Deresuscitation of Patients With Iatrogenic Fluid

Overload Is Associated With Reduced Mortality in

Critical Illness*
Jonathan A. Silversides, MBBCh1,2; Emma Fitzgerald, MBBCh3; Uma S. Manickavasagam, MBBS4;
Stephen E. Lapinsky, MBBCh, MSc5,6; Rosane Nisenbaum, PhD7,8; Noel Hemmings, MBBCh9;
Downloaded from http://journals.lww.com/ccmjournal by BhDMf5ePHKbH4TTImqenVGPFTMFeRXdQU83XL2NAOmcObj7K1RQ0UrNbiWtMGbQzWXPdScxIVSQ= on 09/21/2018

Christopher Nutt, MBBCh10; T. John Trinder, MD11; David G. Pogson, MBBCh3;

Eddy Fan, MD, PhD6,12; Andrew J. Ferguson, MBBCh13; Daniel F. McAuley, MD1,2;
John C. Marshall, MD4,6; for the Role of Active Deresuscitation After Resuscitation (RADAR) Investigators

*See also p. 1692. Grant WKR0-2017-0019, National Institute of Health Research Health
1
Centre for Experimental Medicine, Queen’s University of Belfast, Belfast, Technology Assessment grant application pending, and GlaxoSmithKline,
United Kingdom. and he received other funding from Bayer, Boehringer Ingelheim, Peptin-
novate. Dr. Marshall received funding from the Data and Safety Monitoring
2
Department of Critical Care, Belfast Health and Social Care Trust, Bel- Board of Asahi-Kasei Pharma, Bristol-Myers Squibb (Advisory Board), and
fast, United Kingdom. Regeneron. The remaining authors have disclosed that they do not have
3
Academic Department of Critical Care, Queen Alexandra Hospital, any potential conflicts of interest.
Portsmouth Hospitals NHS Trust, Portsmouth, United Kingdom. For information regarding this article, E-mail: jon.silversides@belfasttrust.
4
Department of Critical Care, Saint Michael’s Hospital, Toronto, ON, Canada. hscni.net
5
Intensive Care Unit, Mount Sinai Hospital, Toronto, ON, Canada.
6
Interdepartmental Division of Critical Care, University of Toronto, Toronto,
ON, Canada. Objectives: To characterize current practice in fluid administra-
7
Centre for Urban Health Solutions, Li Ka Shing Knowledge Institute, tion and deresuscitation (removal of fluid using diuretics or renal
Saint Michael’s Hospital, Toronto, ON, Canada replacement therapy), the relationship between fluid balance,
8
Dalla Lana School of Public Health, University of Toronto, Toronto, ON, deresuscitative measures, and outcomes and to identify risk fac-
Canada.
tors for positive fluid balance in critical illness.
9
Intensive Care Unit, Altnagelvin Area Hospital, Western Health and
Social Care Trust, Londonderry, United Kingdom. Design: Retrospective cohort study.
Intensive Care Unit, Antrim Area Hospital, Northern Health and Social
10 Setting: Ten ICUs in the United Kingdom and Canada.
Care Trust, Antrim, United Kingdom. Patients: Adults receiving invasive mechanical ventilation for a
Intensive Care Unit, Ulster Hospital, South-Eastern Health and Social
11 minimum of 24 hours.
Care Trust, Dundonald, United Kingdom. Interventions: None.
Intensive Care Unit, Mount Sinai Hospital, Toronto, ON, Canada.
12
Measurements and Main Results: Four-hundred patients were
Intensive Care Unit, Craigavon Hospital, Southern Health and Social
13
included. Positive cumulative fluid balance (fluid input greater than
Care Trust, Portadown, United Kingdom.
output) occurred in 87.3%: the largest contributions to fluid input
A full list of Role of Active Deresuscitation After Resuscitation (RADAR)
Investigators is listed in Appendix 1. were from medications and maintenance fluids rather than resus-
Supplemental digital content is available for this article. Direct URL citations citative IV fluids. In a multivariate logistic regression model, fluid
appear in the printed text and are provided in the HTML and PDF versions balance on day 3 was an independent risk factor for 30-day mor-
of this article on the journal’s website (http://journals.lww.com/ccmjournal). tality (odds ratio 1.26/L [95% CI, 1.07–1.46]), whereas negative
Supported, in part, by a doctoral fellowship award to Dr. Silversides from the fluid balance achieved in the context of deresuscitative measures
Northern Ireland Health and Social Care Research and Development Division.
was associated with lower mortality. Independent predictors of
Dr. Silversides’ institution received funding from British Journal of Anaes-
thesia (BJA)/Royal College of Anaesthetists (RCoA) project grant WKR0- greater fluid balance included treatment in a Canadian site.
2017-0019. Dr. Lapinsky received other support from research grants Conclusions: Fluid balance is a practice-dependent and poten-
from the Chest Foundation and Ontario Lung Association for an unrelated tially modifiable risk factor for adverse outcomes in critical illness.
research project. Dr. McAuley’s institution received funding from v Divi-
sion (Doctoral Fellowship grant for Dr. Silversides), BJA/RCoA Project Negative fluid balance achieved with deresuscitation on day 3 of
Copyright © 2018 by the Society of Critical Care Medicine and Wolters ICU stay is associated with improved patient outcomes. Minimiza-
Kluwer Health, Inc. All Rights Reserved. tion of day 3 fluid balance by limiting maintenance fluid intake and
DOI: 10.1097/CCM.0000000000003276 drug diluents, and using deresuscitative measures, represents a

