Skip Navigation
Skip to contents

ACC : Acute and Critical Care

OPEN ACCESS
SEARCH
Search

Articles

Page Path
HOME > Acute Crit Care > Forthcoming articles > Article
Original Article
Rapid response system
Clinical characteristics and risk factors for intensive care unit admission and mortality in rapid response team-activated ward patients: a retrospective cohort study at a tertiary hospital in South Korea

DOI: https://doi.org/10.4266/acc.002782
Published online: October 15, 2026

1Department of Nursing, Keimyung University Dongsan Hospital, Daegu, Korea

2College of Nursing, Keimyung University, Daegu, Korea

Corresponding author: Gaeun Kim College of Nursing, Keimyung University, 1095 Dalgubeol-daero, Dalseo-gu, Daegu 42601, Korea Tel: +82-53-580-3928 Email: eun0325@kmu.ac.kr
• Received: March 24, 2026   • Revised: June 11, 2026   • Accepted: July 6, 2026

© 2026 The Korean Society of Critical Care Medicine

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

  • 729 Views
  • 14 Download
  • Background
    Factors driving intensive care unit (ICU) transfer decisions and mortality after rapid response team (RRT) activation remain unclear. We identified independent predictors of ICU admission, 28-day mortality, and in-hospital mortality in RRT-activated ward patients.
  • Methods
    This retrospective cohort study included adults with first RRT activation from September 2023 to August 2025 at a Korean tertiary hospital (n=549). Multivariable logistic regression with backward elimination was used.
  • Results
    Of 549 patients, 191 (34.8%) were transferred to the ICU. ICU admission was independently predicted by procalcitonin >0.5 ng/ml (adjusted odds ratio [aOR], 2.833; P=0.021) and lower hemoglobin (aOR, 0.808 per g/dl; P=0.040), whereas solid tumor was associated with lower transfer likelihood (aOR, 0.142; P<0.001). Mortality did not differ significantly between ICU and ward groups at 28 days (27.7% vs. 27.7%, P=0.937) or during hospitalization (46.1% vs. 40.2%, P=0.162). Declining peripheral oxygen saturation (SpO₂)/fraction of inspired oxygen (FiO₂) ratio, rising lactate, and hyperkalemia predicted both mortality outcomes. Lower platelet count uniquely predicted 28-day mortality, whereas solid tumor, chronic kidney disease, and deep learning-based continuous cardiac arrest risk score (DeepCARS) additionally predicted in-hospital mortality. ICU admission itself was not independently associated with mortality.
  • Conclusions
    ICU transfer after RRT activation appears to reflect acute reversibility, clinical trajectory, and possible at/after-activation goals-of-care processes rather than severity alone. Mortality was more closely associated with oxygenation failure, hypoperfusion, and electrolyte instability than with disposition.
Clinical deterioration in general ward patients is a major contributor to preventable in-hospital deaths. The increasing prevalence of complex, multimorbid patients on general wards has amplified the frequency of unexpected clinical deterioration, unplanned intensive care unit (ICU) transfers, and in-hospital cardiac arrest [1,2]. The rapid response system (RRS) was introduced to mitigate these events by enabling early identification and intervention before irreversible organ failure [3,4]. The operational arm of the RRS—the rapid response team (RRT)—monitors high-risk patients through automated screening algorithms and direct clinician referral, provides bedside assessment and intervention, and guides subsequent triage decisions regarding ICU transfer or continued ward management [5,6].
Following RRT activation, the clinical trajectory diverges markedly: approximately one-third of patients require ICU admission, while the remainder are managed on the ward with ongoing observation [7-9]. This triage decision is clinically consequential. ICU transfer provides access to invasive monitoring, mechanical ventilation, and vasoactive support, but carries risks of nosocomial infection, delirium, psychological distress, and substantial resource consumption [10,11]. Conversely, delayed or withheld ICU admission in patients who require intensive support is associated with increased mortality [12]. The tension between over-triage and under-triage underscores the need for objective, evidence-based criteria to inform disposition decisions at the point of RRT activation.
Prior RRT studies have predominantly focused on the macroscopic effectiveness of RRS implementation—examining reductions in hospital-wide cardiac arrest rates or overall mortality before and after system introduction [13,14]. Fewer studies have examined the clinical characteristics that differentiate ICU-transferred from ward-remaining patients within the same RRT-activated cohort. Among those that have, cross-sectional comparisons at the time of RRT activation have identified lower oxygenation indices, hemodynamic instability, and elevated lactate as features of the ICU group [7,8]. However, the independent predictive value of individual biomarkers—particularly newer markers such as procalcitonin and artificial intelligence (AI)-based composite scores—for ICU admission decisions has not been systematically evaluated in the RRT context.
Equally important, but less studied, is the question of what determines mortality after RRT activation. A common assumption is that ICU transfer is itself a marker of worse prognosis. Yet the relationship between ICU admission and mortality is confounded by indication: sicker patients are transferred to the ICU precisely because they need intensive support, and this support may partially offset their higher baseline risk. Whether ICU transfer independently predicts mortality after adjusting for physiological severity remains unclear. Furthermore, few studies have simultaneously examined risk factors for both 28-day (short-term) and in-hospital (longer-term) mortality in RRT-activated cohorts, or assessed whether these risk profiles differ—a distinction that could inform the temporal allocation of monitoring and intervention resources.
This study addressed these gaps using data from a tertiary academic hospital in South Korea that operates a 24-hour nurse-led RRT integrated with the deep learning-based continuous cardiac arrest risk score (DeepCARS) AI early warning system. We had three objectives: (1) to compare clinical and physiological characteristics between ICU-transferred and ward-remaining patients at the time of RRT activation; (2) to identify independent risk factors for ICU admission; and (3) to identify independent risk factors for 28-day and in-hospital mortality. We present this study in accordance with the STrengthening the Reporting of OBservational studies in Epidemiology (STROBE) reporting guideline [15].
The study was approved by the Institutional Review Board of Keimyung University Dongsan Hospital (No. DSMC 2025-09-028) and conducted in accordance with the Declaration of Helsinki. Written informed consent was waived owing to the retrospective design and use of anonymized electronic health record (EHR) data.
Study Design and Setting
This was a single-center retrospective cohort study conducted at a tertiary academic hospital (≥900 beds) in Daegu, South Korea. The institution operates a 24-hour nurse-led RRT that responds to automated alerts generated by the DeepCARS deep learning algorithm—a system that continuously computes a cardiac arrest risk score from real-time vital sign data—as well as to direct clinician calls. DeepCARS derives a continuous 0–100 score from four routinely measured vital signs (systolic blood pressure, heart rate [HR], respiratory rate [RR], and body temperature [BT]) together with patient age, and has been developed and externally validated for predicting in-hospital cardiac arrest within 24 hours [16,17].
Participants
All adult inpatients (≥18 years) on general wards who experienced a first RRT activation between September 1, 2023 and August 31, 2025 were screened. Exclusion criteria were: (1) RRT activated during an ongoing cardiopulmonary resuscitation event; and (2) a do-not-resuscitate or treatment-limitation order documented before RRT activation. For patients with multiple RRT activations during a single admission, only the first event was analyzed. Patients were categorized into two groups based on the clinical decision following RRT evaluation: (1) ICU admission group—patients transferred to an ICU within 24 hours of activation; and (2) ward observation group—patients who remained on the general ward. Throughout this report, the terms “ICU admission” and “ICU transfer” are used synonymously to denote transfer from the general ward to an ICU within 24 hours of RRT activation; “ICU admission” is used as the formal outcome label.
Data Collection and Variables
Data were extracted retrospectively from the EHR by trained research nurses using a standardized data collection form. The following variables were recorded for each patient.

Demographic and baseline characteristics

Age, sex, admitting department (medical, surgical, other), and underlying comorbidities (solid tumor, diabetes mellitus, chronic lung disease, cardiovascular disease, hepatobiliary disease, cerebrovascular disease, chronic kidney disease [CKD], and solid organ transplantation) were recorded.

