Abstract
Background
Specialty intensive care units (ICUs) provide diagnosis-specific patient care. Nonetheless, the advantage or disadvantage of ICU specialization remains debatable. We conducted a multicenter retrospective cohort study to evaluate the clinical outcomes of specialty ICUs in critical care.
Methods
This retrospective study reviewed adult ICU patients in the eICU Collaborative Research Database (eICU-CRD), a United States (US)-based multicentered ICU cohort database. Patients over 16 years old with documented Acute Physiology and Chronic Health Evaluation (APACHE) IV scores were enrolled and grouped by their admitted ICUs. Multilevel logistic regression models were performed to assess the relationships between hospital mortality and ICU type/ICU triage conditions in patients admitted with a diagnosis of cerebrovascular accident (CVA), acute coronary syndrome (ACS), coronary artery bypass graft (CABG), pneumonia (PNA) and other respiratory diseases, and trauma. A parallel multiple mediation analysis was performed to examine whether interventions in the ICU mediated the association between ICU triage and hospital mortality.
Results
A total of 136,236 patients admitted to 271 ICUs from 186 hospitals were enrolled. In multilevel analysis, admission to cardiac ICU (OR: 0.81, p < 0.001) and surgical ICU (OR: 0.88, p = 0.041) were associated with lower mortality risks, and admission to Neuro-ICU (OR: 1.32, p < 0.001) was linked to a higher risk of mortality compared to general ICU in total cohort. In patients without trauma, general ICU or discordant triage was not associated with a higher in-hospital mortality risk than concordant ICU triage. In trauma patients, discordant ICU triage was associated with higher hospital mortality (OR: 1.87, p = 0.005), partially mediated by mechanical ventilation (OR = 1.007, p < 0.001).
Conclusion
Specialized concordant ICUs were not associated with lower hospital mortality risk compared to general ICUs after adjusting for confounders. Under resource-constrained settings, discordant ICU triage may be a viable option.
Keywords: critical care, ICU triage, intensive care units, multilevel model, specialization
1. Introduction
Over the recent three decades, intensive care medicine has become a featured specialty of clinical medicine, along with the increase in the number of intensive care units (ICUs) and their capacity to carry out critical care services to meet the mounting demand. It is reported that the ICU-to-hospital-bed ratio in the United States has increased from 13.5 to 16.2% in the first decade of the 21st century (1, 2). The capacity and expertise of ICUs to diagnose and manage critical illnesses have now been widely regarded as an indicator of the overall service quality of a hospital. Therefore, establishing and optimizing the critical care system under limited medical resources for the best treatment outcomes has become a significant challenge in hospital management. One of the most discussed trends is the specialization of ICUs. A specialized ICU, namely, cardiology ICU, medical ICU, neurology ICU, etc., refers to an ICU where a designated service specialty concentrates its medical resources and provides diagnosis-specific care for selected patients within the scope of its expertise (3). Specialized ICUs manage a narrower spectrum of diseases, thus they may better understand patients’ primary conditions and disease progression, and be able to implement more tailored monitoring and therapy, intuitively improving the clinical outcomes (4). It was found that the mortality rate of patients with cerebral hemorrhage treated in the Neuro-ICU was significantly lower than in general ICU (5–7). Some similar results have been shown in earlier studies of admitted to specialty care units for shock, cardiovascular diseases (7–10). On the contrary, in a retrospective cohort study with a larger sample (84,182 patients), Lott et al. demonstrated no advantage from an ideal (“diagnosis-appropriate”) specialty ICU over general ICU in improving risk-adjusted survival or shortening ICU length of stay (LOS) (11). As comorbidities and multiple organ failure are commonly presented in critically ill patients, general ICU may be better at balancing the risk and benefit of various treatments, distinguishing primary and secondary diagnoses, and overseeing the big picture of an ICU encounter. As the average age of ICU patients continues to rise and multimorbidity become more prevalent (12), there is an urgent need to discuss whether general ICUs facilitate a more integrated and comprehensive approach to managing complex, multi-system failures.
The previous studies focused on specific specialty ICUs and a few medical conditions, limiting their conclusions to a particular field. Therefore, we aim to first conduct a comprehensive analysis of ICU patients from a large multicenter database with a wide range of diagnoses who were admitted to various specialty ICUs. Secondly, to investigate the effect of a concordant ICU triage versus a discordant ICU specialty or general ICU on in-hospital mortality risk, we selected five specialized diseases with relatively high case volumes across each specialized ICU for subgroup analysis, including CVA (cerebrovascular accident), ACS (acute coronary syndrome), pneumonia and other respiratory disease, CABG (coronary artery bypass graft), and trauma.
2. Methods
2.1. Data source
The retrospective cohort study was based on data from the eICU Collaborative Research Database (eICU-CRD), a multi-center intensive care unit database with high granularity de-identified health data for over 200,000 ICU admissions across the United States between 2014 and 2015 (13). The up-to-date version 2.0 of eICU-CRD was employed. After completing the Collaborative Institutional Training Initiative training course “Data or Specimens Only Research,” our group gained access to the eICU-CRD (certification number: 2093226) and was approved by the Massachusetts Institute of Technology Affiliates to employ the above data. Informed consent was not applicable since the patient’s identity had been de-identified.
