Lung transplantation (LT) has become an established therapeutic option for patients with end-stage lung disease over the past few decades, and it results in good functional outcomes and improves quality of life [1]. Surgical outcomes have improved with advances in surgical techniques and perioperative critical care; however, the survival rate after LT is lower than that of other organ transplants [2].

Patients who were readmitted to the intensive care unit (ICU) showed higher mortality and longer hospital length of stay compared to those who were not readmitted [3,4]. Previous studies have shown that factors associated with ICU readmission can be related to the extent of the procedure or procedure-related complications rather than preoperative underlying disease status [5-7]. This implies the importance of optimizing postoperative patient conditions for posttransplant recovery. In addition to the aforementioned conditions, postoperative infection is a particularly important issue in transplant patients, who are highly vulnerable to infection due to the inevitable use of immunosuppressants. Several studies have found that disease severity or vital signs at first ICU discharge are associated with ICU readmission [3,4]; however, consistent factors related to ICU readmission have not been established. There are studies on risk factors for rehospitalization or readmission to the ICU after hospital discharge regarding the mid- and long-term prognoses after LT. However, there is a paucity of data on the risk factors for ICU readmission during the hospitalization period immediately after LT. Therefore, the primary aim of this retrospective study was to investigate the risk factors for ICU readmission during index hospitalization period after LT, especially with regard to patient condition posttransplant.

A total of 166 patients underwent LT between October 2012 and August 2017 (Figure 1). Twenty patients were excluded as follows: 4 patients were adolescents, 6 patients had undergone concurrent other surgical procedures such as heart transplantation and coronary artery bypass grafting, and 10 patients had received LT due to GVHD after bone marrow transplantation. We also excluded 14 patients who died without recovery during the first ICU stay and two patients who died on the GW due to unexpected cardiac arrest after initial ICU discharge. Finally, the data obtained from 130 adult patients who received unilateral (n = 8) or bilateral LT (n = 122) were analyzed. The median follow-up period was 47 days (interquartile range, 29−83). Among eligible patients, 32 (24.6%) were readmitted to the ICU 47 times during index hospitalization period after LT. Among patients readmitted to the ICU, 18 were readmitted to the ICU once, 13 were readmitted to the ICU twice, and 1 was readmitted to the ICU thrice. Throughout the study, we compared the readmission group (n=32) and the control group (n = 98).

Baseline characteristics of recipients are shown in Table 1. Compared to the control group, the readmission group was older (55 [45−60] vs. 62 [53−65], P = 0.003), had more unilateral LT patients (3 [3.1%] vs. 5 [15.6%], P = 0.022), and had a lower percentage of highly educated people (82 [83.7%] vs. 20 [62.5%], P = 0.024) and the proportion of unilateral LT was higher (3 [3.1%] vs. 5 [15.6%], P=0.022) in the readmission group. Forced expiratory volume in 1 second (FEV1; %) was higher (41 [28−50] vs. 51 [37−59], P = 0.013) in the readmission group; however, the FEV1 to forced vital capacity ratio was not different between the groups (P=0.815). Other preoperative characteristics, such as performance status, diagnosis for transplantation, were not different between the groups. There were no donor-related factors that differed between the groups (Table 2).

Perioperative patient data are shown in Table 3. The proportion of preoperative intubation was higher (22 [22.4%] vs. 14 [43.8%], P = 0.024) and preoperative albumin (g/dl) was lower (3.5 [2.9−4.0] vs. 3.1 [2.5−3.6], P = 0.015) in the readmission group, while the intraoperative data did not differ between the groups. Among the recipient features at initial ICU discharge, SOFA score (3 [2−4] vs. 4 [3−6], P = 0.016), serum creatinine (mg/dl; 0.42 [0.32−0.59] vs. 0.52 [0.39−0.85], P = 0.049), and the proportion of tracheostomy (34 [34.7%] vs. 19 [59.4%], P = 0.014) were higher, length of ICU stay (day) was longer (8 [6−12] vs. 10 [6−23], P = 0.017), and pH (7.460 [7.431−7.482] vs. 7.446 [7.371−7.465], P = 0.012) and albumin (g/dl; 3.1 ± 0.3 vs. 3.0 ± 0.3, P = 0.037) were lower in the readmission group (Table 4). The SOFA score (9 [7−11] vs. 10 [8−11], P = 0.295) and APACHE II score (21 [18−24] vs. 21 [17−24], P = 0.807) did not significantly differ between the groups when the patients were admitted to the ICU immediately after LT.