1600 www.ccmjournal.org October 2018 • Volume 46 • Number 10

Copyright © 2018 by the Society of Critical Care Medicine and Wolters Kluwer Health, Inc. All Rights Reserved.
 Clinical Investigations

potentially beneficial therapeutic strategy which merits investiga- Participants

tion in randomized trials. (Crit Care Med 2018; 46:1600–1607) We included patients over 16 years old who received invasive
Key Words: critical illness; diuretics; fluid therapy; furosemide; mechanical ventilation in a participating ICU for a minimum
water-electrolyte balance of 24 hours. Exclusion criteria were failure to survive for more
than 24 hours; lack of a commitment to full active care at ICU
admission; drug overdose (fluid management unlikely to be
relevant to outcome); transfer from another ICU or ICU case

F
luid therapy is widely used to optimize tissue perfusion records unavailable; and those with a diagnosis of subarach-
in sepsis and other critical illness states, supported by noid hemorrhage or diabetic ketoacidosis, conditions where
international guidelines (1, 2). In addition to resuscita- specific considerations are relevant to fluid management.
tion fluid boluses, significant volumes of fluid are administered
as nutrition, maintenance fluid, and a diluent for medications. Data Collection and Definitions
In the context of endocrine influences and acute kidney injury Data collected included demographic factors, diagnosis, sever-
(AKI) predisposing to fluid retention, critically ill patients ity of illness, furosemide and RRT use, and fluid input and
commonly accumulate a positive fluid balance (2–5). output. Data were collected by trained study personnel using
The association between positive fluid balance and mor- a standard data dictionary. The primary outcome was 30-day
tality in a range of critical illness states is now well established mortality. Secondary outcomes included duration of ICU stay
(2–8). Although fluid accumulation is most evident clinically as in survivors, censored at day 30, and duration of mechanical
peripheral and pulmonary edema, other negative physiologic ventilation in survivors, defined as the number of ICU days
consequences such as renal congestion have been ­postulated (9). from intubation until successful extubation (> 24 hr without
Two complementary strategies have evolved to prevent or invasive ventilatory support).
mitigate fluid accumulation: restrictive fluid administration Study day 1 was from ICU admission until 07:00 or 8:00 am
(10, 11) or active removal of accumulated fluid using diuret- depending on unit practice, whereas subsequent study days
ics or renal replacement therapy (RRT) when clinically stable were consecutive 24-hour periods, or part thereof on the day
(deresuscitation) (12). Our recent systematic review and meta- of ICU discharge. Daily fluid balance was the total delivered
analysis found low-quality evidence in favor of a conservative input (IV fluids, nutrition, medications) minus total measured
or deresuscitative fluid strategy compared with liberal fluid output (e.g., urine, ultrafiltrate, output from surgical drains,
administration or usual care (13), but considerable heteroge- etc) as recorded in case records. Cumulative fluid balance for a
neity was evident with regard to interventions: these ranged given day was the total fluid input minus the total output from
from restriction of resuscitation fluid alone (10) to deresus- ICU admission at the end of that day. Deresuscitative mea-
citation using diuretics and “hyperoncotic” albumin (14). sures were defined as any use of furosemide or fluid removal
Further work is needed to define the optimum fluid strategy to using RRT.
be tested in future large-scale randomized trials. We calculated Multiple Organ Dysfunction Scores (MODS)
We undertook a multicenter cohort study of critically (15) daily. We estimated missing central venous pressure values
ill patients in Canada and the United Kingdom, aiming to as 8 mm Hg, as previously described (16). Daily delta-MODS
increase our understanding of variability in current practice was defined as the change in MODS between that day and the
in fluid administration and deresuscitation, to characterize subsequent day. We estimated AKI severity using the Kidney
the relationship between fluid balance, deresuscitation, and Disease: Improving Global Outcomes (KDIGO) consensus cri-
patient outcomes and to identify risk factors for positive fluid teria for the definition and classification of AKI (17) daily.
balance with a view to informing the design of future trials.
We hypothesized that a more positive fluid balance would be Statistical Analysis
associated with mortality, that drivers of positive fluid balance Between-group comparisons were made using univariate
would include practice-dependent variables, and that the use logistic regression, Kruskal-Wallis test, or chi-square as appro-
of deresuscitative measures to achieve a negative fluid balance priate. We constructed multivariate logistic regression mod-
would be associated with better outcomes. els for 30-day mortality. Negative binomial regression was
used for the outcomes of duration of ICU stay and duration
of mechanical ventilation. A generalized estimating equation
METHODS
model was used to account for time and clustering within
Design and Setting patients when investigating a possible association between
Participating centers were selected on the basis of expressed daily fluid balance and the outcome of daily delta-MODS,
interest in the study and were asked to provide data retro- using an identity link and exchangeable correlation structure.
spectively for up to 50 consecutive patients admitted between Multivariate linear regression modeling was used for explora-
June 2012 and June 2014 who met prespecified inclusion and tion of a possible association between day 3 fluid balance and
exclusion criteria. A formal sample size calculation was not potential predictor variables.
performed. Research ethics approval was obtained, but the Four criteria were used to select variables for inclusion
requirement for individual patient consent was waived. in multivariate models: relevance to underlying hypothesis,