2.2. Study population and data extraction
All patients 16 years and older with documented APACHE IV scores and admission diagnosis in the eICU database were enrolled in the study (14). Exclusion criteria were not the first admission. Patients were stratified based on their initially-admitted ICU specialty, including general ICU (GICU), surgical ICU (SICU), medical ICU (MICU), neurological ICU (Neuro-ICU), and CICU, which consists of cardiac ICU, CCU-CTICU (Coronary Care Unit/Cardiothoracic Intensive Care Unit), CSICU (Cardiac Surgical Intensive Care Unit), and CTICU (Cardiothoracic Intensive Care Unit).
The demographic and clinical data, including sex, age, ethnicity, APACHE-IV score, diagnosis, and comorbidities, were retrieved by PostgreSQL (version 9.6; PostgreSQL Global Development Group). The implementation of ICU interventions (mechanical ventilation, dialysis, vasopressor drug usage, and antibiotics use) was used to evaluate the demand for aggressive medical service. Hospital-level covariates, including geographic region, teaching status, and bed capacity, were also collected. No missing values were present in the variables included in this study; therefore, no imputation was performed.
The study focused on a primary outcome of in-hospital mortality.
2.3. Statistical analysis
Baseline demographic and clinical features were individually compared between different ICU units. Pearson’s chi-square test was performed for categorical data, and continuous data were compared by pairwise Student’s t-test or one-way analysis of variance (ANOVA) (parametric data) across groups, or Wilcoxon rank-sum test or Mann–Whitney U-test (non-parametric data).
We employed multilevel mixed-effects logistic regression models to compare the in-hospital mortality risks across different ICU types. We adjusted for the fixed effects of confounders such as gender, age, APACHE-IV score, comorbidities (hypertension, diabetes, CKD, COPD, heart failure, cancer), and hospital characteristics (teaching status, bed capacity compared to beds ≥500), and the random effects at the hospital level to account for between-hospital variability.
Given the significant heterogeneity in admission diagnoses across ICU types, we further focused on five sub-cohorts that were diagnosed with the following common conditions at admission: CVA (cerebrovascular accident), ACS (acute coronary syndrome), pneumonia and other respiratory disease, CABG (coronary artery bypass graft), and trauma. We aimed to test whether the initial ICU specialty, that is, a concordant versus discordant ICU triage, was associated with distinct in-hospital mortality risks for these highly specialized conditions. Consequently, we re-classified their initially-admitted ICU specialty into concordant ICU triage, discordant ICU triage, and general ICU according to the consistency between admission diagnosis and ICU specialty. The ICU triage grouping details are displayed in Table 1.
Table 1.
Cohort re-grouping based on ICU triage.
| Diseases | Concordant ICU triage | Discordant ICU triage | General ICU |
|---|---|---|---|
| CVA | Neuro-ICU | CICU, MICU, SICU | GICU |
| ACS | CICU | Neuro-ICU, MICU, SICU | GICU |
| CABG | CICU | Neuro-ICU, MICU, SICU | GICU |
| PNA/Other respiratory diseases | MICU | Neuro-ICU, CICU, SICU | GICU |
| Trauma | SICU | CICU, MICU, Neuro-ICU | GICU |
CVA, cerebrovascular accident; ACS, acute coronary syndrome; CABG, coronary artery bypass graft; PNA, pneumonia; CICU, cardiac ICU; GICU, general ICU; MICU, medical ICU; Neuro ICU, neurology ICU; SICU, surgical ICU.
To describe causal assumptions among ICU types, hospital mortality, confounders, and mediators, we constructed a directed acyclic graph (directed acyclic graphs) following the latest guidelines (15). A parallel multiple mediation analysis was performed to examine whether ICU interventions (including mechanical ventilation, vasopressor use, dialysis, and antibiotic therapy) mediated the association between ICU triage and hospital mortality. To determine effect sizes and obtain 95% confidence intervals, we assessed direct and indirect effects using the bootstrap method with 1,000 iterations. R Package “bruceR” was used in analysis, which estimates all mediation paths simultaneously while controlling for potential confounders. For clinical interpretability, regression coefficients from logistic regression models were converted to odds ratios (ORs) with 95% confidence intervals (CIs) using the formula OR = exp(β). This analysis was restricted to diagnostic subgroups in which ICU triage was significantly associated with hospital mortality in multilevel mixed-effects models accounting for hospital-level random effects.
The above analyses were computed with results visualized by R (version 4.4.3, R Foundation for Statistical Computing, Vienna, Austria).1 A two-tailed p-value <0.05 was considered statistically significant.
3. Results
3.1. Overview of ICU settings
The participant enrollment flowchart is illustrated in Figure 1. This study included 136,236 patients with documented APACHE IV scores from 271 ICUs across 190 hospitals in the eICU database. More than 50% of patients were admitted to general ICUs (GICU, 54.96%), followed by cardiac ICUs (CICU, 22.03%), medical ICUs (MICU, 8.73%), neurology ICUs (neuro ICU, 7.80%), and surgical ICUs (SICU, 6.48%).
Figure 1.
