Abstract
-
Background
- In Korea, intensive care units (ICUs) are defined mainly by administrative categories, which do not reflect functional capability. This study examined the relationship between hospital capacity and functional performance by analyzing nationwide patterns of volume, capability, and institutional variation.
-
Methods
- We analyzed adult emergency department-to-ICU admissions in 378 hospitals from 2016 to 2023. Admission and life-support intervention (LSI) volumes were measured; organ-supportive therapies such as mechanical ventilation and renal replacement therapy were considered LSIs, and the LSI rate (LSI admissions/total admissions) was used as a proxy for functional capability. Generalized additive models assessed non-linear relationships with bed capacity, and residual-based analysis was used to classify hospitals into functional phenotypes.
-
Results
- Critical care delivery was highly concentrated; the top 8% of hospitals provided 50% of LSIs. Admission and LSI volumes increased with hospital size but plateaued beyond approximately 1,300 beds, while the LSI rate continued to rise until approximately 1,700 beds. Hospitals of similar size showed wide variation in both volume and capability. Based on residuals, 42.4% of hospitals formed a central reference group, while others were classified as high-volume/high-capability (13.7%), low-volume/high-capability (15.0%), low-volume/low-capability (12.0%), or high-volume/low-capability (16.9%).
-
Conclusions
- The relationship between structural capacity and functional performance in Korean ICUs is complex and non-linear. Hospital size alone is insufficient to predict emergency critical care delivery, underscoring the importance of operational models alongside structural resources. These findings provide a foundation for capability-based ICU stratification to guide resource allocation and strengthen system resilience.
-
Key Words: critical care; hospital bed capacity; intensive care units; life support care
INTRODUCTION
The intensive care unit (ICU) is an organized system that provides enhanced monitoring and multi-modal organ support for patients with life-threatening physiological instability [1,2]. Consequently, ICU performance cannot be assessed by structural indicators like bed capacity alone; a multidimensional approach considering operational resource management and outcomes is necessary [3]. To address this complexity, the "levels of care" (LoC) framework, which characterizes ICU capability based on intervention intensity rather than physical location, was developed. The World Federation of Societies of Intensive and Critical Care Medicine and the United Kingdom Intensive Care Society describes tiered systems ranging from basic monitoring to advanced multi-organ support [1,2]. While LoC classifications align resources with patient needs, they do not directly measure care delivery [1,4]. However, in regions like Ontario, Canada, these systems are operationalized using life-support intervention (LSI) thresholds for objective stratification [5].
In Korea, ICUs are classified by administrative attributes such as hospital type or bed capacity, with no reference to functional capability frameworks. Policy efforts focus on institution-wide quality assessments and value-based payments, which influence staffing and outcomes like mortality [6,7]. While supporting accountability, these institutional-level initiatives fail to reflect variation in functional resource deployment. Consequently, empirical evidence describing the functional landscape of Korean ICUs remains limited.
Establishing a capability-based classification requires empirical evidence of differences in functional performance. Before defining LoC criteria, it is necessary to map ICU activities and assess if structural capacity explains delivery variation. For Korea to develop an internationally comparable framework, a consistent reference population is needed. Since approximately 60% of Korean ICU admissions are through the emergency department (ED), which represents unplanned acute illness [8], we aimed to characterize the functional landscape of Korean critical care by examining patient volume and capability, providing foundational evidence for a capability-aware ICU classification system.
MATERIALS AND METHODS
The Ethics Committee of the Chungnam National University Sejong Hospital approved this study (No. CNUSH-2025-01-003). The requirement for informed consent was waived because the identities of all patients in the National Emergency Department Information System (NEDIS) were concealed.
Data Source
We used data from the NEDIS, a nationwide database established under Article 15 of the Emergency Medical Service Act to support Korea’s emergency and critical care system. NEDIS collects standardized, visit-level data from approximately 400 participating EDs across the country. The dataset includes information on patient demographics, hospital identifiers, ED level, mode of arrival, triage acuity, timestamps of arrival and disposition, diagnoses, procedures, and clinical outcomes. Data submission is mandatory for all participating institutions and is conducted electronically, with automated validation systems in place to ensure completeness and consistency. During the study period, the annual participation rate remained high, ranging from 98.8% to 100%. Details on the database structure and quality assurance procedures have been described previously [9-13].
Study Population and Patient-Level Variables
We included patients aged ≥18 years who were admitted to an ICU from the ED between January 1, 2016, and December 31, 2023. Eligible patients were identified using standardized ED disposition codes for ICU admission. Because NEDIS captures only ED-originating encounters, the study population was restricted to patients admitted to an ICU directly from the ED. Patients admitted through non-ED pathways (e.g., elective surgery, ward transfers) were not represented in this dataset and were therefore not considered in the analyses.
For all included visits, we obtained patient-level variables from NEDIS. Demographic variables included age and sex. We also collected information on mode of ED arrival (walk-in or emergency medical services [EMS]), interhospital transfer status, and in-hospital mortality. Comorbidity burden was assessed using the Charlson Comorbidity Index (CCI) derived from discharge diagnoses and categorized as 0, 1, 2, or ≥3.
