Racial and Ethnic Inequalities Among Survivors of Critical Illness in the MIMIC-IV Database

Abstract

Background:

Racial and ethnic disparities in healthcare outcomes are well-documented, but less is known about how these disparities manifest among survivors of critical illness. We examined whether Black and Hispanic ICU survivors experience different rates of 90-day and 1-year mortality and hospital readmission compared to White survivors, and whether these associations vary by age or Medicaid insurance status.

Methods:

We conducted a retrospective cohort study using the Medical Information Mart for Intensive Care-IV (MIMIC-IV) database. Primary outcomes were 90-day and 1-year mortality; secondary outcomes included 90-day and 1-year hospital readmissions. We used Cox proportional hazards, Accelerated Failure Time (AFT), and Fine-Gray competing risk models, adjusting for age, sex, and Medicaid status. Prespecified subgroup analyses were performed among patients aged ≥60 and those admitted to surgical ICUs.

Results:

Among 46 640 ICU survivors (mean age 63.2 years; 55.6% male; 11.8% Black; 4.6% Hispanic), Black patients had lower survival at 90 days (absolute difference (AD): −0.85% (95% CI: −1.47%, −0.23%) and 1 year (AD: −1.42% (−2.46%, −0.40%) compared to White patients. Hispanic patients had higher survival (90-day AD: 1.33% (0.39%, 2.31%); 1-year AD: 2.31% (0.67%, 4.03%). Differences were more pronounced among patients ≥60 years. Black (1-year SDHR: 1.29 (1.23, 1.34)) and Hispanic patients (SDHR: 1.22 (1.14, 1.30)) had higher readmission rates. Medicaid coverage was more common among Black (aOR: 2.26 (2.10, 2.43)) and Hispanic patients (aOR: 4.23 (3.82, 4.68)). Adjustment for Medicaid was associated with smaller survival differences between Black and White patients, with limited effect on other estimates.

Conclusions:

In this cohort, Black ICU survivors had lower long-term survival, and both Black and Hispanic patients had higher readmission rates compared to White patients. Differences were more pronounced among older adults. Variation in Medicaid coverage may contribute to observed disparities and warrants further investigation.

Introduction

Survivors of critical illness suffer from an increased likelihood of healthcare utilization and mortality.^1-8 An increase in care complexity is likely to be associated with worsened outcomes for intensive care unit (ICU) survivors: One out of every eight patients do not survive beyond one year, and one third of patients are re-admitted within 90 days.^1-3 Previous studies have shown higher case fatality rates among Black patients as compared to Hispanic and White patients as well as higher rates of severe sepsis among Black and Hispanic patients.⁹ Among the factors which have been associated with healthcare disparities are illness acuity, hospital location, or inequalities in interventions performed in the intensive care settings such as restraint use and sedation practices.^10-12 Recent studies have shown worse functional and cognitive outcomes among older adults at 12 months by area deprivation index (ADI), with racial minorities showing greater disadvantage.^13,14 However, less is known about racial and ethnic differences in survival following hospital discharge for critical illness.¹⁰

Upon hospital discharge, patients re-enter a setting which may have contributed to their presenting illness. Among the few studies previously performed, similar or improved rates of 30-day mortality was associated with race in previous analyses of earlier databases of ICU survivors including Medical Information Mart for Intensive Care-II (MIMIC-II),^15,16 and 60-day mortality was similar among Black ARDS survivors when adjusting for illness severity.¹⁷ However, fewer studies have examined hospital readmissions, which was recently identified as a marker for hospital-level disparities.¹⁸ Moreover, disparities among subgroups such as age and surgical status—identified in other health conditions^19,20—warrant further exploration among ICU survivors.

Here we aim to address gaps in the literature by describing differences in outcomes by race and ethnicity among ICU survivors. We evaluated 90-day and 1-year hospital readmissions and mortality among ICU survivors in the MIMIC-IV database. We additionally explored prespecified subgroup analyses among patients older than 60 years and among those undergoing surgical care. By identifying differences in readmission and survival among ICU survivors by race and ethnicity and exploring associated structurally modifiable risk factors such as Medicaid insurance status, restraint use, and sedation practices, we lay the groundwork for post-ICU interventions and healthcare policy initiatives aimed at improving outcomes for survivors of critical illness.