Flowchart of study cohort selection.
Of the 192 hospitals, 160 (84.21%) had a GICU (84.21%), followed by CICU in 23.68% of hospitals. The number of ICU types was associated with both the hospital’s bed capacity and its status as a teaching hospital. All hospitals with fewer than 100 beds hosted only one ICU specialty, but over half (68.18%) of the hospitals with more than 500 beds had three or more ICU specialties. Besides, nearly 75% of teaching hospitals had two or more ICUs, compared to less than 25% in non-teaching hospitals (Figure 2).
Figure 2.
Overview of ICU settings. (A) The coverage ratio of specialty ICUs in US hospitals; (B) the number of ICU types in hospitals with different bed capacities; (C) the number of ICU types in teaching or non-teaching hospitals.
3.2. Patient characteristics
Baseline characteristics of the total cohort are summarized in Table 2. The overall cohort had a median age of 65 years (IQR: 53–76), 54% male, and a median APACHE IV score of 51 (IQR: 37–68). APACHE IV scores varied significantly by ICU type, with the lowest in Neuro-ICU (43, IQR: 31–58) and the highest in MICU (56, IQR: 41–75) (p < 0.001). ICU admission diagnoses included sepsis (13%), CVA (7%), cardiac arrest (7%), ACS (6%), and gastrointestinal bleeding (5%), with significant different distribution across ICU types (p < 0.001). Comorbidities varied by ICU type, with Neuro-ICU patients exhibiting the fewest. Cancer prevalence was highest in SICU (10% vs. 5% in GICU, p < 0.001). In ICU interventions, mechanical ventilation, dialysis, vasopressors, and antibiotics were administered more frequently in MICU than in Neuro-ICU (p < 0.001). The overall ICU mortality rate was 6%, highest in MICU (8% vs. 4% in neuro ICU, p < 0.001), and the median ICU length of stay was 42 h (IQR: 23–76), shortest in GICU (41 h, IQR: 22–73). Hospital characteristics analysis showed that 27% of patients were admitted to teaching hospitals (59% in neuro ICU vs. 16% in GICU), while 39% were treated in large hospitals with ≥500 beds (79% in neuro ICU vs. 20% in GICU).
Table 2.
Demographic, clinical, and admitted hospital information characteristics of patients.
| Variables | Total | GICU | CICU | MICU | Neuro ICU | SICU | p-value |
|---|---|---|---|---|---|---|---|
| n | 136,236 | 74,872 | 30,010 | 11,895 | 10,629 | 8,830 | |
| Age | 65 (53, 76) | 65 (52, 76) | 66 (55, 76) | 65 (52, 76) | 63 (50, 75) | 64 (52, 75) | <0.001 |
| Gender, male | 73,611 (54) | 69,472 (54) | 17,493 (58) | 6,182 (52) | 5,481 (52) | 5,011 (57) | <0.001 |
| APACHE IV | 51 (37, 68) | 51 (37, 69) | 50 (38, 67) | 56 (41, 75) | 43 (31, 58) | 50 (37, 67) | <0.001 |
| Admission diagnosis | <0.001 | ||||||
| ACS | 8,343 (6) | 3,434 (5) | 4,289 (14) | 339 (3) | 71 (1) | 210 (2) | |
| ARF | 1932 (1) | 1,305 (2) | 273 (1) | 265 (2) | 15 (0) | 74 (1) | |
| Asthma | 3,948 (3) | 2,561 (3) | 687 (2) | 512 (4) | 65 (1) | 123 (1) | |
| CABG | 4,771 (4) | 1,324 (2) | 3,044 (10) | 1 (0) | 5 (0) | 397 (4) | |
| Cardiac arrest | 9,135 (7) | 4,833 (6) | 3,158 (11) | 731 (6) | 136 (1) | 277 (3) | |
| Chest pain unknown | 761 (1) | 407 (1) | 295 (1) | 36 (0) | 5 (0) | 18 (0) | |
| CHF | 5,592 (4) | 3,006 (4) | 1844 (6) | 491 (4) | 64 (1) | 187 (2) | |
| Coma | 2082 (2) | 1,358 (2) | 265 (1) | 233 (2) | 147 (1) | 79 (1) | |
| CVA | 9,758 (7) | 4,202 (6) | 865 (3) | 552 (5) | 3,475 (33) | 664 (8) | |
| Other CV | 3,593 (3) | 1,393 (2) | 1,607 (5) | 250 (2) | 167 (2) | 176 (2) | |
