Cancer is one of the leading causes of morbidity and mortality worldwide. It is projected that there will be 21.9 million new cases by 2025.1 Advances in oncological diagnosis and treatment have considerably improved patient prognosis, allowing the implementation of more precise and personalized therapies.2 However, the clinical course of these patients remains heterogeneous, especially in cases of acute decompensation.

Deciding to admit cancer patients to an Intensive Care Unit (ICU) is a significant clinical challenge, particularly when it comes to identifying those patients who would benefit from an ICU stay.3 Traditionally, ICU admission criteria have been restrictive for oncology patients due to the high mortality rates reported in some studies and significant resource consumption.4 However, recent research suggests that in-hospital mortality for this group is not greater than for other groups with severe comorbidities.5 Likewise, ICU admission does not necessarily imply the application of aggressive therapeutic measures. In this regard, multiple admission policies focused on individualized assessment of the patient's condition upon admission have been developed — the most widespread protocol being the time-limited trial (ICU-TLT).6,7

Currently, there is a lack of studies that comprehensively address the factors that determine the admission of cancer patients to the ICU. Most of the available research focuses on specific cohorts or has limitations in the analysis of the interaction between performance status, comorbidities, and recent cancer treatments.8,9 Furthermore, predicting ICU mortality in these patients remains challenging,10 and there is scant literature on why critical oncological patients are rejected for ICU admission, even though the percentage has traditionally been siginificant.11 This gap in the literature prevents the establishment of more objective and standardized criteria for ICU admission decisions, which may affect equity in access to intensive care for these patients.

The present study aimed to analyze and compare the clinical, functional and oncological characteristics of ICU-admitted and non-admitted cancer patients, identify factors associated with admission decisions, and to develop predictive models to estimate ICU admission probability. We hope these findings will contribute to a better understanding of the ICU patient selection process and provide tools to facilitate more objective, evidence-based decision-making.

Fig. 1 shows the patient selection process. During the study period, a total of 1375 patients were recruited. Of those patients, 34 were excluded due to an inability to collect their data. A total of 1341 patients were analyzed; 1177 (87.8%) were admitted to the ICU, while 164 (12.2%) were rejected. Table 1 shows the general characteristics of the patients, including those admitted to the ICU and those rejected for admission. Admitted patients had a mean age of 66.2 years (SD: 12.2), whereas non-admitted patients had a mean age of 69.8 years (SD: 10.8). The most frequent tumors were those of gastrointestinal origin (22.6%), followed by genitourinary (14.9%) and pulmonary tumors (12.7%). Hodgkin lymphoma frequently resulted in ICU admission (11.6%). Analyzing the status of the hematological disease revealed that 36.5% of the patients had a recent diagnosis, and 26.47% were experiencing progression.

Figure 1.

Flow chart.

Table 2 describes the variables that were independently associated with ICU admission. Neutropenia (OR: 2.38; 95%CI: 1.12−5.05; p = 0.024), gastrointestinal tumors (OR: 2.42; 95%CI: 1.37−4.30; p = 0.002), ear, nose and throat (ENT) malignancies (OR: 3.65; 95%CI: 1.49−8.96; p = 0.005), acute myeloid leukemia (OR: 4.89; 95%CI: 1.11−21.54; p = 0.036), non-Hodgkin lymphoma (OR: 3.03; 95%CI: 01.46−6.30; p = 0.003), multiple myeloma (OR: 2.81; 95%CI: 1.06−7.41; p = 0.037), or having received radiotherapy within the three months before assessment was associated with an increased likelihood of admission (OR: 1.46; 95%CI: 1.01−2.11; p = 0.045).

Predictive models of ICU admission were developed for the total cohort, as well as for patients with solid and hematological tumors (Fig. 2). Upon analyzing the types of neoplasms, Model 1, the full model, showed the best predictive capacity, with an area under the curve (AUC) of 0.79 (95%CI: 0.75−0.84) for solid malignancies and 0.82 (95%CI: 0.76−0.88) for hematological tumors.

Figure 2.

Predictive models of ICU admission.

A: Predictive models for all the patients: Model 1: Initial ECOG score, smoking status, age, neutropenia, alcohol consumption, radiotherapy, oncological treatment within the last three months; Model 2: Initial ECOG score, smoking status, age, neutropenia, alcohol consumption, radiotherapy; Model 3: Initial ECOG score, smoking status, age, neutropenia, alcohol consumption; Model 4: Initial ECOG score, smoking status, age, neutropenia.

B: Predictive models for the patients with solid organ tumors: Model 1: Initial ECOG score, smoking status, alcohol consumption, metastasis, oncological treatment within the last three months, ENT tumor, gastrointestinal tumor, neutropenia, radiotherapy; Model 2: Initial ECOG score, smoking status, alcohol consumption, metastasis, oncological treatment within the last three months, ENT tumor, neutropenia, radiotherapy; Model 3: Initial ECOG score, smoking status, alcohol consumption, metastasis, oncological treatment within the last three months, ENT tumor, neutropenia; Model 4: Initial ECOG score, smoking status, alcohol consumption, metastasis, oncological treatment in last three months, ENT tumor.

