This retrospective study was conducted in a 30-bed medical-surgical ICU of a tertiary teaching hospital in Beijing, China (Beijing Friendship Hospital, Capital Medical University). A retrospective review of the medical records of patients admitted to this ICU from January 2015 to December 2018 was performed. The patients included were those admitted to the ICU with septic AKI and undergoing CRRT. Septic AKI was defined as the simultaneous presence of sepsis and AKI. The patients with the following characteristics were excluded: pre-existing chronic kidney disease (estimated glomerular filtration rate [GFR] < 20 ml/min/1.73 m² for at least 1 year); ICU stay of less than 48 h; and missing data on the fluid status and body weight. Data were collected only from the first ICU admission if a patient underwent multiple ICU admissions that required CRRT during the study period. This study was approved by the Bioethics Committee of Beijing Friendship Hospital, Capital Medical University (2020-P2-210-01).

The definition of AKI was based on the Kidney Disease: Improving Global Outcomes (KDIGO) clinical practice guidelines for AKI. Thus, AKI was defined by the presence of at least one of the following three criteria: an increase in the serum creatinine (sCr) level to at least 0.3 mg/dl (26.5 μmol/L) within 48 h; an increase in the sCR level to at least 1.5 times the baseline level that was known or was presumed to have occurred within the previous 7 days; or urine volume below 0.5 ml/kg/h for 6 h (23). The diagnostic criteria for sepsis and septic shock were in accordance with the 2016 International Sepsis Definitions (24). The baseline eGFR was calculated by the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) formula (25). This eGFR calculation used the sCr value closest to the date of hospitalization, but not more than 1 year prior to the hospitalization, or the lowest sCr value documented during the current hospitalization if no other value was available. Chronic kidney disease (CKD) was defined as the eGFR below 60 ml/min/1.73 m².

For each patient, the CFB was expressed as a percentage (%) and calculated using the following equation.

The CFB was assessed at 48 and 72 h after initiation of CRRT. A negative FB was defined as a weight-adjusted CFB less than 0%, an even FB as a weight-adjusted CFB of 0% to less than 5%, and a positive FB as a weight-adjusted CFB of 5% or more (26). In addition, we collected data on input and output from the hospital admission to CRRT initiation to calculate the CFB and weight-adjusted CFB at the initiation of CRRT.

The primary outcome was all-cause mortality at 28 days after CRRT initiation. The secondary outcomes were all-cause mortality at 60 days, length of stay in the ICU, mechanical ventilation-free days, and vasopressor-free days within 28 days from the CRRT initiation.

At baseline, the following characteristics of the enrolled patients were recorded at the time of CRRT initiation: demographic data (age, sex, body weight, and body mass index [BMI]); clinical data (admission type, comorbidities, septic shock, the CFB before CRRT initiation, and indication for CRRT); infection data (site of infection and infection category); laboratory data (white blood cells, hemoglobin, platelets, eGFR, sCr, albumin, and lactate). For assessment of disease severity, the Sequential Organ Failure Assessment (SOFA) score and the Acute Physiology and Chronic Health Evaluation (APACHE) II score were determined at the time of CRRT initiation. The organ support measures at the time of CRRT initiation, such as the need for mechanical ventilation and vasopressor support, were recorded.

For comparisons, the patients were stratified into three groups based on 48 h weight-adjusted CFB: positive FB, negative FB, and even FB. The baseline values of the continuous variables were reported as means and SDs or medians and interquartile ranges (IQR) for normally and non-normally distributed variables, respectively. The categorical variables were presented as numbers and percentages. The categorical variables were compared using the chi-squared test and continuous variables using a one-way ANOVA and the Kruskal–Wallis test.

Survival analysis was performed using a Kaplan–Meier method and the log-rank test. The multivariable Cox regression models were used to examine the association of 28-day mortality with CFB. The Cox regression was first adjusted for demographic data (age, sex, and BMI; Model 1) and then additionally adjusted for severity of illness (SOFA score and APACHE II score; Model 2). The clinical variables with p values below 0.03 in the univariate analysis and previous variables were entered into the final model (Model 3). If two variables were strongly correlated, only one of these variables was retained and added to the multivariable model. The weight-adjusted CFB was also treated as a continuous variable in a Cox regression model that was used to calculate the unadjusted and adjusted hazard ratio (HRs); these HRs were plotted using restricted cubic spline models to assess the potential nonlinear associations between the CFB and 28-day mortality.