The variables at initial ICU discharge with P-value <0.1, which included PaCO₂ (mm Hg; P = 0.051) and duration of mechanical ventilation (hour; P = 0.072), were analyzed using univariable logistic regression (Table 5). In this regression, SOFA score (odds ratio [OR], 1.547; 95% confidence interval [CI], 1.164−2.056; P = 0.003), pH (OR, 0.885; 95% CI, 0.815−0.960; P = 0.003; when the pH value increases by 0.01), PaCO₂ (OR, 1.044; 95% CI, 1.003−1.087; P=0.035), albumin (OR, 0.222; 95% CI, 0.051−0.964; P = 0.045), tracheostomy (OR, 2.751; 95% CI, 1.213−6.240; P = 0.015), and length of ICU stay (OR, 1.056; 95% CI, 1.013−1.100; P = 0.010) were independent risk factors for readmission to the ICU during the hospitalization period after LT (Table 5). Variables with P-value <0.05 in univariable logistic regression were analyzed with multivariable logistic regression using stepwise selection. The multivariable model with SOFA score (OR, 1.464; 95% CI, 1.083−1.978; P = 0.013), pH (OR, 0.884; 95% CI, 0.813−0.962; P = 0.004; when the pH value increases by 0.01), and albumin (P = 0.102) yielded the best predictive power, with an AUROC of 0.739 (95% CI, 0.631−0.847; P < 0.001). The SOFA score (OR, 1.527; 95% CI, 1.065−2.190; P = 0.021) and pH (OR, 0.843; 95% CI, 0.764−0.929; P = 0.001; when the pH value increases by 0.01) were significant risk factors when adjusting for the predetermined confounding factors. Further, the predictability of the adjusted multivariable model increased to an AUROC of 0.836 (95% CI, 0.737−0.934; P < 0.001). According to the Jonckheere–Terpstra test, as the number of ICU readmissions increased, the SOFA score increased, and pH and albumin decreased (P = 0.018, P = 0.013, P = 0.034, respectively, respectively).

If each readmission to the ICU was considered an independent event, the total number of ICU readmissions was 47 in this study. The most common causes for ICU readmission were sepsis and pneumonia (25.5% for both), followed by pulmonary complications such as intrapulmonary bleeding and bronchopulmonary fistula (17%) and cardiac dysfunction (8.5%). These were analyzed based on the number of ICU readmissions (n = 47) (Figure 2). The median interval between initial ICU discharge and first ICU readmission was 10 days. Among 32 readmitted recipients, 11 were readmitted to the ICU within 5 days for the following causes: sepsis (45.4%), pneumonia (27.3%), and cardiac dysfunction, pulmonary thromboembolism, and renal dysfunction (9.1% for each).

Among the 130 patients included, 13 (10%) died during index hospitalization period after LT due to sepsis (61.5%), hypovolemic shock due to gastrointestinal bleeding (23.1%), massive lung hemorrhage (7.7%), and multiorgan failure (7.7%) (Figure 3). In the Cox proportional hazard model with time-varying covariates, the number of ICU readmissions was a significant risk factor for in-hospital mortality (hazard ratio [HR], 21.34; 95% CI, 2.34−2,831.02; P = 0.003 for ICU readmission once; HR, 34.12; 95% CI, 4.10−4,450.90; P < 0.001 for ICU readmission twice or thrice), as was a lower education level (HR, 3.45; 95% CI, 1.38−9.28; P = 0.008).

In this study, 32 patients (24.6%) were readmitted to the ICU before hospital discharge after LT, which was lower than reported in an earlier study that showed an ICU readmission rate of 44% during the first 90 days after LT [8]. However, the readmission rate is higher than that observed in other studies [4,9]. Thirteen (10%) patients died during hospitalization, and the number of ICU readmissions was significantly associated with in-hospital mortality (HR, 21.34 for ICU readmission once; HR, 34.12 for ICU readmission twice or thrice) at a much stronger level than in previous reports [3,4]. The association we found suggests that it is important to prevent ICU readmission through early postoperative management in LT recipients. Therefore, it is essential to identify risk factors for ICU readmission and focus on those factors in LT patients to improve prognosis.

