Buffered solutions versus 0.9% saline for resuscitation in critically ill adults and children
Abstract
Background
Fluid therapy is one of the main interventions provided for critically ill patients, although there is no general consensus regarding the type of solution. Among crystalloid solutions, 0.9% saline is the most commonly administered. Buffered solutions may offer some theoretical advantages (less metabolic acidosis, less electrolyte disturbance), but the clinical relevance of these remains unknown.
Objectives
To assess the effects of buffered solutions versus 0.9% saline for resuscitation in critically ill adults and children.
Search methods
We searched the following databases to July 2018: CENTRAL, MEDLINE, Embase, CINAHL, and four trials registers. We checked references, conducted backward and forward citation searching of relevant articles, and contacted study authors to identify additional studies. We imposed no language restrictions.
Selection criteria
We included randomized controlled trials (RCTs) with parallel or cross‐over design examining buffered solutions versus intravenous 0.9% saline in a critical care setting (resuscitation or maintenance). We included studies on participants with critical illness (including trauma and burns) or undergoing emergency surgery during critical illness who required intravenous fluid therapy. We included studies of adults and children. We included studies with more than two arms if they fulfilled all of our inclusion criteria. We excluded studies performed in persons undergoing elective surgery and studies with multiple interventions in the same arm.
Data collection and analysis
We used Cochrane's standard methodological procedures. We assessed our intervention effects using random‐effects models, but when one or two trials contributed to 75% of randomized participants, we used fixed‐effect models. We reported outcomes with 95% confidence intervals (CIs).
Main results
We included 21 RCTs (20,213 participants) and identified three ongoing studies. Three RCTs contributed 19,054 participants (94.2%). Four RCTs (402 participants) were conducted among children with severe dehydration and dengue shock syndrome. Fourteen trials reported results on mortality, and nine reported on acute renal injury. Sixteen included trials were conducted in adults, four in the paediatric population, and one trial limited neither minimum or maximum age as an inclusion criterion. Eight studies involving 19,218 participants were rated as high methodological quality (trials with overall low risk of bias according to the domains: allocation concealment, blinding of participants/assessors, incomplete outcome data, and selective reporting), and in the remaining trials, some form of bias was introduced or could not be ruled out.
We found no evidence of an effect of buffered solutions on in‐hospital mortality (odds ratio (OR) 0.91, 95% CI 0.83 to 1.01; 19,664 participants; 14 studies; high‐certainty evidence). Based on a mortality rate of 119 per 1000, buffered solutions could reduce mortality by 21 per 1000 or could increase mortality by 1 per 1000. Similarly, we found no evidence of an effect of buffered solutions on acute renal injury (OR 0.92, 95% CI 0.84 to 1.00; 18,701 participants; 9 studies; low‐certainty evidence). Based on a rate of 121 per 1000, buffered solutions could reduce the rate of acute renal injury by 19 per 1000, or result in no difference in the rate of acute renal injury. Buffered solutions did not show an effect on organ system dysfunction (OR 0.80, 95% CI 0.40 to 1.61; 266 participants; 5 studies; very low‐certainty evidence). Evidence on the effects of buffered solutions on electrolyte disturbances varied: potassium (mean difference (MD) 0.09, 95% CI ‐0.10 to 0.27; 158 participants; 4 studies; very low‐certainty evidence); chloride (MD ‐3.02, 95% CI ‐5.24 to ‐0.80; 351 participants; 7 studies; very low‐certainty evidence); pH (MD 0.04, 95% CI 0.02 to 0.06; 200 participants; 3 studies; very low‐certainty evidence); and bicarbonate (MD 2.26, 95% CI 1.25 to 3.27; 344 participants; 6 studies; very low‐certainty evidence).
Authors' conclusions
We found no effect of buffered solutions on preventing in‐hospital mortality compared to 0.9% saline solutions in critically ill patients. The certainty of evidence for this finding was high, indicating that further research would detect little or no difference in mortality. The effects of buffered solutions and 0.9% saline solutions on preventing acute kidney injury were similar in this setting. The certainty of evidence for this finding was low, and further research could change this conclusion. Patients treated with buffered solutions showed lower chloride levels, higher levels of bicarbonate, and higher pH. The certainty of evidence for these findings was very low. Future research should further examine patient‐centred outcomes such as quality of life. The three ongoing studies once published and assessed may alter the conclusions of the review.
PICOs
Plain language summary
Buffered solutions versus 0.9% saline for resuscitation in critically ill adults and children
Background
Intravenous fluid therapy serves as the cornerstone of treatment for a wide spectrum of severe illnesses. Knowing its impact in terms of clinical outcomes is an important issue. There are some doubts as to whether the use of 0.9% saline may cause higher mortality among inpatients or a relevant worsening of their kidney function.