Life-Support Interventions
LSIs were defined as organ-supportive therapies delivered in response to acute physiological failure. Following previous literature, LSIs included invasive arterial pressure monitoring, high-flow nasal cannula (HFNC), invasive or non-invasive mechanical ventilation, renal replacement therapy (RRT), and extracorporeal membrane oxygenation (ECMO) [1,5,14]. This definition aligns with international classification frameworks where aggregated LSI data serve as a validated proxy for determining institutional critical care capability [5]. Each intervention was counted if administered during the ICU stay following ED admission, including those initiated in the ED and continued in the ICU. Patients receiving at least one LSI were classified as having received high-intensity critical care.
Hospital-Level Aggregation
To examine institutional patterns, patient-level data were aggregated to construct a hospital-level dataset. For each hospital, we calculated the total number of ED-to-ICU admissions. For other characteristics, including patient demographics, arrival modes, receipt of LSIs, and in-hospital mortality, we computed both absolute counts and their proportions relative to each hospital’s admission volume. Procedure-specific rates were also calculated for each of the five LSI components. The LSI rate—the proportion of ED-to-ICU patients receiving at least one intervention—was used as a proxy for each hospital’s functional capability.
Hospital-Level Structural Characteristics
Hospital-level structural information was obtained directly from institutional fields embedded in the NEDIS dataset. Variables included hospital type (tertiary, general, or community hospital), ED designation (regional emergency medical center, local emergency medical center, or local emergency medical institution), and total licensed bed count. The number of licensed beds was used as a standardized proxy for hospital resources, as it generally reflects parallel increases in ICU and ED capacity, specialist coverage, and service scope. To account for year-to-year variation, the average licensed bed count for each hospital was calculated using December registry data from 2016 to 2023. This approach reduces bias from temporal fluctuations, providing a stable representation of each institution’s structural scale. ICU bed capacity was categorized into five groups (<300, 300–599, 600–799, 800–999, and ≥1000 beds), based on statutory criteria for general hospitals in Korea and a previously validated nationwide classification associated with differences in clinical outcomes, including ED length of stay and mortality [9]. Hospitals that closed or lost emergency care designation before 2023 were excluded, leaving 378 hospitals that participated throughout the study period and had at least one ED-to-ICU admission in the final analysis.
Statistical Analysis
Hospitals were categorized into five groups based on their average number of licensed beds, and group-level differences in ICU-related indicators were compared. Variables of interest included LSIs, CCI distributions, referral characteristics, and in-hospital mortality. Values were summarized as medians with interquartile ranges, and group comparisons were performed using the Kruskal-Wallis test. Ordered trends across bed capacity groups were assessed using the Jonckheere-Terpstra test. Supplementary analyses used hospital type and ED level as alternative grouping variables, applying the same approach.
To evaluate the concentration of ICU care delivery, hospitals were ranked by their volumes of ED-to-ICU admissions, total LSIs, each LSI component, and key referral-related indicators. Cumulative distribution curves were generated, and the degree of concentration was quantified using the Herfindahl-Hirschman Index (HHI) and Gini coefficient. The HHI, defined as the sum of squared market shares, reflects the extent to which ICU services are concentrated in a small number of hospitals, with higher values indicating greater concentration. The Gini coefficient ranges from 0 to 1 and measures inequality in the distribution of services, where 0 represents perfect equality and 1 represents maximal concentration within a single institution.
Associations between ICU bed capacity and outcomes were modeled using generalized additive models (GAMs) with thin-plate smoothing splines. To account for overdispersion in count data, a negative binomial family with a log link was used. For rate-based outcomes, a quasibinomial family with a logit link was applied to constrain values to the 0–1 range while handling extra-binomial variation. These models served to: (1) characterize non-linear associations and inflection points relative to structural capacity, and (2) generate residuals representing deviations from expected performance. Model fit was evaluated using the proportion of deviance explained (analogous to adjusted R²), and smoothing parameters were estimated using restricted maximum likelihood to limit overfitting and improve generalizability.
For in-hospital mortality, funnel plots were used to assess whether variation across hospitals could be attributed to random fluctuation. Plots were constructed with the admission-weighted national mean as the reference, and 95% and 99.8% binomial control limits were applied, truncated to the [0,1] range.
Residuals were standardized to z-scores to ensure comparability. To identify distinct functional patterns, we defined a central reference region using the minimum covariance determinant (MCD) estimator [15]. The MCD identifies the tightly clustered subset of residuals representing the national baseline, ensuring the reference region reflects typical behavior unaffected by extreme structural outliers. The reference region was defined as the 90% bivariate tolerance region based on Mahalanobis distances relative to the robust MCD center. Assuming a chi-square distribution with two degrees of freedom, this threshold delineates typical hospitals [16]. Hospitals outside this region were classified into four functional quadrants (I–IV) based on whether their volume and capability residuals fell above or below model-based expectations. Sensitivity analysis using a 95% region was performed to assess robustness to the threshold choice.
All analyses were performed using SAS 9.4 (SAS Institute) for data preparation and R version 4.3.3 (R Foundation for Statistical Computing) for statistical analysis and visualization.