Methods

Study Design and Population

We performed a retrospective cohort study using the Medical Information Mart for Intensive Care (MIMIC)-IV database, a publicly available deidentified database of patients admitted to the emergency department or ICU Beth Israel Deaconess Medical Center in Boston, MA during the years 2008-2022; IRB approval for the database and waiver of informed consent as a deidentified database has been previously described.^21-23 Adults (≥18) who were admitted to the ICU and survived to discharge were included in our study. As described, mortality data were obtained via hospital and state records using an algorithm based on date of birth, social security number, and name.²¹ We calculated readmissions from the index admission for each individual by a unique identifier. STROBE criteria were followed (Supplementary Table S1).²⁴ Approval for data collection was previously obtained.

Outcomes

We examined the association between Black or White race, Hispanic ethnicity, and 90-day and 1-year survival as our primary outcome. We included only those patients who fell within one of the three race and ethnicity categories who survived hospitalization in our analysis. Our secondary outcome was hospital readmission following hospital discharge.

Statistical Analysis

Choice of Accelerated Failure Time (AFT) Model.

Cox Proportional Hazards were initially evaluated for mortality but noting the violation of the proportional hazards assumption (Supplementary Figure S1), we examined accelerated failure time (AFT) models. Akaike Information Criterion revealed that the log-normal distribution outperformed other distributions. The inverse time ratio (ITR, or 1/exp(β)), defined as the acceleration factor (AF) for ease of interpretation, was used to depict survival. Non-parametric analysis was performed using Kaplan-Meier survival curves. We examined absolute and relative differences in the probability of mortality at 90-days and 1-year based on survival curves generated using the AFT models.

Readmissions.

We explored readmissions within 90-days and 1-year by racial or ethnic subgroup as a primary outcome using the first index admission as the reference point. AFT was used to model survival to readmission. We controlled for age and sex as confounders. We accounted for the competing risk of mortality at both time points using a Fine-Gray model (via R package “cmprsk”). Survival curves were modeled by both unadjusted Kaplan-Meier curves and age- and sex-controlled cumulative incidence function (CIF) curves based on our Fine-Gray model.

Exposure, Covariates, and Confounders.

Race and Hispanic ethnicity were obtained from electronic health record data, which comprised a combination of self and personnel reports. We identified and defined our cohort within “White patients” to include all patients with such designation in the dataset (ie “White”, “White-Brazilian”, etc); “Black patients” to include “Black – African”, “Black – African” etc, and patients of Hispanic ethnicity to include “Hispanic or Latino”, “Hispanic/Latino – Dominican” etc A complete list of race and ethnicity designations are listed in the Supplement (Table S2). We control for age and sex in our models, and incorporate Medicaid insurance status as an incomplete measure of SDOH. We adjust for other covariates including comorbidities and 24-h SOFA in our sensitivity analyses.^9,10,25,26

Medicaid insurance status was determined using publicly available data from the MIMIC-IV database. Percentages and age- and sex-adjusted odds ratios were calculated for each race or ethnicity using logistic regression modeling.

We explored the inclusion of the Charlson Comorbidity Index (CCI) and SOFA in detail in our sensitivity analysis. P-values for the interaction of race were obtained for age and sex.

Subgroup Analysis.

Because of the contributions of age and surgery to hospital readmission and mortality, we elected to perform prespecified subgroup analyses on individuals in the database characterized as (1) age and (2) surgical status. We found a significant interaction between age and Black race (P = 0.035) (Table S3), prompting visual inspection of our data and defining subgroups by age: <60 and ≥60 (Figure S2). Furthermore, as differences in survival by race were previously reported among surgical patients, we examined mortality among a subgroup of patients receiving care in a surgical floor or surgical-affiliated ICU during their hospitalization and compared these to the non-surgical cohort by AFT model adjusting for age and sex.

Handling of Missing Data.