Clinical and physiological parameters at RRT activation (T0)

Oxygen therapy status; DeepCARS score; vital signs (mean arterial pressure [MAP], HR, and peripheral oxygen saturation [SpO2]); and 11 laboratory parameters—SpO₂/fraction of inspired oxygen (FiO2) ratio (S/F ratio), arterial pH, bicarbonate (HCO3–), lactate, hemoglobin (Hb), platelet count, neutrophil-to-lymphocyte ratio (NLR), potassium, creatinine, C-reactive protein (CRP), and procalcitonin—were recorded. The S/F ratio was selected in preference to the partial pressure of arterial oxygen (PaO₂)/FiO₂ ratio because it can be derived entirely from non-invasive, continuously available SpO₂ and FiO₂ data without arterial blood sampling—an approach previously validated for predicting ICU transfer in RRT-activated ward patients [18]. For each patient, laboratory values represent the most recent result available within the 24 hours preceding or at the time of RRT activation (T0). Procalcitonin was not measured according to a standardized RRT protocol; it was obtained at the discretion of the treating team as part of routine clinical care, generally when infection was clinically suspected, and the value recorded at T0 was the most recent result available at or before activation. Because procalcitonin assay turnaround time precludes immediate point-of-activation measurement in most cases, the recorded values predominantly reflect testing performed before RRT activation rather than real-time sampling at the moment of triage.