| DKA | 4,384 (3) | 3,157 (4) | 503 (2) | 562 (5) | 54 (1) | 108 (1) | |
| GI bleed | 7,277 (5) | 4,768 (6) | 885 (3) | 968 (8) | 116 (1) | 540 (6) | |
| GI obstruction | 1,232 (1) | 823 (1) | 114 (0) | 72 (1) | 15 (0) | 208 (2) | |
| Neuro | 4,640 (3) | 2028 (3) | 330 (1) | 276 (2) | 1804 (17) | 202 (2) | |
| Overdose | 4,268 (3) | 3,073 (4) | 528 (2) | 485 (4) | 97 (1) | 85 (1) | |
| PNA | 4,577 (3) | 3,180 (4) | 639 (2) | 583 (5) | 69 (1) | 106 (1) | |
| Other respiratory diseases | 7,970 (6) | 4,726 (6) | 1,266 (4) | 1,309 (11) | 229 (2) | 440 (5) | |
| Sepsis | 18,087 (13) | 12,119 (16) | 2,554 (9) | 2,436 (20) | 306 (3) | 672 (8) | |
| Trauma | 5,884 (4) | 2,972 (4) | 289 (1) | 154 (1) | 1,376 (13) | 1,093 (12) | |
| Valvular disorders | 2,795 (2) | 697 (1) | 1921 (6) | 0 (0) | 2 (0) | 175 (2) | |
| Other | 25,207 (19) | 13,506 (18) | 4,654 (16) | 1,640 (14) | 2,411 (23) | 2,996 (34) | |
| Comorbidities | |||||||
| Hypertension | 19,878 (15) | 10,837 (14) | 5,135 (17) | 1,057 (9) | 1,644 (15) | 1,205 (14) | <0.001 |
| Diabetes | 15,623 (11) | 9,486 (13) | 3,625 (12) | 1,151 (10) | 636 (6) | 725 (8) | <0.001 |
| CKD | 11,869 (9) | 6,508 (9) | 2,869 (10) | 1,282 (11) | 426 (4) | 784 (9) | <0.001 |
| COPD | 10,544 (8) | 6,822 (9) | 2030 (7) | 1,039 (9) | 257 (2) | 396 (4) | <0.001 |
| Heart failure | 11,361 (8) | 6,522 (9) | 3,202 (11) | 994 (8) | 209 (2) | 434 (5) | <0.001 |
| Cancer | 7,336 (5) | 4,080 (5) | 992 (3) | 564 (5) | 783 (7) | 917 (10) | <0.001 |
| ICU intervention | |||||||
| Mechanical ventilation | 41,293 (30) | 22,150 (30) | 9,838 (33) | 4,148 (35) | 2,218 (21) | 2,939 (33) | <0.001 |
| Dialysis | 5,722 (4) | 3,251 (4) | 1,215 (4) | 775 (7) | 146 (1) | 335 (4) | <0.001 |
| Vasopressor drug usage | 16,227 (12) | 8,489 (11) | 4,581 (15) | 1,602 (13) | 502 (5) | 1,053 (12) | <0.001 |
| Antibiotics use | 32,640 (24) | 20,695 (28) | 6,173 (21) | 2,716 (23) | 1,195 (11) | 1861 (21) | <0.001 |
| Outcomes | |||||||
| Hospital death | 12,061 (8.9) | 6,650 (8.9) | 2,384 (7.9) | 1,458 (12.3) | 849 (8) | 720 (8.2) | <0.001 |
| ICU death | 7,619 (6) | 4,204 (6) | 1,601 (5) | 898 (8) | 463 (4) | 453 (5) | <0.001 |
| ICU LOS, hours | 42 (23, 76) | 41 (22, 73) | 43 (24, 76) | 46 (25, 88) | 44 (24, 87) | 46 (25, 89) | <0.001 |
| Hospital information | |||||||
| Bed | <0.001 | ||||||
| <100 | 8,072 (6) | 6,966 (9) | 567 (2) | 192 (2) | 0 (0) | 347 (4) | |
| ≥500 | 52,508 (39) | 14,795 (20) | 15,607 (52) | 7,584 (64) | 8,384 (79) | 6,138 (70) | |
| 100–249 | 29,199 (21) | 22,945 (31) | 4,185 (14) | 993 (8) | 652 (6) | 424 (5) | |
| 250–499 | 32,177 (24) | 22,209 (30) | 6,388 (21) | 1,507 (13) | 631 (6) | 1,442 (16) | |
| Unknown | 14,280 (10) | 7,957 (11) | 3,263 (11) | 1,619 (14) | 962 (9) | 479 (5) | |
| Teaching status | 37,355 (27) | 12,040 (16) | 10,622 (35) | 4,296 (36) | 6,231 (59) | 4,166 (47) | < 0.001 |
Data are presented as number (percentage) or median (interquartile range). APACHE IV, Acute Physiology and Chronic Health Evaluation IV; ACS, acute coronary syndrome; ARF, acute renal failure; CABG, Coronary Artery Bypass Graft; CHF, Congestive heart failure; CVA, Cerebrovascular accident; CV, Cerebrovascular; DKA, Diabetic Ketoacidosis; GI, Gastrointestinal; PNA, pneumonia; CKD, Chronic kidney disease; COPD, Chronic Obstructive Pulmonary Disease; GICU, Medical-Surgical ICU; MICU, Medical ICU; Neuro ICU, Neurology ICU; SICU, Surgical ICU; LOS, length of stay.