Missing data were imputed using multiple imputations by changed equations (MICE) including data for 16.2% observations of the CCI and 21.0% of 24-h maximum SOFA scores. Twenty imputations were generated with age and sex as clinically relevant variables. Predictive mean matching (PMM) was employed to reflect realistic data distributions. Insurance data was missing for 624 patients (1.33%) and restraint use data was missing for 12 (0.03%) patients. given the small number, a complete case analysis was chosen for the incorporation of Medicaid in our AFT models and restraint use in our sensitivity analysis. Among a subgroup of patients who were mechanically ventilated beyond 24 h (n = 3624), sedation data as defined by the Riker Sedation-Agitation Scale (SAS) was missing for nearly half (46.7%) of patients. Given the risk and questionable validity of imputation for missing data of this size, we opted for complete case analysis and acknowledge this as a significant limitation in our sensitivity analysis incorporating this variable.

Sensitivity Analysis.

We included moderate-to-high 24-h SOFA score (as defined by 24-h SOFA ≥ 6) and moderate-to-high comorbidities (CCI > 2) in our AFT models for our subgroups to assess the sensitivity of our results. We further sought to incorporate the significance of deep sedation on the association between race, Hispanic ethnicity and mortality outcomes among patients who underwent mechanical ventilation for ≥24 h with available Sedation-Agitation Scale (SAS) data, in addition to the association of restraint use as defined by ICD-9 code V4987 (“Physical restraints status”) and ICD-10 code Z78.1 (“Physical restraint status”).

Analyses were performed using R version 4.4.1 (2024-06-14 ucrt).

Results

Descriptive Statistics

Our study 46 640 ICU survivors (25 912 (55.6%) male, mean age 63.2 (SD 16.76); Black race 5498 (11.8%), Hispanic ethnicity 2129 (4.6%); mean age Black patients: 58.14 (SD 17.12), Hispanic patients (53.89 (17.27), White patients 64.43 (SD 16.39). Black patients (25.1%) and Hispanic patients (40.7%) were more likely to be insured by Medicaid than White patients (10.6%). Average Charlson Comorbidity Index (CCI), mean 24-h SOFA score, and surgical ICU status are shown in Table 1.

Table 1.

Patient Characteristics by Race.

Variable	Overall	Black	Hispanic	White
n	46 640	5498	2129	39 013
Age (mean (SD))	63.2 (16.8)	58.14 (17.1)	53.89 (17.3)	64.43 (16.4)
Male (%)	25 912 (55.6)	2609 (47.5)	1211 (56.9)	22 092 (56.6)
Charlson Comorbidity Index (mean (SD))	5.1 (3.2)	5.39 (3.5)	4.43 (3.5)	5.11 (3.2)
Surgical (%)	19 696 (42.2)	2418 (44.0)	993 (46.6)	16 285 (41.7)
24-h SOFA (mean (SD))	4.15 (3.1)	3.94 (3.1)	4.07 (3.2)	4.18 (3.1)

Primary Outcome: 90-day and 1-year Mortality

An accelerated failure time (AFT) model with log-normal distribution was fitted to examine the effect of race, Hispanic ethnicity, age, and sex on 90-day and 1-year mortality after discharge from a hospitalization that included an ICU stay. The model showed that overall, individuals characterized as Black race had decreased likelihood of survival as compared to those of White race (acceleration factor (AF) 1.21 (1.05, 1.40; P = 0.007)). Absolute differences (AD) in survival were significant at both 90-days (−0.85% (−1.47%, −0.23%)) and 1 year (−1.42% (−2.46%, −0.40%)). Overall, Hispanic patients had improved survival duration (AF 0.71 (0.56, 0.91; P = 0.006), with 90-day AD: 1.33% (0.39%, 2.31%) and 1-year AD: 2.31% (0.67%, 4.03%) compared to White patients (Table S4). Logistic regression modeling showed that Black race and Hispanic ethnicity were associated with increased likelihood of being enrolled in Medicaid insurance (aOR for Black patients: 2.26 (2.10, 2.43; P < 0.001); for Hispanic patients: 4.23 (3.82, 4.68; P < 0.001)) compared to referents of White race. The inclusion of Medicaid insurance status in multivariate adjustment attenuated the association of survival duration with Black race to nonsignificance (AF 1.14 (0.99, 1.31; P = 0.075), while the association with prolonged survival time was retained for Hispanic patients (AF 0.62 (0.48, 0.80; P = 0.002) (Table S5). Unadjusted Kaplan-Meier curves and survival curves based on the AFT model, adjusted for age and sex, are shown in Figure 1.

