|Home | About | Journals | Submit | Contact Us | Français|
In pediatric intensive care unit (PICU) patients, fluid overload (FO) at initiation of continuous renal replacement therapy (CRRT) has been reported to be an independent risk factor for mortality. Previous studies have calculated FO based on daily fluid balance during ICU admission, which is labor intensive and error prone. We hypothesized that a weight-based definition of FO at CRRT initiation would correlate with the fluid balance method and prove predictive of outcome.
This is a retrospective single-center review of PICU patients requiring CRRT from July 2006 through February 2010 (n = 113). We compared the degree of FO at CRRT initiation using the standard fluid balance method versus methods based on patient weight changes assessed by both univariate and multivariate analyses.
The degree of fluid overload at CRRT initiation was significantly greater in nonsurvivors, irrespective of which method was used. The univariate odds ratio for PICU mortality per 1% increase in FO was 1.056 [95% confidence interval (CI) 1.025, 1.087] by the fluid balance method, 1.044 (95% CI 1.019, 1.069) by the weight-based method using PICU admission weight, and 1.045 (95% CI 1.022, 1.07) by the weight-based method using hospital admission weight. On multivariate analyses, all three methods approached significance in predicting PICU survival.
Our findings suggest that weight-based definitions of FO are useful in defining FO at CRRT initiation and are associated with increased mortality in a broad PICU patient population. This study provides evidence for a more practical weight-based definition of FO that can be used at the bedside.
Acute kidney injury (AKI) occurs commonly in pediatric intensive care unit (PICU) patients and is associated with increased mortality of up to 50% [1–5]. Over the past two decades, continuous renal replacement therapy (CRRT) has emerged as a standard therapy for patients in the PICU with AKI [6–8]. Despite such technological advances there remains significant room for improvement in the outcomes of pediatric patients with AKI.
A growing body of literature suggests that the degree of fluid overload at initiation of renal replacement therapy is associated with increased mortality and may represent an important target for intervention. While first reported in pediatric bone marrow transplant patients [9–11], the importance of the concept of fluid overload has been extended to broader PICU patient populations requiring CRRT [12–17]. The American College of Critical Care Medicine practice guidelines for pediatric and neonatal septic shock have recognized a threshold of 10% volume overload as a key time to act, although they do not specify interventions . The clinical significance of fluid overload has also been demonstrated in adult patients with sepsis , respiratory distress syndrome [20, 21], and acute kidney injury [22–25].
Several important questions remain regarding fluid overload as a clinical marker, including the optimal method for calculating fluid overload. To date, most pediatric studies have relied on the definition put forth by Goldstein et al. in 2001, which is based on measuring fluid input and output from PICU admission . The accuracy of this formula relies on precise accounting of fluid balance on a daily basis and can be cumbersome for the clinician to calculate and error prone. In clinical practice, weight is frequently used as a surrogate to define fluid overload, yet there are no studies comparing weight-based definitions of fluid overload and their association with mortality in pediatric patients requiring CRRT. There remains no consensus as to what the optimal clinical definition of fluid overload should be.
The aims of this study are to confirm the association between fluid overload and outcome in a large and diverse PICU population undergoing CRRT and to assess the predictive value and correlation of weight-based methods versus the fluid balance method for calculating fluid overload. We hypothesized that a practical weight-based definition of fluid overload at CRRT initiation would correlate with the fluid balance method and prove predictive of outcome in pediatric patients requiring CRRT.
We conducted a retrospective review of all pediatric patients undergoing CRRT between July 2006 and February 2010 in C.S. Mott Children’s Hospital at the University of Michigan. Patients were admitted to the neonatal intensive care unit, cardiothoracic unit or the PICU. Patients were excluded if they were in the PICU for >2 months prior to initiation of CRRT. Patients who had multiple courses of CRRT separated by >24 h were included for the first course of CRRT only. This study was approved by the institutional investigational review board.
The primary modality of CRRT performed was continuous venovenous hemodiafiltration (CVVHDF) for patients not requiring extracorporeal life support (ECLS). In patients receiving CVVHDF, CRRT was performed using the Gambro Prismaflex system (Gambro, Lund, Sweden). Until December 2008, patients requiring ECLS received continuous venovenous hemodialysis by having a dialysis filter placed in line with the ECLS circuit. After December 2008, we began to perform CVVHDF using the Prismaflex system for patients on ECLS. For patients weighing <25 kg, CRRT was performed using an AN-69 M60 filter (Hospal, France). A polysulfone HF 1000 (Hospal, France) was used for the remainder of patients, who were not on ECLS. Polysulfone HF 400 (Renalflo II, MN) or Optiflux (Fresenius, Germany) filters were used in line for patients requiring ECLS based on body surface area. Anticoagulation was performed using a standardized regional citrate protocol . Prior to 2008, heparin was used for anticoagulation in patients requiring CRRT in line with ECLS.