3.3. Multilevel models of hospital mortality ICU admission patients
Our multilevel mixed-effects logistics regression model identified the following independent predictors for increased hospital mortality risks among all enrolled participants: admission to Neuro-ICU (odds ratio, OR: 1.32, 95% confidence interval, CI: 1.19–1.47, p < 0.001), APACHE-IV score (OR: 1.05, 95%CI: 1.05–1.05, p < 0.001), age (OR: 1.011, 95% CI: 1.01–1.01, p < 0.001), COPD (OR: 1.10, 95%CI: 1.02–1.19, p = 0.01), heart failure (OR: 1.3, 95%CI: 1.24–1.42, p < 0.001), cancer (OR: 1.77, 95%CI: 1.63–1.92, p < 0.001). However, admission to cardiac ICU (OR: 0.81, 95%CI: 0.75–0.88, p < 0.001) and surgical ICU (OR: 0.88, 95% CI: 0.78–1.00, p = 0.041) were associated with lower mortality risks compared to general ICU. (Figure 3).
Figure 3.
Multilevel logistic regression model for risk of death in the total cohort.
3.4. Subgroup analysis of five common specialized conditions
Next, we conducted subgroup analysis on five common conditions that might benefit from treatment in a concordant, specialized ICU using multilevel mixed-effects regression models, including CVA, ACS, CABG, trauma, PNA and other respiratory diseases. In patients diagnosed with CVA at admission, discordant ICU triage was associated with lower hospital mortality compared to concordant ICU triage (OR: 0.69, 95%CI: 0.51–0.94, p = 0.018) (Figure 4). Contrarily, patients diagnosed with trauma who were initially admitted to a discordant ICU triage showed significantly higher in-hospital mortality risks compared to a concordant ICU specialty (OR: 1.87, 95%CI: 1.2–2.9, p = 0.005) (Figure 5).
Figure 4.
Multilevel logistic regression model for risk of death in patients diagnosed with CVA.
Figure 5.
Multilevel logistic regression model for risk of death in patients diagnosed with trauma.
Regarding patients diagnosed with ACS, pneumonia and other respiratory diseases, and CABG, there were no significant statistical differences between concordant ICU triage, discordant ICU triage, and general ICU (Figures 6, 7, 8). APACHE-IV score and age were generally associated with a higher hospital mortality across all diseases.
Figure 6.
Multilevel logistic regression model for risk of death in patients diagnosed with ACS.
Figure 7.
Multilevel logistic regression model for risk of death in patients diagnosed with PNA and other respiratory diseases.
Figure 8.
Multilevel logistic regression model for risk of death in patients diagnosed with CABG.
3.5. Mediation analysis of CVA and trauma patient
Mediation analysis confirmed the mediating effect relationship between hospital mortality, ICU types and ICU intervention, including mechanical ventilation, dialysis, vasopressor drug usage and Antibiotics use. In the trauma subgroup, ICU triage demonstrated a significant direct effect on hospital mortality (OR = 0.817, 95% CI: 0.690–0.968; p = 0.02), whereas no significant direct effect was observed in the CVA subgroup (OR = 1.000, 95% CI: 0.990 to 1.007; p = 0.87). Regarding indirect effects, vasopressor use significantly mediated the association between ICU triage and hospital mortality in the CVA subgroup (OR = 0.999, 95% CI: 0.998 to 1.000; p = 0.029). In contrast, in the trauma subgroup, mechanical ventilation showed a significant mediating effect, with ICU triage associated with reduced hospital mortality through increased use of mechanical ventilation (OR = 1.007, 95% CI: 1.002 to 1.013; p < 0.001). No statistically significant indirect effects were observed for dialysis or antibiotic use in either the CVA or trauma subgroups, and vasopressor use did not demonstrate a significant mediating effect in the trauma subgroup (p > 0.05) (Figures 9–11).
Figure 9.
DAG of with exposure to ICU types, outcome of hospital mortality, confounder, and mediators.
Figure 11.
![Diagram illustrating the mediation analysis of ICU triage effects on hospital mortality with four pathways: mechanical ventilation, dialysis, vasopressor drug usage, and antibiotics use. Mechanical ventilation pathway shows statistical significance with ab = 1.007 [1.002, 1.013] and c’ = 0.817 [0.690, 0.968], indicating mediation effects on mortality. Numeric values detail odds ratios along each arrow connecting variables.](https://www.ncbi.nlm.nih.gov/core/lw/2.0/html/tileshop_pmc/tileshop_pmc_inline.html?title=Click%20on%20image%20to%20zoom&p=PMC3&id=13018113_fmed-13-1785019-g011.jpg)
Mediation effects of ICU triage on hospital mortality in trauma subgroup. The figure illustrates the mediation analysis with ICU triage as the independent variable, and hospital mortality as the dependent variable. The numbers above the arrows indicate the odds ratio (OR) values for the respective paths. The indirect effect (ab) is represented by the path from ICU triage (concordant vs. discordant) to hospital mortality through the mediators, ICU interventions. The direct effect (c’) represents the direct relationship between ICU triage and hospital mortality, excluding the mediators. The values in brackets [] indicate the Bootstrap 95% Confidence Interval (CI) for each effect OR values. * p < 0.5, ** p < 0.01, *** p < 0.001.
Figure 10.