RESULTS
Hospital and Patient Characteristics by Bed Capacity
A total of 378 hospitals met inclusion criteria. A detailed flowchart of the hospital inclusion process is provided in Figure 1. Average licensed bed capacity was as follows: 209 hospitals with <300 beds, 90 with 300–599, 37 with 600–799, 27 with 800–999, and 15 with ≥1,000 beds. Table 1 summarizes hospital characteristics and treatment capability by bed group. Median ED-to-ICU admission volume increased steadily with hospital size (P<0.001; P for trend<0.001). Larger hospitals admitted a higher proportion of male patients (P<0.001; P for trend<0.001) and a lower proportion of those aged ≥65 years (P<0.001; P for trend=0.001). Referral patterns also varied: EMS arrivals showed modest differences across groups (P=0.038) without a clear trend, whereas interhospital transfers were substantially more frequent in larger hospitals (P<0.001 for both group comparison and trend). Comorbidity burden rose with hospital size. The proportion of patients with CCI 0 declined, while the proportion of patients with CCI 1, 2, or ≥3 increased in a stepwise manner (all P<0.001). Treatment capability showed graded differences. The overall LSI rate increased across bed groups (P<0.001; P for trend<0.001), and procedure-specific rates for invasive arterial monitoring, HFNC, mechanical ventilation, RRT, and ECMO were all higher in larger hospitals (all P<0.001). Notably, the complexity of care intensified with hospital size; while 90.0% of patients in hospitals with <300 beds required no LSIs, only 48.8% of those with ≥1,000 beds did so, with the latter group showing a substantially higher proportion of patients requiring multiple simultaneous interventions (Supplementary Figure 1). In-hospital mortality did not differ significantly between groups (P=0.099), but showed a modest upward trend across ordered categories (P for trend=0.011). Results were consistent when hospitals were stratified by type under the Medical Service Act and by ED designation (Supplementary Tables 1 and 2, respectively).
Distribution and Variability of ICU Activity
Figure 2 shows the distributions of ED-to-ICU admission volume, total LSI volume, and the proportion of ED-to-ICU patients receiving an LSI, stratified by hospital bed group. Across all panels, median values increased monotonically with bed capacity, yet considerable within-group variability persisted, as indicated by overlapping interquartile ranges between adjacent categories. The degree of overlap suggests that bed capacity alone does not fully explain variation in these measures. Corresponding distributions by hospital type under the Medical Service Act and by ED designation are provided in Supplementary Figures 2 and 3, respectively.
Concentration of Critical Care Delivery
Figure 3 shows Pareto curves for hospital-level ED-to-ICU admissions and total LSI volume. In both panels, cumulative shares rose steeply among the highest-ranked hospitals. Concentration indices confirmed these patterns, with total LSI delivery more concentrated than ED-to-ICU admissions (Gini, 0.776 vs 0.628, respectively; HHI, 123.6 vs 72.3, respectively). Half of all admissions occurred in the top 50 hospitals (13.2%), while half of all LSIs were delivered by the top 31 hospitals (8.2%). The top 103 and 62 hospitals accounted for 75% of national admissions and LSIs.
Non-linear Relationship with Hospital Bed Capacity
Figure 4 shows GAM fits for the association between average licensed bed capacity and ED-to-ICU admission volume, total LSI volume, and LSI rate. In all panels, predicted values increased with bed capacity before reaching inflection points clustered around 1,300–1,700 beds, depending on the metric (ED-to-ICU admission volume approximately 1,300; LSI volume approximately 1,400; LSI rate approximately 1,700). Markers on each curve indicate the bed capacities at which the fitted value reaches 40%, 50%, 60%, and 90% of its panel-specific peak; across panels, these thresholds corresponded to approximately 479–1,097 beds. Across panels, 95% confidence bands were narrowest at mid-range capacities and widened toward both extremes. The institutional density was highest between 400 and 1,000 beds, whereas observations above 1,500 beds were sparse, resulting in significantly broader confidence intervals in the extreme upper range. Model fit, expressed as explained variance, was 0.644 for ED-to-ICU admission volume, 0.658 for total LSI volume, and 0.543 for LSI rate. As a supplementary analysis, a funnel plot of in-hospital mortality against ED-to-ICU admission volume showed a national mean of 14.7%, with 95% and 99.8% control limits narrowing as hospital volume increased (Supplementary Figure 4).
Residual-Based Functional Phenotypes of Hospitals
Applying the residual-based quadrant analysis (Figure 5), we identified distinct functional phenotypes based on deviations from model-based expectations. Overall, 42.4% of hospitals fell within the 90% MCD ellipse (central reference region), indicating functional activity consistent with their structural scale. The remaining hospitals were categorized into four quadrants based on whether their volume and LSI rate were higher or lower than expected given their bed capacity: quadrant I (high volume/high LSI rate, 13.7%), quadrant II (low volume/high LSI rate, 15.0%), quadrant III (low volume/low LSI rate, 12.0%), and quadrant IV (high volume/low LSI rate, 16.9%).
Large hospitals (≥600 beds) were more frequently located in quadrants I, II, or IV, whereas small hospitals (<300 beds) comprised the central reference group (83.7%). Among 600–799 bed hospitals, quadrant II was the most common phenotype (37.8%), characterized by a pattern of high-acuity care delivery despite lower-than-expected admission volumes. In contrast, quadrant III was most frequent among 300–599 bed hospitals (23.3%). As a sensitivity analysis, we repeated the classification using a 95% MCD coverage region to evaluate the robustness of the boundaries; results were consistent (Supplementary Figure 5). Detailed distributions by hospital bed group are provided in Supplementary Table 3.
