Permissive fluid volume in adult patients undergoing extracorporeal membrane oxygenation treatment

Patients who received ECMO at Yonsei University Health System, Seoul National University Hospital, and Seoul National University Bundang Hospital from January 2005 to May 2016 were initially screened (n = 1499). Patients who met the following criteria were excluded (n = 776): (1) age under 18 years, (2) ECMO support for less than 24 h, (3) death within 3 days from ECMO initiation, (4) switching of ECMO modality, (5) end-stage renal disease with dialysis at admission, (6) transfer after initiating ECMO in other hospitals, and (7) missing data. As a result, 406 patients who underwent ECMO for cardiovascular causes (cardiovascular disease [CVD] group) and 317 patients who underwent ECMO for non-cardiovascular causes (non-CVD group) were included in the final analysis (Fig. 1). Although myocarditis is caused by infection, patients with myocarditis were included in the CVD group on the basis of pathophysiologic considerations. The study protocol was approved by the institutional review boards of Yonsei University Health System and Seoul National University Bundang Hospital. Informed consent was waived by the institutional review boards owing to the retrospective study design.

The patients’ demographic and laboratory data at the time of ECMO initiation were retrieved from electronic medical records. The data included age, sex, height, weight, underlying medical conditions, cause of ECMO, mode of ECMO (venoarterial [VA]-ECMO or venovenous [VV]-ECMO), ECMO settings, and total fluid input and output on each day of ECMO therapy. Laboratory data, including white blood cell count, hemoglobin, sodium, potassium, bicarbonate, albumin, bilirubin, C-reactive protein, blood urea nitrogen, creatinine, and arterial blood gas analysis on the day of ECMO implantation, were also collected.

Acute Physiology and Chronic Health Evaluation (APACHE) II scores were calculated on the basis of medical records. The ENCOURAGE (Prediction of cardiogenic shock outcome for AMI patients salvaged by VA-ECMO) and PRESERVE (Predicting death for severe ARDS on VV-ECMO) scores were also calculated for patients in the CVD and non-CVD groups, respectively [26, 27]. The Modification of Diet in Renal Disease equation was used to determine the estimated glomerular filtration rate (eGFR) [28]. Baseline serum creatinine was defined as the average serum creatinine within 3 months prior to ICU admission. For those whose baseline creatinine levels were not obtainable, urine output was used for AKI diagnosis. Development of AKI within 72 h of ECMO implantation was evaluated using the Kidney Disease: Improving Global Outcomes (KDIGO) AKI criteria [29]. The amount of fluid administered was decided by the intensive care specialists. Crystalloid fluids were used to maintain fluid therapy for all patients. The low-chloride fluid resuscitation strategy had not been widely practiced in the institutions during the study period. Cumulative input was defined as total administration of fluid within 3 days of ECMO initiation, which was divided by patients’ body weight at the time of ECMO initiation. In addition, cumulative total output was defined as the sum of all fluid drainage from patients, and cumulative urine output as the sum of the total amount of urine during 3 days from ECMO initiation divided by body weight. CFB was defined as cumulative input minus cumulative total output. Daily fluid balance before ECMO was defined as total fluid input minus total output from ICU admission to ECMO initiation, which was divided by ICU length of stay (LOS) and patients’ body weight. If patients were admitted to the ICU after surgery, fluid balance during surgery was also included in daily fluid balance before ECMO.

The decision to initiate ECMO therapy was made by the treating intensive care specialist or the attending cardiothoracic surgeons. Indication for ECMO was classified into two categories. Indications for ECMO in the CVD group were as follows:

Indications for ECMO in the non-CVD group were as follows:

Cannulations for all ECMO supports were performed by cardiovascular surgeons with limited cut-down using the Seldinger technique [30]. Either the CAPIOX EBS (Terumo Co., Ltd., Tokyo, Japan) or the QUADROX PLS (Maquet GmbH, Rastatt, Germany) system was used in all patients. The sweep gas flow was set to maintain a partial pressure of carbon dioxide target of 35–45 mmHg under lung rest ventilation strategy. The target ECMO blood pump speed was 3.0–4.5 L/min. Systemic anticoagulation with unfractionated heparin was titrated to maintain activated clotting times (ACT) between 180 and 220 s and between 160 and 180 s for VA-ECMO and VV-ECMO, respectively. ACT was measured using STA R MAX® hemostasis analyzer (Diagnostica Stago, Inc., Parsippany, NJ, USA). Hemolysis markers, such as free plasma hemoglobin and haptoglobin, were monitored every day. Careful inspections were made by the intensive care specialist to determine the cause of hemolysis when plasma hemoglobin levels were greater than 50 mg/dl. ECMO circuit change was performed when ECMO-related technical problems were suspected as the cause of hemolysis. Red blood cell (RBC) transfusion was performed to maintain hemoglobin target between 10 and 11 g/dl. Antithrombin (AT) replacement was performed when AT activities were less than 70%. Other specific ECMO run protocols are delineated in Additional file 1: Table S1.