The SOFA score and pH on initial ICU discharge were associated with readmission to the ICU during index hospitalization. The SOFA score is a widely used tool in assessing clinical condition and response to treatment of critically ill patients [10]. The effectiveness of the SOFA score on stratification of morbidity and mortality was reconfirmed through this study. In the present study, pH was statistically lower in the readmission group (7.460 vs. 7.446; P = 0.012), though the difference was not clinically significant. PaCO2 (mm Hg), one of the factors that can affect pH, was investigated; however, it was within normal range and not statistically different between the groups (35 vs. 39; P = 0.051). Sepsis was one of the major ICU readmission causes (25.5%) and can cause metabolic acidosis. However, there were no differences in pH between patients readmitted to the ICU due to sepsis and patients readmitted to the ICU due to other causes. The lower was the pH, the higher was the ICU readmission rate; however, this finding should be interpreted cautiously, and further studies are warranted to identify the clinical usefulness and cut-off values and to clarify the factors that affect pH in posttransplant patients.

In our study, the rate of ICU readmission was higher among eight patients who received unilateral LT than among those who received bilateral LT. The patients received unilateral LT because of unfavorable conditions such as old age, small thoracic cavity with severe adhesion, and fatal arrhythmia during transplantation, which can lead to suboptimal conditions after transplantation, resulting in a high ICU readmission rate. However, the lengths of the initial and total ICU stays did not differ between patients with unilateral vs. bilateral LT (P=0.903 and P = 0.279, respectively), and in-hospital mortality did not occur in patients with unilateral LT. Patients with a lower education level were readmitted to the ICU at a higher rate, and a lower education level was a significant risk factor for in-hospital mortality (HR, 3.45) along with the number of ICU readmissions, which is in line with previous results. Level of education can be a proxy for socioeconomic status, which can influence survival and might be related to willingness to be actively treated [11,12].

Sepsis was one of the leading causes of ICU readmission along with pneumonia (25.5% for both) and the main cause of in-hospital mortality (61.5%), which was consistent with previous studies [13-15]. The results suggest that post-lung transplant infection is the most crucial factor for ICU readmission and mortality during index hospitalization. Posttransplant infection risk in LT recipients regarding immunosuppressive therapy, prophylaxis, and the infection status of donors and recipients is similar to that of other organ-transplant recipients. However, factors such as constant exposure to the various pathogens present in the environment could increase vulnerability of LT recipients to infection [16]. Therefore, post-LT infection control in patients with disrupted lymphatics and short- to long-term lung denervation, including removal of superfluous in-dwelling catheters, optimal dosing of antibiotics, and overall surveillance regarding infection control measures, is of importance to improve patient prognosis [17].

This study is subject to the following limitations. First, it was a single-center study with a relatively small study population. Therefore, our study population might not be representative of the entire LT recipient population. Second, due to the retrospective nature of the study, we cannot rule out selection bias and information bias. Moreover, the retrospective design of the study allowed us to distinguish between pneumonia and rejection; however, this is difficult to perform in clinical practice. Third, surgical techniques and teamwork on the LT team might have improved over time, which could have confounded interpretation of the data. Fourth, GVHD patients were excluded because we considered their underlying disease status as a strong confounding factor in this study. Further, we excluded patients who died during the initial ICU stay or in the GW after the initial ICU discharge; that is, patients who did not have the opportunity for ICU readmission were excluded since it would not be appropriate to include such patients in analysis of the primary outcome. However, this can be considered as excluding the patients in the worst condition.

There have been steady efforts to develop prediction models for early unpredicted ICU readmission after initial ICU discharge in various patient groups [9,18-21]. However, no consensus model is commonly used. Although this retrospective study includes a small population, it is valuable since it can help to develop a predictive model of ICU readmission after initial ICU discharge in patients undergoing LT.

In conclusion, the SOFA score and pH were independently associated with increased risk of ICU readmission after initial ICU discharge during index hospitalization after LT. Careful optimization considering these factors and thorough postsurgical infection control can reduce the ICU readmission rate and increase patient survival. Further large-scale studies are warranted to establish a predictive model that considers a larger number of factors.