DISCUSSION
In this nationwide analysis, we show that while critical care volume and functional capability generally scale with hospital size, the relationship is distinctly non-linear, plateauing beyond 1,300–1,700 beds. Crucially, significant functional overlap exists across bed capacity groups and administrative categories, confirming that nominal size or administrative status are imprecise indicators of performance. By isolating this variation via residual analysis, we moved beyond structural proxies to identify four distinct functional phenotypes, ranging from high-volume/high-capability to low-volume/low-capability. This classification reveals that distinct institutional factors, rather than physical scale alone, determine a hospital's role within the unplanned emergency critical care ecosystem.
The distribution of these phenotypes offers insight into diverse operational patterns. While the central reference group (42.4%) aligned with structural expectations, off-diagonal quadrants highlighted divergent behaviors. Quadrant II hospitals (low volume/high capability), many with beds in the 600–799 range, likely reflect selective high-acuity care hospitals that function as specialized hubs, concentrating resources on smaller cohorts rather than maximizing throughput. Conversely, Quadrant IV (high volume/low capability) hospitals likely have lower admission thresholds and prioritize monitoring over active organ support. This pattern parallels prior findings showing that high ICU utilization without commensurate intensity of care often reflects discretionary admission thresholds and heterogeneous triage practices, rather than consistent gains in patient outcomes [17,18].
The substantial variation among hospitals of comparable capacity suggests functional performance is shaped as much by institutional practices as by structural resources. This extends our previous finding, namely that the survival benefit of larger hospitals is likely mediated by unmeasured organizational factors rather than scale alone [9]. As conceptualized by Seymour et al., ICU utilization is driven not only by patient severity and structural hospital characteristics, but also by discrete hospital practices [19]. Our findings empirically support this framework, demonstrating that distinct functional phenotypes exist independent of nominal capacity. This aligns with evidence that hospital-level variation in ICU use persists even after rigorous patient-level adjustment [20]. Similarly, the volume-outcome relationship often diminishes when adjusting for organizational features like staffing and care protocols, suggesting that these operational factors drive outcomes more than nominal capacity alone [21]. Beyond structure, managerial elements, including executive strategies and incentive structures, play a critical role [22,23]. Furthermore, organizational culture is decisive. Studies show that admission thresholds and team organization vary significantly after adjustment [7,17], and weaker safety climates are associated with higher mortality [24]. Consistent with frameworks viewing ICU performance as the interplay of structure and shared norms [25], our findings suggest this residual variation reflects distinct cultural and operational priorities rather than simple resource constraints.
Superimposed on these institutional effects, we observed distinct flexion points depending on the metric. While ED-to-ICU admission volumes leveled off at around 1,300 beds, the LSI rate continued to rise to approximately 1,700 beds. We hypothesize that this divergence reflects a structural shift in service-line composition inherent to mega-hospitals. Although we could not directly quantify elective volume based on our dataset, which was restricted to ED admissions, institutions of this scale typically expand their role in quaternary care, including high-complexity interventions such as organ transplantation and heart valve replacement [26-28]. As the proportion of these scheduled, high-complexity cases increases, the relative share of unplanned ED-based admissions may stabilize, even as overall technical infrastructure, reflected in LSI rates, continues to advance. However, trends in the ≥1,000-bed tier warrant caution given wide confidence intervals. Because most of these institutions fell outside the central reference group, the apparent plateau may conceal divergence into distinct operational strategies, such as quaternary care dominance versus functioning as specialized regional emergency centers, rather than a uniform ceiling in capability.
Korea lacks a capability-based ICU classification system. Hospital-level designations, such as tertiary status and ED level, serve as implicit structural surrogates. However, our findings reveal substantial functional variation within these categories and across bed capacity groups, suggesting that current structural classifications may not adequately capture emergency critical care delivery in practice. Taken together, these findings underscore the limitations of administratively defined structural classifications and suggest the value of moving toward capability-aware frameworks. The observed disconnect between bed capacity and functional output indicates that administrative designations alone may not reliably reflect how critical care resources are deployed in practice. From a policy perspective, our functional phenotypes offer a practical roadmap for system redesign. For instance, Quadrant II (low-volume/high-LSI rate) hospitals could be explored as specialized regional centers, while Quadrant IV (high-volume/low-LSI rate) institutions might benefit from initiatives focused on admission triage and protocol standardization. For hospital administration, these results suggest that strategies to enhance performance should prioritize organizational models and shared norms alongside physical expansion. These phenotypes, which should be viewed as a descriptive framework rather than prescriptive performance categories, need to be further validated using more granular clinical and administrative data. Future studies incorporating patient-level case-mix, staffing levels, and care protocols are essential to elucidate the mechanisms underlying institutional variation and inform the development of robust capability-aware tiering systems. Such evidence would support capability-based ICU stratification that considers both structural resources and demonstrated functional performance.