Critical Care Medicine www.ccmjournal.org 1601

Copyright © 2018 by the Society of Critical Care Medicine and Wolters Kluwer Health, Inc. All Rights Reserved.
Silversides et al

clinical plausibility, previously reported associations in the lit- (IQR, 0.25–1.75 L) to 3.50 L (IQR, 1.0–9.0 L) and in mainte-
erature, and univariate associations with a p value of less than nance fluids administered (ranging from one site where no
0.2. Where candidate variables were correlated at a level greater maintenance fluid was given to a median 5.77 L [IQR, 4.66–
than 0.5, the most clinically plausible was selected. 6.37 L] in another) (Fig. 2 and Table E1, Supplement Digital
Sensitivity analyses were conducted using several fluid bal- Content 1, http://links.lww.com/CCM/D719).
ance time points to identify that most strongly associated with
the primary outcome. We also separated cumulative fluid bal- Use of Deresuscitative Measures
ance into “early” (days 1–2) and “later” time points, on the basis Deresuscitative measures were used on at least 1 day while in
that early fluid balance may be more reflective of resuscitation ICU in 209 patients (52.3%), most commonly for the first time
fluid administration and thus potentially more confounded on days 2 or 3. Deresuscitative measures were used in 125 of
by severity of illness. Other sensitivity analyses were used to 400 patients (31.3%) on days 1–3 and in 140 of 320 patients
explore the relationship between deresuscitative measures and (43.8%) on days 4–5. This varied considerably between sites
mortality. from 16.7% to 50.0% of patients on days 1–3 (Fig. 2) and from
Statistical analyses were performed using Stata Version 18.2% to 60.6% on days 4–5.
14 (StataCorp LLC, College Station, TX). Two-sided p val-
ues of less than 0.05 were considered to represent statistical Fluid Balance and Mortality
significance. In univariate analysis, 30-day mortality was associated with
greater fluid balance on days 1–7. The strength of the associa-
tion decreased over days 1–7 and increased when “early” fluid
RESULTS
balance (days 1 and 2) was excluded, so that the association
Patient Characteristics between fluid balance and 30-day mortality was strongest for
Data were collected on 400 patients from 10 ICUs (three in fluid balance on day 3 alone (odds ratio [OR], 1.32; 95% CI,
Canada and the remaining seven in the United Kingdom); the 1.17–1.50 per liter) (Table E2, Supplement Digital Content 1,
minimum number of patients per site was 21. During the study http://links.lww.com/CCM/D719). The dose-response rela-
period, a total of 2,473 patients were admitted to participating tionship between day 3 fluid balance and unadjusted 30-day
ICUs. Baseline characteristics are shown in Table 1. mortality is shown in Figure E1 (Supplement Digital Content
Mortality at day 30 following ICU admission was 31.0% 1, http://links.lww.com/CCM/D719).
(124/400), with a median length of ICU stay of 7 days (inter- In a multivariate logistic regression model for 30-day mor-
quartile range [IQR], 4–13 d) for survivors. Nonsurvivors dif- tality, we found daily fluid balance on day 3 to be independently
fered from survivors with respect to age, source and type of associated with 30-day mortality (Table E3, Supplement Digital
admission, baseline creatinine, Acute Physiology and Chronic Content 1, http://links.lww.com/CCM/D719). Sensitivity anal-
Health Evaluation II scores, MODS, and KDIGO scores and yses confirmed that the association was strongest for daily fluid
in vasopressor use, blood pressure, and oxygenation (Table 1). balance on day 3 alone rather than other fluid balance time
points (e.g., days 1–3, days 1–5).
Fluid Balance Over Time In a sensitivity analysis, we treated day 3 fluid balance as a
Mean daily fluid balances and mean cumulative fluid balances ternary variable, defined as positive balance (> 500 mL), even
from day 1 to day 7 are shown in Figure 1. Mean daily fluid bal- balance (between 500 mL negative and 500 mL positive), or
ance was significantly greater for nonsurvivors than survivors negative balance (> 500 mL). Both an “even balance” (adjusted
on days 1–5 and 7 (maximum difference in means, 0.98 L [95% OR, 0.40; 95% CI, 0.21–0.80) and “negative balance” (adjusted
CI, 0.57–1.37 L] on day 3). Mean cumulative fluid balance was OR, 0.17; 95% CI, 0.07–0.40) were strongly protective com-
significantly higher in nonsurvivors for days 1–7 (maximum pared with positive balance (Table E4, Supplement Digital
difference in means, 2.38 L [95% CI, 0.98–3.78 L] on day 5). Content 1, http://links.lww.com/CCM/D719).
Most patients still in ICU at day 7 were in a positive fluid bal-
ance (148/237; 62.45%). Deresuscitative Measures and Mortality
Of the 123 patients in whom fluid balance on day 3 was nega-
Sources of Fluid Input tive, a negative balance developed spontaneously (without
The largest component of fluid input over days 1–3 was from diuretics or RRT) in 70 (56.9%). Patients who achieved a spon-
medications (1,148.0 of a total 3,325.5 L; 34.5%), whereas taneous negative fluid balance on day 3 of ICU stay were less
maintenance and bolus IV fluids comprised 26.5% and 24.4%, severely ill than either those with a positive fluid balance or
respectively. There was considerable variability between sites those who achieved negative fluid balance using deresuscita-
in both total volume of fluid input and in sources of that input tive measures (Table E5, Supplement Digital Content 1, http://
(Fig. 2 and Table E1, Supplement Digital Content 1, http:// links.lww.com/CCM/D719); however, crude mortality was
links.lww.com/CCM/D719). Over days 1–7, proportionately lower whether a negative fluid balance occurred in the context
lower contributions were from bolus and maintenance IV of deresuscitative measures (p = 0.02) or spontaneously (p <
fluid. Striking numerical differences were present between 0.01) compared with patients who had a positive fluid balance
study sites in fluid bolus volumes (ranging from median 0.75 L on day 3. However, patients who had a positive fluid balance in