Data collection included demographic data, comorbidity data, laboratory data, indication for CRRT defined by the pediatric nephrology note, and characteristics of each subject’s hospital and ICU course (e.g., length of stay, requirement for vasoactive agents, requirement for mechanical ventilation, etc.). Pediatric Risk of Mortality (PRISM) III scores were calculated at ICU admission .
AKI was classified by the RIFLE and pRIFLE criteria based on serum creatinine, estimated creatinine clearance (eCCl), and urine output in the 24 h prior to initiation of CRRT  . The pRIFLE was modified slightly to exclude the “failure” component of eCCl < 35 ml/min/1.73 m2 for children ≤14 days. Estimated glomerular filtration rate (GFR) was calculated using the Schwartz equation in patients <18 years old  and the Modification of Diet in Renal Disease (MDRD) formula in patients >18 years old .
For fluid status determination we recorded weight upon hospital admission, weight at ICU admission, weight upon CRRT initiation, fluid intake from ICU admission until CRRT initiation, and fluid output from ICU admission until CRRT initiation. It is standard of care at our institution to weigh patients on ECLS daily. Fluid intake included blood products, intravenous fluids and flushes, medications, and all forms of nutritional support. Fluid output included urine output, drain output, blood loss, nasogastric tube output, stool volume, and wound drainage. For each patient, the daily flow charts were reviewed and 24 h totals of fluid intake and output were recorded for each patient for every day on the intensive care unit prior to CRRT initiation. These daily totals were then used to calculate the degree of fluid overload as described by Goldstein et al. :
This method was then compared with two weight-based formulas. These formulas calculated fluid overload based upon ICU admission weight and hospital admission weight:
The primary outcome was all-cause ICU mortality.
Due to skewness of several variables, continuous variables are represented as median (25th percentile, 75th percentile). A small amount of missing data in explanatory variables was handled via multiple imputation. Univariate comparisons were made using the nonparametric Mann–Whitney U and Kruskal–Wallis tests, as well as chi-square and Fisher’s exact tests, as appropriate.
Methods of calculating fluid overload were compared directly via Pearson correlation coefficients. The distribution of each method was also examined and compared, and differences between each patient’s fluid-based and weight-based methods were assessed.
Due to high correlations between the methods for fluid overload determination, separate multivariate logistic regression models predicting ICU mortality were conducted for all three fluid overload calculation methods. Each set of models adjusted for the same covariates, including PRISM III score and variables with p < 0.10 after backward selection on models containing method 1. Regression models were then compared using receiver–operator characteristic (ROC) curves, and the area under the curve (AUC) was calculated.
Subanalyses were conducted to assess the impact of fluid overload on survival within ECLS and non-ECLS patients. Models were refitted without extreme outliers, with no substantial difference in results. Significance for all statistical tests was set at two-sided α = 0.05. Analyses were conducted using SAS 9.1 (SAS Institute, Cary, NC).
During the study period a total of 116 patients underwent CRRT in our institution. Three patients were excluded from our data analysis because of prolonged intensive care unit stays (>2 months) prior to CRRT initiation, resulting in a final study population of 113 patients. Median patient age was 19 months [interquartile range (IQR) 0.2, 181 months]. This included 37 patients 1 month of age or younger (30 patients on ECLS). Median number of hospital days prior to CRRT initiation was 6 (2, 16), and median number of PICU days prior to CRRT initiation was 3 (2, 6). Ninety-eight (87%) patients were on mechanical ventilation at time of CRRT initiation, and 85 (75%) patients were receiving vasoactive agents at CRRT initiation. Fifty (44%) patients were receiving ECLS at CRRT initiation. Nineteen (17%) patients received therapeutic plasma exchange (TPE) while on CRRT. Baseline characteristics of the study population, both overall and by ECLS status, are summarized in Table 1.
Median PRISM score at ICU admission was 13 (7,18). Fluid overload was the primary indication for initiation of CRRT in 73 patients (65%, Table 2). The most common underlying diagnosis was heart disease, in 41 patients (36%, Table 3).