A key strength of this study is our use of a comprehensive, nationwide dataset spanning eight years. While restricted to ED-to-ICU admissions, our focus was intended to encompass the most representative population of unplanned critical illnesses. As established previously, this pathway accounts for over 200,000 admissions annually, serving as the primary conduit for community-acquired critical illness [9]. Distinct from the controlled scheduling of elective interventions, emergency critical care serves as an essential 'safety net' for unpredictable, life-threatening instability [29,30]. Characterizing these admissions reflects the system's core functional capacity. Using LSIs as a functional proxy moves beyond structural metrics to provide insights into how hospitals meet the demands of acute physiological failure. This approach aligns with international frameworks, notably Ontario's tiered system, which explicitly utilizes aggregated LSI data to designate and regulate adult ICU levels [5].
This study also has several limitations that should be acknowledged. First, we did not directly evaluate patient outcomes in relation to ICU capacity and capability. Although crude mortality patterns were analyzed, the absence of standardized physiological scores precluded robust case-mix adjustment. Consequently, we could not definitively distinguish whether variations in LSI rates stemmed from intrinsic patient severity or differing institutional practices. Second, while LSI metrics served as key indicators, they should be interpreted with caution. Rather than measuring "capability" per se, these rates likely reflect the outcomes of specific operational models, influenced by admission policies, patient mix, and cultural practices regarding the intensity of organ-supportive therapies. Third, focusing exclusively on ED-to-ICU admissions restricted the scope of our analysis. While this approach effectively isolated unplanned critical illness, the exclusion of elective surgical admissions and inter-unit transfers limits the generalizability of our findings. Fourth, residuals reflected all influences beyond structural capacity, not purely operational quality. Thus, the quadrant positions reported represent relative "operational phenotypes" rather than absolute performance grades. This approach is best interpreted as exploratory phenotyping, rather than a normative judgment of hospital quality. Fifth, reliance on administrative data introduced the risk of coding errors; defining LSIs based on procedure codes may not have fully captured intervention timing or completeness, particularly regarding end-of-life care decisions. Finally, the cross-sectional aggregation of data prevented causal inference and assumed stability over time despite potential temporal fluctuations.
In conclusion, we found that hospital performance, defined by functional volume and capability, had a complex, non-linear relationship with structural capacity in Korea. Substantial variation, unexplained by scale alone, underscores that operational models are as critical as physical resources in shaping the ICU’s role. These findings provide empirical foundations for a capability-aware classification system that integrates functional metrics with organizational assessments. Future research should link these phenotypes to patient outcomes, costs, and operational practices to inform targeted policies for capability-based intensive care classification.
KEY MESSAGES
▪ In Korea, delivery of critical care is highly concentrated: half of all admissions from the emergency department to intensive care units occurred in only 13 percent of hospitals, and half of all life-support interventions were delivered by 8 percent of hospitals.
▪ Hospitals with comparable licensed bed capacity often differed markedly in admission and intervention volumes, indicating that structural capacity alone cannot explain differences in emergency critical care performance.
▪ Residual-based analysis revealed four functional patterns, demonstrating that institutional practices, not structural capacity alone, shape critical care performance.
NOTES
-
CONFLICT OF INTEREST
No potential conflict of interest relevant to this article was reported.
-
FUNDING
This research was supported by a grant from the Korean ARPA-H Project through the Korea Health Industry Development Institute, funded by the Ministry of Health & Welfare, Republic of Korea (RS-2024-00512375).
-
ACKNOWLEDGMENTS
None.
-
AUTHOR CONTRIBUTIONS
Conceptualization: HKS, JYM. Data curation: HKS. Methodology: HKS, MO. Formal analysis: HKS. Visualization: HKS. Funding acquisition: JYM. Writing - review & editing: JL, MO, JYK, YKC, JYM. All authors read and agreed to the published version of the manuscript.
SUPPLEMENTARY MATERIALS
Supplementary materials can be found via https://doi.org/10.4266/acc.004225.
Supplementary Figure 2.
Distributions of hospital type emergency department (ED)-to-intensive care unit (ICU) admission volume, life-sustaining intervention (LSI) volume, and LSI rate by hospital type. (A) ED-to-ICU admission volume, (B) Total LSI volume, and (C) Proportion of ED-to-ICU patients receiving an LSI, by hospital bed capacity group. Violin plots depict the distribution of hospital-level values within each bed-capacity group. Boxes indicate the median and interquartile range, and grey points represent individual hospitals.
acc-004225-Supplementary-Figure-2.pdf
Supplementary Figure 3.
Distributions of hospital-level emergency department (ED)-to-intensive care unit (ICU) admission volume, life-sustaining intervention (LSI) volume, and LSI rate by emergency department designation. (A) ED-to-ICU admission volume, (B) total LSI volume, and (C) proportion of ED-to-ICU patients receiving an LSI, by hospital bed capacity group. Violin plots depict the distribution of hospital-level values within each bed-capacity group. Boxes indicate the median and interquartile range, and grey points represent individual hospitals.
acc-004225-Supplementary-Figure-3.pdf
Supplementary Figure 4.
Unadjusted funnel plot of in-hospital mortality by hospital volume. Each point represents a hospital’s unadjusted mortality plotted against its emergency department (ED)-to-intensive care unit (ICU) admissions during 2016–2023. The solid line marks the national mean (14.73%) and dashed/dotted curves indicate the 95% and 99.8% control limits under binomial variation (e.g., approximately 7.8%–21.7% at n=100 and approximately 14.3%–15.1% at n=30,000). No risk adjustment was applied; interpretation should consider differences in case mix.
acc-004225-Supplementary-Figure-4.pdf
Supplementary Figure 5.
