Introduction

Early, rapid fluid resuscitation to restore hemodynamic integrity is the cornerstone to managing shock [1] and a strategy which has been championed by the early goal-directed therapy of the Surviving Sepsis Campaign [2]. However, fluid administered to critically ill children during the resuscitative phase has the potential to accumulate for many reasons, including increased capacity of the intravascular space, third-spacing and a decreased ability of the kidneys to excrete the excess fluid. There has been limited evaluation of the impact of such early fluid accumulation in this patient population.

Fluid accumulation can eventually lead to fluid overload and resulting organ dysfunction [3–5]. Many recently published observational studies have shown that higher fluid overload at the time of intervention [e.g. renal replacement therapy (RRT)] is associated with increased mortality and morbidity [6–10]. Other studies, both observational and experimental, have evaluated the impact of liberal versus restrictive fluid management strategies on physiological and clinical outcomes [11–16]. More recently, a clinical trial in African children with severe infection has provocatively challenged the benefits of fluid boluses in early resuscitation [17]. However, studies establishing the link between early fluid accumulation resulting from liberal fluid management and the eventual development of fluid overload and poor outcomes are lacking.

The objective of our study was to evaluate the association between early fluid accumulation and mortality in critically ill children admitted to the intensive care unit (ICU) with sepsis and shock states. We monitored fluid accumulation after ICU admission and defined early fluid overload as the peak cumulative fluid accumulation of ≥10 % of body weight during the initial 3 days in the ICU. We hypothesized that the presence, severity and duration of early fluid overload are associated with increased ICU mortality.

Methods

Setting and study design

This study was conducted in a tertiary level multi-disciplinary ICU over a 7-month period, from September 2009 through March 2010. Approval from the Institutional Review Board was obtained prior to initiation of the study, with waiver of consent. It was designed as a case–control study, nested in a cohort of children admitted to the ICU with a diagnosis of sepsis or shock states during the study period. Children with early fluid overload were designated as cases and those without, as controls. The two groups of patients were compared with respect to primary and secondary outcomes.

Study population and data collection

Patients for recruitment to the study were identified by querying the ICU database for patients (age ≤18 years) admitted with the following diagnoses: systemic inflammatory response syndrome, sepsis, severe sepsis, septic shock and other shock states. Children who were <48 h in the ICU, either due to death or transfer out of the ICU, were excluded to avoid patients at the extremes of the severity of the illness spectrum. Premature neonates and children with post-operative congenital heart disease were also excluded to minimize the heterogeneity of the cohort. ICU mortality was designated as the primary outcome a priori and defined as all-cause mortality occurring during the ICU stay, including deaths resulting from withdrawal or limitation of care. The secondary outcomes evaluated included length of ICU stay and need for and duration of organ support [extra-corporeal life support (ECLS), mechanical ventilation, vasopressors/inotropes and RRT among survivors].

We retrospectively reviewed electronic medical records and abstracted data into standardized data collection forms. The study duration was defined as the entire length of ICU stay if it was <7 days and as the initial 7 days if the ICU stay if it was ≥7 days. For each patient, daily fluid data for the study duration and outcome data for the entire ICU stay were obtained. Other data collected included demographic data, diagnosis-related data and severity of illness and organ dysfunction/support data. Primary diagnoses were grouped into infectious and non-infectious diseases. Reasons for ICU admission were grouped into respiratory distress/failure, hemodynamic instability and others. The Pediatric Index of Mortality 2 (PIM 2) score was used as the measure of illness severity at ICU admission and calculated retrospectively [18, 19]. Respiratory failure was defined as the need for respiratory support (non-invasive or invasive ventilation), and shock was defined as need for vasopressor support (other than dopamine ≤5 mcg/kg/min) during the initial 24 h of admission. Severity of renal dysfunction was categorized based on urine output and estimated glomerular filtration rate (Schwartz method) using the pediatric Risk–Injury–Failure–Loss of Function–End Stage (pRIFLE) criteria [20, 21].

Fluid assessment

Fluid status for the study duration was characterized by the presence, severity and duration of fluid overload. Fluid accumulation was measured as the net balance between the intake and output and expressed as a percentage of the body weight at ICU admission [22]. Cumulative fluid accumulation for any given day was defined as fluid accumulation from the time of ICU admission to the day of interest. The presence of early fluid overload was defined as cumulative fluid accumulation of ≥10 % during the initial 3 days. Severity and duration of fluid overload were measured as peak cumulative fluid accumulation and the number of days with fluid overload, respectively, during the study duration.