In the event of the development of AKI, the decision to start continuous renal replacement therapy (CRRT) was made by the attending nephrologist. Generally, CRRT was applied to patients with AKI who had uncontrolled fluid overload, hyperkalemia, or metabolic acidosis. Continuous venovenous hemodiafiltration was applied using Prisma (Gambro Co., Ltd., Hechingen, Germany), Prismaflex (Gambro Co., Ltd.), or multiFiltrate (Fresenius Medical Care GmbH, Bad Homburg, Germany) machines. Biocompatible AN69 ST membranes were used for Prisma and Prismaflex, and polysulfone membranes were used for multiFiltrate. Bicarbonate-containing replacement fluid was administered by the predilution method. CRRT machines were routinely applied via separate lines from the ECMO circuit. However, when additional venous access was not obtainable, CRRT was performed over the same circuit with ECMO. In these cases, CRRT was applied between ECMO pump and oxygenator to avoid air entrainment.

The patients were followed until their last visit at our centers or death. The primary endpoint was death from any cause within 90 days after ECMO initiation. Survival data were collected through electronic medical records of the hospital and outpatient clinics.

Patients in each group were categorized into quartiles according to the CFB of each group. Continuous variables were expressed as mean ± SD. Comparisons of continuous variables were conducted with one-way analysis of variance, and linear trend was analyzed among groups. The normality of the distribution was analyzed using the Kolmogorov-Smirnov test. Variables not normally distributed were expressed as median and IQR and compared with the Jonckheere-Terpstra test. Categorical variables were expressed as percentages and compared with the chi-square test. Cumulative survival curves were derived using the Kaplan-Meier method, and statistical differences were analyzed using the log-rank test. To evaluate the association between outcomes and clinical parameters, a Cox proportional hazards model was used, and data were expressed as HR with 95% CI. This model included propensity scores. The propensity score was obtained by using a multivariate logistic regression analysis to determine whether the CFB of each patient was higher than the median amount. The covariates for adjustment were selected using a stepwise procedure when the P value for each variable was less than 0.2. All variables, except those that were already included in the APACHE II score, were included for propensity score calculation. Those who were lost to follow-up were treated as censored in the survival analysis. Cubic spline curve analysis was conducted as previously described by Canaud et al. [31]. The Cox model for 90-day mortality was used, and cubic spline curves had four equally distributed knots. The threshold of CFB was defined as the point where the lower limit of the 95% CI was higher than 1.0. Analyses were conducted for the CVD and non-CVD groups. Additional evaluations were also conducted after dividing the non-CVD group into a respiratory group and an others group. The respiratory group was defined as patients who underwent ECMO for lung transplant, ARDS, or other pulmonary diseases. All other patients in the non-CVD group were defined as the others group. Statistical significance was defined as P < 0.05. All analyses were conducted using SPSS version 23.0 software (IBM, Armonk, NY, USA) and R language (version 3.3.1; R Foundation for Statistical Computing, Vienna, Austria).

The baseline characteristics of the patients in the CVD group are presented in Table 1. The mean age was 58.4 ± 17.7 years, and 68.2% were male. The most common cause of ECMO implantation was nonoperative CVD, and 98% of patients underwent VA-ECMO. The median CFB was 64.7 ml/kg, whereas the median daily fluid balance throughout the entire period of ECMO was 26.2 ml/kg/d. The median duration of ECMO treatment was 3 days. The age of the patients tended to increase in patients with higher CFB quartiles. Body weight tended to decrease in higher CFB quartiles, but body mass index (BMI) was comparable among the quartiles. In the non-CVD group, the mean age was 55.7 ± 15.7 years, and 65.3% of patients were male. Among these patients, 201 (63.8%) patients received ECMO treatment for respiratory disease, and VA-ECMO was applied for 23% of the patients. The median CFB was 53.5 ml/kg, and median daily fluid balance during the entire ECMO period was 15.9 ml/kg. Patients’ weight and BMI were comparable among the CFB quartiles. However, APACHE II scores tended to increase in patients with higher CFB quartiles (Table 2). In the non-CVD group, there were no significant differences in Charlson comorbidity index (P = 0.616), APACHE II score (P = 0.302), and CFB (P = 0.206) between VA-ECMO- and VV-ECMO-treated patients. Regarding the resuscitation volume differences during the study duration, further analyses were made by categorizing the patients into three groups according vintage: 2005–2008, 2009–2012, and 2013–2016. When the CFB quartiles were compared among the vintage groups, no statistical differences were found (Additional file 2: Figure S1).