Outcome variables

Outcome variables were ICU admission within 24 hours of RRT activation, 28-day mortality, and in-hospital mortality.
Statistical Analysis
Descriptive statistics are presented as mean±standard deviation for continuous variables and frequency with percentage for categorical variables. Univariate between-group comparisons used independent-samples t-test or Mann-Whitney U-test for continuous variables and chi-squared or Fisher’s exact test for categorical variables, as appropriate.
To identify independent risk factors for ICU admission, 28-day mortality, and in-hospital mortality, multivariable logistic regression was performed for each outcome. Candidate variables were selected in two stages: first, all variables reaching P≤0.20 in univariate analysis were included; second, clinically important variables identified from prior literature were added to prevent omission of relevant predictors [19,20]. Backward elimination was applied to derive the final model for each outcome. Procalcitonin was dichotomized at the clinically established threshold of 0.5 ng/ml and was available in a subset of 130 patients at T0; this variable was entered into the ICU admission model given its clinical relevance but was excluded from the mortality models owing to the small subsample size. Because procalcitonin was available in only 130 patients, the multivariable model for ICU admission that incorporated procalcitonin was estimated by complete-case analysis restricted to these 130 patients; no imputation was undertaken for the remaining patients, and the corresponding goodness-of-fit and discrimination statistics in Table 1 refer to this 130-patient subsample. We did not compute the diagnostic performance of procalcitonin (area under the receiver operating characteristic curve, sensitivity, specificity, or a data-derived cutoff); the 0.5 ng/ml threshold was prespecified from established literature rather than derived from the present data. Multicollinearity was assessed using the variance inflation factor, with all values confirmed below 10. Model fit was evaluated using the Hosmer-Lemeshow goodness-of-fit test. Results are reported as adjusted odds ratios (aORs) with 95% CIs. All analyses were performed with IBM SPSS Statistics version 23.0 (IBM Corp.). Statistical significance was set at two-sided P<0.05.
Study Population and Baseline Characteristics at RRT Activation
A total of 549 patients met the inclusion criteria (Figure 1): 191 (34.8%) were transferred to the ICU and 358 (65.2%) remained on the ward. Baseline characteristics and clinical parameters at the time of RRT activation (T0) are presented in Table 2. The ICU group was significantly younger than the ward group (68±14 vs. 71±12 years, P=0.025). Sex distribution did not differ between groups (P=0.426). Medical department patients comprised a higher proportion in the ward group, with significant between-group differences in departmental distribution (P=0.028). Solid tumor (45.8% vs. 27.7%, P<0.001) and chronic lung disease (20.9% vs. 10.5%, P=0.002) were more prevalent in the ward group, whereas hepatobiliary disease (16.8% vs. 8.9%, P=0.007) and solid organ transplantation (6.3% vs. 2.2%, P=0.016) were more prevalent in the ICU group.
At the time of RRT activation, the ICU group exhibited significantly worse physiological profiles: higher oxygen therapy use (92.7% vs. 80.2%, P<0.001), lower MAP (65.52±39.82 vs. 80.09±26.14 mm Hg, P<0.001), lower SpO₂ (90.78±11.49% vs. 93.37±10.02%, P=0.016), a markedly lower S/F ratio (167.05±130.15 vs. 260.26±138.40, P<0.001), and higher lactate (4.02±3.90 vs. 2.98±2.95 mmol/L, P=0.002). The proportion of patients with procalcitonin >0.5 ng/ml was significantly higher in the ICU group (65.2% vs. 42.9%, P=0.015). DeepCARS score was paradoxically higher in the ward group than in the ICU group (91.12±10.19 vs. 89.14±12.40, P=0.047), likely because the algorithm’s vital-sign-based architecture generates different risk-score phenotypes for the hemodynamic profiles predominant in each group. Other laboratory parameters, including pH, HCO₃⁻, Hb, platelet count, NLR, potassium, creatinine, and CRP, did not differ significantly between groups.
Mortality Outcomes
Overall 28-day mortality was 27.7% (152/549) and in-hospital mortality was 42.3% (232/549). Neither outcome differed significantly between the ICU and ward groups: 28-day mortality was 27.7% vs. 27.7% (P=0.937) and in-hospital mortality was 46.1% vs. 40.2% (P=0.162), respectively.
Risk Factors for ICU Admission
Multivariable logistic regression results for ICU admission are presented in Table 1. The model showed adequate fit (Hosmer-Lemeshow χ²=5.295, df=8, P=0.726; Nagelkerke R²=0.291). Three variables independently predicted ICU admission. Procalcitonin >0.5 ng/ml was the strongest positive predictor, with a nearly three-fold increase in ICU admission risk (aOR, 2.833; 95% CI, 1.169–6.866; P=0.021). Hb was inversely associated, with each 1 g/dl increase reducing admission risk by approximately 19% (aOR, 0.808; 95% CI, 0.659–0.991; P=0.040). Solid tumor history was the strongest inverse predictor, reducing the adjusted odds of ICU admission by 86% (aOR, 0.142; 95% CI, 0.055–0.371; P<0.001). Variables that were significant in univariate analysis—including oxygen therapy, S/F ratio, and lactate—did not retain significance after multivariable adjustment.
Risk Factors for 28-Day Mortality
Multivariable logistic regression results for 28-day mortality are presented in Table 3. The model showed adequate fit (Hosmer-Lemeshow χ²=9.044, df=8, P=0.339; Nagelkerke R²=0.398). Four variables were independently associated with 28-day mortality. Lactate was associated with a 26.5% increase in mortality risk per 1 mmol/L increment (aOR, 1.265; 95% CI, 1.053–1.519; P=0.012). Potassium showed a nearly two-fold risk increase per 1 mmol/L (aOR, 1.972; 95% CI, 1.034–3.760; P=0.039). Higher S/F ratio was protective (aOR, 0.993; 95% CI, 0.989–0.997; P=0.001), as was higher platelet count (aOR, 0.995; 95% CI, 0.991–0.999; P=0.012). ICU admission itself was not an independent predictor of 28-day mortality (univariate odds ratio [OR], 1.046; P=0.823).
Risk Factors for In-Hospital Mortality
Multivariable logistic regression results for in-hospital mortality are presented in Table 4. The model showed adequate fit (Hosmer-Lemeshow χ²=8.966, df=8, P=0.345; Nagelkerke R²=0.400). Six variables were independently associated with in-hospital mortality. Solid tumor history carried the highest risk (aOR, 5.342; 95% CI, 1.900–15.019; P=0.001), followed by CKD (aOR, 5.100; 95% CI, 1.257–20.700; P=0.023). Lactate (aOR, 1.211; 95% CI, 1.018–1.440; P=0.031) and potassium (aOR, 2.190; 95% CI, 1.096–4.376; P=0.026) were consistent risk factors. S/F ratio was protective (aOR, 0.996; 95% CI, 0.993–0.999; P=0.010). Notably, DeepCARS score was independently associated with in-hospital mortality, with each one-point increase conferring a 5.7% risk increment (aOR, 1.057; 95% CI, 1.001–1.117; P=0.048). ICU admission was not an independent predictor of in-hospital mortality (univariate OR, 1.312; P=0.133).
This study provides, to our knowledge, the first simultaneous analysis of ICU admission predictors and dual-timeframe mortality risk factors in a cohort of RRT-activated ward patients integrated with an AI-based early warning system. Three principal findings emerged. First, ICU admission decisions were associated with factors reflecting acute reversibility (procalcitonin, Hb) and disease trajectory (solid tumor), rather than physiological severity alone. Second, mortality after RRT activation was most consistently associated with core physiological derangements—oxygenation failure, tissue hypoperfusion, and electrolyte instability—across both disposition groups. Third, ICU transfer itself was not independently associated with either 28-day or in-hospital mortality after multivariable adjustment. Given the divergent baseline risk profiles of the two groups, this association-level finding should be read as a caution against equating disposition with prognosis, not as evidence that ICU transfer has no causal effect on outcome.
The observed ICU transfer rate of 34.8% is consistent with prior single-center [7] and multicenter [8] studies in South Korea reporting rates of 35.5% and 41.1%, respectively, confirming that approximately one in three RRT-activated patients requires ICU-level care. At the time of activation, the ICU group displayed the expected profile of acute organ dysfunction: lower MAP, lower S/F ratio, higher lactate, and higher procalcitonin, consistent with acute respiratory failure and hemodynamic instability as primary drivers of ICU transfer [7,8,21].