Residual-based quadrant classification of hospitals using a 95% minimum covariance determinant coverage region. Hospitals are plotted by standardized residuals from generalized additive models for emergency department (ED)-to-intensive care unit (ICU) admission volume (x-axis) and life-sustaining intervention (LSI) rate (y-axis). The dashed red lines indicate zero residuals. The solid black ellipse represents the 95% tolerance region estimated by the minimum covariance determinant, with the dotted ellipse showing the classical 95% coverage for sensitivity comparison. Points represent individual hospitals, color-coded by bed capacity group (≥1000 beds, 800–999, 600–799, 300–599, <300).
acc-004225-Supplementary-Figure-5.pdf
Figure 1.Flowchart of the hospital inclusion process. ED: emergency department; ICU: intensive care unit.
Figure 2.Distributions of hospital-level emergency department (ED)-to-intensive care unit (ICU) admission volume, life-support intervention (LSI) volume, and LSI rate by hospital bed capacity group. (A) ED-to-ICU admission volume, (B) total LSI volume, and (C) proportion of ED-to-ICU patients receiving an LSI, by hospital bed capacity group. Violin plots depict the distribution of hospital-level values within each bed-capacity group. Boxes indicate the median and interquartile range, and grey points represent individual hospitals.
Figure 3.Concentration of critical care delivery by hospital rank, 2016–2023. (A) Emergency department (ED)-to-intensive care unit (ICU) admissions. (B) Life-support interventions (LSIs). Pareto charts illustrate the distribution of (A) total admissions from the emergency department to the intensive care unit and (B) total LSIs across 378 hospitals. Hospitals are ranked by their respective volumes along the x-axis. Blue bars indicate absolute counts for each hospital (left y-axis), and the red curve represents the cumulative percentage of national totals (right y-axis).
Figure 4.Non-linear associations between hospital bed capacity and critical care measures. (A) Emergency department (ED)-to-intensive care unit (ICU) admission volume. (B) Life-support intervention (LSI) volume. (C) Proportion of ED-to-ICU patients receiving an LSI. Scatter points represent hospital-level observations. The solid line is the generalized additive model smooth with 95% confidence bands (shaded). Diamonds mark the peak of the fitted curve, and colored markers indicate bed capacities at which the fitted value reaches 40%, 50%, 60%, and 90% of the peak. Explained variance is the deviance explained by the generalized additive model (1−residual/null deviance), analogous to R².
Figure 5.Residual-based quadrant classification of hospitals by emergency department (ED)-to-intensive care unit (ICU) admission volume and life-support intervention (LSI) rate. Hospitals are plotted by standardized residuals from generalized additive models for ED-to-ICU admission volume (x-axis) and LSI rate (y-axis). The dashed red lines indicate zero residuals, dividing the plot into four quadrants: I (high volume, high LSI rate), II (low volume, high LSI rate), III (low volume, low LSI rate), and IV (high volume, low LSI rate). The black ellipse represents the 90% tolerance region estimated by the minimum covariance determinant. Points represent individual hospitals, color-coded by bed capacity group.
Table 1.Hospital-level characteristics and life-support intervention patterns by hospital bed capacity group, Korea, 2016–2023
|
Hospital bed capacity |
|
|
|
<300 (n=209) |
300–599 (n=90) |
600–799 (n=37) |
800–999 (n=27) |
≥1000 (n=15) |
P-value |
P for trend |
|
No. of ED-to-ICU admissions |
800 (95–1,802) |
4,032 (1,644–6,448) |
9,458 (5,891–14,111) |
14,235 (8,786–16,038) |
15,307 (11,947–20,445) |
<0.001 |
<0.001 |
|
Male (%) |
52.1 (46.3–57.1) |
57.5 (54.2–59.4) |
59.5 (58.1–61.6) |
60 (59.2–62.2) |
61.3 (61–61.9) |
<0.001 |
<0.001 |
|
Age ≥65 yr (%) |
73.5 (64.1–82.6) |
63.7 (60.2–70.5) |
58.8 (54.7–61.5) |
57.6 (54.9–61.1) |
57.3 (52.2–59.8) |
<0.001 |
0.001 |
|
EMS arrival (%) |
48.2 (36.6–55.8) |
47.9 (42.9–54.6) |
50.1 (45.1–53.7) |
45.8 (38.4–48.8) |
38.4 (32.3–46.2) |
0.038 |
0.325 |
|