In the CVD group, AKI occurred in 306 (75.4%) patients and CRRT was commenced in 127 (31.3%) patients during the initial 3 days of ECMO initiation. The occurrence rate of AKI tended to increase in quartiles with higher CFB values. In addition, in the CVD group, median ICU LOS was 8 days. A total of 207 deaths (51.0%) occurred within 90 days of ECMO initiation, and the mortality rate tended to increase in higher CFB quartiles (P < 0.001). In the non-CVD group, 184 (70.0%) patients underwent AKI, and CRRT was applied in 92 (29.0%) patients during first 3 days of ECMO commencement. Both AKI occurrence (P = 0.011) and CRRT application rates (P = 0.011) tended to increase in patients with higher CFB. Patients treated with VA-ECMO had a significantly higher incidence of AKI than those treated with VV-ECMO (P = 0.014). The median ICU LOS was 16 days. During the 90 days of ECMO initiation, a total of 209 (65.9%) deaths occurred (Table 3).

The Kaplan-Meier curves revealed that the cumulative survival rates of CFB quartiles 3 (P < 0.001) and 4 (P < 0.001) were significantly lower than those of CFB quartiles 1 and 2 in the CVD group (Fig. 2a). The non-CVD group was further divided into those initiating ECMO for respiratory failure (respiratory group) and other causes (others group). The characteristics of the respiratory group are presented in Additional file 1: Table S2. In patients in the respiratory group, the cumulative survival rate of CFB quartile 4 was significantly lower than quartile 1 (P = 0.047) (Fig. 2b). In addition, in the others group, CFB quartile 4 showed a significantly lower cumulative survival rate than quartile 1 (P = 0.004) and quartile 2 (P = 0.001) (Fig. 2c). When patients in the non-CVD group were divided into those treated with VA-ECMO or VV-ECMO, those treated with VA-ECMO had significantly lower cumulative survival rates than those treated with VV-ECMO (P = 0.002).

In addition, when the association between CFB and mortality was further evaluated using multivariate Cox proportional hazards regression analyses, in the CVD group, the risk of mortality significantly increased in patients in CFB quartiles 3 (HR, 2.58; 95% CI, 1.62–4.11; P < 0.001) and 4 (HR, 2.11; 95% CI, 1.26–3.54; P = 0.004) compared with that of patients in CFB quartile 1. However, the mortality risk did not show a significant increase among patients in CFB quartile 2. Similarly, in the non-CVD group, the mortality risk of CFB quartiles 3 (HR, 1.66; 95% CI, 1.06–2.59; P = 0.026) and 4 (HR, 1.69, 95% CI, 1.05–2.72; P = 0.03) was significantly higher than quartile 1, whereas mortality risk did not significantly differ between CFB quartiles 1 and 2 (Table 4). These findings imply that a CFB below the level of CFB quartile 3 in both the CVD and non-CVD groups may not increase the risk of mortality. When CFB was considered as a continuous variable, log-transformed CFB was independently associated with 90-day mortality in both the CVD group (HR, 1.76; 95% CI, 1.37–2.27; P < 0.001) and the non-CVD group (HR, 1.46; 95% CI, 1.17–1.83; P < 0.001) (Table 5).

No significant interactions between APACHE II scores and CFBs were found in both the CVD and non-CVD groups (P = 0.521 for CVD group, P = 0.166 for non-CVD group). When further adjustments were made for daily fluid balance before ECMO commencement, the relationship between CFB and risk of mortality was preserved in both groups. In addition, cumulative input was independently associated with 90-day mortality even after adjustments were made for cumulative total output and urine output (Table 5).

When multivariate Cox models were analyzed with adjustment for ENCOURAGE score instead of APACHE II score in the CVD group, the relationship between CFB and 90-day mortality remained significant. This relationship was also maintained when adjustments were made for PRESERVE score instead of APACHE II score in patients who had undergone VV-ECMO in the non-CVD group (Additional file 1: Table S3).

To further investigate the CFB threshold level that considerably increases the risk of mortality in ECMO patients, the relationship between CFB and mortality risk was evaluated by cubic spline analyses. In the CVD group, the relative HR of mortality started to increase significantly when CFB values were greater than 82.3 ml/kg (Fig. 3a). The relative HR increased significantly above CFB levels of 189.6 ml/kg in the respiratory group (Fig. 3b). In the others group, relative HR tended to increase with higher CFB amounts without statistical significance (Fig. 3c).