Methods
Types of studies
We will include:
1.
All clinical trials (randomised or quasi-randomised) comparing two or more fluid regimens in which different volumes of fluid are administered or different fluid balance is achieved, regardless of techniques used to guide fluid therapy
2.
Observational studies in which fluid management is a major focus of the study and in which the study reports the relationship between the volume of fluid administered or fluid balance and one or more of the outcomes above
We will exclude case series or reports, observational studies in which the total number of subjects is less than 50, and studies subject to post-publication investigation or retraction.
Types of participants
We will include studies of adult and paediatric patients. Since ‘critical illness’ is a poorly defined entity, we will include critical illnesses for which widely accepted definitions exist: ARDS, sepsis and SIRS in this review. We will exclude studies involving animals, or in which the patient population is made up predominantly of neonates, post-cardiac surgery or heart failure patients or perioperative patients.
Types of intervention(s) and comparators
Studies will be included if the intervention involved minimisation of fluid intake or active deresuscitation using diuretics or ultrafiltration and if the comparator was standard care or liberal fluid strategy.
We will exclude studies in which fluid management is only one aspect of a haemodynamic intervention strategy or in which the main comparison is between two or more different types of fluid.
Types of outcome measures
2.
Secondary outcomes:
-
Length of ICU stay, length of hospital stay
-
Organ dysfunction (multiple organ dysfunction score (MODS), sequential organ failure score (SOFA) or simplified acute physiology score (SAPS II))
-
Renal dysfunction (AKI incidence, renal replacement therapy use)
-
Respiratory dysfunction (incidence of new ARDS, ventilator-free days)
-
Cognitive function (author-defined)
-
Long-term mortality (>90 days)
Search strategy
We will search the following databases for relevant studies from 1980 onwards with no language constraints: Medline, EMBASE, and Cochrane Central Register of Controlled Trials (CENTRAL). In addition, we will undertake a manual search of indexed abstracts from the following key conference proceedings from 2009 to the present: the American Thoracic Society, Society of Critical Care Medicine, and European Society of Intensive Care Medicine Annual Congresses and the International Symposium on Intensive Care and Emergency Medicine.
Key search terms will include the following: fluid therapy, diuretics, furosemide, ultrafiltration, water-electrolyte balance, systemic inflammatory response syndrome, sepsis, acute lung injury, respiratory distress syndrome and adult. Our Medline search strategy (Additional file
1) will be adapted for searches in other databases.
Citation management and screening
Citations will be stored using Microsoft Excel software (Microsoft Corporation, Redmond, WA), and duplicates will be removed. Studies will be screened initially according to title and abstract by two authors independently, and those not meeting the criteria will be discarded. Disagreements will be resolved by discussion and referral to a third author if necessary.
After this initial stage, the full text of all remaining studies will be reviewed by two authors independently for inclusion or exclusion in the final study. As before, disagreements will be resolved by discussion and referral to a third author if necessary.
Data abstraction
For interventional studies, we will record information on the type and setting of the study, number of subjects, patient characteristics, nature of the intervention(s) in each group and outcomes as described above. After piloting, data will be extracted independently by two authors using standardised report forms (Additional file
2). Adjudication by a third author will be used if necessary.
For observational studies, a minimal dataset will include study details, patient characteristics, description of the intervention(s), main outcomes and adverse events. Efforts will be made to contact the authors of primary studies to provide missing data where necessary.
Assessment of risk of bias
The quality of included studies will be independently assessed by two authors and verified by a third if necessary. Study quality of RCTs will be assessed using the domain-based evaluation recommended by the Cochrane Collaboration [
17]. The domains include:
1.
Random sequence generation
For each domain, we will assign a judgement regarding the risk of bias as high, low or unclear [
17]. We will attempt to contact the trial corresponding author for clarification when insufficient detail is reported to assess risk of bias. Once we have consensus on the quality assessment of the six domains for eligible studies, we will assign them to the following categories:
1.
Low risk of bias: describes studies for which all domains are scored as ‘yes’
2.
Moderate risk of bias: describes studies for which one or more domains are scored as unclear or one domain is scored as ‘no’
3.
High risk of bias: describes studies for which more than one domain is scored as ‘no’
We will construct a ‘Risk of Bias’ table to present the results. We will use the assessment of risk of bias to perform sensitivity analyses based on methodological quality as necessary.