1602 www.ccmjournal.org October 2018 • Volume 46 • Number 10

Copyright © 2018 by the Society of Critical Care Medicine and Wolters Kluwer Health, Inc. All Rights Reserved.
 Clinical Investigations

TABLE 1. Comparison of Patient Characteristics Between Survivors and Nonsurvivors,

and Candidate Predictor Variables for 30-Day Mortality
All Survivors Nonsurvivors
Demographics (n = 400) (n = 276) (n = 124) p (Univariate)

Age, yr, median (IQR) 63 (51–74) 60 (48–71) 69 (57–79) < 0.01

Male sex, n (%) 243 (60.8) 73 (58.9) 170 (61.6) 0.61
Admission source, n (%) 0.01
  Emergency department 117 (29.3) 89 (32.3) 28 (22.6)
  Operating room/recovery 75 (18.8) 64 (23.2) 11 (8.9)
  Ward (same hospital) 139 (34.8) 79 (28.6) 60 (48.4)
  Other hospital (not ICU) 66 (16.5) 41 (14.9) 25 (20.2)
 Other 3 (0.8) 3 (1.1) 0 (0.0)
Admission type, n (%) 0.01
  Elective surgical 31 (7.8) 28 (10.1) 3 (2.4)
  Emergency surgical 81 (20.3) 65 (23.6) 16 (12.9)
 Medical 267 (66.8) 162 (58.7) 105 (84.7)
 Trauma 21 (5.3) 21 (7.6) 0 (0.0)
Baseline creatinine, μmol/L, median (IQR), (n = 230) 77 (60–100) 75 (58–96) 77 (67–113) 0.05
Admission diagnosis, n (%) 0.22
 Cardiovascular 64 (16.0) 37 (13.4) 27 (21.8)
 Respiratory 132 (33.0) 91 (33.0) 41 (33.1)
 Gastrointestinal 85 (21.3) 62 (22.5) 23 (18.6)
 Neurologic 36 (9.0) 28 (10.1) 8 (6.5)
 Other 83 (20.8) 58 (21.0) 25 (20.2)
Acute Physiology And Chronic Health Evaluation II score, 18.5 (7.2) 16.8 (6.6) 22.4 (6.9) 0.01
mean (sd)
Congestive cardiac failure, n (%) 37 (9.3) 21 (7.6) 16 (12.9) 0.09
End-stage renal disease, n (%) 12 (3.0) 8 (2.9) 4 (3.2) 0.86
Highest MOD score (days 1–2), mean (sd) 4.6 (2.5) 4.1 (2.3) 5.8 (2.6) < 0.01
Highest Kidney Disease: Improving Global Outcomes stage < 0.01
(days 1–2), n (%)
 0 182 (45.5) 148 (53.6) 34 (27.4)
 1 99 (24.8) 67 (24.3) 32 (25.8)
 2 11 (2.8) 7 (2.5) 4 (3.2)
 3 108 (27) 54 (19.6) 54 (43.6)
Lowest Pao2/Fio2 ratio (days 1–2) (n = 397), median (IQR) 174.5 (118–258) 193 (128–271) 137 (91–227) < 0.01
Lowest mean arterial pressure, days 1–2 (mm Hg), mean (sd) 73.7 (13.3) 75.0 (12.8) 70.9 (13.9) < 0.01
Vasopressors, days 1–2, n (%) 250 (62.5) 158 (57.2) 92 (74.2) < 0.01
Highest serum lactate, days 1–2 (mmol/L), mean (sd) 3.8 (3.2) 3.3 (2.6) 4.9 (4.0) < 0.01
RRT, days 1–2 (n = 399), n (%) 40 (10) 24 (8.7) 16 (12.9) 0.20
RRT with fluid removal, days 1–3, n (%) 40 (10.0) 24 (8.7) 16 (12.9) 0.20
Use of furosemide, days 1–3, n (%) 111 (27.8) 76 (27.5) 35 (28.2) 0.89
Total furosemide dose, days 1–3 (mg), median (IQR) 0 (0–20) 0 (0–20) 0 (0–20) 0.81
IQR = interquartile range, RRT = renal replacement therapy.

Critical Care Medicine www.ccmjournal.org 1603

Copyright © 2018 by the Society of Critical Care Medicine and Wolters Kluwer Health, Inc. All Rights Reserved.
Silversides et al

In a multivariate model, the association between day 3 fluid

balance and mortality was robust after dichotomization of
patients with a negative fluid balance, in that both spontane-
ous negative fluid balance (adjusted OR, 0.21; 95% CI, 0.08–
0.56) and negative fluid balance with deresuscitative measures
(adjusted OR, 0.29; 95% CI, 0.12–0.69) were independently
associated with lower mortality relative to a positive fluid bal-
ance (Table 2).
In a further sensitivity analysis limited to patients who were
in a positive fluid balance on day 3, neither furosemide use nor
RRT with fluid removal was associated with increased 30-day
mortality (Table E6, Supplement Digital Content 1, http://
links.lww.com/CCM/D719).

Fluid Balance and Organ Function Scores

Missing central venous pressure values were imputed as 8 mm
Hg for 971 of 1,869 measurements (52.0%) for the purpose of
calculating MODS. A greater fluid balance on day 3 was associ-
ated with worsening organ dysfunction reflected in an increas-
ing MOD score over the subsequent 2 days, after adjustment
Figure 1. Cumulative and daily fluid balances over time for survivors and
nonsurvivors. *p < 0.05. for change in MOD score over days 1–3 (coefficient 0.2/L; 95%
CI, 0.1–0.3). This association was not identified in any specific
organ system subscore.
We found a weakly positive association between delta MOD
score over any 1-day period and daily fluid balance over the
same period (number of days = 2,115; coefficient = 0.06/L
[95% CI, 0.03–0.08]). This association was similar whether the
fluid balance occurred in the context of deresuscitative mea-
sures or not.

Fluid Balance and Other Outcomes

In multivariate models, greater fluid balance on day 3 was asso-
ciated with longer duration of ICU stay (Table E7, Supplement
Digital Content 1, http://links.lww.com/CCM/D719) and with
increased duration of invasive mechanical ventilation in sur-
vivors (Table E8, Supplement Digital Content 1, http://links.
lww.com/CCM/D719). Negative fluid balance in the context
of deresuscitation was associated with shorter duration of
mechanical ventilation and a shorter length of ICU stay in sur-
vivors. Fluid balance on day 3 was not associated with change
in KDIGO AKI stage over the following 2 days (days 3–5) of
ICU admission.