Statistical analysis

Survivors and non-survivors were compared with respect to fluid accumulation characteristics and other confounders. Two-tailed Student t tests and Wilcoxon rank sum tests were used to analyze continuous variables, and Fisher exact tests were used for categorical variables. Predictors of mortality were assessed using multivariable logistic regression models. Since there was significant correlation between the different fluid accumulation characteristics, each of these was tested individually in different multivariable regression models. Stepwise regression procedures were utilized to produce the final models (entry significance level of 0.10 and stay significance level of 0.05).

Cases and controls were compared with respect to ICU mortality. A subset of cases and controls who survived admission to the ICU were compared with respect to secondary outcomes. Cox proportional hazards models were used to calculate hazard ratios (HR) for the risk of death during ICU stay (with controls as the reference group). All potential confounders were included in the initial model. The final model was generated using stepwise selection. Since cases and controls differed with respect to severity of illness, further matched analysis was performed. Cases and controls were matched for age group, severity of illness (PIM 2 score), presence of respiratory failure, need for vasopressor support and need for ECLS using a ‘nearest neighbor’ 1:1 matching algorithm [23]. Matched cases and controls were compared with respect to ICU mortality and other secondary outcomes. Finally, similar matched analysis was performed in a subgroup of non-ECLS patients in the cohort. Statistical analyses were performed using R version 3.1.1 [24, 25]. Power calculations were performed using OpenEpi [26].

Results

In total, 179 children were admitted to the ICU with sepsis or shock during the study period. Among these, 28 children with a short ICU stay (<48 h) and 37 children admitted primarily for management of prematurity or post-operative congenital heart disease were excluded from the study. Mortality rate among the 114 children ultimately included in the final analysis was 13 % (15/114). Baseline characteristics of the cohort are given in Table 1. The median age was 1.1 years, with about 47 % of the children being younger than 1 year. There was a slight male preponderance (59 %). A significant proportion of the patients required mechanical ventilation (77 %) and vasopressor support (49 %) during the initial 24 h of their stay in the ICU. Infection was the most common primary diagnosis (83 %), and the most common underlying illness was a diagnosis of cancer (13 %). The majority of the patients were admitted to the ICU due to respiratory issues (57 %) and a smaller proportion with hemodynamic instability (27 %). The majority of the patients had some degree of renal dysfunction at the time of ICU admission (64 %).

Table 1 Baseline characteristics of the entire study cohort, survivors and non-survivors
Full size table

Survivors vs. non-survivors

Comparison of the baseline patient characteristics of the survivor group and the non-survival group is shown in Table 1. No significant differences were noted between these two groups with respect to age, gender and weight. Compared to the survivors, non-survivors were clearly more ill (based on PIM 2 score) and more likely to require organ support. Renal dysfunction at the time of ICU admission was more common among the non-survivors. The two groups did not differ significantly with respect to primary diagnoses, underlying co-morbidities or reason for ICU admission.

Figure 1 shows the mean daily cumulative fluid accumulation for the two groups. During the initial 3 days in the ICU, the median peak cumulative fluid accumulated was higher for non-survivors than for survivors (14 vs. 5 %; p < 0.001). This trend continued for the 7-day period (20 vs. 7 %; p < 0.001). A larger proportion of non-survivors compared to survivors had early fluid overload (73 vs. 24 %; p < 0.001), with 80 and 60 % of non-survivors having at least 1 and ≥2 days of fluid overload, respectively. In comparison, only 25 and 14 % of survivors had at least 1 and ≥2 days of fluid overload, respectively.

Fig. 1
figure 1

Fluid accumulation during the first 7 days (Day 0 to Day 7) of admission to the intensive care unit (ICU) in survivors vs. non-survivors. Plot on left Mean cumulative fluid accumulation, plot on right mean daily fluid accumulation. Fluid accumulated is shown on the y-axis and expressed as a percentage of admission body weight. Black curve Survivors, gray curve non-survivors. Error bars indicate standard errors of mean. Survivors and non-survivors are compared using the Student t test (significance:**α = 0.001, *α = 0.05

Full size image

Predictors of mortality: multivariable logistic regression analysis

The odds ratios (OR) for the various fluid accumulation parameters are given in Table 2. There was significant correlation between the different fluid accumulation parameters and hence they were tested individually in different multivariable models [Electronic Supplementary Material (ESM) Table 1]. An increase in mortality of about 14 % was noted with every 1 % (of body weight) increase in peak cumulative fluid accumulated during the initial 3 ICU days (adjusted OR 1.14, p 0.001). Similarly, a threefold increase in mortality was noted with every additional day of fluid overload during the initial 7 days (adjusted OR 3.13, p < 0.001). A significant interaction was noted between the severity (peak early fluid accumulation) and duration of fluid overload (p = 0.01). Introduction of the interaction term (peak cumulative fluid accumulation × duration of fluid overload) into the multivariable model increased the lethality of both, with the adjusted OR for peak fluid increasing to 1.38 (p 0.007) and the adjusted OR for duration of fluid overload increasing to 9.80 (p = 0.005). Other independent predictors identified in this model were presence of infectious diagnosis, severity of illness score and duration of hospitalization prior to ICU admission.