Figure 1.

Post-discharge survival by race and Hispanic ethnicity. A. Kaplan-Meier 1-year survival curve and B. Accelerated Failure Time (AFT)-based 1-year survival curves adjusted for age and gender.

Secondary Outcome: 90-day and 1-year Readmissions

When accounting for the competing risk of mortality, individuals of Black race had increased risk of hospital readmission at 90-days (subdistribution hazard ratio (SDHR) 1.21 (95%CI: 1.15, 1.27; P < 0.001)) and 1-year (1.29 (1.23, 1.34; P < 0.001)) as compared to those of White race. Similarly, a higher rate of readmission was seen among Hispanic patients at 90 days (SDHR 1.22 (1.13, 1.32; P < 0.001)) and 1 year (1.22 (1.14, 1.30; P < 0.001)) compared to individuals characterized as White race. When we included Medicaid insurance status as a covariate in multivariate adjustment, we observed that individuals of Black race and Hispanic ethnicity had higher rates of readmission at 90 days (Black patients: SDHR 1.20 (1.14, 1.26; P < 0.001); Hispanic: 1.21 (1.12, 1.31; P < 0.001)) and 1 year (Black patients: SDHR 1.27 (1.22, 1.33; P < 0.001; Hispanic: 1.20 (1.12, 1.29; P < 0.001)) as compared to White patients (Table S6). Cumulative incidence function curves for Fine-Gray Models are shown in Figure 2.

Figure 2.

Fine-gray cumulative incidence of 90-day and 1-year readmission and mortality following ICU discharge, stratified by race and ethnicity. A. 90-day readmission and mortality; B. 1-year readmission and mortality. Dotted lines represent hospital readmission and solid lines represent mortality. Mortality was treated as a competing risk for readmission.

Subgroup Analysis by Age Group

We conducted subgroup analyses among patients <60 and ≥60 years of age (Table S7). Among patients ≥60 years, Black race was associated with lower survival (AF 1.37 (1.16, 1.63; P < 0.001); 90-day AD: −1.89% (−3.01%, −0.76%); 1-year AD: −2.91% (−4.57%, −1.18%)) and Hispanic ethnicity was associated with improved survival (AF 0.70 (0.51, 0.96; P = 0.028); 90-day AD: 1.90% (0.28%, 3.33%); 1-year AD: 3.05% (0.45%, 5.46%)). No significant difference was seen among Black patients <60. Hispanic ethnicity <60 was associated with improved survival (90-day AD: 0.82% (0.03%, 1.59%); 1-year AD: 1.54% (0.05%, 3.07%)). When we adjusted for Medicaid insurance status, we noted that the association for Black (Acceleration Factor: 1.33 (1.12, 1.58; P = 0.002) and Hispanic patients (AF 0.65 (0.47, 0.90; P = 0.010)) (Table S8). Figure 3 presents a Forest plot to compare age subgroups.

Figure 3.

Forest plots with subgroup comparison to overall mortality. A. Comparison between overall acceleration factor (AF) for Black patients and Hispanic patients with subgroups age ≥60 and surgical subgroups. B. 1-Year subgroup comparison between overall AF by race and Hispanic ethnicity by age ≥60 and surgical subgroups. additional results are summarized in the text.

Subgroup Analysis by Surgical ICU Status

We sought to evaluate whether any mortality difference might be seen among patients receiving care in a surgical floor or surgical-affiliated ICU during their hospitalization. Among non-surgical patients, Black race was associated with lower survival at both time points (AF 1.51 (1.25, 1.83; P < 0.001); 90-day AD: −1.71% (−2.56%, −0.84%); 1-year AD: −2.89% (−4.29%, −1.43%)). Hispanic race was associated with no difference in survival at either time point (AF 0.82 (0.58, 1.14; P = 0.238); 90-day AD: 0.73% (−0.47%, 1.81%); 1-year AD: 1.28% (−0.81%, 3.24%)) among non-surgical patients. In contrast, Black race was not associated with improved survival among surgical patients (AF 0.93 (0.75, 1.14; P = 0.479)). Hispanic race was associated with improved survival among surgical patients (AF 0.60 (0.42, 0.86; P = 0.005)) at both 90 days (AD: 2.14% (0.79%, 3.51%)) and 1 year (AD: 3.67% (1.35%, 6.13%)) (Table S9). When Medicaid insurance was adjusted for, the shorter survival associated with Black race in the non-surgical group was retained (AF 1.45 (1.20, 1.76; P < 0.001)) as was the prolonged survival associated with Hispanic race among surgical patients (0.53 (0.37, 0.76; P < 0.001)) (Table S10). Forest plot comparison by surgical subgroup is shown in Figure 3.

