This retrospective study was conducted at a single center. The participants were patients admitted to the ICU due to COVID-19 infection between March 2021 and February 2022. The center had eight beds in the ICU for patients with severe COVID-19 since March 2021. Among the severe COVID-19 patients in the ICU, those who showed a low probability of pneumonia related to COVID-19 infection (bacterial pneumonia, malignancy) among those who received oxygen therapy at the HFNC level or above were excluded through medical review (Figure 1).

ICU admission and discharge criteria for all patients included in this study were defined in accordance with government quarantine policies. All patients admitted to the ICU were administered oxygen therapy at the HFNC level or above at the time of admission; MV or ECMO was applied depending on the stage of pneumonia and ARDS. Additionally, all patients were treated according to the clinical practice guidelines for ARDS and the Sepsis-Surviving Campaign guidelines [15,16]. Treatment included administering remdesivir and steroids, as well as antibiotics and anticoagulants, to prevent secondary bacterial pneumonia. For steroid therapy, 6 mg dexamethasone was administered once a day for 10 days. For ARDS aggravation, despite steroid therapy, dosage was increased to 20 mg once a day for 5 days, with tapering depending on patient status [17]. Tocilizumab 8mg/kg was administered when the oxygen demand increased within 24–48 hours and C-reactive protein (CRP) level was ≥5.0 mg/dL even after starting steroid administration [18]. When the patient showed a decrease in oxygen requirement following ICU care, treatment was switched to a 4 L/min nasal prong. Patients were transferred to the general ward when they were deemed stable based on clinical symptoms, blood test results, and X-ray results. However, considering the possibility of further deterioration, some cases were transferred after 1–2 days of follow-up observation at the discretion of the attending physician.

To identify the factors associated with prolonged ICU LOS ≥2 weeks, patient age, sex, body weight, underlying disease, and smoking history were investigated. Sequential Organ Failure Assessment (SOFA) and Acute Physiology and Chronic Health Evaluation (APACHE) II scores were determined immediately after hospitalization to assess severity. Following ICU admission, blood test results (neutrophil-lymphocyte ratio (NLR), CRP level, D-dimer level, and PaO₂ to FiO₂ ratio (P/F ratio) on days 3 and 5 of admission were compared to analyze the association between prolonged ICU LOS and changes in blood parameters.

Patient characteristics and laboratory results were summarized as frequencies with percentages for categorical variables. Normality was assessed using the Shapiro-Wilk test. If the assumption of normality was satisfied, continuous variables were presented as mean±standard deviation, and a t-test was used. If the normality assumption was not satisfied, continuous variables were presented as median (range) and the Wilcoxon rank-sum test was used. A logistic regression model was used to identify the prognostic factors associated with a prolonged ICU stay. In the multivariable logistic model, we included all risk factors for prolonged ICU stay with a P-value <0.05, and a risk factor for prolonged ICU stay was selected using backward elimination in the final model. All analyses were performed using SAS software ver. 9.4 (SAS institute).

A total of 118 patients were admitted to the COVID-19-designated ICU, and 96 patients were included in the study after excluding those who did not meet the study criteria (Figure 1). After HFNC treatment, 65 patients showed improvement, while 31 received MV care. Most MV care (28/31, 90%) and continuous renal replacement therapy (CRRT) were performed within 3 days of initial admission. ECMO was performed in five cases on admission to the ICU, and for the remaining three cases, ECMO treatment was applied due to ARDS exacerbation even after 5–10 days of MV care. A neuromuscular (NM) blocking agent was injected in 61.2% (19/31) of patients who received MV care, while 32.3% (10/31) were placed in a prone position. Eight patients (8.3%) showed ARDS aggravation despite MV care, and ECMO was applied. Twenty-three patients (24%) received treatment in the ICU for 2 weeks or more. Of the 96 patients, four died in the ICU due to sepsis from pneumonia aggravation (n=3) and pulmonary hemorrhage during ECMO for ARDS (n=1). All four deceased patients died after being in the ICU for more than 2 weeks. The median ICU LOS was 14 days as a cut off for prolonged ICU LOS (Table 1).

Tracheostomy was performed on nine patients. Of these, seven underwent tracheostomy after undergoing ECMO treatment and two patients due to prolonged intubation (more than 2 weeks). Tracheostomy was performed by an otolaryngologist in the ICU by an open method.

Univariable analyses showed that SOFA score at the time of ICU admission, CRP level, high-dose steroids, MV care, CRRT, ECMO, and prone position was significantly correlated (Table 2) with ≥2 weeks of ICU care (prolonged ICU stay). Multivariable analysis showed that MV care and P/F ratio on hospital day 5 were independent factors for prolonged ICU LOS (Table 3).

CRP level showed a decrease between days 3 and 5 in both groups, but D-dimer showed a decrease only in the non-ICU LOS group with LOS <2 weeks. The P/F ratio showed an increase between days 3 and 5 in both groups. CRP level was significantly higher in the prolonged ICU LOS patients at 6.2 mg/dl (range, 0.1–19.1) vs. 13.4 mg/dL (range, 1.8–25.6) (P=0.029) at the time of ICU admission; however, the level was reduced on hospital days 3 and 5. For the D-dimer level, no significant variation was observed at ICU admission; however, after days 3 and 5, significant between-group variation was detected (1.0 [range, 0.2–20.0] vs. 3.2 [range, 0.3–20.0], P=0.007). Finally, for the P/F ratio, an increasing trend was observed in the non–ICU LOS group with LOS <2 weeks on days 3, but no significant difference was observed, and after 5 days a significant difference between the two groups was noted (158.6 [range, 75.6–433.3] vs. 121.6 [range, 55.9–255.0], P=0.001) (Table 4).

For MV care, dexmedetomidine was preferentially administered as a sedative, and midazolam or propofol was added when additional deep sedation was required during the first few days of MV care according to the attending physician’s discretion. Administration of midazolam (38.5% vs 94.4%, P=0.001) and NM blocking agent (38.5% vs 77.8%, P=0.027) was significantly associated with prolonged ICU LOS. Duration of sedative use (3 [2–5] vs. 7 [3–12], P=0.030) and NM blocking agents (2 [1–3] vs. 6 [4–9], P=0.011) were also statistically significant with prolonged ICU LOS (Table 5).