Timing and Clinical Significance of Fluid Overload in Pediatric Acute Respiratory Distress Syndrome

Abstract

Objective:

Fluid overload is associated with worse outcomes in adult and pediatric acute respiratory distress syndrome (ARDS). However, the time-course of fluid overload and its relationship to outcome has not been described. We aimed to determine the relationship between the timing of fluid overload and outcomes over the first 7 days after ARDS onset in children.

Design:

Retrospective cohort study.

Setting:

Single tertiary care pediatric intensive care unit (PICU).

Patients:

Intubated children with ARDS between 2011 and 2019.

Interventions:

None.

Measurements and Main Results:

Daily and cumulative total fluid intake, total output, urine output, and fluid balance were collected for each 24-hour period from days 1 to 7 after ARDS onset. We tested the association between daily cumulative fluid metrics with PICU mortality and probability of extubation by 28 days using multivariable logistic and competing risk regression, respectively. In a subset of children, plasma was collected on day 1 and day 3 of ARDS and angiopoietin 2 (Ang2) quantified. Of 723 children with ARDS, 132 (18%) died. In unadjusted analysis, non-survivors had higher cumulative fluid balance starting on day 3. In multivariable analysis, a positive cumulative fluid balance on days 5 through 7 was associated with increased mortality. Higher cumulative fluid balance on days 4 to 7 was associated with lower probability of extubation. Elevated Ang2 on day 1 predicted early (within 3 days) fluid overload ≥ 10%, and elevated Ang2 on day 3 predicted late (between days 4 and 7) fluid overload.

Conclusions:

Fluid overload after day 4 of ARDS, but not before, was associated with worse outcomes. Higher Ang2 predicted subsequent fluid overload. Our results suggest that future interventions aimed at managing fluid overload may have differential efficacy depending on when in the time-course of ARDS they are initiated.

BACKGROUND

Acute respiratory distress syndrome (ARDS) is a disease of acute hypoxemia, with 18–35% mortality in pediatrics (1–3). Increased alveolar-capillary permeability contributes to pulmonary edema (4). One proposed pathogenic mediator is angiopoietin-2 (Ang2), which correlates with increased mortality in pediatric ARDS (5, 6). In adult ARDS, conservative fluid management resulted in improved oxygenation and more ventilator-free days (VFDs) at 28 days (7). Ang2 predicted worse outcomes in a subset of these patients (8). In retrospective analyses of pediatric ARDS, a positive cumulative fluid balance at day 3 of illness and beyond has been associated with fewer VFDs (9), longer length of stay (10), and increased mortality (10–13).

While the correlation between fluid overload and worse outcomes is established in pediatric ARDS, it is unclear when in the illness time-course the relationship between fluid overload and outcomes becomes relevant, as most studies simply report cumulative balance at 72 hours. Specifically, it is unclear whether fluid overload during the initial resuscitation period in early ARDS, the period of de-resuscitation after initial stabilization, or both, carries a stronger association with outcome. A granular understanding of the temporal relationship of fluid overload with poor outcomes could clarify the timing of de-resuscitation in future studies. Furthermore, existing studies are limited by small sample sizes (9, 11–13) and low mortality rates, precluding inclusion of several potential confounders. Therefore, in order to assess the relationship between the timing of fluid overload and outcomes, we leveraged an ongoing prospective cohort and abstracted detailed fluid intake, output, urine output (UOP), and cumulative fluid balance for the first 7 days after ARDS onset. Ang2 was measured in a subset of patients to assess for utility as a predictor of fluid overload.

METHODS

Study Design and Setting

This was a retrospective analysis of a prospective cohort (July 2011 to June 2019) from the Children’s Hospital of Philadelphia’s (CHOP) pediatric intensive care unit (PICU), approved by the Institutional Review Board (#14-011201) with requirement for informed consent waived. Ventilator escalation, fluid management, and use of diuretics or continuous renal replacement therapy (CRRT) were not protocolized. Sedation and ventilator weaning were protocolized, although the decision to extubate was also at the discretion of the attending. A subset of subjects after July 2014 were enrolled in an observational study with biomarker measurements on day 1 and 3 after ARDS onset, for which informed consent was obtained.