Sensitivity Analysis

For sensitivity analyses we evaluated the inclusion of moderate to high comorbidities, 24-h SOFA score, or both into AFT models for the primary outcome for mortality and for the Fine-Gray models for readmission. The significant association between Black race and Hispanic ethnicity and mortality was retained on inclusion of either Charlson Comorbidity Index (CCI) score, 24-h SOFA score, or both (Tables S11-S13). When accounting for both CCI and 24-h SOFA, Black race continued to be associated with a shorter survival duration (AF 1.17 (1.01, 1.34; P = 0.033) and Hispanic race continued to be associated with an improved survival duration (AF 0.67 (0.55, 0.90; P = 0.005). Among patients who were age 60 and older, Black race continued to be associated with a decreased survival duration (1-year AF 1.35 (1.14, 1.61; P < 0.001)) (Table S14), which also held true for non-surgical patients (AF 1.43 (1.18, 1.73; P < 0.001) (Table S15). Notably, the significant increase in the likelihood of 90-day and 1-year readmission was retained even when CCI and SOFA were included for both Black patients and Hispanic patients at 90-days (1.17 (1.11, 1.23) for Black patients, P < 0.001; 1.20 (1.11, 1.30) for Hispanic patients; P < 0.001) and 1 year (1.25 (1.20, 1.30) for Black patients, P < 0.001; 1.20 (1.12, 1.28) for Hispanic patients, P < 0.001) (Table S16).

We additionally sought to incorporate depth of sedation as well as restraint use during the ICU stay as an potential mediators of healthcare inequities. Among patients who underwent mechanical ventilation for ≥ 24 h with available Sedation-Agitation Scale (SAS) data (n = 1932 (11.6% Black, 4.5% Hispanic; 55.5% male)), we found no significant difference in odds of deep sedation (pairwise OR for Black patients vs White patients: 1.25 (0.70, 2.14; P = 0.392); for Hispanic versus White patients: 1.26 (0.48, 2.82; P = 0.506)). Similarly, no significant difference was seen by race or Hispanic ethnicity whether or not deep sedation was incorporated into our AFT model for survival (Table S17-S18). Further, among patients with available ICD data (n = 46 628 (11.8% Black, 4.6% Hispanic; 55.6% male)) we found that the odds of restraint use as defined by ICD codes was significantly higher in Black patients but not Hispanic patients in both adjusted and unadjusted models (OR 1.13 (1.01, 1.27; P = 0.031); aOR 1.16 (1.04, 1.30; P = 0.009). The incorporation of restraint use did not attenuate the significance of the association between race and mortality outcomes previously described (Table S19-S20). The variable of restraint was significantly associated with accelerated mortality (AF 2.54 (2.15, 3.00; P < 0.001)).

Discussion

In a large retrospective cohort of survivors of intensive care with heterogeneous conditions, we observed that Black race was associated with lower survival, and Hispanic ethnicity was associated with higher survival up to 1 year after hospitalization that included an ICU stay as compared to White patients. Additionally, both Black race and Hispanic ethnicity were associated with a higher likelihood of hospital readmission at 90 days and 1 year. Our findings may reflect broader challenges racial and ethnic minorities may face in recovery from hospital illness owing to structural disparities including decreased access to community resources.^5,6,27 Following discharge, patients often return to a setting in which their health can be adversely impacted by structural factors, many of which may have contributed to their initial presentation to the ICU. Patients who survive critical illness can require complex care which can further strain such resources and can lead to worse outcomes. Our findings portray how complex care burden in ICU survivors may intersect with racial constructs, while proposing the importance of Medicaid insurance status as a means to alleviate healthcare inequities.