Table 2 Results of the multivariable logistic regression analysis
Full size table

‘No early fluid overload’ group versus ‘early fluid overload’ group

There were 72 patients (controls) in the ‘no early fluid overload’ (Non-EFO) group and 42 patients (cases) in the ‘early fluid overload’ (EFO) group. Table 3 summarizes the comparison between the two groups. Patients in the two groups were similar with respect to age, gender and size. A higher proportion of patients in the EFO group required mechanical ventilation (90 vs. 69 %; p 0.01) and vasopressor support (57 vs. 44 %; p 0.24). Similar trends were noted with respect to severity of illness score (PIM 2 score). No significant differences were noted with respect to reason for ICU admission, primary diagnosis and underlying co-morbidities. The two patient groups also did not significantly differ with respect to degree of renal dysfunction. Mortality was higher in the EFO group (26 vs. 6 %; p 0.003), and this difference was statistically significant at a two-sided α = 0.05, with a power of 0.86. A higher proportion of the patients in the EFO group required ECLS and RRT. Survivors in the EFO group also showed a trend toward longer ICU stay (8.5 vs. 7 days; p 0.08) and a higher number of ventilator days (6 vs. 2; p 0.004).

Table 3 Comparison of ‘no early fluid overload’ group and ‘early fluid overload’ group
Full size table

Matched nested case–control analysis

Matched analysis was performed to account for the differences between the EFO and Non-EFO groups, especially with respect to severity of illness indicators. Thirty cases from the EFO group were compared with 30 matched controls from the Non-EFO group. This sample size provided a power of 0.92 at two-sided α = 0.05. The two groups were similar with respect to all of the severity of illness indicators after matching (ESM Table 2). Mortality remained higher in the EFO group (37 vs. 3 %; p = 0.002). The comparison between the two groups with respect to other outcomes is summarized in Table 4. A significantly higher proportion of the 30 patients of the EFO group required ECLS in spite of ECLS being used as a factor in the matching algorithm. A similar matched analysis in a subgroup of non-ECLS patients continued to show a difference in mortality between EFO and Non-EFO groups (Table 4, ESM Table 3). The EFO group also showed a trend towards increased need for RRT in both the primary matched analysis and the non-ECLS subgroup analysis.

Table 4 Matched case–control analysis: ‘no early fluid overload’ (controls) versus ‘early fluid overload’ (cases)
Full size table

Survival analysis

The univariate Cox-proportional hazards model showed a higher hazard ratio of death for patients of the EFO group compared to those of the Non-EFO group (HR 4.17, p = 0.02). Early fluid overload was adjusted for age group, severity of illness (PIM 2 score), respiratory failure, vasopressor need, infectious diagnosis, reason for ICU admission and oncology diagnosis in a multivariable model. Significant factors in the final model after the stepwise procedure included early fluid overload (adjusted HR 7.27, p 0.003), severity of illness (adjusted HR 1.04, p = 0.001), infectious diagnosis (adjusted HR 13.65, p 0.03) and underlying oncology diagnosis (adjusted HR 16.98, p 0.03). ESM Fig. 1 shows the adjusted survival curves for the EFO and Non-EFO groups.

Discussion

Recent years has seen an increasing awareness of the impact of fluid overload on outcomes, resulting in the management of fluid overload becoming an important target for intervention in critically ill patients. However, once fluid overload sets in, mortality remains high, irrespective of the intervention strategy [8, 27, 28], suggesting that it is the prevention of fluid overload which may be essential to improved outcomes. Fluid resuscitation at the time of ICU admission is a potential source of excessive fluid administration that may lead to overload. There is significant variability and heterogeneity in the approach to initial fluid resuscitation [29], with the type and volume of resuscitation fluid, specific immediate goals for fluid therapy and approaches adapted to monitoring hemodynamic and fluid status varying from patient to patient and from practitioner to practitioner [13]. Macro-circulatory goals are currently recognized as poor indicators of micro-circulation, especially in sepsis and septic shock [30].