C: Predictive models in the patients with hematological tumors: Model 1: Hematological malignancy status, initial ECOG score, age, smoking status, alcohol consumption, neutropenia, radiotherapy, acute myeloid leukemia; Model 2: Hematological malignancy status, initial ECOG score, age, smoking status, alcohol consumption, neutropenia, radiotherapy; Model 3: Hematological malignancy status, initial ECOG score, age, smoking status, alcohol consumption, neutropenia; Model 4: Hematological malignancy status, initial ECOG score, age, smoking status, alcohol consumption.

For solid tumors, the model included the following variables: ECOG score at admission, smoking and alcohol consumption habits, metastasis status, cancer treatment within the previous three months, ENT or gastrointestinal tumor status, neutropenia status, and radiotherapy status. For hematological malignancies, the variables included in the model were hematological malignancy status, initial ECOG score, age, smoking and alcohol consumption history, neutropenia, radiotherapy, and acute myeloid leukemia.

Table 3 shows the characteristics of the patients who were admitted after evaluation by the DICM. The most common reasons for ICU admission were respiratory failure (n = 273, 25.1%) and hemodynamic instability (n = 246, 22.7%). Regarding the treatments applied in the ICU, 55% (n = 647) of the patients required ventilatory support, and 58.6% (n = 690) required vasoactive drugs. Of those admitted, 21.4% (n = 251) died during their hospital stay. The most frequent cause of death was multiorgan failure (64.3%). In 22.9% (n = 210) of the patients, measures to limit therapeutic efforts were applied in the ICU.

Table 4 shows the characteristics of the patients whose admission to the ICU was rejected. The most frequent reasons were futility (66%, n = 101) and chronic disease (n = 69, 45.4%). Of those patients not admitted to the ICU, 87 (53.7%) died during their hospital stay.

This study deepens our understanding of the profile of cancer patients who require ICU admission following an acute event. We generated predictive models of ICU admission for patients with solid organ tumors and hematological malignancies, taking into account the degree to which each variable contributes to the model’s overall predictive capacity.

Through a comparative analysis of ICU-admitted patients and those rejected for admission, we have gained a better understanding of the determinants of this critical decision. To date, only one comparative study has been conducted between oncology patients admitted to the ICU versus those not admitted. The results of the present study suggest a probable loosening of ICU admission criteria for cancer patients over the last two decades (90% admitted in our series of 2025, compared to 51% in 2005), though potential investigator selection bias cannot be excluded (e.g., possible non-inclusion of patients who were not included in the sample). Furthermore, Thiery et al. published a study16 in which 26% of the patients deemed too ill to benefit from ICU admission were still alive on day 30, and 16.7% were still alive on day 180. In our study, 46.63% of the patients who were not admitted survived their hospital stay.

Regarding the factors that influence ICU admission, age has consistently been considered an important factor in ICU admission decisions for cancer patients.17,18 Studies have been carried out on oncology patients over 65 years of age, reporting ICU mortality rates similar to those found in the non-cancer population.19 In our sample, there were significant differences in the mean age of the patients admitted to the ICU versus those not admitted (66 versus 69 years; p < 0.001). Despite this, the results confirm a greater probability of ICU admission for non-elderly patients. Similarly, the mean age of the patients admitted to the ICU was seen to be lower than in previous studies conducted in our setting,20 which may reflect the impact of advances in early cancer detection.21

As already noted by Bos et al., although more women than men with cancer are registered, a higher proportion of male patients are admitted to the ICU.22 The exact reason why severe events occur more frequently in men is unclear, but it may be related to variations in epidemiology, pathophysiology, presentation, severity, and response to treatment, due to the significant impact of sex hormones on the immune and cardiovascular systems.23

All the studied basal performance status scales (Clinical Frailty Scale, Karnofsky-PS, and ECOG-PS) have limitations. In our study, however, we incorporated the ECOG scale into all the predictive models, since it improves the overall ability to predict ICU admission. The ECOG-PS is an easy-to-use tool that is widely used in cancer patients. However, some studies have reported a high degree of interprofessional variability in its application.24 There is currently a need to develop dynamic and more objective tools to assess functionality in cancer patients.25,26 In our work, we have proposed creating predictive models by selecting variables according to their joint contribution to the overall predictive power. Our findings encourage us to advance toward precision oncology27 and explore new scales, such as the Clinical Frailty Scale.28