Healthcare inequalities by race are well-described across a spectrum of disease conditions and care settings and are likely the result of multiple patient-, health care system-, neighborhood-, and societal-level factors.^28-32 In the acute critical care setting, differences in survival by race and ethnicity have also been described, but few studies have explored what happens to patients after hospital discharge.¹⁰ Our findings align with the known literature describing disparities among Black patients,^28-32 including a recent study by Black et al showed that Black patients who survive septic shock have higher odds of 90-day mortality than non-Hispanic White patients.²⁸ Further, while our findings are discordant with previously reported mortality disparities in Hispanic patients,³³ other studies have reported similar findings to ours, including in severe sepsis incidence⁹ and in sepsis and acute respiratory failure hospitalization length of stay.³⁴ Proposed reasons include a small sample size, residual confounding,^9,35-37 and a categorization method that inadequately captures a diverse population of Hispanic patients comprised of Mexican, Puerto Rican, Cuban, Dominican, Central and South American patients, which differ significantly in terms of immigration history, language preference, geographic concentration, socioeconomic status, insurance status,³⁸ and access to care—as has been described in numerous studies^38-41—and which may in turn impact differences in mortality and readmissions. It should be additionally noted that while other studies have found improved health outcomes among Hispanic populations in the past, this phenomenon may be changing, particularly in the context of the recent COVID-19 pandemic and its disproportionate impact on minorities.^42,43

In our study we attempted to account for several acute critical illness-related factors in our sensitivity analysis, examining SOFA scores, restraint use, and sedation. Among these, restraint use was found to be more likely among Black patients, as has been shown in other studies.^44,45 We further identified an association between use of restraints and mortality among ICU survivors. While inclusion of restraint use did not significantly attenuate the association of race and 1-year mortality, our findings suggest that processes of care during critical illness may contribute to or exacerbate health care inequities. Current restraint protocols often employ subjective assessments of agitation or non-adherence, which can be influenced by communication barriers (including due to low health literacy), implicit bias, and inadequate staffing or training in alternative de-escalation techniques.^46-48 The inequitable application of restraints can trigger a cascade of downstream harms: For example, restrained patients may receive less frequent turning and repositioning, leading to a higher risk of pressure injuries, while enforced immobility can exacerbate ICU-acquired weakness and may worsen the very delirium restraints were meant to manage. Interventions aimed at reducing both restraint use and sedation include addressing pain, agitation and delirium, which often complicate care, through such mechanisms as multicomponent educational interventions.⁴⁷ Multicomponent educational interventions may include multidisciplinary rounds, decision support tools, and provision of information on alternatives to physical restraints such as re-orientation and sleep quiet time optimization.^47,49 Institutional interventions may also include enhanced standardization of restraint practices such as checklists proposed by Jin et al to alleviate any implicit bias⁴⁶ while also addressing any communication barriers that may influence or impede decision-making.

In addition to acute factors, long-term and chronic exposures have been shown to influence outcomes among ICU survivors. For instance, the association between an increase in area deprivation index (ADI) and adverse 1-year functional outcomes among older ICU survivors explored by Falvey et al highlights the important role that neighborhood socioeconomic disadvantage may play in mediating racial disparities.¹³ However, ADI carries limitations in that it is an aggregate value⁵⁰ that is likely influenced by such factors as median home value, failing to capture high variation in housing prices,⁵¹ while also failing to fully capture such key elements as access to health care such as density of healthcare providers or transportation infrastructure, at least directly.⁵² Another study among ICU survivors in France pointed to the importance of social deprivation as an important risk factor for adverse physical and psychiatric symptoms at 1-year using an alternative but similar socioeconomic scale.⁵³ Social support may also be a mediator of varying outcomes by race and ethnicity, including among ICU survivors. Non-Hispanic Black patients, for instance, may have lower levels of perceived social support (such as family support),⁵⁴ which may decline further among ICU survivors.⁵⁵ Importantly, MIMIC-IV focuses primarily on clinical data, with little information available regarding social determinants of health (SDOH) such as socioeconomic status or education level. Importantly, many of these same social and structural elements that contribute to worse outcomes may underlie inequalities in comorbidities and measures of critical illness. However, Medicaid and overall insurance status have been explored as a component of healthcare disparities⁵⁶ and a surrogate for SDOH.^57,58 In our study, the inclusion of Medicaid status as a covariate in our survival model attenuated the observed mortality difference among Black patients but did not impact readmissions. Our study builds on prior research showing that federal or state insurance coverage mediated the impact of race on functional, cognitive,¹⁴ and mortality differences among older survivors of critical illness.⁵⁹