Participants

Intubated children meeting American-European Consensus Conference criteria for acute lung injury (two consecutive PaO₂/FIO₂ ≤ 300 separated by ≥ 1 hour with bilateral infiltrates) were included. As the study was initiated prior to the Berlin definition (14) minimum positive end-expiratory pressure (PEEP) was not specified; however, as all PEEP was ≥ 5 cmH₂O, all patients met Berlin criteria. Similarly, as the study was initiated prior to the Pediatric Acute Lung Injury Consensus Conference (PALICC) definition of pediatric ARDS (15) we did not screen using oxygenation index (OI); however, all but one patient met PALICC criteria by OI.

Variables and Definitions

Demographics, comorbidities, and ventilator settings for the first 72 hours from ARDS diagnosis were recorded prospectively, as the aim of the parent study was identification of determining early predictors of mortality in pediatric ARDS. Fluid variables (standardized to admission body weight) over the first 7 days were collected retrospectively. Intake and output were recorded per 24-hour period (7 AM to 7 AM) for the first 7 days from ARDS onset. “Day 1” was the day a subject met all ARDS criteria, irrespective of time. Total intake, output, UOP, and combination output (mixed urine and stool) from all previous 24-hour periods were summed to give cumulative daily values. Cumulative fluid balance was the primary exposure. If a patient died or was transferred out of the PICU prior to day 7, data was recorded through last 24-hour period with reliable data. If the final day had < 8 hours of data, values were excluded; if there was ≥ 8 hours of data, hourly rates were extrapolated to complete 24 hours.

Severity of illness score was Pediatric Risk of Mortality (PRISM) III score at 12 hours (16). Shock was quantified using vasopressor score (17). Non-pulmonary organ failures were identified using pediatric definitions (18). “Immunocompromised” required an immunocompromising diagnosis (oncologic, immunologic, rheumatologic, or transplant) and active immunosuppressive chemotherapy, or a congenital immunodeficiency (19, 20). Acute kidney injury (AKI) as Kidney Disease Improving Global Outcomes (KDIGO) stage 2 or 3: UOP < 0.5 ml/kg/hour, creatinine ≥ 2x baseline, or use of CRRT.

The primary outcome was PICU mortality, which was assigned as either being caused primarily by hypoxemia, multisystem organ failure (MSOF), or neurologic failure. The secondary outcome was VFDs at 28 days. Non-invasive support was not counted toward ventilator days. Liberation from ventilation ≥ 24 hours defined duration of ventilation. Patients requiring re-initiation of invasive ventilation had the extra days counted towards total ventilator days. VFDs were determined by subtracting total ventilator days from 28 in survivors. Patients with ≥ 28 ventilator days and PICU non-survivors were assigned VFD = 0.

Plasma Collection and Angiopoietin-2 Measurements

In the subset of subjects enrolled in the biomarker study, blood was collected on days 1 and 3 of ARDS in citrated tubes (Becton, Dickinson and Co; Franklin Lakes, NJ), centrifuged within 30 minutes (2000 g, 20 minutes, 20C) to generate platelet-poor plasma, aliquoted to prevent freeze/thaw cycles, and stored at −80C until analysis. Ang2 was measured in duplicate using an enzyme-linked immunosorbent assay (ELISA; R&D Systems, Minneapolis, MN).

Statistical methods

Analyses were performed using Stata 16.1 SE (StataCorp, College Station, TX). Univariate comparisons were made using Wilcoxon rank-sum or Fisher exact tests. For our primary analyses, we tested the association of each cumulative fluid variables (cumulative intake, cumulative output, cumulative UOP, cumulative fluid balance for each day up to day 7) with mortality. We performed multivariable logistic regression adjusting a priori for OI, number of organ failures, immunocompromised status, and vasopressor score on day 1 of illness. These confounders were chosen as they comprise the variables in a validated mortality prediction model for pediatric ARDS (21), and are all baseline variables reflecting ARDS severity, organ failure, and shock. Results are presented as adjusted odds ratios (aORs) per change in 100 mL/kg, with analyses adjusted for multiple comparisons using Bonferroni’s method (nominal p ≤ 0.007) given separate tests for days 1 through 7. To assess association with VFDs, we tested the association of these same cumulative fluid variables with probability of extubation, treating death as a competing risk, using competing risk regression adjusting for the same confounders above (22). Censoring at day 28 makes this analysis comparable to VFDs at 28 days.