One relevant finding of our study is that the patients who received cancer treatment within three months prior to an acute event were more likely to be admitted to the ICU. This observation underscores the importance of close monitoring and preventive management for these patients because recently receiving cancer-specific treatment seems to predispose them to developing serious complications. Some studies have focused on the early detection of complications in cancer patients and have found that daily joint rounds (DICM-Oncology and/or Hematology) and early warning systems can identify complications before they require ICU admission.29 Similarly, studies have shown that mortality increases by 1.05% for each hour of delay in critical patients.30 Thus, optimizing algorithms based on vital signs and biomarkers, providing personnel with specific training in the detection of clinical deterioration, and establishing rapid response teams could significantly improve the prognosis of these patients.31,32 The predictive models developed in our study to predict the need for ICU admission constitute valuable tools for identifying high-risk patients, optimizing resource allocation, and improving clinical decision-making in the ICU.

Cancer patients are particularly susceptible to developing serious infections due to multiple factors, including immunosuppression resulting from the cancer treatments and associated organ dysfunction. However, diagnosing sepsis in these patients can be difficult because conditions such as cytokine release syndrome after CAR-T therapy or tumor lysis syndrome can present similarly. In the present study, 16.58% of the patients experienced a drug-related complication, with 2.62% suffering a CAR-T therapy-related complication. These findings underscore the urgent need for more precise diagnostic tools to differentiate between these conditions, thereby optimizing management and treatment.33

Upon analyzing the reasons for rejecting ICU admission, we found that futility and poor estimated quality of life were the reasons most frequently cited by intensivists. The ADENI-UCI study,34 carried out in 62 Spanish ICUs, analyzed the variables associated with decisions to reject ICU admission as a limitation of life-support treatment measure. Twenty-eight percent of the patients had some type of neoplasm, and poor estimated quality of life was also one of the most frequent reasons for rejection.

The mortality rate among patients admitted to the ICU was 21.45%, a figure similar to that reported in the ONCO-ENVIN registry,9 where the mortality rate was 27.5% among cancer patients admitted for medical reasons. Among patients not admitted to the ICU, 53.37% died during their hospital stay. Thus, a significant subgroup of patients survived their hospital stay without ICU admission. This underscores the complexity of the decision-making processes, which should be carried out by multidisciplinary teams, in both the initial assessment of acute deterioration and over follow-up during hospitalization.35

One of the most important limitations of our study is the difference between the total number of patients included in the ICU admission group and the number of patients not included. This discrepancy could be explained by the inherent difficulty of asking patients who have been rejected, either by themselves or their families, to participate in a study. It could also be due to the progressive relaxation of the ICU admission criteria for cancer patients in our country. The exact number of excluded patients who, despite meeting the inclusion criteria, could not be recruited because they did not sign the informed consent form is not known.

In conclusion, in the present study, we have developed predictive models of ICU admission by selecting variables according to their joint contribution to the overall predictive power, not in isolation. The so-called complete model (Model 1) demonstrated the greatest predictive capacity, with an area under the curve (AUC) of 0.79 (95%CI: 0.75−0.84) for solid tumors and 0.82 (95%CI: 0.76−0.88) for hematological malignancies. The most frequent cause of ICU admission was respiratory failure, and more than half of the patients required ventilatory support. Gastrointestinal malignancies among solid tumors and Hodgkin lymphoma among hematological neoplasms were the diagnoses that most frequently resulted in ICU admission.

Study design: Elena Cuenca Fito and Inés Gómez-Acebo.

Statistical analysis: Inés Gómez-Acebo.

Data acquisition: Elena Cuenca Fito, Eric Mayor-Vázquez, Cándido Díaz Lagares, Bárbara Vidal Tegedor, Noelia Isabel Lázaro Martín, Alba López Fernández, Laura Sanchez Montori, Íñigo Isern, Amparo Cabanillas Carrillo, Jorge Sánchez Gómez, Maria Luisa Gómez Grande, Alba Fernández Rodríguez, Anastasio Espejo, Domingo Díaz Díaz, Alejandra García Roche, Margarita Márquez, Mireya Molina Cortés, Natalia Valero, Manuel Gracia Romero, Fernando Eiras Abalde, Juan Higuera Lucas, Noelia Albalá Martínez, María Del Mar Jiménez Quintana, Alberto Orejas Gallego, Rosa María de la Casa, Diana Monge Donaire, Amanda Lesmes González de Aledo, Ariadna Bellès Casas, Lucía Larrañaga Sigwald, Sandra Portillo Sánchez, Alfredo Bardal Ruiz, Inés Lipperheide, Jimena Luján Varas, Carola Gimenez-Esparza Vich, Paula Recena Pérez and Maricela Jiménez-López.

Draft of the article, critical review of the intellectual content, and final approval of the presented version of the manuscript: Elena Cuenca Fito, Cándido Díaz Lagares, Eric Mayor-Vázquez, and Inés Gómez Acebo.

No AI was used.

The present study has received no funding.