Overall survival to ICU discharge was 44% and was significantly lower for patients treated with ECLS compared with those patients who did not receive ECLS (32% versus 54%, p = 0.0195). On univariate analysis, several variables were significantly different between survivors and nonsurvivors (Table 4). These included age, number of hospital days prior to CRRT initiation, number of ICU days prior to CRRT initiation, presence of vasoactive medications, number of vasoactive medications, and mechanical ventilation at CRRT initiation. The degree of fluid overload at initiation of CRRT was significantly greater among nonsurvivors when compared with survivors, regardless of the method used to calculate fluid overload (Table 5, p < 0.05). This remained true for patients on ECLS (Table 5, p < 0.05). Univariate analysis yielded a significant association between fluid overload at CRRT initiation and mortality (Table 6, p < 0.05).
Multivariate logistic regression analysis found patient age, number of hospital days prior to CRRT initiation, presence of ECLS, pRIFLE score of failure, and number of vasoactive agents present at initiation to predict survival. While trending toward predicting mortality, fluid overload did not reach statistical significance on multivariate analysis (Table 6). Using method 1, the odds ratio for a 1% increase in fluid overload was 1.04 (95% CI 1.00–1.07, p = 0.0529). Method 2 gave an odds ratio of 1.03 (95% CI 0.99–1.07, p = 0.0829), and method 3 also gave an odds ratio of 1.03 (95% CI 0.99–1.06, p = 0.10). Additional analysis using fluid overload as a categorical variable (cutoff of either 10% or 20% fluid overload) did not demonstrate statistically significant associations either (data not shown).
There was a high degree of correlation between methods 1 and 2 (Pearson’s coefficient 0.77), methods 1 and 3 (Pearson’s coefficient 0.75), and methods 2 and 3 (Pearson’s coefficient 0.92). All three methods shared similar predictive ability as assessed by the construction of ROC curves. The AUC for the multiple logistic regression models for method 1 was 0.881 compared with 0.858 for method 2 and 0.855 for method 3.
This study is the first to systematically evaluate weight-based methods as a means to calculate fluid overload in pediatric patients at initiation of CRRT and its association with mortality. Our study is consistent with previous reports showing an association between fluid overload at CRRT initiation and mortality [9, 12–14] while providing a more practical weight-based approach to determining degree of fluid overload. Our study extends this finding to a broader pediatric patient population that includes a significant number of neonatal and ECLS patients.
In their recent analysis of the multicenter Prospective Pediatric CRRT (ppCRRT) registry of 297 patients, Sutherland and colleagues reported an adjusted mortality odds ratio of 1.03 (95% CI 1.01–1.05)  associated with increasing fluid overload at time of CRRT initiation. Using both fluid balance and weight-based methods to define fluid overload, we found nearly identical results. We observed high correlation and comparable predictive values (AUC) between the methods, suggesting that they may be used interchangeably. We chose to assess the weight-based method because this is a more practical and less labor-intensive approach compared with calculating cumulative fluid balance. In addition, studies have reported that methods utilizing fluid balance to determine fluid overload are often inaccurate and unreliable when compared with daily weights in ICU [32, 33] or general floor settings . Fluid balance calculations are generally unable to account for insensible fluid losses, and so weight-based calculations could provide for improved control of this variable. It is worth noting that daily weights also have the potential for inaccuracy related to the use of differing scales and techniques for weighing patients. There are potential safety issues around weighing ECLS patients daily; if done improperly, weighing patients on ECLS can pose dangers to the patient, including decannulation.
Nearly all previous studies examining fluid overload have defined baseline weight at the time point of ICU admission. However, fluid overload can begin to develop during the hospital stay prior to ICU admission, and thus ICU admission weights may underestimate the degree of fluid overload. We compared fluid overload definitions based on hospital admission (method 3) and ICU admission (method 2) weights. As expected, fluid overload was consistently higher using method 3. Interestingly, both methods demonstrated significant differences in percentage fluid overload between survivors and nonsurvivors, and both had similar predictive value for mortality. This was in part due to the fact that the values were identical (i.e., hospital admission was directly to the ICU) in 72% of cases. In addition, both methods correlated with the fluid balance method. Based on our limited data, we cannot conclude whether one approach is superior in providing prognostic information, and future studies are needed to explore the optimal definition of baseline weight.
In previous studies, neonates and patients requiring ECLS were underrepresented or excluded when investigating the association of fluid overload at CRRT initiation and mortality. At our institution, patients less than 1 month of age accounted for a large proportion of our CRRT population. Symons and colleagues previously reported a high mortality in patients less than 10 kg requiring CRRT , but the contribution of fluid overload at CRRT initiation was not investigated. We strengthen the findings of the ppCRRT group about the association of fluid overload at CRRT initiation and increased mortality in patients 1 month of age or younger [12, 36]. By the inclusion of these patients we extend the clinical importance of fluid overload to a previously understudied patient population.