We performed multiple sensitivity analyses. First, due to bias from patients either improving rapidly or dying before 7 days, we repeated analyses restricted to cases with data for all 7 days. Second, to assess whether there was bias by analyzing UOP rather than combination output, we repeated analyses using combination output. Third, to assess the impact of temporal trends, we repeated the analysis adjusting for year. Fourth, to assess whether etiology affected our conclusions, we performed an analysis stratifying subjects according to primary diagnosis (pneumonia, non-pulmonary sepsis, or other). To assess whether the cause of death affected our conclusions, we performed an analysis stratified by cause of death (hypoxemia, MSOF, or neurologic). Finally, to test whether poor outcomes associated with fluid overload were mediated by AKI or shock, we performed mediation analysis using structural equation modeling assessing whether AKI or maximum vasopressor score by day 3 mediated the association between late (days 4 to 7) fluid overload > 100 mL/kg fluid balance (> 10% fluid overload) and mortality.

RESULTS

Description of the Cohort

Over the study period, 723 children had ARDS with 132 (18%) non-survivors (Table 1), with most deaths caused by neurologic failure or MSOF. Survivors had lower PRISM III, organ failures, and vasopressor score than non-survivors. Survivors had a higher diuretic use after day 2 (Supplementary Table 1), although this did not reach statistical significance after multiple testing correction (all nominal p > 0.007). Non-survivors had a greater CRRT use than survivors on every day. Over the study period, there was a trend towards higher use of PEEP (Supplementary Table 2), and lower daily fluid intake on days 1 and 2 (Supplementary Table 3).

Table 1:

Demographics of the ARDS cohort stratified by survival status.

Variables	All patients (n=723)	Survivors (n=591)	Non-survivors (n=132)	p value
Age in years	5.1 [1.6, 12.9]	4.7 [1.5, 12.5]	7.2 [2.9, 13.8]	0.016
Female (%)	311 (43)	255 (43)	56 (42)	0.923
Severity of illness
PRISM III at 12h	11 [5, 18]	9 [5, 16]	17 [12, 30]	< 0.001
Non-pulmonary organ failure	2 [1, 3]	1 [1, 2]	3 [2, 4]	< 0.001
Vasopressor score	10 [0, 20]	8 [0, 16]	20 [7, 50]	< 0.001
Immunosuppressed (%)	154 (21)	89 (15)	65 (49)	< 0.001
Stem cell transplant (%)	65 (9)	29 (5)	36 (27)	< 0.001
Co-morbidities (%)
None	275 (38)	231 (39)	44 (33)	0.235
Prematurity	84 (12)	82 (14)	2 (2)	< 0.001
Chronic kidney disease	3 (0.4)	3 (0.5)	0 (0)	1.000
Chronic liver disease	14 (2)	8 (1)	6 (5)	0.028
Cause of ARDS (%)
Pneumonia	366 (51)	325 (55)	41 (31)	< 0.001
Aspiration	98 (14)	79 (13)	19 (14)
Sepsis	160 (22)	116 (20)	44 (33)
Trauma	42 (6)	34 (6)	8 (6)
Other	57 (8)	37 (6)	20 (15)
At ARDS onset
FIO2	0.5 [0.4, 0.9]	0.5 [0.4, 0.8]	0.65 [0.4, 1.0]	0.018
PaO2/ FIO2	156 [105, 217]	160 [111, 217]	141 [76, 225]	0.067
OI	10.4 [7.1, 17.6]	10.3 [7.1, 16]	12.1 [7.4, 25.7]	0.003
PEEP (cmH2O)	10 [8, 12]	10 [8, 12]	10 [8, 12]	0.004
ΔP (cmH2O)	21 [17, 25]	20 [16, 24]	21 [18, 26]	0.101
24h after ARDS onset
FIO2	0.4 [0.3, 0.5]	0.35 [0.3, 0.45]	0.45 [0.4, 0.7]	< 0.001
PaO2/ FIO2	225 [162, 284]	232 [171, 292]	182 [111, 248]	< 0.001
OI	6.9 [4.8, 11.5]	6.6 [4.7, 10.6]	9.9 [5.7, 21.6]	< 0.001
PEEP (cmH2O)	10 [8, 12]	10 [8, 10]	10 [8, 12]	< 0.001
ΔP (cmH2O)	17.5 [14, 22]	17 [14, 22]	20 [16, 23]	0.001
Ancillary Therapies in first 72h (%)
Inhaled nitric oxide	261 (36)	194 (33)	67 (51)	< 0.001
Nonconventional ventilator mode	202 (28)	156 (26)	46 (35)	0.054
ECMO	32 (4)	23 (4)	9 (7)	0.158
Corticosteroids	387 (54)	299 (51)	88 (67)	0.001
Neuromuscular blockade	346 (48)	282 (48)	64 (48)	0.923
Prone positioning	25 (3)	20 (3)	5 (4)	0.793
Cause of death in non-survivors (%)
Multisystem organ failure			54 (41)
Neurologic			52 (39)
Hypoxemia			26 (20)