For observational studies, the Newcastle-Ottawa Scale will be used to assess study quality. Domains used to judge the quality of cohort studies include:
1.
Representativeness of the exposed cohort
2.
Selection of the non-exposed cohort
3.
Ascertainment of exposure
4.
Demonstration that outcome of interest was not present at the start of the study
5.
Comparability of cohorts on the basis of the design or analysis
Case-control studies will be assessed using the appropriate version of this scale. Reporting of all observational studies will include this assessment of study quality.
Data synthesis
If two or more randomised controlled trials are available for an outcome, their results will be combined in a meta-analysis using Review Manager (RevMan 5.3) software (Nordic Cochrane Centre, Cochrane Collaboration) on an intention-to-treat basis if appropriate to do so. We will not attempt meta-analysis of observational studies. If heterogeneity is very high (i.e. >90 %) and is unexplained by pre-planned subgroup analysis, we will report results narratively. If there are data from only one study for an outcome, the results will be reported narratively. For continuous data, we will use the inverse variance method, while the Mantel-Haenszel method will be employed for dichotomous data. For key outcomes, we will evaluate our confidence in the body of evidence according to the GRADE scale ([
18]).
Measures of effect for dichotomous data
Where possible, the treatment effect for dichotomous outcomes will be expressed as odds ratios with 95 % confidence intervals. If time-to-event data is available, we will use hazard ratios rather than risk ratios. We will use the Peto method to estimate the odds ratios for rare outcomes.
Measures of effect for continuous data
The treatment effect for continuous outcomes will be expressed as mean difference with 95 % confidence intervals. Where different scales are used, standardised mean difference (SMD) with 95 % confidence intervals will be used to express treatment effect.
Measures of effect for ordinal data
We will analyse longer ordinal scales as continuous data, while shorter ordinal scales will be abbreviated to dichotomous data. Treatment effects will then be assessed as described for each type of data.
Unit of analysis issues
Individual participants in each trial arm will comprise the unit of analysis. We anticipate that all trials included in the meta-analysis will have a parallel group design, and thus, no adjustment will be necessary for crossover or clustering.
Missing data
Where possible, we will contact trial authors to request missing data.
Assessment of heterogeneity
We will assess qualitative heterogeneity of the included studies and only combine studies whose participants are sufficiently similar, such that combining data across studies leads to a meaningful result. Statistical heterogeneity will be informally evaluated from forest plots of the study estimates and more formally using the chi
2 test (
P value <0.1, significant heterogeneity) and
I
2 statistic (
I
2 > 50 % = significant heterogeneity) [
17].
Assessment of reporting biases
If a sufficient number of studies is identified (n > 10), we will investigate reporting biases through the use of a funnel plot.
Sensitivity and subgroup analysis
If appropriate, we will investigate the influence of methodological quality on results by examining the effect of excluding studies at high or unclear risk of bias, and if necessary, we will undertake sensitivity analysis to assess the effect of excluding individual studies.
If sufficient studies are available, we will undertake subgroup analyses for adults, children, all patients with ALI or ARDS and all patients with sepsis, SIRS or septic shock.
Standards
Reporting will conform to Preferred Reporting Items for Systematic Reviews and Meta Analyses (PRISMA) standards (Additional file
3) [
19]. This systematic review has been registered with PROSPERO, an international prospective register of systematic reviews (
http://www.crd.york.ac.uk/PROSPERO/).
Discussion
While fluid management has been extensively studied and discussed in the critical care literature, no systematic review to date has attempted to summarise the evidence for a greater or lesser fluid volume or balance at different time points in critical illness, and this remains an area of debate among clinicians and researchers. It is hoped that in summarising the current evidence, this systematic review will inform practice and future trial design.
Acknowledgments
The authors wish to thank Ms. Melanie Anderson, information specialist at University Health Network library, for assistance with the development of the search strategy.
This work is being conducted as part of a doctoral research fellowship awarded to JS, funded by the Northern Ireland Health and Social Care Research and Development Division.
Authors’ contributions
JCM and JS conceived the idea for the review. JS drafted this protocol under the supervision of AJF, DM, BB, JCM and EF. All authors read and approved the final manuscript. Neither the funding body, sponsor nor the institutions had any involvement in development of the protocol.