In this study, we evaluated the impact of early fluid accumulation on ICU mortality and other outcomes in critically children. We study group included only those who were admitted to the ICU for least 48 h, with the intent to focus on a group which could potentially benefit from appropriate early fluid management. Fluid balance may not impact outcomes in patients at extremes of the severity of illness spectrum, especially those in the high-risk category for whom the risk of mortality is high irrespective of fluid balance [31]. We defined early fluid accumulation as the peak fluid accumulation occurring within the initial 3 days of ICU admission because we found that in our cohort, both survivors and non-survivors, fluid accumulation peaked on the second day. Further, we performed a matched analysis by matching for various confounders that could potentially impact mortality. The association between early fluid accumulation and mortality remained significant even after matching. Analysis of a subgroup of non-ECLS patients also continued to show this association.

In their study of critically ill adults with acute kidney injury (AKI), Bouchard et al. evaluated the relationship between fluid accumulation at the time of AKI diagnosis and mortality [32]. These investigators found higher mortality among patients in the fluid overload group compared to those in the non-fluid overload group (30-day mortality 37 vs. 25 %; p 0.02). However, this difference did not remain statistically significant after adjustment for the severity of illness score (APACHE III). In contrast, in our cohort, we continued to observe a difference between the EFO and non-EFO groups even after matching for severity of illness indicators. In another study in critically ill adults with septic shock, Boyd et al. noted a similar difference in mortality with fluid accumulation at both 12 h and 4 days after ICU admission [33]. These authors were also able to demonstrate a relationship between early fluid accumulation at 12 h and central venous pressure. In another trial using calfactant, a surfactant extract from calf lungs, in children with acute respiratory distress syndrome, post hoc analysis demonstrated the association between early fluid accumulation and mortality [34]. Interestingly, early fluid accumulation and fluid overload occurred in this trial, despite a conservative fluid management algorithm built into the study design.

In many studies, adequate early fluid resuscitation during the initial few hours after the recognition of shock has been associated with a decrease in mortality [35–38]. In a multicenter trial involving adult patients with septic shock, mortality was lower when at least 1 L of fluid was administered during the first hour after onset of hypotension followed by vasopressor initiation between 1 and 6 h post-hypotension onset [36]. Similarly, in another adult study of critically ill patients whose ICU stay was at least 3 days, an association was noted between higher fluid balance on the first day in the ICU and lower mortality (adjusted OR 0.71) [2]. In contrast, in our study, survivors had lower fluid balance even on their first day in the ICU compared to non-survivors, both in the unmatched cohort (2 vs. 7 %, p = 0.02) and matched subgroup (3 vs. 8 %, p = 0.04). The inclusion in our analysis of non-surviving ECLS patients, who received significantly higher amounts of fluid than the other subgroups in our cohort, partly explained this difference. When ECLS patients were excluded, first-day fluid balance was similar between survivors and non-survivors (2 vs. 3 %; p = 0.54).

Another potential explanation for this discrepancy is that the initial fluid resuscitation—to the extent that adequate intravascular volume is achieved—is essential, while further fluid resuscitation will only lead to accumulation and overload. The results of recent studies support this hypothesis [2, 35]. To the contrary, the FEAST trial in African children challenges the need for fluid resuscitation even during the initial phase [17]. The authors of this study found both a higher early mortality (<48 h) and a higher late mortality (4 weeks) among children who received fluid boluses in response to impaired perfusion compared to those who did not receive fluid boluses. Interestingly, in the post hoc analysis, fluid boluses did provide short-term benefit in terms of resolution of shock state, but children who received fluid boluses had increased mortality due to progression of cardiovascular dysfunction after this initial improvement [39, 40].

A potential limitation to our study is its single-center, retrospective observational design. We performed a matched analysis to minimize the impact of confounders on the outcomes. The single-center design and the short (6-month) study period may minimize the impact of center-based heterogeneity as well as the impact of secular/temporal changes in management and outcomes. Further larger, prospective, multicenter studies are required to confirm our findings.

Conclusions

An association between severity and duration of early fluid accumulation and ICU mortality was noted in this cohort of critically ill children with sepsis and in shock states. This association remained significant after adjusting for age, severity of illness and various other confounders. Randomized controlled trials are needed to further assess the impact of early fluid accumulation. Identifying fluid responsive states prior to fluid administration and the development of reliable, objective clinical goals for fluid therapy during the early resuscitative phase of critical illness are the essential next steps prior to designing such trials.