Mediation effects of ICU triage on hospital mortality in CVA subgroup. The figure illustrates the mediation analysis with ICU triage as the independent variable, and hospital mortality as the dependent variable. The numbers above the arrows indicate the odds ratio (OR) values for the respective paths. The indirect effect (ab) is represented by the path from ICU triage (concordant vs. discordant) to hospital mortality through the mediators, ICU interventions. The direct effect (c’) represents the direct relationship between ICU triage and hospital mortality, excluding the mediators. The values in brackets [] indicate the Bootstrap 95% Confidence Interval (CI) for each effect OR values. * p < 0.5, ** p < 0.01, *** p < 0.001.
4. Discussion
Despite being widely discussed, no consensus was reached on the benefit of ICU specialization. In our study, which included a large cohort of ICU patients who were managed at different levels of hospitals, demonstrated that specialized ICU was not associated with lower hospital mortality risk compared to the general ICU after adjusting for confounders, in fact, Neuro-ICU showed an increased in-hospital mortality risk. Regarding specialized conditions, CVA patients who were initially admitted to a concordant ICU (Neuro-ICU) were not linked with lower mortality risk, while trauma patients admitted to a concordant ICU (SICU) demonstrated lower mortality risk.
This study outlines characteristics of participating hospitals, including bed capacity, teaching status, and ICU unit number. Notably, general ICUs were widely implemented across institutions, while the establishment of specialized ICUs appeared correlated with bed capacity≥300 and academic affiliation. Our findings align with existing evidence suggesting that tertiary care centers and teaching hospitals, which typically possess greater infrastructural resources and organizational maturity, are more likely to adopt specialized ICUs. Furthermore, large medical complexes often implement critical care organizations (CCOs) to coordinate multiple ICUs, a strategy shown to enhance both operational efficiency and cost-effectiveness through centralized governance and resource optimization (16–19).
Compared to prior studies, the present research encompassed a more extensive selection of specialized ICUs and for multi-level hierarchical data, multilevel models effectively control for random effects at the hospital level. In the multilevel analysis including all patients, bed capacity (compared to bed >500) and hospitals’ teaching status had no significant effect on hospital mortality. Admission to cardiac ICU (OR: 0.81, p < 0.001) and surgical ICU (OR: 0.88, p = 0.041) were associated with lower mortality risks, and admission to Neuro-ICU (OR: 1.32, p < 0.001) was linked to a higher risk of mortality compared to general ICU. In the subgroup and mediation analyses, discordant ICU triage was associated with lower in-hospital mortality in the CVA group (OR: 0.69, p = 0.018), and this relationship appears to be mediated by the use of vasopressors (OR = 0.999, 95% CI: 0.998 to 1.000; p = 0.029). In contrast, in the trauma group, concordant ICU triage was associated with lower in-hospital mortality (OR: 1.87, p = 0.005), which was partially mediated by mechanical ventilation (indirect effect = −0.019, p < 0.001). These findings indicate that the impact of different ICU types on mortality is associated with interventions within the ICU. In baseline analysis (Supplementary Tables 2–6), Neuro-ICUs exhibited lower rates of organ support therapy, potentially reflecting their greater focus on specialized conditions (neurological disorders). However, this specialization may also imply reduced or delayed interventions for other complications, thereby influencing mortality rates. The higher utilization of life-support interventions like mechanical ventilation in SICUs compared to discordant ICUs may correlate with lower mortality rates, particularly for trauma patients who often require more advanced life support. It should be noted that discordant screening often involves selection bias, as patients with more severe CVAs (such as those with significant cerebral infarction or cerebral edema) may be more likely to be transferred to the neurocritical care unit for specialized treatment, thereby influencing in-hospital mortality rates.
Recent advancements in evidence-based critical care guidelines and bundled management protocols have enhanced comprehensive patient management in ICUs (20–22). Critically ill patients often develop multi-organ complications, including infections, shock, AKI, and ARDS. This requires the specialty ICU team to acquire more skills and knowledge in different specialties. And it makes the boundaries between specialty and general ICUs gradually blurred (23). General ICUs demonstrate superior resource efficiency through optimized monitoring-intervention workflows, particularly in multi-organ dysfunction management (11). In contrast, specialty ICUs may lead to resource redundancy due to focusing on specialized diseases, especially when admit non-ideal patients, resource utilization efficiency will be further reduced (3, 11). At the same time, specialty ICUs may provide more detailed assessment and monitoring of the corresponding specialty disease. This improves diagnostic accuracy, but may also prolong ICU LOS (24). Therefore, tertiary hospitals require specialty ICUs for academic and operational needs. However, physician training and resource allocation should balance specialty-specific competencies with universal critical care principles. Interdepartmental collaboration must be prioritized to enhance clinical adaptability and establish synergistic resource-sharing mechanisms between general and specialty ICUs.
Several factors limited the present study. Despite utilizing multilevel models to mitigate the impact of random effects at the hospital level, the study subjects still exhibit a certain degree of heterogeneity due to the inherent limitations of retrospective research. Our study based on only one country’s current situation of ICUs. The research conclusions may vary depending on the medical, economic, dietary, and ethnic characteristics of different countries and regions. As the data for this study were sourced from publicly available databases in developed countries, the findings may be more applicable to resource-rich nations capable of establishing specialized ICUs. We also suggest that in resource-limited scenarios, discordant triage may be an acceptable option without compromising patient safety. The follow-up research may need to collect detailed clinical data from more countries and areas to obtain research results that are further consistent with the real world. In addition, to avoid the heterogeneity of the research population, we also need to focus on particular patient populations to verify the conclusions for specific specialties.