The finding that procalcitonin >0.5 ng/ml independently predicted ICU admission (aOR, 2.833) after adjusting for S/F ratio, lactate, and other covariates is clinically noteworthy. Procalcitonin is an established biomarker for bacterial sepsis severity [22,23], and its elevation in emergency settings has been linked to ICU admission and mortality [24,25]. However, its predictive value for ICU disposition specifically in the context of ward-based RRT activation has rarely been evaluated. Our data suggest that elevated procalcitonin at RRT activation identifies a subset of patients with systemic infection requiring vasoactive support and intensive monitoring—a clinical phenotype for which ward-level care is insufficient. This positions procalcitonin as a potential adjunctive marker of a bacterial-sepsis phenotype rather than as a validated standalone triage criterion, and several caveats temper this interpretation. Procalcitonin is characterized by high specificity but comparatively limited sensitivity, aligning it more with confirmation of bacterial infection than with the early, undifferentiated detection that is the core aim of an RRT; values may also remain low in sepsis of fungal, viral, or other non-bacterial origin, creating a risk of under-triage if a “negative” result is over-interpreted. Moreover, assay turnaround time generally precludes real-time measurement at the moment of triage, and we did not evaluate its diagnostic performance (area under the curve, sensitivity, specificity, or an internally derived cutoff). Procalcitonin should therefore be regarded as a hypothesis-generating, supplementary signal that may complement—but not replace—clinical assessment and vital-sign-based criteria; its incremental value as a triage adjunct requires prospective evaluation incorporating formal diagnostic-accuracy analysis.
The inverse association between Hb and ICU admission (aOR, 0.808 per 1 g/dl increase) implicates acute anemia—likely reflecting hemorrhagic shock from gastrointestinal or postprocedural bleeding—as an underappreciated trigger for ICU transfer. Hb decline compromises oxygen delivery, exacerbates tissue hypoxia, and necessitates transfusion and hemodynamic stabilization in a monitored environment [26]. While some prior studies found no association between Hb and ICU transfer [7], international evidence consistently identifies anemia as a predictor of clinical deterioration and ICU admission [27,28], supporting our findings. The discrepancy may reflect differences in institutional transfusion thresholds and transfer protocols, underscoring the need for multicenter validation.
The most striking ICU admission predictor was the inverse association with solid tumor (aOR, 0.142), indicating that cancer patients were approximately 86% less likely to be transferred to the ICU after RRT activation. This finding warrants careful interpretation. It does not indicate that cancer patients are less severely ill; indeed, solid tumor was the strongest independent predictor of in-hospital mortality (aOR, 5.342). Because patients with a do-not-resuscitate or treatment-limitation order documented before RRT activation were excluded, the solid-tumor patients analyzed here were, in principle, candidates for active treatment, and the lower transfer odds are therefore unlikely to be explained by pre-existing care limitations alone. We interpret this association cautiously as reflecting, at least in part, a dynamic goals-of-care process in which treatment goals are re-evaluated at the moment of RRT activation, rather than a fixed values-based exclusion from intensive care. In patients with advanced solid tumors, RRT activation often serves as a catalyst for re-evaluating treatment goals, transitioning from aggressive intervention to comfort-focused or palliative care [29]. This interpretation is supported by multicenter Korean data reporting similar patterns [8] and international literature documenting the role of RRT in facilitating end-of-life discussions [30]. An important caveat is that solid tumor was analyzed as a single category and was not stratified by stage or metastatic status. Early-stage or postoperative cancer patients—who are typically candidates for aggressive intensive care—and patients with advanced metastatic disease, whose goals of care differ fundamentally, were necessarily grouped together. This heterogeneity, combined with the fact that goals-of-care decisions made at or after RRT activation were not systematically recorded (see Limitations), means that the aOR of 0.142 should be regarded as a composite signal requiring confirmation in stage-stratified analyses rather than as definitive evidence of a values-based triage mechanism. From a health system perspective, this finding demonstrates that RRT functions not merely as a “rescue” mechanism but as a clinical decision node that integrates acute physiology with long-term prognosis and patient preferences.
The equivalence of 28-day (27.7% vs. 27.7%) and in-hospital (46.1% vs. 40.2%) mortality between the ICU and ward groups is counterintuitive but clinically plausible. The ICU group entered with significantly worse acute physiology but received intensive support; the ward group was less acutely ill at activation but had a higher burden of irreversible comorbidity, particularly solid tumor. These opposing risk profiles may have produced similar crude mortality. In the multivariable models, ICU admission was not independently associated with either mortality outcome. This result must, however, be interpreted in light of confounding by indication: the sicker ICU group received intensive support that may have partly offset its higher baseline risk, whereas the ward group included patients with irreversible disease for whom escalation may not have been pursued and who may have received ward-based palliative care. The comparable mortality therefore cannot be taken as evidence that ICU care was effective, nor that disposition is unimportant; it is equally consistent with a balance between treatment benefit in the ICU group and a poorer-prognosis subgroup managed conservatively on the ward—a configuration that can introduce selection and potential lead-time bias. Because the two groups had divergent risk profiles and were not exchangeable, residual confounding precludes causal inference about the effect of ICU transfer on mortality. This finding aligns with growing recognition that the prognostic value of RRT disposition lies not in the transfer decision itself, but in the physiological and comorbid substrate of the patient [31,32].
Three physiological derangements—declining S/F ratio, rising lactate, and hyperkalemia—emerged as shared independent risk factors for both 28-day and in-hospital mortality. This convergence identifies a core pathophysiological triad—oxygenation failure, tissue hypoperfusion, and electrolyte instability—that drives mortality irrespective of disposition or time horizon. The S/F ratio, a non-invasive surrogate for PaO₂/FiO₂ ratio, has been validated as a severity marker in acute respiratory failure and correlates with ICU transfer risk [18,21]. Lactate elevation reflects anaerobic metabolism secondary to tissue hypoperfusion and is among the most robust predictors of mortality in critical illness [33,34]. Hyperkalemia signals renal dysfunction, metabolic acidosis, and cardiac conduction risk, and has been independently associated with mortality in ICU populations [35,36]. The concurrent presence of all three at RRT activation defines a high-risk phenotype warranting immediate, aggressive intervention regardless of the disposition decision.