The aim of this Cochrane Review was to find out if fluid therapy with buffered solutions (water‐based salt (saline) solution with a buffer to maintain a constant pH) resulted in fewer hospital deaths and less damage to the kidneys for critically ill adults and children, when compared to 0.9% saline.
Study characteristics
We found 21 studies conducted in both children and adults, with a total of 20,213 participants. These studies compared buffered solutions with 0.9% saline solutions for critically ill adults and children (including those with sepsis, trauma, burns, or shock) who had not had planned surgery. We excluded trials where participants underwent planned (elective) surgery. These studies took place in 13 countries.
Study funding sources
Twelve of the included studies were funded by governments or non‐profit organizations, two received mixed funding, one was funded by a company whose role was not clarified, and six provided no details about trial funding.
Key results
Buffered solutions did not seem to reduce hospital deaths or worsening of renal (kidney) function in critically ill adults and children when compared to 0.9% saline.
The review shows that when critically ill patients received buffered solutions compared to 0.9% saline solutions:
1. buffered solutions made little or no difference to overall mortality (19,664 participants; 14 studies; high‐certainty evidence);
2. buffered solutions probably may make little or no difference in reducing the number of patients with worsening kidney function (18,701 participants; 9 studies; low‐certainty evidence); and
3. we are uncertain whether buffered solutions reduce impairment of other organs (e.g., lung, liver, or brain function), electrolyte disturbances (increasing or decreasing chloride or sodium or other salts), and the need to receive blood transfusions because evidence certainty has been assessed as very low.
None of the studies looked at blood loss, clotting disturbances (concerning risk of bleeding or clots), and quality of life.
Results varied in terms of the time points at which they were reported, the unit of measurement used, and the measures reported. The total amount of fluid given as fluid therapy was not recorded. Only four studies involved children. These children were less sick than participants included in the adult trials, and kidney damage was not reported. The three ongoing studies once published and assessed may alter the conclusions of this review.
How up‐to‐date is this review?
We searched for studies that had been published up to July 2018.
Authors' conclusions
Summary of findings
| Patient or population: critically ill adults and children | |||||
| Outcomes | Anticipated absolute effects* (95% CI) | Odds ratio | № of participants | Certainty of the evidence | |
| Risk with saline 0.9% | Risk with buffered solutions | ||||
| Overall mortality (in‐hospital death) | Study populationa | OR 0.91 | 19,664 | ⊕⊕⊕⊕ | |
| 119 per 1000 | 108 per 1000 | ||||
| Acute renal Injury (defined as RIFLE or AKIN criteria by study authors) | Study population | OR 0.92 | 18,701 | ⊕⊕⊝⊝ | |
| 121 per 1000 | 111 per 1000 | ||||
| Organ system dysfunction (during admission) | Study population | OR 0.80 | 266 | ⊕⊝⊝⊝ | |
| 156 per 1000 | 137 per 1000 | ||||
| Electrolyte disturbances | |||||
| Plasma potassium, mmol/L (during admission) | Median potassium in control groups was 3.55 | MD 0.09 | ‐ | 158 | ⊕⊝⊝⊝ |
| Plasma chloride, mmol/L (during admission) | Median chloride in control groups was 109.75 | MD ‐3.02 | ‐ | 351 | ⊕⊝⊝⊝ |
| Plasma pH (during admission) | Median pH in control groups was 7.36 | MD 0.04 | ‐ | 200 | ⊕⊝⊝⊝ |
| Plasma bicarbonate, mmol/L (during admission) | Median bicarbonate in control groups was 19.75 | MD 2.26 | ‐ | 344 | ⊕⊝⊝⊝ |
| *The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). * The basis for the basal risk assumption in continuous outcomes is the median control group response across studies. | |||||
| GRADE Working Group grades of evidence. | |||||
| aFew included studies were rated as high risk of bias, and the weight of them was 1%. Relevant heterogeneity was not detected. Several studies did not report mortality but publication bias was not suspected upon the funnel plot. Certainty of the outcome was not downgraded. bWe estimated a serious limitation of risk of bias because several studies were not blinded. Certainty of the outcome was downgraded two levels because of risk of bias and publication bias. cThe possibility of publication bias is not excluded because several studies did not report this outcome. dSeveral studies included in this outcome have high or unclear risk of bias for blinding or incomplete reporting. eStudies included few participants and few events. fCertainty was downgraded by three levels because risk of bias, imprecision, and publication bias. gOnly four studies with a total of 258 participants involved children. | |||||
Background
Description of the condition
Intravenous fluid therapy is the intervention that is probably provided most frequently for critically ill patients (Myburgh 2015). More than a third of all hospitalized patients are given intravenous fluid therapy (Finfer 2010). It is estimated that every year, 200 million litres of intravenous fluids are used in the USA (Young 2014), as well as 10 million litres of 0.9% saline in the UK (Awad 2008). Intravenous fluid preparations differ in their physiochemical properties. The ideal intravenous fluid should keep electrolytes and pH at physiological levels and should have the ability to expand intravascular volume (Myburgh 2013). For many clinical scenarios, such as sepsis, acute pancreatitis, and severe trauma, intravenous fluid therapy serves as the cornerstone of treatment.