Predictors of Positive Fluid Balance

Independent predictors of a positive fluid balance on day 3 of
ICU admission were vasopressor use, serum creatinine, blood
product transfusion, and treatment in a Canadian site, while
furosemide use predicted a less positive fluid balance (Table 3).

DISCUSSION
Figure 2. Median fluid input and output and use of deresuscitative In this binational study of a broad cohort of critically ill
measures for days 1–3 by study site. RRT = renal replacement therapy.
patients, we found that a negative fluid balance on day 3 of
ICU stay is associated with lower 30-day mortality whether
the context of deresuscitative measures had the highest crude occurring spontaneously or achieved with deresuscitative
mortality (Fig. E2, Supplement Digital Content 1, http://links. measures; that medication diluents, rather than IV fluids, rep-
lww.com/CCM/D719). resent the single largest component of fluid input; and that

1604 www.ccmjournal.org October 2018 • Volume 46 • Number 10

Copyright © 2018 by the Society of Critical Care Medicine and Wolters Kluwer Health, Inc. All Rights Reserved.
 Clinical Investigations

TABLE 2. Multivariable Logistic Regression Model for Outcome of 30-Day Mortality, Fluid
Balance on Day 3 Trichotomized as Positive (Reference Group), Spontaneously Negative,
or Negative With Deresuscitative Measures
Variables OR (95% CI) p

Age (/yr) 1.04 (1.02–1.06) < 0.01

Acute Physiology and Chronic Health Evaluation II score (per point) 1.04 (0.99–1.09) 0.09
Congestive cardiac failure 0.86 (0.34–2.21) 0.76
Multiple Organ Dysfunction Scores (highest on days 1 and 2, per point) 1.16 (1.02–1.31) 0.02
Kidney Disease: Improving Global Outcomes stage (highest on days 1 and 2, per stage) 1.10 (0.88–1.39) 0.40
Maximum serum lactate (days 1 and 2, per mmol/L) 1.08 (0.99–1.18) 0.08
Days of vasopressors (days 1–3) 1.14 (0.89–1.44) 0.30
Admission source
 Ward Reference
  Emergency department 0.32 (0.16–0.65) < 0.01
  Operating room/recovery room 0.69 (0.23–2.02) 0.50
  Other hospital 0.72 (0.34–1.55) 0.41
Surgical admission type 0.22 (0.09–0.53) < 0.01
Early fluid balance (up to 2 d, per liter) 1.08 (0.99–1.17) 0.07
Positive fluid balance on day 3 Reference
Spontaneous negative fluid balance on day 3 0.21 (0.08–0.56) < 0.01
Negative fluid balance with deresuscitative measures on day 3 0.29 (0.12–0.69) < 0.01
OR = odds ratio.
n = 377, Pseudo r2 = 0.28. Pearson goodness of fit, p = 0.88.

TABLE 3. Multivariate Linear Regression Model for the Outcome of Day 3 Fluid Balance
Variables Coefficient (95% CI) p

Acute Physiology And Chronic Health Evaluation II score (per point) 0.01 (–0.02 to 0.04) 0.37
Canadian site 0.91 (0.39–1.42) < 0.05
Change in Multiple Organ Dysfunction Scores, days 1– 3 (per point) 0.06 (–0.03 to 0.14) 0.30
Transfusion of blood products, day 3 0.65 (0.06–1.24) 0.13
Maximum lactate, day 3 (per mmol/L) 0.13 (–0.00 to 0.27) 0.35
Vasopressors, day 3 0.53 (0.14–0.92) 0.03
Pao2/Fio2 ratio, day 3 (per 10 mm Hg) –0.01 (–0.03 to 0.01) 0.27
Creatinine, day 3 (per μmol/L) 0.02 (0.00–0.04) 0.03
Cumulative fluid balance, days 1–2 0.14 (0.07–0.20) < 0.01
Total furosemide dose, day 3 (per 10 mg) –0.13 (–0.18 to –0.08) < 0.01
Renal replacement therapy, day 3 –0.08 (–0.72 to 0.56) 0.70

marked practice variability exists in the use of fluid therapy be unhelpful or even harmful. Alternatively, several consider-
and ­deresuscitative measures. ations support our hypothesis that fluid balance represents an
There are two possible interpretations of our findings. First, iatrogenic factor in critical illness which may be modified by
positive fluid balance may be a biomarker of greater severity, deresuscitative measures. First, rigorous attempts were made to
complexity, or duration of critical illness, in which case attempts minimize confounding by illness severity, comorbidities, and
to modify fluid balance using diuretics or ultrafiltration may early illness course and to adjust for early fluid balance which