The differential mortality predictors between the two time horizons are clinically informative. Twenty-eight-day mortality was uniquely associated with lower platelet count (aOR, 0.995 per ×10³/μl), a marker of consumptive coagulopathy, disseminated intravascular coagulation, and sepsis-related organ failure [37,38]. This finding identifies thrombocytopenia as a possible early signal of the acute physiological crisis—the immediate threat to survival within the first month. By contrast, in-hospital mortality was additionally associated with non-modifiable comorbidities (solid tumor, CKD) and the DeepCARS score. Solid tumor carried the highest mortality risk (aOR, 5.342), consistent with its role as a marker of disease irreversibility and potential treatment-limitation decisions [39]. CKD (aOR, 5.100) reflects the cumulative burden of metabolic derangement, fluid imbalance, and cardiovascular risk that compounds mortality over a longer hospital stay [40]. Notably, the direction of the CKD association reversed between univariate (OR, 0.713) and multivariable (aOR, 5.100) analysis, a suppression effect in which CKD’s independent mortality risk was masked in the unadjusted model. The principal suppressor was solid tumor: in this cohort, patients with CKD were less likely to carry a concurrent solid-tumor diagnosis—the single strongest driver of in-hospital mortality (aOR, 5.342)—so that the lower tumor burden among CKD patients plausibly pulled the crude estimate below unity (OR, 0.713). Once solid tumor and the acute physiological markers retained in the final model (S/F ratio, lactate, and potassium) were adjusted for, this negative confounding was removed and CKD emerged as an independent predictor (aOR, 5.100). Because all variance inflation factors in the in-hospital mortality model were below the conventional threshold of 10 and tolerance values were >0.5, this univariate-to-multivariable reversal is more consistent with negative confounding/suppression by solid tumor than with estimation instability arising from collinearity. The independent predictive value of DeepCARS for in-hospital mortality (aOR, 1.057 per point) is notable: despite not significantly predicting ICU admission in the multivariable model, the AI score captured a longer-term deterioration signal that transcended the immediate triage decision. This suggests that AI-based early warning systems may be most valuable not for binary disposition decisions but for continuous risk stratification throughout the hospital stay [16,17,41]. This interpretation also addresses the paradoxical observation that DeepCARS scores were lower in the ICU group despite its greater acute severity. In our institution, the triage decision after RRT activation is physician-led and is driven primarily by indicators of acute, reversible organ dysfunction—hypoxemia and a falling S/F ratio, hemodynamic instability and elevated lactate, acute anemia, and the anticipated need for mechanical ventilation or vasoactive support—rather than by the DeepCARS score, which is not applied as a binary transfer threshold. Because DeepCARS is a vital-sign-based algorithm trained to predict cardiac arrest, it may under-weight the acute organ-failure phenotypes (e.g., hypoxemic respiratory failure or hemorrhagic shock) that prompt immediate ICU transfer but that do not yet present as the vital-sign pattern most predictive of imminent arrest; conversely, ward patients with persistent vital-sign derangement may accrue high scores without meeting transfer criteria. DeepCARS is therefore best interpreted as a complementary instrument for continuous risk stratification and escalation prompting, used alongside explicit assessment of acute organ failure, rather than as a standalone triage tool. This interpretation is directly corroborated by our companion longitudinal study [42], in which the DeepCARS score showed a significantly steeper rise over the 24 hours preceding activation among ICU-bound patients (time×group interaction: B=0.073, P=0.028), despite a lower absolute value 24 hours before activation, indicating that the algorithm captures the velocity of deterioration rather than its absolute level.
Several limitations should be acknowledged. First, this is a single-center retrospective study at a tertiary academic hospital in South Korea with a DeepCARS-integrated RRT; generalizability to institutions with different RRT models, patient populations, or early warning systems requires multicenter validation. In particular, the disposition patterns observed here are partly shaped by this institution’s specific RRT operating protocol, bed-availability constraints, and local transfusion and escalation thresholds, which may not transfer to other settings. Second, the retrospective design precludes causal inference; the associations identified are predictive, not causative. Third, procalcitonin was available in only a subset of patients (130 at T0), which may have limited statistical power and introduced selection bias if procalcitonin was preferentially ordered in patients with suspected infection. We also did not assess the diagnostic performance of procalcitonin (area under the curve, sensitivity, or specificity) and applied a literature-based rather than a data-derived threshold, so its incremental value as a triage adjunct remains to be established. Fourth, we did not capture the intensity or timing of interventions administered between RRT activation and outcome assessment; such treatments (fluid resuscitation, vasopressors, antibiotics) may have modified the natural trajectory. Fifth, the cross-sectional analysis at the moment of RRT activation captures a snapshot rather than a trajectory; the companion study examining 24-hour physiological trajectories provides complementary longitudinal insights [42]. Sixth, treatment-limitation decisions and transitions to palliative care during the hospital stay were not systematically captured, which may have confounded the in-hospital mortality analysis, particularly for the solid tumor subgroup. Seventh, DeepCARS is validated for cardiac arrest prediction; its application to mortality prediction represents an extrapolation that requires prospective validation. Eighth, the comparison of mortality between the ICU and ward groups is subject to confounding by indication; because patients were not randomly allocated to disposition and the groups had divergent baseline risk profiles, the absence of an independent association between ICU transfer and mortality cannot be interpreted as evidence of a causal—or null—treatment effect. Ninth, the overall sample size, and the smaller numbers within key subgroups, limited statistical power, produced wide confidence intervals for several estimates (notably the comorbidity ORs), and precluded planned subgroup analyses such as procalcitonin stratified by infection source or solid tumor stratified by stage; these findings should accordingly be regarded as hypothesis-generating and require validation in larger, multicenter cohorts.
In conclusion, in this cohort of 549 RRT-activated ward patients, ICU admission was associated with markers of acute reversible crisis—elevated procalcitonin suggesting bacterial sepsis and lower Hb suggesting possible acute blood loss—whereas solid tumor was associated with lower transfer likelihood, possibly reflecting disease trajectory and at/after-activation goals-of-care processes. Mortality was more closely associated with three core physiological derangements—oxygenation failure (S/F ratio), tissue hypoperfusion (lactate), and electrolyte instability (potassium)—than with treatment location in the adjusted models. Comorbid burden (solid tumor, CKD) and the DeepCARS score added independent prognostic value for in-hospital mortality, while lower platelet count uniquely predicted short-term death.
These findings carry three implications for clinical practice. First, RRT triage protocols may benefit from considering procalcitonin and Hb as adjuncts to—rather than replacements for—traditional vital-sign criteria, recognizing that procalcitonin has limited sensitivity and requires prospective evaluation, including formal diagnostic-accuracy assessment, before any routine adoption as a triage criterion. Second, risk-factor–directed management at the point of RRT activation—targeting oxygenation, perfusion, and electrolyte correction—should be prioritized irrespective of the disposition decision, as these derangements determine mortality in both groups. Third, the observed decoupling between ICU transfer and mortality challenges the use of ICU admission rates as a quality metric for RRT performance; mortality-based outcome measures stratified by baseline risk may provide more meaningful benchmarks. Prospective validation of these predictors in multicenter cohorts, integration into automated triage algorithms, and evaluation of their impact on patient outcomes should be priorities for future research.
▪ In 549 rapid response team (RRT)-activated patients, intensive care unit (ICU) admission was independently predicted by elevated procalcitonin, lower hemoglobin, and absence of solid tumor, suggesting that triage reflects acute reversibility, disease trajectory, and possible at/after-activation goals-of-care processes rather than severity alone.
▪ Mortality was more closely associated with physiological deterioration—declining peripheral oxygen saturation (SpO₂)/fraction of inspired oxygen (FiO₂) ratio, rising lactate, and hyperkalemia—than with ICU disposition in the adjusted models.
▪ Clinical practice should prioritize risk-factor–directed management at RRT activation and incorporate artificial intelligence-based scores such as deep learning-based continuous cardiac arrest risk score (DeepCARS) for continuous risk stratification rather than binary triage.

