and interpretation; manuscript drafting; and critical revisions. All authors attest to the accuracy and integrity of the analyses performed and presented herein.
This article has an online supplement, which is accessible from this issue’s table of contents at www.atsjournals.org.
Author disclosures are available with the text of this article at www.atsjournals.org.
References
- 1. Martin GS, Mannino DM, Eaton S, Moss M. The epidemiology of sepsis in the United States from 1979 through 2000. N Engl J Med . 2003;348:1546–1554. doi: 10.1056/NEJMoa022139. [DOI] [PubMed] [Google Scholar]
- 2. Jones JM, Fingar KR, Miller MA, Coffey R, Barrett M, Flottemesch T, et al. Racial disparities in sepsis-related in-hospital mortality: using a broad case capture method and multivariate controls for clinical and hospital variables, 2004-2013. Crit Care Med . 2017;45:e1209–e1217. doi: 10.1097/CCM.0000000000002699. [DOI] [PubMed] [Google Scholar]
- 3. Barnato AE, Alexander SL, Linde-Zwirble WT, Angus DC. Racial variation in the incidence, care, and outcomes of severe sepsis: analysis of population, patient, and hospital characteristics. Am J Respir Crit Care Med . 2008;177:279–284. doi: 10.1164/rccm.200703-480OC. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Taylor SP, Karvetski CH, Templin MA, Taylor BT. Hospital differences drive antibiotic delays for Black patients compared with White patients with suspected septic shock. Crit Care Med . 2018;46:e126–e131. doi: 10.1097/CCM.0000000000002829. [DOI] [PubMed] [Google Scholar]
- 5. Bime C, Poongkunran C, Borgstrom M, Natt B, Desai H, Parthasarathy S, et al. Racial differences in mortality from severe acute respiratory failure in the United States, 2008-2012. Ann Am Thorac Soc . 2016;13:2184–2189. doi: 10.1513/AnnalsATS.201605-359OC. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Cochi SE, Kempker JA, Annangi S, Kramer MR, Martin GS. Mortality trends of acute respiratory distress syndrome in the United States from 1999 to 2013. Ann Am Thorac Soc . 2016;13:1742–1751. doi: 10.1513/AnnalsATS.201512-841OC. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Stefan MS, Shieh MS, Pekow PS, Rothberg MB, Steingrub JS, Lagu T, et al. Epidemiology and outcomes of acute respiratory failure in the United States, 2001 to 2009: a national survey. J Hosp Med . 2013;8:76–82. doi: 10.1002/jhm.2004. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Danziger J, Ángel Armengol de la Hoz M, Li W, Komorowski M, Deliberato RO, Rush BNM, et al. Temporal trends in critical care outcomes in U.S. minority-serving hospitals. Am J Respir Crit Care Med . 2020;201:681–687. doi: 10.1164/rccm.201903-0623OC. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Halpern SD. ICU capacity strain and the quality and allocation of critical care. Curr Opin Crit Care . 2011;17:648–657. doi: 10.1097/MCC.0b013e32834c7a53. [DOI] [PubMed] [Google Scholar]
- 10. Kerlin MP, Harhay MO, Vranas KC, Cooney E, Ratcliffe SJ, Halpern SD. Objective factors associated with physicians’ and nurses’ perceptions of intensive care unit capacity strain. Ann Am Thorac Soc . 2014;11:167–172. doi: 10.1513/AnnalsATS.201306-141OC. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Kohn R, Harhay MO, Bayes B, Mikkelsen ME, Ratcliffe SJ, Halpern SD, et al. Ward capacity strain: a novel predictor of 30-day hospital readmissions. J Gen Intern Med . 2018;33:1851–1853. doi: 10.1007/s11606-018-4564-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Hua M, Halpern SD, Gabler NB, Wunsch H. Effect of ICU strain on timing of limitations in life-sustaining therapy and on death. Intensive Care Med . 2016;42:987–994. doi: 10.1007/s00134-016-4240-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Weissman GE, Gabler NB, Brown SE, Halpern SD. Intensive care unit capacity strain and adherence to prophylaxis guidelines. J Crit Care . 2015;30:1303–1309. doi: 10.1016/j.jcrc.2015.08.015. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Gabler NB, Ratcliffe SJ, Wagner J, Asch DA, Rubenfeld GD, Angus DC, et al. Mortality among patients admitted to strained intensive care units. Am J Respir Crit Care Med . 2013;188:800–806. doi: 10.1164/rccm.201304-0622OC. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Wagner J, Gabler NB, Ratcliffe SJ, Brown SE, Strom BL, Halpern SD. Outcomes among patients discharged from busy intensive care units. Ann Intern Med . 2013;159:447–455. doi: 10.7326/0003-4819-159-7-201310010-00004. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Anesi GL, Gabler NB, Allorto NL, Cairns C, Weissman GE, Kohn R, et al. Intensive care unit capacity strain and outcomes of critical illness in a resource-limited setting: a 2-hospital study in South Africa. J Intensive Care Med . 2020;35:1104–1111. doi: 10.1177/0885066618815804. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Anesi GL, Chowdhury M, Small DS, Delgado MK, Kohn R, Bayes B, et al. Association of a novel index of hospital capacity strain with admission to intensive care units. Ann Am Thorac Soc . 2020;17:1440–1447. doi: 10.1513/AnnalsATS.202003-228OC. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Anesi GL, Liu VX, Gabler NB, Delgado MK, Kohn R, Weissman GE, et al. Associations of intensive care unit capacity strain with disposition and outcomes of patients with sepsis presenting to the emergency department. Ann Am Thorac Soc . 2018;15:1328–1335. doi: 10.1513/AnnalsATS.201804-241OC. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Anesi GL, Liu VX, Chowdhury M, Small DS, Wang W, Delgado MK, et al. American Thoracic Society International Conference. ; 2021. [Google Scholar]
- 20. Anesi GL, Liu VX, Chowdhury M, Small DS, Wang W, Delgado MK, et al. Association of intensive care unit admission and outcomes in sepsis and acute respiratory failure. Am J Respir Crit Care Med . 2021;205:520–528. doi: 10.1164/rccm.202106-1350OC. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Brown SE, Rey MM, Pardo D, Weinreb S, Ratcliffe SJ, Gabler NB, et al. The allocation of intensivists’ rounding time under conditions of intensive care unit capacity strain. Am J Respir Crit Care Med . 2014;190:831–834. doi: 10.1164/rccm.201406-1127LE. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Lieberman HR, Tharion WJ, Shukitt-Hale B, Speckman KL, Tulley R. Effects of caffeine, sleep loss, and stress on cognitive performance and mood during U.S. Navy SEAL training. Sea-air-land. Psychopharmacology (Berl) . 2002;164:250–261. doi: 10.1007/s00213-002-1217-9. [DOI] [PubMed] [Google Scholar]
- 23. Sandström A, Rhodin IN, Lundberg M, Olsson T, Nyberg L. Impaired cognitive performance in patients with chronic burnout syndrome. Biol Psychol . 2005;69:271–279. doi: 10.1016/j.biopsycho.2004.08.003. [DOI] [PubMed] [Google Scholar]
- 24. Santry HP, Wren SM. The role of unconscious bias in surgical safety and outcomes. Surg Clin North Am . 2012;92:137–151. doi: 10.1016/j.suc.2011.11.006. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Burgess DJ. Are providers more likely to contribute to healthcare disparities under high levels of cognitive load? How features of the healthcare setting may lead to biases in medical decision making. Med Decis Making . 2010;30:246–257. doi: 10.1177/0272989X09341751. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Burgess DJ, Fu SS, van Ryn M. Why do providers contribute to disparities and what can be done about it? J Gen Intern Med . 2004;19:1154–1159. doi: 10.1111/j.1525-1497.2004.30227.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Cichowitz C, Loevinsohn G, Klein EY, Colantuoni E, Galiatsatos P, Rennert J, et al. Racial and ethnic disparities in hospital observation in Maryland. Am J Emerg Med . 2021;46:532–538. doi: 10.1016/j.ajem.2020.11.010. [DOI] [PubMed] [Google Scholar]
- 28. Lo AX, Flood KL, Biese K, Platts-Mills TF, Donnelly JP, Carpenter CR. Factors associated with hospital admission for older adults receiving care in U.S. emergency departments. J Gerontol A Biol Sci Med Sci . 2017;72:1105–1109. doi: 10.1093/gerona/glw207. [DOI] [PubMed] [Google Scholar]
- 29. Liu V, Kipnis P, Gould MK, Escobar GJ. Length of stay predictions: improvements through the use of automated laboratory and comorbidity variables. Med Care . 2010;48:739–744. doi: 10.1097/MLR.0b013e3181e359f3. [DOI] [PubMed] [Google Scholar]