Critical Care Medicine www.ccmjournal.org 1605

Copyright © 2018 by the Society of Critical Care Medicine and Wolters Kluwer Health, Inc. All Rights Reserved.
Silversides et al

is most likely to be impacted by resuscitation fluid volumes. deresuscitative measures, is associated with lower mortality.
Second, negative fluid balance was independently associated We found considerable practice variation in the use of IV fluids
with lower mortality even when occurring in the context of and in the use of deresuscitative measures and demonstrated
deresuscitative measures and not spontaneously. Third, we that the largest source of fluid intake is from maintenance flu-
found a positive association between daily fluid balance and ids, leading us to infer that fluid balance is a practice driven and
other outcomes such as delta MOD score, and this associa- therefore modifiable variable. We identified day 3 as potentially
tion was present whether deresuscitative measures were used a pivotal time point for intervention.
or not. Finally, much of the variation in fluid input derives An approach involving limited maintenance fluid intake,
from potentially modifiable practice factors, particularly the and deresuscitative measures if required to minimize fluid
use of maintenance fluids, rather than from bolus fluids which balance on day 3 of ICU stay onwards, therefore, represents a
are likely to reflect perceived clinical need. It is reassuring promising therapeutic strategy which merits investigation in
that even in those patients in whom deresuscitative measures randomized controlled trials.
were “unsuccessful,” in that fluid balance remained positive,
there was no evidence of worse outcomes after adjustment for
potential confounders.
ACKNOWLEDGMENTS
We acknowledge the invaluable contributions of Orla Smith,
We attempted to identify the time point at which fluid bal-
Elaine Caon, Tanya Longmuir, Paul Caddell, and Griania White
ance was most closely associated with outcome. Reasoning
to the administration of the study.
that much of the fluid administered early in the ICU stay may
constitute fluid boluses for resuscitation, we separated these
into “early” and “late” fluid balance variables. We found the REFERENCES
strongest associations with outcome to be with fluid balance 1. Rhodes A, Evans LE, Alhazzani W, et al: Surviving Sepsis Campaign:
International guidelines for management of sepsis and septic shock:
on day 3 alone, after adjustment for early (days 1–2) fluid bal- 2016. Intensive Care Med 2017; 43:304–377
ance. This was also the most common point at which spon- 2. Marik PE, Linde-Zwirble WT, Bittner EA, et al: Fluid administration
taneous negative fluid balance occurred, and the first use of in severe sepsis and septic shock, patterns and outcomes: An
deresuscitative measures was most commonly on day 2 or analysis of a large national database. Intensive Care Med 2017;
43:625–632
3. Although patients are unlikely to all follow the same time
3. Payen D, de Pont AC, Sakr Y, et al; Sepsis Occurrence in Acutely Ill
trajectory, it is possible that day 3 of ICU stay represents a Patients (SOAP) Investigators: A positive fluid balance is associated
pivotal point at which many patients begin to demonstrate with a worse outcome in patients with acute renal failure. Crit Care
spontaneous negative fluid balance or are perceived to be suit- 2008; 12:R74
4. Rosenberg AL, Dechert RE, Park PK, et al; NIH NHLBI ARDS Net-
able for deresuscitative measures. Independent predictors of work: Review of a large clinical series: Association of cumulative fluid