We show here that both Black race and Hispanic ethnicity was associated with higher rates of 90-day and 1-year readmissions than White patients. This finding aligns with the literature, including a study in a healthcare system in California that showed increased 7-day readmissions for Black and Asian patients³⁰ as well as other studies in Medicare and Medicare Advantage-insured patients as well as in patients with diabetes that showed an increase in 30-day readmissions among Black patients.^31,32 Our study is among the first to report an increase in hospital readmissions by Black race among ICU survivors, highlighting an increase in healthcare utilization that may be amenable to early interventions such as structured post-ICU care.¹⁰ Importantly, our findings held even with the inclusion of the Charlson Comorbidity Index and SOFA scores suggesting that these factors do not adequately explain inequities in survivors of critical illness.

To further explore effect modifiers of healthcare inequities, we prespecified a subgroup analysis among patients age 60 and older, and found that in this age group Black race was associated with higher 1-year mortality than White patients. Our study aligns with similar findings in a longitudinal cohort showing a change in cardiometabolic risk over four years that was worse among Black patients as compared to White patients and Hispanic patients.¹⁹

Further, we prespecified a subgroup analysis among patients who received care in a surgical setting. We observe increased mortality associated with Black race among those patients who did not receive care in a surgical unit and decreased mortality associated with Hispanic ethnicity among those patients who underwent care in a surgical unit as compared to White patients. While our study aligns with other studies in medical populations,¹⁰ previous research among surgical patients suggested that Black patients have higher in-hospital mortality for procedures such as minimally invasive surgeries for benign surgical conditions⁶⁰ as well as a higher odds of readmission and re-operation after general surgery.⁶¹ Potential reasons for the discrepancy in our results may include the designation of surgical patients by floor or hospital type rather than the surgery itself and unmeasured confounders related to the type of surgery. For instance, one study comparing elective to emergency surgery showed that among patients with colorectal cancer, Black non-Hispanic patients were more likely to undergo emergency surgery than White patients and that emergency surgery was associated with an increased risk of postoperative complications such as mortality.⁶² Further research on surgical subgroups of ICU survivors may benefit from incorporating types of surgery to help offset these potential confounders. Of note, our findings of decreased mortality in Hispanic patients mirror previously reported findings among surgical patients in the American College of Surgeons National Surgical Quality Improvement Program database.⁶³

The identification of healthcare inequities is an important first step in improving the lives of survivors of critical illness. In addition to optimizing equitable care delivery to patients receiving acute care through education aiming towards equitable restraint and sedation practices, interventions aimed at reducing care complexity can be implemented among ICU survivors, including among older patients, those with primarily medical admissions, and patients on Medicaid insurance. Others have proposed incorporating socioeconomic disadvantage as a component of post-intensive care syndrome (PICS) and examining structural vulnerabilities as they specifically relate to survivors of critical illness.⁶⁴ Disaggregating social and area deprivation indexes will likely be needed to identify which factors could be targeted to best reduce disparities, as suggested by the finding that education but not socioeconomic status was associated with a decreased odds of PICS.⁶⁵ Among the elements associated with SDOH, expansion of Medicaid under the Affordable Care Act has been shown to improve insurance coverage for Black and Hispanic patients, which translated into increased use of preventative services; reductions in medical debt, poverty, and home eviction; and reduced mortality among patients with low income.⁶⁶ Implementation of transitional care models for ICU survivors may help alleviate the compounded complex care burden that may be experienced by demographic minorities by reducing readmissions and medical expenditures while enhancing early primary care follow-up, particularly in Medicaid patients.⁶⁷ Post-ICU clinics represent one such transitional care model, and while data are limited, may help to reduce healthcare disparities, particularly as they meet patients within the context of the community and associated day-to-day structural factors with which a patient may interact. Incorporating adequate post-ICU care in healthcare facilities such as Federally Qualified Health Centers (FQHCs)⁶⁸ may help improve outcomes among diverse survivors of critical illness, particularly given a high proportion of Medicaid-covered patients treated at such facilities.