5. Conclusion
Our study demonstrates that specialized concordant ICU care does not associate with an in-hospital survival advantage over general ICU care after adjusting for confounders. Consequently, discordant triage may be acceptable in resource-limited environments without compromising patient safety. However, trauma patients derived a survival benefit from concordant triage, potentially mediated by differences in the mechanical ventilation application. These findings further emphasize the importance of prioritizing specialized trauma centers in medical resource planning. Future prospective studies are needed to determine if specific patient subgroups might still benefit from specialized triage.
Glossary
Glossary
- ICUs
Intensive care units
- eICU-CRD
eICU Collaborative Research Database
- US
United States
- APACHE
Acute Physiology and Chronic Health Evaluation
- GICU
general ICU
- AKI
acute kidney injury
- CVA
cerebrovascular accident
- ACS
acute coronary syndrome
- CICU
cardiology ICU
- MICU
medical ICU
- Neuro ICU
neurology ICU
- LOC
length of stay
- SICU
surgery ICU
- CTICU
cardiothoracic ICU
- CSICU
cardiac surgical ICU
- COPD
chronic obstructive pulmonary disease
- ANOVA
one-way analysis of variance
- CCO
critical care organization
- ARDS
acute respiratory distress syndrome
- CABG
coronary artery bypass graft
Funding Statement
The author(s) declared that financial support was received for this work and/or its publication. The research was supported by Guangdong Clinical Research Center for Critical Care Medicine (Grant No. 2020B1111170005).
Edited by: Gabriele Melegari, University Hospital of Modena, Italy
Reviewed by: Camila Vantini Capasso Palamim, Sao Francisco University, Brazil
Denny Ruku, Mount Klabat College, Indonesia
Data availability statement
Publicly available datasets were analyzed in this study. This data can be found at: https://eicu-crd.mit.edu/gettingstarted/access/.
Ethics statement
Ethical approval was not required for the study involving humans in accordance with the local legislation and institutional requirements. Written informed consent to participate in this study was not required from the participants or the participants’ legal guardians/next of kin in accordance with the national legislation and the institutional requirements.
Author contributions
RD: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Software, Validation, Visualization, Writing – original draft, Writing – review & editing. JX: Conceptualization, Investigation, Methodology, Supervision, Writing – review & editing. KL: Conceptualization, Methodology, Supervision, Validation, Writing – review & editing. XS: Data curation, Resources, Software, Supervision, Validation, Writing – review & editing. XH: Methodology, Validation, Visualization, Writing – review & editing. YZ: Formal analysis, Resources, Supervision, Validation, Writing – review & editing. CC: Conceptualization, Formal analysis, Methodology, Supervision, Validation, Writing – review & editing.
Conflict of interest
The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Correction note
This article has been corrected with minor changes. These changes do not impact the scientific content of the article.
Generative AI statement
The author(s) declared that Generative AI was not used in the creation of this manuscript.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
Supplementary material
The Supplementary material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmed.2026.1785019/full#supplementary-material
References
- 1.Halpern NA, Pastores SM. Critical care medicine in the United States 2000-2005: an analysis of bed numbers, occupancy rates, payer mix, and costs. Crit Care Med. (2010) 38:65–71. doi: 10.1097/CCM.0b013e3181b090d0, [DOI] [PubMed] [Google Scholar]
- 2.Halpern NA, Goldman DA, Tan KS, Pastores SM. Trends in critical care beds and use among population groups and Medicare and Medicaid beneficiaries in the United States: 2000-2010. Crit Care Med. (2016) 44:1490–9. doi: 10.1097/CCM.0000000000001722, [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Nguyen YL, Milbrandt EB. Are specialized ICUs so special? Crit Care. (2009) 13:314. doi: 10.1186/cc8116, [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Shortell SM, Zimmerman JE, Rousseau DM, Gillies RR, Wagner DP, Draper EA, et al. The performance of intensive care units: does good management make a difference? Med Care. (1994) 32:508–25. doi: 10.1097/00005650-199405000-00009, [DOI] [PubMed] [Google Scholar]
- 5.Mirski MA, Chang CW, Cowan R. Impact of a neuroscience intensive care unit on neurosurgical patient outcomes and cost of care: evidence-based support for an intensivist-directed specialty ICU model of care. J Neurosurg Anesthesiol. (2001) 13:83–92. doi: 10.1097/00008506-200104000-00004, [DOI] [PubMed] [Google Scholar]