a positive fluid balance on day 3 included markers of hemo- balance on outcome in acute lung injury: A retrospective review of
dynamic instability and renal dysfunction. The strongest risk the ARDSnet tidal volume study cohort. J Intensive Care Med 2009;
24:35–46
factor, however, after adjustment for patient and disease char-
5. Acheampong A, Vincent JL: A positive fluid balance is an independent
acteristics was treatment in one of the three Canadian ICUs prognostic factor in patients with sepsis. Crit Care 2015; 19:251
in the cohort. Marked variability in practice patterns between 6. Boyd JH, Forbes J, Nakada TA, et al: Fluid resuscitation in septic
sites highlights the potentially modifiable nature of fluid shock: A positive fluid balance and elevated central venous pres-
accumulation. sure are associated with increased mortality. Crit Care Med 2011;
39:259–265
Our study has limitations. The sample size is modest, and 7. Vaara ST, Korhonen AM, Kaukonen KM, et al; FINNAKI Study Group:
practice variability is evaluated in only 10 separate sites: the Fluid overload is associated with an increased risk for 90-day mortal-
results may not therefore be representative, and practice vari- ity in critically ill patients with renal replacement therapy: Data from
the prospective FINNAKI study. Crit Care 2012; 16:R197
ability may be underrecognized. The retrospective nature
8. Silversides JA, Pinto R, Kuint R, et al: Fluid balance, intradialytic
of our study made it impossible to determine indications or hypotension, and outcomes in critically ill patients undergoing renal
clinician intent for practice variables. Our definition of dere- replacement therapy: A cohort study. Crit Care 2014; 18:624
suscitative measures, therefore, was pragmatic, representing 9. Prowle JR, Echeverri JE, Ligabo EV, et al: Fluid balance and acute
“presumed attempted deresuscitation” as we captured only kidney injury. Nat Rev Nephrol 2010; 6:107–115
the use of furosemide and no other diuretics or RRT with any 10. Hjortrup PB, Haase N, Bundgaard H, et al; CLASSIC Trial Group;
Scandinavian Critical Care Trials Group: Restricting volumes of
removal of fluid. We lacked hourly urine output data, limiting resuscitation fluid in adults with septic shock after initial management:
our ability to distinguish KDIGO stages 2 and 3 using urine The CLASSIC randomised, parallel-group, multicentre feasibility trial.
output criteria. In the absence of prospective screening, we Intensive Care Med 2016; 42:1695–1705
11. Chen C, Kollef MH: Targeted fluid minimization following initial
lacked data on excluded patients, and in addition, we lacked resuscitation in septic shock: A pilot study. Chest 2015; 148:1462–
data on pre-ICU fluid balance. Finally, despite adjustment for 1469
illness severity, we cannot exclude the possibility that the asso- 12. Wiedemann HP, Wheeler AP, Bernard GR, et al; National Heart, Lung,
ciation between positive fluid balance and poor outcomes rep- and Blood Institute Acute Respiratory Distress Syndrome (ARDS)
Clinical Trials Network: Comparison of two fluid-management strate-
resents residual confounding by indication. gies in acute lung injury. N Engl J Med 2006; 354:2564–2575
In conclusion, we demonstrate that negative fluid bal- 13. Silversides JA, Major E, Ferguson AJ, et al: Conservative fluid man-
ance, whether occurring spontaneously or achieved through agement or deresuscitation for patients with sepsis or acute respira-