Interhospital transfer (%) |
17.2 (7.1–24.7) |
23.5 (17.9–30.0) |
28.2 (25.6–33.9) |
36.2 (31.0–44.2) |
37.0 (32.8–41.9) |
<0.001 |
<0.001 |
|
CCI 0 (%) |
49.7 (41.6–60.4) |
41.9 (36.3–47.1) |
39.5 (34.4–43.5) |
37.7 (32.3–41.8) |
36.0 (32.4–40.1) |
<0.001 |
<0.001 |
|
CCI 1 (%) |
28.0 (22.7–32.4) |
33.8 (30.6–37.2) |
35.5 (33.6–38.0) |
38.4 (35.6–40.0) |
37.3 (29.6–39.0) |
<0.001 |
<0.001 |
|
CCI 2 (%) |
13.0 (8.5–16.8) |
14.5 (12.5–16.8) |
14.5 (13.8–16.9) |
15.3 (12.5–18.3) |
17.5 (14.0–19.3) |
0.001 |
<0.001 |
|
CCI ≥3 |
6.6 (3.4–10.0) |
8.0 (6.0–11.1) |
10.4 (6.4–13.9) |
9.7 (6.7–13.0) |
8.7 (7.3–12.0) |
<0.001 |
<0.001 |
|
LSI rate (%) |
2.6 (0.0–9.7) |
15.0 (8.1–29.1) |
40.5 (21.1–53.5) |
45.2 (30.4–55.8) |
54.3 (35.3–64.3) |
<0.001 |
<0.001 |
|
IAM (%) |
0.0 (0.0–1.6) |
6.0 (0.6–16.0) |
35.4 (8.5–51.7) |
35.9 (22.4–50.5) |
51.2 (24.4–61.8) |
<0.001 |
<0.001 |
|
HFNC (%) |
0.4 (0.0–3.3) |
4.9 (1.4–8.8) |
9.1 (5.4–14.9) |
13.0 (7.2–17.3) |
12.9 (10.5–14.3) |
<0.001 |
<0.001 |
|
MV (%) |
0.2 (0.0–1.9) |
4.1 (1.3–9.3) |
16.5 (5.8–23.5) |
19.2 (14.3–24.1) |
21.6 (19.2–26.3) |
<0.001 |
<0.001 |
|
RRT (%) |
1.0 (0.0–2.8) |
5.2 (2.3–8.0) |
8.6 (5.8–10.3) |
10.7 (5.3–13.7) |
9.5 (7.5–13.4) |
<0.001 |
<0.001 |
|
ECMO (%) |
0.0 (0.0–0.0) |
0.0 (0.0–0.1) |
0.6 (0.2–1.0) |
0.5 (0.3–0.7) |
0.7 (0.4–1.4) |
<0.001 |
<0.001 |
|
In-hospital mortality (%) |
12.9 (7.0–18.8) |
14.5 (12.3–17.7) |
14.8 (13.3–16.8) |
15.3 (13.5–17.9) |
14.3 (13.9–16.3) |
0.099 |
0.011 |
REFERENCES
- 1. Intensive Care Society. Levels of adult critical care: second edition consensus statement [Internet]. Intensive Care Society 2021 [cited 2026 Mar 4]. Available from: https://ics.ac.uk/resource/levels-of-care.html
- 2. Marshall JC, Bosco L, Adhikari NK, Connolly B, Diaz JV, Dorman T, et al. What is an intensive care unit? A report of the task force of the World Federation of Societies of Intensive and Critical Care Medicine. J Crit Care 2017;37:270-6.ArticlePubMed
- 3. Al-Dorzi HM, Arabi YM. Quality indicators in adult critical care medicine. Glob J Qual Saf Healthc 2024;7:75-84.ArticlePubMedPMCPDF
- 4. Kluge GH, Brinkman S, van Berkel G, van der Hoeven J, Jacobs C, Snel YE, et al. The association between ICU level of care and mortality in the Netherlands. Intensive Care Med 2015;41:304-11.ArticlePubMedPDF
- 5. Critical Care Services Ontario. Adult critical care levels of care: guidance document (version 1.0) [Internet]. Critical Care Services Ontario 2020 [cited 2026 Mar 4]. Available from: https://criticalcareontario.ca/resources/adult-critical-care-levels-of-care-guidance-document/
- 6. Han KT, Kim S, Kim GO, Lee S, Kwon YU. Quality control efforts of medical institutions: the impacts of a value-based payment system on medical staff and healthcare-associated infections. J Hosp Infect 2024;153:3-13.ArticlePubMed
- 7. Kim S, Kim GO, Lee S, Kwon YU. Effects of intensive care unit quality assessment on changes in medical staff in medical institutions and in-hospital mortality. Hum Resour Health 2024;22:12.ArticlePubMedPMCPDF
- 8. Oh TK, Kim HG, Song IA. Epidemiologic study of intensive care unit admission in South Korea: a nationwide population-based cohort study from 2010 to 2019. Int J Environ Res Public Health 2022;20:81.ArticlePubMedPMC
- 9. Lee KS, Min HS, Moon JY, Lim D, Kim Y, Ko E, et al. Patient and hospital characteristics predict prolonged emergency department length of stay and in-hospital mortality: a nationwide analysis in Korea. BMC Emerg Med 2022;22:183.ArticlePubMedPMCPDF
- 10. Min HS, Chang HJ, Sung HK. Emergency department utilization of adult cancer patient in Korea: a nationwide population-based study, 2017-2019. Cancer Res Treat 2022;54:680-9.ArticlePubMedPDF
- 11. Lee KS, Han C, Min HS, Lee J, Youn SH, Kim Y, et al. Impact of the early phase of the COVID-19 pandemic on emergency department-to-intensive care unit admissions in Korea: an interrupted time-series analysis. BMC Emerg Med 2024;24:51.ArticlePubMedPMCPDF
- 12. Sung HK, Cho YH, Jeong IS, Kim HS, Kim SJ, Lee JH, et al. Trends and institutional patterns of extracorporeal cardiopulmonary resuscitation for out-of-hospital cardiac arrest in Korea: a nationwide analysis, 2016-2023. Resuscitation 2025;215:110692.ArticlePubMed