Data are presented as frequencies (percentages) or medians [interquartile ranges].

ARDS: acute respiratory distress syndrome; ECMO: extracorporeal membrane oxygenation; OI: oxygenation index; PRISM III: Pediatric Risk of Mortality III; PEEP: positive end-expiratory pressure; ΔP: driving pressure (peak pressure minus PEEP)

Unadjusted Analysis

In unadjusted analysis (Supplementary Figure 1), there was no difference in daily intake between survivors and non-survivors until day 7 (median 83 ml/kg vs 101 ml/kg on day 7). Daily output was not different on any day; UOP was always higher in survivors. When assessing cumulative fluid metrics (Figure 1), there was no difference between survivors and non-survivors in cumulative intake until day 7 (median 690 mL/kg vs 774 mL/kg on day 7). Survivors had a slower rate of fluid accumulation (median 81 mL/kg/day) relative to non-survivors (92 mL/kg/day; p = 0.002). Beginning on day 3, non-survivors had a greater cumulative balance than survivors (median 79 ml/kg vs 109 ml/kg on day 3), which was maintained through day 7.

Figure 1:

Comparison between survivors (white) and non-survivors (gray) without adjustment for severity of illness, of cumulative fluid intake, cumulative fluid output, cumulative urine output, and cumulative fluid balance in ml/kg for days 1 through 7 (each day’s metrics adding upon the previous). Box plots represent median (line), lower and upper quartiles (boxes), and range (error bars). Comparisons with statistical significance p < 0.05 after Bonferroni correction (nominal p ≤ 0.007) are depicted with an asterisk (*). The number of subjects in each group available for comparison over all 7 days are provided.

Adjusted for Severity of Illness

In multivariable analysis (Figure 2), cumulative intake on day 7 was associated with increased mortality, while cumulative output and UOP were not. Higher cumulative balances on days 5 to 7 were associated with increased mortality (aOR 1.34 per 100 ml/kg on day 5, 95% CI 1.11 to 1.61). Higher cumulative intake and cumulative balance on days 4 to 7 were also associated with a lower likelihood of successful extubation (Figure 3). Findings were consistent when analyzing complete cases (Supplementary Figures 2 and 3), when analyzing combination output (Supplementary Figure 4), and when adjusting for study year (Supplementary Figure 5).

Figure 2:

Adjusted association (adjusted for OI, number of non-pulmonary organ failures, immunocompromised status, and vasopressor score) of the association between cumulative fluid intake, cumulative fluid output, cumulative urine output, and cumulative fluid balance with mortality over the first 7 days of ARDS (each day’s metrics adding upon the previous). Adjusted odds ratios are plotted per day and are scaled per increase in 100 mL/kg. Comparisons with statistical significance p < 0.05 after Bonferroni correction (nominal p ≤ 0.007) are depicted with an asterisk (*).