- 6.Diringer MN, Edwards DF. Admission to a neurologic/neurosurgical intensive care unit is associated with reduced mortality rate after intracerebral hemorrhage. Crit Care Med. (2001) 29:635–40. doi: 10.1097/00003246-200103000-00031, [DOI] [PubMed] [Google Scholar]
- 7.Patel HC, Menon DK, Tebbs S, Hawker R, Hutchinson PJ, Kirkpatrick PJ. Specialist neurocritical care and outcome from head injury. Intensive Care Med. (2002) 28:547–53. doi: 10.1007/s00134-002-1235-4, [DOI] [PubMed] [Google Scholar]
- 8.Fagerberg B, Blomstrand C. Do stroke units save lives? Lancet. (1993) 342:992. doi: 10.1016/0140-6736(93)92039-V, [DOI] [PubMed] [Google Scholar]
- 9.Killip T, 3rd, Kimball JT. Treatment of myocardial infarction in a coronary care unit. A two year experience with 250 patients. Am J Cardiol. (1967) 20:457–64. doi: 10.1016/0002-9149(67)90023-9, [DOI] [PubMed] [Google Scholar]
- 10.Evans R, 3rd, LeBailly S, Gordon KK, Sawyer A, Christoffel KK, Pearce B. Restructuring asthma care in a hospital setting to improve outcomes. Chest. (1999) 116:210s–6s. doi: 10.1378/chest.116.suppl_2.210S, [DOI] [PubMed] [Google Scholar]
- 11.Lott JP, Iwashyna TJ, Christie JD, Asch DA, Kramer AA, Kahn JM. Critical illness outcomes in specialty versus general intensive care units. Am J Respir Crit Care Med. (2009) 179:676–83. doi: 10.1164/rccm.200808-1281OC, [DOI] [PubMed] [Google Scholar]
- 12.Zimmerman JE, Kramer AA, Knaus WA. Changes in hospital mortality for United States intensive care unit admissions from 1988 to 2012. Crit Care. (2013) 17:R81. doi: 10.1186/cc12695, [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Pollard TJ, Johnson AEW, Raffa JD, Celi LA, Mark RG, Badawi O. The eICU collaborative research database, a freely available multi-center database for critical care research. Sci Data. (2018) 5:180178. doi: 10.1038/sdata.2018.178 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Zimmerman JE, Kramer AA, McNair DS, Malila FM. Acute physiology and chronic health evaluation (APACHE) IV: hospital mortality assessment for today's critically ill patients. Crit Care Med. (2006) 34:1297–310. doi: 10.1097/01.CCM.0000215112.84523.F0, [DOI] [PubMed] [Google Scholar]
- 15.Feeney T, Hartwig FP, Davies NM. How to use directed acyclic graphs: guide for clinical researchers. BMJ. (2025) 388:e078226. doi: 10.1136/bmj-2023-078226 [DOI] [PubMed] [Google Scholar]
- 16.Leung S, Gregg SR, Coopersmith CM, Layon AJ, Oropello J, Brown DR, et al. Critical care organizations: business of critical care and value/performance building. Crit Care Med. (2018) 46:1–11. doi: 10.1097/CCM.0000000000002696, [DOI] [PubMed] [Google Scholar]
- 17.Moore JE, Oropello JM, Stoltzfus D, Masur H, Coopersmith CM, Nates J, et al. Critical care organizations: building and integrating academic programs. Crit Care Med. (2018) 46:e334–41. doi: 10.1097/CCM.0000000000002917, [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Pastores SM, Halpern NA, Oropello JM, Kostelecky N, Kvetan V. Critical care organizations in academic medical centers in North America: a descriptive report. Crit Care Med. (2015) 43:2239–44. doi: 10.1097/CCM.0000000000001200, [DOI] [PubMed] [Google Scholar]
- 19.Lilly CM, Oropello JM, Pastores SM, Coopersmith CM, Khan RA, Sessler CN, et al. Workforce, workload, and burnout in critical care organizations: survey results and research agenda. Crit Care Med. (2020) 48:1565–71. doi: 10.1097/CCM.0000000000004552, [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.De Backer D, Deutschman CS, Hellman J, Myatra SN, Ostermann M, Prescott HC, et al. Surviving sepsis campaign research priorities 2023. Crit Care Med. (2024) 52:268–96. doi: 10.1097/CCM.0000000000006135, [DOI] [PubMed] [Google Scholar]
- 21.Levy MM, Evans LE, Rhodes A. The surviving Sepsis campaign bundle: 2018 update. Crit Care Med. (2018) 46:997–1000. doi: 10.1097/CCM.0000000000003119, [DOI] [PubMed] [Google Scholar]
- 22.Umemura Y, Abe T, Ogura H, Fujishima S, Kushimoto S, Shiraishi A, et al. Hour-1 bundle adherence was associated with reduction of in-hospital mortality among patients with sepsis in Japan. PLoS One. (2022) 17:e0263936. doi: 10.1371/journal.pone.0263936, [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Santacruz CA, Pereira AJ, Celis E, Vincent JL. Which multicenter randomized controlled trials in critical care medicine have shown reduced mortality? A systematic review. Crit Care Med. (2019) 47:1680–91. doi: 10.1097/CCM.0000000000004000, [DOI] [PubMed] [Google Scholar]
- 24.Toptas M, Sengul Samanci N, Akkoc İ, Yucetas E, Cebeci E, Sen O, et al. Factors affecting the length of stay in the intensive care unit: our clinical experience. Biomed Res Int. (2018) 2018:1–4. doi: 10.1155/2018/9438046 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
Publicly available datasets were analyzed in this study. This data can be found at: https://eicu-crd.mit.edu/gettingstarted/access/.