A subgroup analysis was performed to examine primary outcomes in the patients with fluid overload before CRRT initiation and patients with septic shock. To explore the robustness of the results, three sensitivity analyses were performed. First, the multivariable logistic regression models were constructed to calculate the unadjusted and adjusted odds ratios (ORs) to test the robustness of the findings from the Cox models. Second, to account for potential survivorship bias, the effect of CFB on 28-day mortality in a subgroup of patients who survived at least 3 days after CRRT initiation was examined. Third, to reduce the effect of selection bias and potential confounding, a propensity score representing the probability that a patient would be in a fluid balance group was developed based on the following variables: age, sex, BMI, SOFA score, APACHE II score, septic shock, CKD, hospital-acquired infection, lactate, eGFR, and weight-adjusted CFB before CRRT. This score was calculated using logistic regression and additionally adjusted in the Cox regression model. Another propensity score was calculated using the baseline variables that differed significantly among the three groups and adjusted in the Cox regression.

Data were analyzed using IBM SPSS software version 19.0 (IBM, NY, USA), R version 4.0.1 (Austria), and STATA version 14.1 (Stata Corp LLC, TX, USA). A two-sided p-value below 0.05 was considered significant.

During the 4-year study period, 3,413 critically ill adult patients were admitted to the ICU, and 287 patients developed severe septic AKI that required CRRT (Figure 1). After the exclusion of patients based on pre-defined criteria, we included 227 patients in this study.

We recorded the demographic, clinical, and laboratory characteristics of all the patients with stratification by CFB at baseline (Table 1). Overall, most of the patients were male (64.3%) and the mean age was 62.4 ± 18.3 years. There were 104 patients (45.8%) in the negative FB group, 79 (34.8%) in the even FB group, and 44 (19.3%) in the positive FB group. The history of CKD was more prevalent in the negative and even FB groups. The positive FB group had a higher APACHE-II score, a higher prevalence of receiving vasopressor support and septic shock, a lower level of albumin, and a higher level of lactate. The CFB at the initiation of CRRT was significantly higher in the positive FB group. The negative FB group had a lower level of hemoglobin and smaller proportions of men and patients with hospital-required infections. The most common indications for CRRT were oliguria and anuria, followed by worsening azotemia and fluid overload. There were no significant differences among the three groups.

We assessed the CFB at 48 and 72 h after initiation of CRRT. At these two times, the positive FB group had median FBs of 5,562 and 6,607 ml, and the negative FB group had median FBs of −1,550 and –2,043 ml.

Among all the 227 patients, 123 patients (54.2%) died within 28 days after CRRT initiation (Table 2). The CFB at 48 h was 204 ml (range: −1,505, 1,491) in the survivors and 379 ml (range: −1,394, 3,274) in the non-survivors. The non-survivors also had a higher CFB at 72 h after CRRT initiation, but this difference was not statistically significant.

The 28-day mortality was significantly lower in the even FB group (43.0%, p = 0.007, Table 3) than in the positive FB group (72.7%) and the negative FB group (54.8%). A Kaplan–Meier analysis and the log-rank test (Figure 2) indicated significantly longer survival in the even FB group than in the positive FB group, but no significant difference between the even FB and negative FB groups.

We initially used the univariate Cox models to identify the factors associated with all-cause mortality. Relative to the even FB group, the positive FB group (but not the negative FB group) had an increased adjusted HR for mortality in all the Cox regression models (Table 4). In addition, the multivariate Cox model indicated that the 28-day mortality was significantly associated with the female gender, higher SOFA score, and higher eGFR (Table 4). We then used a restricted cubic spline procedure to examine the relationship of the HR for 28-day all-cause mortality with CFB, which was treated as a continuous variable (Figure 3). There was a marginal J-shaped association between the CFB and 28-day all-cause mortality in the unadjusted model (p for non-linearity = 0.0435), but this relationship was not significant in the adjusted model (p for non-linearity = 0.1165).

In addition, the three groups had no significant differences in all the four secondary outcome measures—RRT dependence in survivors, vasopressor-free days, mechanical ventilation-free days, and length of ICU stay (Table 3).

Additionally, we collected detailed data on the cause of death (Table 5). Septic shock was the most common cause of death, followed by refractory cardiogenic shock.

A logistic regression, with even FB as the comparator, indicated that positive FB was associated with the 28-day mortality (aOR: 3.68, 95% CI, 1.34–10.13; Table 6). This finding confirmed the robustness of our results.

To account for the potential survivorship bias, we used the univariate and multivariate Cox models to analyze the 207 patients who survived beyond 72 h to assess the effect of FB on the 28-day mortality (Table 7). Similar to the above results, the positive FB group had greater 28-day mortality than the even FB group, but there was no significant difference between the even FB and negative FB groups.

Furthermore, the positive FB group still had a higher risk of death at 28 days when we used the propensity score as a covariate in the two other sensitivity analyses (Tables 8, 9).

We performed a subgroup analysis using the univariate and multivariate Cox models to assess the association of CFB with 28-day mortality in the patients who had septic shock (Table 10). The univariate analysis showed that the positive FB and negative FB groups had higher mortality rates than the even FB group, but only the positive FB group had a greater mortality rate in the multivariate analysis. We conducted another subgroup analysis in the patients with fluid overload at CRRT initiation, in which the fluid overload was defined as a weight-adjusted CFB (from hospital admission to CRRT initiation) more than 5%. In line with the previous results, the positive FB group had a significantly higher mortality rate than the even FB group (Table 11).