- 13. Sung HK, Lee KS, Lee J, Oh MH, Kim JY, Choi YK, et al. Delayed at the door: impact of pandemic response policies on emergency and critical care in Korea: an interrupted time-series analysis. J Korean Med Sci 2026;41:e64. ArticlePubMedPMCPDF
- 14. Donnelly S, Barnett CF, Bohula EA, Chaudhry SP, Chonde MD, Cooper HA, et al. Interhospital variation in admissions managed with critical care therapies or invasive hemodynamic monitoring in tertiary cardiac intensive care units: an analysis from the critical care cardiology trials network registry. Circ Cardiovasc Qual Outcomes 2024;17:e010092. ArticlePubMed
- 15. Fauconnier C, Haesbroeck G. Outliers detection with the minimum covariance determinant estimator in practice. Stat Methodol 2009;6:363-79.Article
- 16. Hardin J, Rocke DM. The distribution of robust distances. J Comput Graph Stat 2005;14:928-46.Article
- 17. Anesi GL, Dress E, Chowdhury M, Wang W, Small DS, Delgado MK, et al. Among-hospital variation in intensive care unit admission practices and associated outcomes for patients with acute respiratory failure. Ann Am Thorac Soc 2023;20:406-13.ArticlePubMedPMCPDF
- 18. Admon AJ, Seymour CW, Gershengorn HB, Wunsch H, Cooke CR. Hospital-level variation in ICU admission and critical care procedures for patients hospitalized for pulmonary embolism. Chest 2014;146:1452-61.ArticlePubMedPMC
- 19. Seymour CW, Iwashyna TJ, Ehlenbach WJ, Wunsch H, Cooke CR. Hospital-level variation in the use of intensive care. Health Serv Res 2012;47:2060-80.PubMedPMC
- 20. Treggiari MM, Martin DP, Yanez ND, Caldwell E, Hudson LD, Rubenfeld GD, et al. Effect of intensive care unit organizational model and structure on outcomes in patients with acute lung injury. Am J Respir Crit Care Med 2007;176:685-90.ArticlePubMedPMCPDF
- 21. Checkley W, Martin GS, Brown SM, Chang SY, Dabbagh O, Fremont RD, et al. Structure, process, and annual ICU mortality across 69 centers: United States Critical Illness and Injury Trials Group Critical Illness Outcomes Study. Crit Care Med 2014;42:344-56.PubMedPMC
- 22. Gasperino J. The Leapfrog initiative for intensive care unit physician staffing and its impact on intensive care unit performance: a narrative review. Health Policy 2011;102:223-8.ArticlePubMed
- 23. Elbus LM, Mostafa MG, Mahmoud FZ, Shaban M, Mahmoud SA. Nurse managers' managerial innovation and it's relation to proactivity behavior and locus of control among intensive care nurses. BMC Nurs 2024;23:485.ArticlePubMedPMCPDF
- 24. Soni K, Minturn JS, Davis BS, Bukowski LA, Kahn JM, Barbash IJ, et al. Variation in corticosteroid prescribing practices for patients with septic shock. Crit Care Explor 2025;7:e1196. ArticlePubMedPMC
- 25. Huang DT, Clermont G, Kong L, Weissfeld LA, Sexton JB, Rowan KM, et al. Intensive care unit safety culture and outcomes: a US multicenter study. Int J Qual Health Care 2010;22:151-61.ArticlePubMedPMC
- 26. Yoo S, Jang EJ, Yi NJ, Kim GH, Kim DH, Lee H, et al. Effect of institutional case volume on in-hospital mortality after living donor liver transplantation: analysis of 7073 cases between 2007 and 2016 in Korea. Transplantation 2019;103:952-8.ArticlePubMed
- 27. Oh HW, Jang EJ, Kim GH, Yoo S, Lee H, Lim TY, et al. Effect of institutional kidney transplantation case-volume on post-transplant graft failure: a retrospective cohort study. J Korean Med Sci 2019;34:e260. ArticlePubMedPMCPDF
- 28. Nam K, Jang EJ, Jo JW, You J, Park JB, Ryu HG, et al. Institutional case volume and mortality after aortic and mitral valve replacement: a nationwide study in two Korean cohorts. J Cardiothorac Surg 2022;17:190.ArticlePubMedPMCPDF
- 29. Yun BJ, Singh MK, Reznek MA, Buehler G, Wolf SJ, Vogel L, et al. Strengthening essential emergency departments: transforming the safety net. Health Aff Sch 2025;3:qxaf044.ArticlePubMedPMCPDF
- 30. Mermiri M, Mavrovounis G, Chatzis D, Mpoutsikos I, Tsaroucha A, Dova M, et al. Critical emergency medicine and the resuscitative care unit. Acute Crit Care 2021;36:22-8.ArticlePubMedPMCPDF
Citations
Citations to this article as recorded by
KSCCM
PubReader
ePub Link
Cite
XML Download