In parallel with institutional efforts, policymakers may also consider the impact of adjustments to Medicaid on minority and non-minority demographics when developing healthcare policy (while simultaneously working to address disparities within Medicaid-insured healthcare delivery)^66,69 and consider increasing resources towards enhancing specialty and post-intensive care access within FQHCs.

Limitations

Our results suggest differences in patient outcomes by race and ethnicity. We recognize that such differences are the result of extensive, historical social and structural factors that contribute to individual- and population-level differences in access to care, optimal preventive practices, and the outcomes identified here. Such factors are not available in the database we used to conduct the analysis and hence not accounted for by us, emphasizing the importance of including such factors in future databases to address this important knowledge gap. We consequently recognize the significant residual confounding that results when reporting differences by race and ethnicity. We would advocate for more comprehensive identification and characterization of social factors to improve our understanding of the mechanisms and drivers of disparities. Further, while we explore differences among Black patients, White patients, and those of Hispanic ethnicity, we recognize that more work is needed to identify disparities across all races and ethnicities. We further recognize the potential overlap among Hispanic ethnicity and race, while retaining these classifications as they were originally entered in the MIMIC-IV database. We also note that the exact proportion of self-reported race/ethnicity is unknown.⁷⁰ Moreover, our study uses a single-center database in Boston with a lower proportion of Black and Hispanic patients than the national average,⁷¹ which may not adequately reflect regional inequalities in healthcare outcomes and delivery and may not capture patients readmitted outside of the healthcare system.

Social determinants of health (SDOH) have been long recognized as key factors through which healthcare disparities may both manifest or be impacted by. The inclusion of Medicaid insurance in our models carries key limitations, not the least of which is potential collinearity. Such collinearity reflects underlying structural factors that may have influences differential access to employment, insurance, and healthcare, and controlling for Medicaid may attenuate the impact of such factors, and therefore should be treated with caution. Nonetheless, Medicaid insurance status is one of the few variables available within MIMIC-IV that exists within the complex structural framework of healthcare inequities. With its inclusion, rather than seeking to explain healthcare inequities as being due to insurance status, we seek to open the door to further exploration of a component of healthcare inequities through further research among ICU survivors. Additional residual confounding includes covariates that would more accurately capture healthcare access and economic deprivation. Further, inadequate data on sedation and sedation depth as well as limited data on restraint use may lead to residual confounding by key variables such as delirium, particularly as both of these factors have become increasingly recognized as a mediators of healthcare disparities by race.^45,72,73 Subsequent work may seek to evaluate how such covariates comorbidities may mediate the association of race and ethnicity with mortality and readmissions at all stages of the data collection and analysis process, though including through engagement and recruitment of minority research participants to more accurately reflect a diverse healthcare population, ensuring self-identification of race and ethnicity to the extent feasible,⁷⁴ inclusion of SDOH-related variables such as ADI,¹³ measures of healthcare access such as transportation and regional concentration of primary care providers,⁵² and adequate documentation of use sedating medications, depth of sedation, and restraint use.

Conclusion

Black race was associated with reduced survival to 1 year as compared to White race after a hospitalization for critical illness. Black race and Hispanic ethnicity were associated with an increased likelihood of hospital readmission at 90 days and 1 year. Among a prespecified subgroup of patients ≥age 60 and medical patients, Black patients had higher mortality. Incorporating Medicaid status attenuated the mortality difference among Black patients within the overall cohort. Further research should explore the social and structural factors that contribute to racial and ethnic differences in survival to develop meaningful interventions to reduce healthcare inequities among survivors of critical illness.

Supplementary Material

Supplemental material for this article is available online.

Data Availability

The MIMIC-IV database has been made publicly available by Alistair Johnson et al (https://physionet.org/content/mimiciv/3.1/)