CONFLICT OF INTEREST

No potential conflict of interest relevant to this article was reported.

FUNDING

None.

ACKNOWLEDGMENTS

The authors thank the nursing staff of the Rapid Response Team at Keimyung University Dongsan Hospital for their dedication to patient safety and the integrity of clinical records.

AUTHOR CONTRIBUTIONS

Conceptualization: GK. Methodology: GK. Formal analysis: SG. Data curation: SG. Writing - original draft: SG, BK. Writing - review & editing: BK, GK. All authors read and agreed to the published version of the manuscript.

Figure 1.
STrengthening the Reporting of OBservational studies in Epidemiology (STROBE) participant flow diagram. Of 632 patients initially assessed, 83 were excluded (cardiopulmonary resuscitation [CPR] ongoing at activation, n=21; pre-existing do-not-resuscitate [DNR] order, n=62). The final analytical cohort comprised 549 patients: 191 (34.8%) transferred to the intensive care unit (ICU) and 358 (65.2%) managed on the ward.
acc-002782f1.jpg
Table 1.
Risk factors for ICU admission after RRT activation: univariate and multivariable logistic regression (n=549)
Variable Univariate OR (95% CI) P-value Multivariable aOR (95% CI) P-value
Solid tumor 0.438 (0.299–0.641) <0.001 0.142 (0.055–0.371) <0.001
Oxygen therapy 3.099 (1.696–5.664) <0.001 - -
S/F ratio 0.995 (0.993–0.996) <0.001 0.998 (0.995–1.001) 0.116
Lactate (mmol/L) 1.094 (1.035–1.156) 0.001 - -
Solid organ transplantation 2.958 (1.188–7.367) 0.020 - -
Procalcitonin >0.5 ng/ml 2.500 (1.187–5.266) 0.016 2.833 (1.169–6.866) 0.021
Age (yr) 0.985 (0.971–0.998) 0.024 - -
Hepatobiliary disease 1.926 (1.139–3.258) 0.015 - -
Chronic lung disease 0.445 (0.263–0.756) 0.003 - -
Hemoglobin (g/dl) 1.022 (0.989–1.056) 0.201 0.808 (0.659–0.991) 0.040
DeepCARS score 0.984 (0.969–1.000) 0.046 - -

The reference group was "no" for categorical variables and ≤0.5 ng/ml for procalcitonin. Hosmer-Lemeshow χ²=5.295 (df=8, P=0.726); Cox & Snell R²=0.213; Nagelkerke R²=0.291. The multivariable model incorporating procalcitonin was estimated by complete-case analysis restricted to the 130 patients with available procalcitonin (no imputation was performed for the remaining patients); the goodness-of-fit and discrimination statistics above refer to this 130-patient subsample, whereas univariate ORs were computed using all available data for each variable.

ICU: intensive care unit; RRT: rapid response team; OR: odds ratio; aOR: adjusted odds ratio; S/F ratio: SpO₂/FiO₂ ratio; SpO₂: peripheral oxygen saturation; FiO₂: fraction of inspired oxygen; DeepCARS: deep learning-based continuous cardiac arrest risk score.

Table 2.
Clinical characteristics and laboratory parameters at RRT activation (T0) (n=549)
Variable All patients (n=549) ICU group (n=191) Ward group (n=358) χ² or t P-value
Demographic and baseline characteristics
 Age (yr) 70±13 68±14 71±12 −2.250 0.025
 Male sex 323 (58.8) 108 (56.5) 215 (60.1) 0.643 0.426
Admitting department 7.168 0.028
 Medical 376 (68.5) 117 (61.3) 259 (72.3)
 Surgical 108 (19.7) 47 (24.6) 61 (17.0)
 Other 65 (11.8) 27 (14.1) 38 (10.6)
Underlying comorbidity
 Solid tumor 217 (39.5) 53 (27.7) 164 (45.8) 17.213 <0.001
 Diabetes mellitus 225 (41.0) 71 (37.2) 154 (43.0) 1.759 0.185
 Chronic lung disease 95 (17.3) 20 (10.5) 75 (20.9) 9.557 0.002
 Cardiovascular disease 339 (61.7) 118 (61.8) 221 (61.7) <0.001 0.991
 Hepatobiliary disease 64 (11.7) 32 (16.8) 32 (8.9) 7.387 0.007
 Cerebrovascular disease 139 (25.3) 50 (26.2) 89 (24.9) 0.114 0.735
 Chronic kidney disease 109 (19.9) 43 (22.5) 66 (18.4) 1.301 0.254
 Solid organ transplantation 20 (3.6) 12 (6.3) 8 (2.2) 5.814 0.016
Clinical characteristics at RRT activation
 Oxygen therapy 464 (84.5) 177 (92.7) 287 (80.2) 14.878 <0.001
 DeepCARS score 90.44±11.02 89.14±12.40 91.12±10.19 −1.990 0.047
Vital sign
 MAP (mm Hg) 75.02±32.30 65.52±39.82 80.09±26.14 - <0.001
 HR (bpm) 108.78±28.69 106.56±28.59 109.78±28.72 - 0.246
 SpO₂ (%) 92.57±10.56 90.78±11.49 93.37±10.02 - 0.016
Laboratory parameter
 S/F ratio 227.83±142.60 167.05±130.15 260.26±138.40 - <0.001
 pH 7.35±0.36 7.33±0.16 7.37±0.44 - 0.281
 HCO₃⁻ (mEq/L) 22.93±8.03 22.26±9.18 23.33±7.22 - 0.151
 Lactate (mmol/L) 3.37±3.37 4.02±3.90 2.98±2.95 - 0.002
 Hemoglobin (g/dl) 10.58±7.34 11.26±12.01 10.21±2.20 - 0.234
 Platelet (×10³/μl) 188.28±125.13 180.86±114.80 192.31±130.38 - 0.309
 NLR 15.44±19.65 14.28±20.60 16.06±19.12 - 0.316
 Potassium (mmol/L) 4.23±0.93 4.23±0.99 4.22±0.90 - 0.924
 Creatinine (mg/dl) 1.79±1.94 1.96±2.09 1.70±1.86 - 0.134
 CRP (mg/dl) 10.72±9.88 9.76±8.58 11.26±10.52 - 0.108
Procalcitonin (ng/ml)a) 5.946 0.015
 ≤0.5 64 (49.2) 16 (34.8) 48 (57.1)
 >0.5 66 (50.8) 30 (65.2) 36 (42.9)
Outcome
 28-Day mortality 152 (27.7) 53 (27.7) 99 (27.7) 0.006 0.937
 In-hospital mortality 232 (42.3) 88 (46.1) 144 (40.2) 1.960 0.162

Values are presented as mean±standard deviation or number (%).

RRT: rapid response team; ICU: intensive care unit; DeepCARS: deep learning-based continuous cardiac arrest risk score; MAP: mean arterial pressure; HR: heart rate; SpO₂: peripheral oxygen saturation; S/F ratio: SpO₂/FiO₂ ratio; FiO₂: fraction of inspired oxygen; HCO₃⁻: bicarbonate; NLR: neutrophil-to-lymphocyte ratio; CRP: C-reactive protein.

a)Procalcitonin available in 130 patients.

Table 3.
Risk factors for 28-day mortality after RRT activation: univariate and multivariable logistic regression (n=549)
Variable Univariate OR (95% CI) P-value Multivariable aOR (95% CI) P-value
S/F ratio 0.996 (0.994–0.997) <0.001 0.993 (0.989–0.997) 0.001
Lactate (mmol/L) 1.198 (1.125–1.275) <0.001 1.265 (1.053–1.519) 0.012
Potassium (mmol/L) 1.253 (1.030–1.526) 0.024 1.972 (1.034–3.760) 0.039
Platelet (×10³/μl) 0.999 (0.997–1.001) 0.204 0.995 (0.991–0.999) 0.012
Oxygen therapy 3.307 (1.661–6.586) 0.001 - -
Solid tumor 1.667 (1.142–2.433) 0.008 - -
Cardiovascular disease 0.549 (0.376–0.803) 0.002 - -
DeepCARS score 1.016 (0.997–1.036) 0.108 1.062 (0.990–1.139) 0.095
ICU admission 1.046 (0.707–1.547) 0.823 - -

The reference group for categorical variables was "no." Hosmer-Lemeshow χ²=9.044 (df=8, P=0.339); Cox & Snell R²=0.282; Nagelkerke R²=0.398.