Figure 3:

Adjusted association (adjusted for OI, number of non-pulmonary organ failures, immunocompromised status, and vasopressor score) of cumulative fluid intake, cumulative fluid output, cumulative urine output, and cumulative fluid balance with probability of extubation, given the competing risk of death, over the first 7 days of ARDS (each day’s metrics adding upon the previous). Adjusted subdistribution hazard ratios (SHRs) are plotted per day and are scaled per increase in 100 mL/kg. Adjusted SHR < 1 implies slower extubation/fewer VFDs. Comparisons with statistical significance p < 0.05 after Bonferroni correction (nominal p ≤ 0.007) are depicted with an asterisk (*).

Analysis Stratified by ARDS Etiology

We assessed the daily association of cumulative balance with outcomes according to ARDS etiology (Supplementary Figure 6). The association between cumulative fluid balance and mortality was driven by pneumonia, with increased mortality risk starting on day 5. By contrast, the association between cumulative balance and decreased probability of extubation was driven by non-pulmonary sepsis, with lower probability of extubation starting on day 4.

Analysis Stratified by Cause of Death

We assessed whether the etiology of death affected the temporal association of cumulative fluid balance with mortality (Supplementary Figure 7). Subjects assigned MSOF as the primary cause of death demonstrated higher adjusted mortality risk for all 7 days, although statistical significance was only achieved after day 4. Subjects dying of hypoxemia or neurologic failure demonstrated a pattern of increased mortality more similar to the primary analysis, with increased cumulative fluid balance being associated with mortality after day 5; however, with the fewer number of deaths, results were not significant after Bonferroni correction.

Mediation Analysis

Given that cumulative fluid balance was associated with worse outcomes after day 4 of ARDS, we dichotomized early versus late fluid overload at day 4. We then tested whether early AKI or maximum vasopressor score by day 3 mediated the association between late (day 4 to 7) fluid overload > 10% and mortality (Supplementary Figure 8). There was minimal evidence for mediation by either AKI (7% mediated) or vasopressor score (13% mediated), with most of the association between fluid overload and mortality representing a direct effect.

Angiopoietin-2 Predicts Fluid Overload

In the subset of patients in the biomarker study, we tested the association between Ang2 with subsequent > 10% fluid overload. Ang2 on day 1 (n = 333) predicted early (first 3 days) fluid overload (Figure 4), with an area under the receiver operating characteristic curve (AUROC) of 0.61 (95% CI 0.55 to 0.67). Similarly, day 3 Ang2 levels (n = 266) predicted fluid overload on days 4 to 7, with an AUROC 0.62 (95% CI 0.56 to 0.69).

Figure 4:

Comparison of plasma angiopoietin-2 levels in subjects with and without fluid overload measured on day 1 and day 3 of illness. Early fluid overload defined as > 100 ml/kg positive cumulative fluid balance on days 1–3 of illness, late fluid overload defined as > 100 ml/kg positive cumulative fluid balance on days 4–7 of illness.

DISCUSSION

Increasing cumulative fluid balance after day 5 of ARDS was associated with mortality in pediatric ARDS, and increasing fluid intake and balance after day 4 was associated with decreased probability of extubation. There was no association between fluids and outcomes in the first 72 hours. Ang2 predicted subsequent fluid overload on both day 1 and day 3 after ARDS onset. Our data suggest that positive fluid balance later in the illness time-course, rather than earlier, had greater association with outcomes. This association was not primarily mediated by AKI or shock. This suggests that future interventions aimed at managing fluid overload may have differential efficacy depending on when in the time-course of ARDS they are initiated.