1606 www.ccmjournal.org October 2018 • Volume 46 • Number 10

Copyright © 2018 by the Society of Critical Care Medicine and Wolters Kluwer Health, Inc. All Rights Reserved.
 Clinical Investigations

tory distress syndrome following the resuscitation phase of critical 15. Marshall JC, Cook DJ, Christou NV, et al: Multiple organ dysfunction
illness: A systematic review and meta-analysis. Intensive Care Med score: A reliable descriptor of a complex clinical outcome. Crit Care
2017; 43:155–170 Med 1995; 23:1638–1652
14. Martin GS, Moss M, Wheeler AP, et al: A randomized, controlled 16. Marshall JC: Measuring organ dysfunction in the intensive care unit:
trial of furosemide with or without albumin in hypoprotein- Why and how? Can J Anaesth 2005; 52:224–230
emic patients with acute lung injury. Crit Care Med 2005; 33: 17. Kwaja A: KDIGO clinical practice guideline for acute kidney injury.
1681–1687 Nephron Clin Pract 2012; 120:179–184

APPENDIX 1. ROLE OF ACTIVE Espie; Mount Sinai Hospital, Toronto, ON, Canada: Stephen
DERESUSCITATION AFTER RESUSCITATION Lapinsky*, Lindsay Hurlburt, Emma Fitzgerald, Jonathan A.
(RADAR) INVESTIGATORS Silversides, Eleanor Hung; Queen Alexandra Hospital, Ports-
mouth, United Kingdom: David Pogson*, Emma Fitzgerald,
Study sites and investigators (*denotes Principal Investigator): Joe Schrieber, Sarah Birkhelzer, Genevieve Baragwanath; Royal
Altnagelvin Area Hospital, Londonderry, United Kingdom: Victoria Hospital, Belfast, United Kingdom: Jonathan A. Silver-
Noel Hemmings*, Sinead O’Kane; Antrim Area Hospital, Ant- sides*, Claire Shevlin, Padraig Headley, Darryl Stewart; Saint
rim, United Kingdom: Christopher Nutt*, Orla O’Neill, Emma Michael’s Hospital, Toronto, ON, Canada (two sites: Medical-
McKay; Belfast City Hospital, Belfast, United Kingdom: Jona- Surgical ICU and Trauma-Neurosurgical ICU): John C. Mar-
than A. Silversides*, Claire Montgomery, Christopher Murray, shall*, Uma Shankar Manickavasagam, Orla Smith, Thomas
John McCaffrey; Craigavon Area Hospital, Portadown, United Whitelegg; Ulster Hospital, Dundonald, United Kingdom: T.
Kingdom: Andrew J. Ferguson*, Jennifer Cosgrove, Laura John Trinder*, Samantha Hagan, Caroline Riddell.

Critical Care Medicine www.ccmjournal.org 1607

Copyright © 2018 by the Society of Critical Care Medicine and Wolters Kluwer Health, Inc. All Rights Reserved.