- 30. Escobar GJ, Gardner MN, Greene JD, Draper D, Kipnis P. Risk-adjusting hospital mortality using a comprehensive electronic record in an integrated health care delivery system. Med Care . 2013;51:446–453. doi: 10.1097/MLR.0b013e3182881c8e. [DOI] [PubMed] [Google Scholar]
- 31. Escobar GJ, Greene JD, Scheirer P, Gardner MN, Draper D, Kipnis P. Risk-adjusting hospital inpatient mortality using automated inpatient, outpatient, and laboratory databases. Med Care . 2008;46:232–239. doi: 10.1097/MLR.0b013e3181589bb6. [DOI] [PubMed] [Google Scholar]
- 32. Seymour CW, Liu VX, Iwashyna TJ, Brunkhorst FM, Rea TD, Scherag A, et al. Assessment of clinical criteria for sepsis: for the third international consensus definitions for sepsis and septic shock (sepsis-3) JAMA . 2016;315:762–774. doi: 10.1001/jama.2016.0288. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Shankar-Hari M, Phillips GS, Levy ML, Seymour CW, Liu VX, Deutschman CS, et al. Sepsis Definitions Task Force Developing a new definition and assessing new clinical criteria for septic shock: for the third international consensus definitions for sepsis and septic shock (sepsis-3) JAMA . 2016;315:775–787. doi: 10.1001/jama.2016.0289. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Ashana DC, Anesi GL, Liu VX, Escobar GJ, Chesley C, Eneanya ND, et al. Equitably allocating resources during crises: racial differences in mortality prediction models. Am J Respir Crit Care Med . 2021;204:178–186. doi: 10.1164/rccm.202012-4383OC. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Anesi GL, Chelluri J, Qasim ZA, Chowdhury M, Kohn R, Weissman GE, et al. Association of an emergency department-embedded critical care unit with hospital outcomes and intensive care unit use. Ann Am Thorac Soc . 2020;17:1599–1609. doi: 10.1513/AnnalsATS.201912-912OC. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Racial and Ethnic Categories and Definitions for NIH Diversity Programs and for Other Reporting Purposes. 2015. https://grants.nih.gov/grants/guide/notice-files/not-od-15-089.html
- 37. Burgess D, van Ryn M, Dovidio J, Saha S. Reducing racial bias among health care providers: lessons from social-cognitive psychology. J Gen Intern Med . 2007;22:882–887. doi: 10.1007/s11606-007-0160-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38. Burgess DJ, van Ryn M, Crowley-Matoka M, Malat J. Understanding the provider contribution to race/ethnicity disparities in pain treatment: insights from dual process models of stereotyping. Pain Med . 2006;7:119–134. doi: 10.1111/j.1526-4637.2006.00105.x. [DOI] [PubMed] [Google Scholar]
- 39. Zurca AD, Fisher KR, Flor RJ, Gonzalez-Marques CD, Wang J, Cheng YI, et al. Communication with limited English-proficient families in the PICU. Hosp Pediatr . 2017;7:9–15. doi: 10.1542/hpeds.2016-0071. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40. Cohen-Cline H, Li HF, Gill M, Rodriguez F, Hernandez-Boussard T, Wolberg H, et al. Major disparities in COVID-19 test positivity for patients with non-English preferred language even after accounting for race and social factors in the United States in 2020. BMC Public Health . 2021;21:2121. doi: 10.1186/s12889-021-12171-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41. Espinoza Suarez NR, Urtecho M, Nyquist CA, Jaramillo C, Yeow ME, Thorsteinsdottir B, et al. Consequences of suboptimal communication for patients with limited English proficiency in the intensive care unit and suggestions for a way forward: a qualitative study of healthcare team perceptions. J Crit Care . 2021;61:247–251. doi: 10.1016/j.jcrc.2020.10.012. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42. Escher M, Cullati S, Hudelson P, Nendaz M, Ricou B, Perneger T, et al. Admission to intensive care: a qualitative study of triage and its determinants. Health Serv Res . 2019;54:474–483. doi: 10.1111/1475-6773.13076. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43. Galiatsatos P, Brigham EP, Pietri J, Littleton K, Hwang S, Grant MC, et al. The effect of community socioeconomic status on sepsis-attributable mortality. J Crit Care . 2018;46:129–133. doi: 10.1016/j.jcrc.2018.01.008. [DOI] [PMC free article] [PubMed] [Google Scholar]