RRT: rapid response team; OR: odds ratio; aOR: adjusted odds ratio; S/F ratio: SpO₂/FiO₂ ratio; SpO₂: peripheral oxygen saturation; FiO₂: fraction of inspired oxygen; DeepCARS: deep learning-based continuous cardiac arrest risk score; ICU: intensive care unit.

Table 4.
Risk factors for in-hospital mortality after RRT activation: univariate and multivariable logistic regression (n=549)
Variable Univariate OR (95% CI) P-value Multivariable aOR (95% CI) P-value
Solid tumor 1.761 (1.244–2.493) 0.001 5.342 (1.900–15.019) 0.001
Chronic kidney disease 0.713 (0.462–1.101) 0.127 5.100 (1.257–20.700) 0.023
S/F ratio 0.996 (0.995–0.997) <0.001 0.996 (0.993–0.999) 0.010
Lactate (mmol/L) 1.182 (1.106–1.263) <0.001 1.211 (1.018–1.440) 0.031
Potassium (mmol/L) 1.163 (0.967–1.398) 0.109 2.190 (1.096–4.376) 0.026
DeepCARS score 1.025 (1.007–1.043) 0.007 1.057 (1.001–1.117) 0.048
Oxygen therapy 3.454 (1.969–6.058) <0.001 - -
pH 0.051 (0.013–0.196) <0.001 - -
CRP (mg/dl) 1.023 (1.004–1.041) 0.017 1.056 (0.995–1.121) 0.072
ICU admission 1.312 (0.920–1.870) 0.133 - -

The reference group for categorical variables was "no." Hosmer-Lemeshow χ²=8.966 (df=8, P=0.345); Cox & Snell R²=0.300; Nagelkerke R²=0.400. The univariate-to-multivariable reversal of the chronic kidney disease estimate (OR, 0.713 → aOR, 5.100) reflects negative confounding (suppression) by solid tumor, the strongest in-hospital mortality predictor (aOR, 5.342), with which chronic kidney disease was inversely associated in this cohort. Multicollinearity diagnostics for the final model confirmed all variance inflation factors below the conventional threshold of 10 and all tolerance values >0.5, indicating that the reversal is attributable to suppression rather than collinearity.

RRT: rapid response team; OR: odds ratio; aOR: adjusted odds ratio; S/F ratio: SpO₂/FiO₂ ratio; SpO₂: peripheral oxygen saturation; FiO₂: fraction of inspired oxygen; DeepCARS: deep learning-based continuous cardiac arrest risk score; CRP: C-reactive protein; ICU: intensive care unit.

  • 1. Hatem GR, Hatem MG. Non-communicable disease multimorbidity: challenges and solutions. World J Adv Res Rev 2023;17:55-8.Article
  • 2. Ouyang F, Cheng X, Zhou W, He J, Xiao S. Increased mortality trends in patients with chronic non-communicable diseases and comorbid hypertension in the United States, 2000-2019. Front Public Health 2022;10:753861.ArticlePubMedPMC
  • 3. DeVita MA, Hillman K, Bellomo R, Odell M, Jones DA, Winters BD, et al. Textbook of rapid response systems: concept and implementation. 2nd ed. Springer; 2017.
  • 4. Benin AL, Borgstrom CP, Jenq GY, Roumanis SA, Horwitz LI. Republished: defining impact of a rapid response team: qualitative study with nurses, physicians and hospital administrators. Postgrad Med J 2012;88:575-82.ArticlePubMedPMCPDF
  • 5. Lee JR, Jung YK, Hong SB, Huh JW. Predictors of repeat medical emergency team activation in deteriorating ward patients: a retrospective cohort study. J Clin Med 2022;11:1736.ArticlePubMedPMC
  • 6. Eyeington CT, Lloyd-Donald P, Chan MJ, Eastwood GM, Young H, Peck L, et al. Rapid response team review of hemodynamically unstable ward patients: the accuracy of cardiac index assessment. J Crit Care 2019;49:187-92.ArticlePubMed
  • 7. Lee SI, Koh JS, Kim YJ, Kang DH, Lee JE. Characteristics and outcomes of patients screened by rapid response team who transferred to the intensive care unit. BMC Emerg Med 2022;22:18.ArticlePubMedPMCPDF
  • 8. Nam Y, Kang BJ, Hong SB, Jeon K, Lee DH, Kim JS, et al. Characteristics and outcomes of patients screened by the rapid response team and transferred to intensive care unit in South Korea. Sci Rep 2024;14:25061.ArticlePubMedPMCPDF
  • 9. Batterbury A, Douglas C, Coyer F. The illness severity of ward remaining patients reviewed by the medical emergency team: a retrospective cohort study. J Clin Nurs 2023;32:6450-9.ArticlePubMedPDF
  • 10. Dziegielewski C, Skead C, Canturk T, Webber C, Fernando SM, Thompson LH, et al. Delirium and associated length of stay and costs in critically ill patients. Crit Care Res Pract 2021;2021:6612187.ArticlePubMedPMCPDF
  • 11. Santos AC, de Oliveira SL, Macedo VL, Araujo PL, Fraiberg FS, Bastos NF, et al. Intensive care unit prioritization: the impact of ICU bed availability on mortality in critically ill patients who requested ICU admission in court in a Brazilian cohort. J Crit Care 2021;66:126-31.ArticlePubMed
  • 12. Kiekkas P, Tzenalis A, Gklava V, Stefanopoulos N, Voyagis G, Aretha D, et al. Delayed admission to the intensive care unit and mortality of critically ill adults: systematic review and meta-analysis. Biomed Res Int 2022;2022:4083494.ArticlePubMedPMCPDF
  • 13. Lee BY, Hong SB. Rapid response systems in Korea. Acute Crit Care 2019;34:108-16.ArticlePubMedPMCPDF
  • 14. Cho MW, Kim SH, Oh IH, Kim JY, Kim RE, Song JH. Evaluation of rapid response system pilot project and development of a standard model. Health Insurance Review and Assessment Service; 2022.
  • 15. von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP, et al. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet 2007;370:1453-7.Article
  • 16. Kwon JM, Lee Y, Lee Y, Lee S, Park J. An algorithm based on deep learning for predicting in-hospital cardiac arrest. J Am Heart Assoc 2018;7:e008678. ArticlePubMedPMC
  • 17. Lee YJ, Cho KJ, Kwon O, Park H, Lee Y, Kwon JM, et al. A multicentre validation study of the deep learning-based early warning score for predicting in-hospital cardiac arrest in patients admitted to general wards. Resuscitation 2021;163:78-85.ArticlePubMed
  • 18. Kwack WG, Lee DS, Min H, Choi YY, Yun M, Kim Y, et al. Evaluation of the SpO2/FiO2 ratio as a predictor of intensive care unit transfers in respiratory ward patients for whom the rapid response system has been activated. PLoS One 2018;13:e0201632. ArticlePubMedPMC
  • 19. Hosmer DW, Lemeshow S, Sturdivant RX. Applied logistic regression. 3rd ed. Wiley; 2013.
  • 20. Mickey RM, Greenland S. The impact of confounder selection criteria on effect estimation. Am J Epidemiol 1989;129:125-37.ArticlePubMed
  • 21. Lee JR, Choi HR. Analysis of risk factors to predict intensive care unit transfer in medical in-patients. J Korean Biol Nurs Sci 2014;16:259-66.Article
  • 22. Ljungström L, Pernestig AK, Jacobsson G, Andersson R, Usener B, Tilevik D, et al. Diagnostic accuracy of procalcitonin, neutrophil-lymphocyte count ratio, C-reactive protein, and lactate in patients with suspected bacterial sepsis. PLoS One 2017;12:e0181704. ArticlePubMedPMC
  • 23. Malik M, Sreekantan Nair A, Illango J, Siddiqui N, Gor R, Fernando RW, et al. The advancement in detecting sepsis and its outcome: usefulness of procalcitonin in diagnosing sepsis and predicting fatal outcomes in patients admitted to intensive care unit. Cureus 2021;13:e14439. ArticlePubMedPMC
  • 24. Wacker C, Prkno A, Brunkhorst FM, Schlattmann P. Procalcitonin as a diagnostic marker for sepsis: a systematic review and meta-analysis. Lancet Infect Dis 2013;13:426-35.ArticlePubMed
  • 25. Coyne CJ, Castillo EM, Shatsky RA, Chan TC. Procalcitonin as a predictive tool for death and ICU admission among febrile neutropenic patients visiting the emergency department. Medicina (Kaunas) 2022;58:985.ArticlePubMedPMC
  • 26. Rhodes CE, Denault D, Varacallo MA. Physiology, oxygen transport [Internet]. StatPearls Publishing 2022 [cited 2026 Jun 30]. Available from: https://www.ncbi.nlm.nih.gov/books/NBK538336/
  • 27. Alhamdan Q, Alhawas F, Alfandi R, Alsaadi S, Khushayl A, Alzahrani F, et al. Pulmonary embolism and risk factors of ICU admission: systematic review. Int J Med Dev Ctries 2024;8:1517-23.Article
  • 28. Peng H, Su Y, Luo J, Ding N. Association between admission hemoglobin level and prognosis in sepsis patients based on a critical care database. Sci Rep 2024;14:5212.ArticlePubMedPMCPDF
  • 29. Kim JS, Lee MJ, Park MH, Park JY, Kim AJ. Role of the rapid response system in end-of-life care decisions. Am J Hosp Palliat Care 2020;37:943-9.ArticlePubMedPDF
  • 30. Jensen HI, Rasmussen CK, Haberlandt TN, Jensen SS. End-of-life issues experienced by the nurse-led rapid response team: an analysis of extent and experiences. Intensive Crit Care Nurs 2023;76:103411.ArticlePubMed
  • 31. McGaughey J, O'Halloran P, Porter S, Blackwood B. Early warning systems and rapid response to the deteriorating patient in hospital: a systematic realist review. J Adv Nurs 2017;73:2877-91.ArticlePubMedPDF
  • 32. Batterbury A, Douglas C, Coyer F. Patient outcomes following medical emergency team review on general wards: development of predictive models. J Clin Nurs 2024;33:3565-75.ArticlePubMed
  • 33. Jansen TC, van Bommel J, Schoonderbeek FJ, Sleeswijk Visser SJ, van der Klooster JM, Lima AP, et al. Early lactate-guided therapy in intensive care unit patients: a multicenter, open-label, randomized controlled trial. Am J Respir Crit Care Med 2010;182:752-61.ArticlePubMed
  • 34. Nichol AD, Egi M, Pettila V, Bellomo R, French C, Hart G, et al. Relative hyperlactatemia and hospital mortality in critically ill patients: a retrospective multi-centre study. Crit Care 2010;14:R25.ArticlePubMedPMCPDF
  • 35. Hougen I, Leon SJ, Whitlock R, Rigatto C, Komenda P, Bohm C, et al. Hyperkalemia and its association with mortality, cardiovascular events, hospitalizations, and intensive care unit admissions in a population-based retrospective cohort. Kidney Int Rep 2021;6:1309-16.ArticlePubMedPMC
  • 36. Xu C, Luo Y, Chen X, Feng Y. Hyperkalemia is associated with short- and mid-term mortalities in critically ill patients in the MIMIC IV database. Sci Rep 2025;15:6539.ArticlePubMedPMCPDF
  • 37. Ostadi Z, Shadvar K, Sanaie S, Mahmoodpoor A, Saghaleini SH. Thrombocytopenia in the intensive care unit. Pak J Med Sci 2019;35:282-7.ArticlePubMedPMCPDF
  • 38. Setarehaseman A, Mohammadi A, Maitta RW. Thrombocytopenia in sepsis. Life (Basel) 2025;15:274.ArticlePubMedPMC
  • 39. Nasir SS, Muthiah M, Ryder K, Clark K, Niell H, Weir A, et al. ICU deaths in patients with advanced cancer. Am J Hosp Palliat Care 2017;34:173-9.ArticlePubMedPDF
  • 40. Warrens H, Banerjee D, Herzog CA. Cardiovascular complications of chronic kidney disease: an introduction. Eur Cardiol 2022;17:e13. ArticlePubMedPMC
  • 41. Jang H, Yoo W, Hwang S, Lee K. DeepCARS-identified high-risk patients: clinical interventions and outcomes during the Korean healthcare crisis. Medicina (Kaunas) 2025;61:1896.ArticlePubMedPMC
  • 42. Kim B, Gwon S, Kim G. Early physiological changes before rapid response team activation differentiate patients requiring ICU transfer: a retrospective cohort study. J Clin Med 2026;15:3722.ArticlePubMedPMC