It is neither surprising nor novel that fluid balance is associated with worse outcomes. In the Fluid and Catheter Treatment Trial (FACTT) of adult ARDS, patients given less fluid had more VFDs (7). Our results were consistent with this; however, as FACTT was a trial, it was better able to assign causality than our study. In prior pediatric cohorts, day 3 fluid balance was associated with fewer VFDs (9) and increased mortality (11). A re-analysis of 109 subjects from a calfactant trial confirmed this, with non-survivors having a higher fluid balance than survivors over the first 7 days (12). A more recent study found that an association between day 3 fluid balance and mortality in children with AKI (13). Lima et al also observed increased mortality in critically ill children in whom fluid overload peaked between days 3 and 7 of PICU admission (10), confirming greater prognostic importance to later fluid overload.

The timing of this association between fluid overload and outcomes has not previously been as clearly delineated as in our study. Using the ROSE framework of fluid management (resuscitation, optimization, stabilization, and evacuation)(23), our study suggests ongoing fluid overload later in the illness (stabilization and evacuation phases) as having an important role in the association of fluid balance with outcomes. Ang2, which increases endothelial permeability and fluid extravasation, suggests a mechanism for fluid accumulation (24). Thus, Ang2 is a useful marker to identify patients at risk for fluid overload, and is potentially a therapeutic target.

Our study had limitations. It was a retrospective single center cohort, which limits generalizability, albeit similar to other single (25) and multicenter (2, 3) cohorts. Fluid data was retrospectively collected; however, all other data were collected prospectively. Fluid management was not protocolized, and data on individual providers making fluid management decisions was not available, and management subject to practice variability. We did not collect longitudinal ventilator settings or metrics of oxygenation which could inform the mechanism between fluid overload and probability of extubation. Confounders were chosen a priori based on plausibility and the analysis performed in a causal framework. Our primary exposures were continuous, precluding the use of other causal inference methods like propensity matching. However, logistic regression outperforms propensity scores when there are sufficient events and few confounders (26), as in this study. However, we acknowledge that alternative analyses of causal inference, including propensity scores, may give different answers than in our study. We are reassured that sensitivity analyses did not change our conclusions.

Finally, we only incompletely assessed the role of AKI in our study with mediation analysis; however, AKI remains a potential source of residual confounding. Both UOP and fluid balance were likely impacted by AKI, reflecting both illness severity and a mechanism contributing to fluid overload. Analysis of renal function as it relates to fluid overload and outcomes in pediatric ARDS is an important topic for future study, requiring dissection of the complex interplay between AKI, UOP, and fluid overload. We did not collect dosing of diuretics or CRRT, or disentangle the relationship between diuretic use, CRRT, UOP, fluid balance, and outcomes. Our focus was on the timing of the relationship between fluid metrics and outcome, rather than how these metrics were achieved, or the role of AKI in how this occurred. The contribution of mode of fluid removal on outcome are deserving of further study, since how fluid is removed may be as important as when fluid is removed.

CONCLUSIONS

We found no association between early fluid balance and outcomes in pediatric ARDS until after day 4 of illness, providing a nuanced perspective on how timing impacts the relationship between fluid overload and outcomes. Future studies should explicitly disentangle the differential roles of initial fluid resuscitation and subsequent fluid removal on outcome in pediatric ARDS, as well as clarify the optimal methods of de-resuscitation.

REPORT IN CONTEXT.

• While fluid overload is known to be associated with poor outcomes in pediatric ARDS, it is unknown whether this association varies with the timing after ARDS onset.

• This is relevant, as the design of future trials of fluid management may have differential impact depending on when the intervention is initiated.

• Our results suggest that later (after day 4 of ARDS onset) fluid overload, rather than earlier (first 3 days), has a greater association with poor outcomes.

AT THE BEDSIDE.

• Later (after day 4 of ARDS onset) fluid overload, rather than earlier (first 3 days), has a greater association with increased PICU mortality and fewer VFDs.

• Our longitudinal analysis provides a more complete picture of fluid overload in pediatric ARDS by providing granular daily analysis over the first 7 days in a large cohort, adjusting for confounders.

• Our results provide guidance for the design of future trials of fluid management in pediatric ARDS, and suggest that de-resuscitation strategies should be tested starting after day 4 of ARDS.