We assessed the prognostic value of early CFB after CRRT initiation in a homogeneous population of patients with septic AKI. The results indicated that the patients with positive FB had a higher risk of 28-day mortality than the patients with even FB, but the patients with negative FB and even FB had similar risks of 28-day mortality. Furthermore, the survivors in the positive FB, even FB, and negative FB groups had no significant differences in the RRT dependence, mechanical ventilation-free days, vasopressor-free days, or length of ICU stay.

One of the main concerns in the treatment of patients with AKI undergoing CRRT is providing precise control of FB (27, 28). Hyung et al. reported that the CFB at 24 and 72 h after the initiation of CRRT were significantly higher in the 28-day non-survivors than survivors (19); however, we found no significant difference in CFB between the survivor and non-survivor groups. One possible reason for our disparate results is that we included the patients who had simultaneous sepsis and AKI, and fluid resuscitation was a major step in the management of these patients. The previous studies of patients with AKI reported that a positive FB after CRRT initiation increased the risk of adverse outcomes (18, 20). A secondary analysis of the Randomized Evaluation of Normal versus Augmented Level (RENAL) trial clearly showed that the presence of a mean daily positive FB after RRT initiation, even within the first 48 h of RRT, was independently associated with the higher mortality in the critically ill patients with severe AKI (18). A prospective cohort study that assessed the association of FB in the 7 days after RRT initiation reported similar results (20). The findings of our study are consistent with these previous studies, in that a positive FB is associated with an increased risk of death in the patients with septic AKI undergoing CRRT. A difference in our study is that we used an even FB group (rather than the negative FB group) as a comparator. Additionally, the prior studies measured volume status as mean daily FB after CRRT initiation, whereas we used CFB during the first 48 and 72 h after onset of CRRT. Despite this methodologic difference, we found similar associations between the positive FB and unfavorable outcomes. Thus, the determination of appropriate fluid management during the RRT is an important topic for future clinical trials.

A recent large retrospective study by Balakumar et al. (26) demonstrated that positive FB and negative FB before RRT initiation were both associated with higher mortality relative to even FB. Furthermore, the present study found that the 28-day mortality in the negative FB group was higher than in the even FB group; however, our univariate and multivariate Cox analysis indicated that negative FB did not significantly increase the risk of all-cause mortality at day-28 relative to even FB. One possible interpretation of these results is that fluid removal using the RRT may provide benefits to some patients because we calculated CFB during the first 48 h after the CRRT initiation. In contrast, Balakumar et al. (26) calculated CFB before RRT initiation. Moreover, we found a marginal J-shaped relationship (rather than a linear relationship) between the 48 h CFB and 28-day mortality in an unadjusted model, although this relationship was not significant in the adjusted model. It is likely that our finding of a negative effect of FB during RRT differed from some previous studies (29, 30) because of differences in the characteristics of patients. In our study, 60% of the patients received vasopressor support at CRRT initiation and 80% experienced septic shock during their ICU stays. The safe achievement of a negative FB during the late phases of septic shock is considered an effective strategy of fluid management (31), and active fluid removal using RRT may cause hemodynamic instability and lead to a worse outcome. Our results may suggest that achieving a negative FB rapidly after CRRT initiation is potentially harmful in patients with septic AKI. Thus, further research is needed to elucidate the benefits and harms associated with the negative FB in these patients.

Our study also evaluated the relationships of CFB and renal recovery in critically ill adults with septic AKI. We found that negative FB and positive FB after CRRT initiation were unrelated to the renal recovery, consistent with the prior studies (20, 26). Our study, thus, confirmed the recent findings that a substantial percentage of RRT-requiring AKI survivors remain dependent on the RRT even after the acute phase of their illness has resolved (20, 32, 33). This serves as a reminder of the need for the measures that protect and restore kidney functions during and after an episode of AKI that necessitates RRT.

To our knowledge, this is the first study to use an even FB group as a comparator to evaluate the association of CFB status during the CRRT and mortality in patients with septic AKI. Nonetheless, there were several limitations in the current study. First, our study was retrospective, conducted in a single center, and the sample size was relatively small. Thus, the selection bias was possible and we were unable to make causal inferences regarding the FB and outcomes. However, we used propensity scores in the sensitivity analysis to reduce the effects of outcome-selection bias. Second, our choice of measuring CFB at 48 h after initiation of CRRT was somewhat arbitrary. Nevertheless, the timing of this assessment varies among the studies, and there is no consensus on the optimal time for this measurement. In addition, our sensitivity analysis of the patients who survived at least 3 days after CRRT initiation produced similar results. Third, although we adjusted for confounding using robust multivariable regression analysis, residual confounding by unknown factors is possible.