Figure & Data

References

    Citations

    Citations to this article as recorded by  

      • PubReader PubReader
      • ePub LinkePub Link
      • Cite
        CITE
        export Copy
        Close
        Download Citation
        Download a citation file in RIS format that can be imported by all major citation management software, including EndNote, ProCite, RefWorks, and Reference Manager.

        Format:
        • RIS — For EndNote, ProCite, RefWorks, and most other reference management software
        • BibTeX — For JabRef, BibDesk, and other BibTeX-specific software
        Include:
        • Citation for the content below
        Clinical characteristics and risk factors for intensive care unit admission and mortality in rapid response team-activated ward patients: a retrospective cohort study at a tertiary hospital in South Korea
        Close
      • XML DownloadXML Download
      Figure
      • 0
      Related articles
      Clinical characteristics and risk factors for intensive care unit admission and mortality in rapid response team-activated ward patients: a retrospective cohort study at a tertiary hospital in South Korea
      Image
      Figure 1. STrengthening the Reporting of OBservational studies in Epidemiology (STROBE) participant flow diagram. Of 632 patients initially assessed, 83 were excluded (cardiopulmonary resuscitation [CPR] ongoing at activation, n=21; pre-existing do-not-resuscitate [DNR] order, n=62). The final analytical cohort comprised 549 patients: 191 (34.8%) transferred to the intensive care unit (ICU) and 358 (65.2%) managed on the ward.
      Clinical characteristics and risk factors for intensive care unit admission and mortality in rapid response team-activated ward patients: a retrospective cohort study at a tertiary hospital in South Korea
      Variable Univariate OR (95% CI) P-value Multivariable aOR (95% CI) P-value
      Solid tumor 0.438 (0.299–0.641) <0.001 0.142 (0.055–0.371) <0.001
      Oxygen therapy 3.099 (1.696–5.664) <0.001 - -
      S/F ratio 0.995 (0.993–0.996) <0.001 0.998 (0.995–1.001) 0.116
      Lactate (mmol/L) 1.094 (1.035–1.156) 0.001 - -
      Solid organ transplantation 2.958 (1.188–7.367) 0.020 - -
      Procalcitonin >0.5 ng/ml 2.500 (1.187–5.266) 0.016 2.833 (1.169–6.866) 0.021
      Age (yr) 0.985 (0.971–0.998) 0.024 - -
      Hepatobiliary disease 1.926 (1.139–3.258) 0.015 - -
      Chronic lung disease 0.445 (0.263–0.756) 0.003 - -
      Hemoglobin (g/dl) 1.022 (0.989–1.056) 0.201 0.808 (0.659–0.991) 0.040
      DeepCARS score 0.984 (0.969–1.000) 0.046 - -
      Variable All patients (n=549) ICU group (n=191) Ward group (n=358) χ² or t P-value
      Demographic and baseline characteristics
       Age (yr) 70±13 68±14 71±12 −2.250 0.025
       Male sex 323 (58.8) 108 (56.5) 215 (60.1) 0.643 0.426
      Admitting department 7.168 0.028
       Medical 376 (68.5) 117 (61.3) 259 (72.3)
       Surgical 108 (19.7) 47 (24.6) 61 (17.0)
       Other 65 (11.8) 27 (14.1) 38 (10.6)
      Underlying comorbidity
       Solid tumor 217 (39.5) 53 (27.7) 164 (45.8) 17.213 <0.001
       Diabetes mellitus 225 (41.0) 71 (37.2) 154 (43.0) 1.759 0.185
       Chronic lung disease 95 (17.3) 20 (10.5) 75 (20.9) 9.557 0.002
       Cardiovascular disease 339 (61.7) 118 (61.8) 221 (61.7) <0.001 0.991
       Hepatobiliary disease 64 (11.7) 32 (16.8) 32 (8.9) 7.387 0.007
       Cerebrovascular disease 139 (25.3) 50 (26.2) 89 (24.9) 0.114 0.735
       Chronic kidney disease 109 (19.9) 43 (22.5) 66 (18.4) 1.301 0.254
       Solid organ transplantation 20 (3.6) 12 (6.3) 8 (2.2) 5.814 0.016
      Clinical characteristics at RRT activation
       Oxygen therapy 464 (84.5) 177 (92.7) 287 (80.2) 14.878 <0.001
       DeepCARS score 90.44±11.02 89.14±12.40 91.12±10.19 −1.990 0.047
      Vital sign
       MAP (mm Hg) 75.02±32.30 65.52±39.82 80.09±26.14 - <0.001
       HR (bpm) 108.78±28.69 106.56±28.59 109.78±28.72 - 0.246
       SpO₂ (%) 92.57±10.56 90.78±11.49 93.37±10.02 - 0.016
      Laboratory parameter
       S/F ratio 227.83±142.60 167.05±130.15 260.26±138.40 - <0.001
       pH 7.35±0.36 7.33±0.16 7.37±0.44 - 0.281
       HCO₃⁻ (mEq/L) 22.93±8.03 22.26±9.18 23.33±7.22 - 0.151
       Lactate (mmol/L) 3.37±3.37 4.02±3.90 2.98±2.95 - 0.002
       Hemoglobin (g/dl) 10.58±7.34 11.26±12.01 10.21±2.20 - 0.234
       Platelet (×10³/μl) 188.28±125.13 180.86±114.80 192.31±130.38 - 0.309
       NLR 15.44±19.65 14.28±20.60 16.06±19.12 - 0.316
       Potassium (mmol/L) 4.23±0.93 4.23±0.99 4.22±0.90 - 0.924
       Creatinine (mg/dl) 1.79±1.94 1.96±2.09 1.70±1.86 - 0.134
       CRP (mg/dl) 10.72±9.88 9.76±8.58 11.26±10.52 - 0.108
      Procalcitonin (ng/ml)a) 5.946 0.015
       ≤0.5 64 (49.2) 16 (34.8) 48 (57.1)
       >0.5 66 (50.8) 30 (65.2) 36 (42.9)
      Outcome
       28-Day mortality 152 (27.7) 53 (27.7) 99 (27.7) 0.006 0.937
       In-hospital mortality 232 (42.3) 88 (46.1) 144 (40.2) 1.960 0.162
      Variable Univariate OR (95% CI) P-value Multivariable aOR (95% CI) P-value
      S/F ratio 0.996 (0.994–0.997) <0.001 0.993 (0.989–0.997) 0.001
      Lactate (mmol/L) 1.198 (1.125–1.275) <0.001 1.265 (1.053–1.519) 0.012
      Potassium (mmol/L) 1.253 (1.030–1.526) 0.024 1.972 (1.034–3.760) 0.039
      Platelet (×10³/μl) 0.999 (0.997–1.001) 0.204 0.995 (0.991–0.999) 0.012
      Oxygen therapy 3.307 (1.661–6.586) 0.001 - -
      Solid tumor 1.667 (1.142–2.433) 0.008 - -
      Cardiovascular disease 0.549 (0.376–0.803) 0.002 - -
      DeepCARS score 1.016 (0.997–1.036) 0.108 1.062 (0.990–1.139) 0.095
      ICU admission 1.046 (0.707–1.547) 0.823 - -
      Variable Univariate OR (95% CI) P-value Multivariable aOR (95% CI) P-value
      Solid tumor 1.761 (1.244–2.493) 0.001 5.342 (1.900–15.019) 0.001
      Chronic kidney disease 0.713 (0.462–1.101) 0.127 5.100 (1.257–20.700) 0.023
      S/F ratio 0.996 (0.995–0.997) <0.001 0.996 (0.993–0.999) 0.010
      Lactate (mmol/L) 1.182 (1.106–1.263) <0.001 1.211 (1.018–1.440) 0.031
      Potassium (mmol/L) 1.163 (0.967–1.398) 0.109 2.190 (1.096–4.376) 0.026
      DeepCARS score 1.025 (1.007–1.043) 0.007 1.057 (1.001–1.117) 0.048
      Oxygen therapy 3.454 (1.969–6.058) <0.001 - -
      pH 0.051 (0.013–0.196) <0.001 - -
      CRP (mg/dl) 1.023 (1.004–1.041) 0.017 1.056 (0.995–1.121) 0.072
      ICU admission 1.312 (0.920–1.870) 0.133 - -
      Table 1. Risk factors for ICU admission after RRT activation: univariate and multivariable logistic regression (n=549)

      The reference group was "no" for categorical variables and ≤0.5 ng/ml for procalcitonin. Hosmer-Lemeshow χ²=5.295 (df=8, P=0.726); Cox & Snell R²=0.213; Nagelkerke R²=0.291. The multivariable model incorporating procalcitonin was estimated by complete-case analysis restricted to the 130 patients with available procalcitonin (no imputation was performed for the remaining patients); the goodness-of-fit and discrimination statistics above refer to this 130-patient subsample, whereas univariate ORs were computed using all available data for each variable.

      ICU: intensive care unit; RRT: rapid response team; OR: odds ratio; aOR: adjusted odds ratio; S/F ratio: SpO₂/FiO₂ ratio; SpO₂: peripheral oxygen saturation; FiO₂: fraction of inspired oxygen; DeepCARS: deep learning-based continuous cardiac arrest risk score.

      Table 2. Clinical characteristics and laboratory parameters at RRT activation (T0) (n=549)

      Values are presented as mean±standard deviation or number (%).

      RRT: rapid response team; ICU: intensive care unit; DeepCARS: deep learning-based continuous cardiac arrest risk score; MAP: mean arterial pressure; HR: heart rate; SpO₂: peripheral oxygen saturation; S/F ratio: SpO₂/FiO₂ ratio; FiO₂: fraction of inspired oxygen; HCO₃⁻: bicarbonate; NLR: neutrophil-to-lymphocyte ratio; CRP: C-reactive protein.

      Procalcitonin available in 130 patients.

      Table 3. Risk factors for 28-day mortality after RRT activation: univariate and multivariable logistic regression (n=549)

      The reference group for categorical variables was "no." Hosmer-Lemeshow χ²=9.044 (df=8, P=0.339); Cox & Snell R²=0.282; Nagelkerke R²=0.398.

      RRT: rapid response team; OR: odds ratio; aOR: adjusted odds ratio; S/F ratio: SpO₂/FiO₂ ratio; SpO₂: peripheral oxygen saturation; FiO₂: fraction of inspired oxygen; DeepCARS: deep learning-based continuous cardiac arrest risk score; ICU: intensive care unit.

      Table 4. Risk factors for in-hospital mortality after RRT activation: univariate and multivariable logistic regression (n=549)

      The reference group for categorical variables was "no." Hosmer-Lemeshow χ²=8.966 (df=8, P=0.345); Cox & Snell R²=0.300; Nagelkerke R²=0.400. The univariate-to-multivariable reversal of the chronic kidney disease estimate (OR, 0.713 → aOR, 5.100) reflects negative confounding (suppression) by solid tumor, the strongest in-hospital mortality predictor (aOR, 5.342), with which chronic kidney disease was inversely associated in this cohort. Multicollinearity diagnostics for the final model confirmed all variance inflation factors below the conventional threshold of 10 and all tolerance values >0.5, indicating that the reversal is attributable to suppression rather than collinearity.

      RRT: rapid response team; OR: odds ratio; aOR: adjusted odds ratio; S/F ratio: SpO₂/FiO₂ ratio; SpO₂: peripheral oxygen saturation; FiO₂: fraction of inspired oxygen; DeepCARS: deep learning-based continuous cardiac arrest risk score; CRP: C-reactive protein; ICU: intensive care unit.


      ACC : Acute and Critical Care
      TOP