This study is the first to include a significant ECLS patient population (50 patients) in the evaluation of the association between fluid overload at CRRT initiation and mortality. The importance of fluid overload in ECLS patients was reported by Swaniker and colleagues, who noted a significant difference in fluid overload at ECLS initiation between survivors and nonsurvivors in a single-center review of 128 patients ; only 14% of the patients required renal replacement therapy. Paden and colleagues recently reported a large single-center review of 154 pediatric patient on ECLS requiring CRRT with 44% survival, but did not report on fluid overload at CRRT initiation . The importance of fluid overload at CRRT initiation in patients on ECLS has not been investigated. The majority of patients in our study were placed on ECLS for underlying heart disease, and there was an overall high incidence of heart disease as the primary disease. The importance of fluid overload and early initiation of CRRT in patients with cardiorenal syndrome has become increasingly recognized . Our results further emphasize this, as subgroup analysis of the ECLS population showed fluid overload at CRRT initiation to be significantly greater in nonsurvivors compared with survivors by weight-based or fluid balance methods. Future studies examining fluid overload status in patients on CRRT should consider inclusion of patients requiring ECLS.
Some limitations of this study should be noted. The pRIFLE criteria have not been validated in patients <28 days of age, and we modified the criteria slightly to allow for classification of younger patients. This was done to allow for a measure of renal failure in the analysis. Interestingly, the RIFLE and pRIFLE criteria provided very similar rates of “failure” at CRRT initiation in this study. In addition, we were not able to completely stratify for severity of illness or multiple organ dysfunction. The inclusion of a large ECLS population makes severity of illness scores at CRRT initiation unreliable, as Pediatric Logistic Organ Dysfunction (PELOD) scores have not been validated in patients on ECLS and PRISM scores have only been validated at time of ICU admission. We attempted to account for this with a robust multivariate model that included PRISM III score at ICU admission. Lastly, due to heterogeneity of the patient population, our study was underpowered to detect a difference by the multivariate model. With our current sample size, we had 60% power to detect the anticipated odds ratio of 1.03 in a multivariate model, and 250 patients would have been needed to achieve 90% power.
This study confirms the association between fluid overload and mortality in a broad PICU patient population. We show that weight-based definitions of fluid overload can be useful alternatives in predicting increased mortality at CRRT initiation in a broad PICU patient population. We extend the literature on fluid overload by including a significant patient population on ECLS and patients less than 1 month of age. The results of this study support a more practical weight-based definition of fluid overload for use at the bedside to guide medical decision-making.
David T. Selewski is supported by the Cellular and Molecular Biology in Pediatrics Training Program grant (T-32 HD007513-13). Timothy T. Cornell is supported by the Pediatric Critical Care Scientist Development Program (K12HD047349) and an individual Career Development Award (K08HD062142). Neal B. Blatt is supported by a Child Health Research Career Development Award (National Institutes of Health, K12 HD 028820 (T32 DK 065517).
David T. Selewski, Division of Nephrology, Department of Pediatrics and Communicable Diseases, C.S. Mott Children’s Hospital, University of Michigan, Ann Arbor, MI, USA.
Timothy T. Cornell, Division of Critical Care, Department of Pediatrics and Communicable Diseases, C.S. Mott Children’s Hospital, University of Michigan, Ann Arbor, MI, USA.
Rebecca M. Lombel, Division of Nephrology, Department of Pediatrics and Communicable Diseases, C.S. Mott Children’s Hospital, University of Michigan, Ann Arbor, MI, USA.
Neal B. Blatt, Division of Nephrology, Department of Pediatrics and Communicable Diseases, C.S. Mott Children’s Hospital, University of Michigan, Ann Arbor, MI, USA.
Yong Y. Han, Division of Critical Care, Department of Pediatrics and Communicable Diseases, C.S. Mott Children’s Hospital, University of Michigan, Ann Arbor, MI, USA.
Theresa Mottes, Division of Nephrology, Department of Pediatrics and Communicable Diseases, C.S. Mott Children’s Hospital, University of Michigan, Ann Arbor, MI, USA.
Mallika Kommareddi, Division of Nephrology, Department of Internal Medicine, University of Michigan, Ann Arbor, MI, USA.
David B. Kershaw, Division of Nephrology, Department of Pediatrics and Communicable Diseases, C.S. Mott Children’s Hospital, University of Michigan, Ann Arbor, MI, USA.
Thomas P. Shanley, Division of Critical Care, Department of Pediatrics and Communicable Diseases, C.S. Mott Children’s Hospital, University of Michigan, Ann Arbor, MI, USA.
Michael Heung, Division of Nephrology, Department of Internal Medicine, University of Michigan, Ann Arbor, MI, USA.