Available options for fluid therapy vary widely in terms of fluid volume, timing, and choice of solution. However, no standard care intravenous fluid therapy has been universally accepted (Finfer 2010; Singer 2012). The cost of different intravenous solutions has been estimated at three to a hundred times greater than the cost of 0.9% saline (Lyu 2014). Several medical societies and authors have developed consensus papers and have presented evidence‐based guidelines intended to improve decisions about fluids and related outcomes (Garnacho‐Montero 2015; NICE 2013; Raghunathan 2014b; Reinhart 2012). Other guidelines have focused more on critical illness and have provided recommendations concerning fluid therapy (Dellinger 2013; Maraví 2014).
Fluid therapy solutions are classified as colloid or crystalloid solutions. Colloids are solutions composed of large molecules dispersed throughout fluid. In theory, they cannot cross the healthy semi‐permeable endothelial layer owing to their large molecular size, but this is not a criterion to be considered for colloids. Colloids have been described as more effective than crystalloids in increasing intravascular volume (Trof 2010).
However, according to two Cochrane Reviews, colloids offer no benefit over crystalloids and may even increase risks of renal failure and death (Lewis 2018; Mutter 2013). Further, studies addressing glycocalyx function during critical illness have provided some explanation for possible detrimental effects of colloids. New insight into the Starling principle suggests that the glycocalyx is the main determining factor in transcapillary flow (Starling 1896). In disease states, the integrity of the glycocalyx may be compromised, which translates to greater permeability and a greater rise in the oncotic pressure gradient, causing interstitial oedema (Levick 2010; Woodcock 2012). Accordingly, recent evidence does not support the use of colloids for resuscitation in critically ill patients (Hartog 2014; Reinhart 2012).
Although 0.9% saline is a widely used crystalloid solution, it causes hyperchloraemic acidosis with significant consequences for patients identified in several observational studies (Raghunathan 2014a; Shaw 2014; Yunos 2014). As safer crystalloids, buffered solutions have been assessed for their resuscitation capacity. According to a Cochrane Review of the safety and efficacy of buffered fluids in adult patients undergoing elective surgery, those who received buffered solutions did not develop hyperchloraemic acidosis (Bampoe 2017). In another systematic review of elective surgery and critical care, high‐chloride fluids were associated with greater risk of acute kidney; this particular review provided little evidence of effects on mortality (Krajewski 2015). Characteristics of elective surgical patients are different from those of critically ill patients. Fluid therapy prescribed for non‐surgical patients is more targeted at multiple organ dysfunction syndrome. This means that conclusions of reviews that have examined intravenous fluid therapy in elective surgical patients cannot necessarily be transferred to critically ill patients (Bampoe 2017).
In summary, according to recent evidence, use of colloid solutions in critically ill patients for resuscitation purposes generally is not recommended, and the benefits of buffered solutions versus 0.9% saline in this patient subset remain unclear.
Description of the intervention
Crystalloids are aqueous solutions of ions that show different properties according to their ion concentrations (Appendix 1). Because of its low cost, wide user experience, and general availability, 0.9% saline is the most commonly used intravenous fluid (Awad 2008). Use of terms such as normal or physiological saline is not recommended by the authors of this Review because, although 0.9% saline contains sodium and chloride in equal concentrations, it contains higher than physiological levels. The terms normal and physiological reflect historical issues rather than chemical properties (Awad 2008). Hence, in this current review, we used the term 0.9% saline.
Other crystalloid formulations, called balanced or buffered solutions, differ from 0.9% saline in terms of three properties: lower sodium and chloride contents, bringing them closer to normal plasma levels; the presence of other ions such as potassium, calcium, or magnesium, which could have effects on factors such as potassium or lactate levels, or could play a role in liver disease (Orbegozo 2014); and, finally, their contents of anions such as lactate, acetate, and gluconate, which are metabolized to bicarbonate by tissue cells and may exert an additional buffering effect.
How the intervention might work
Data from experimental human and animal studies suggest that an infusion of 0.9% saline may induce greater hyperchloraemic acidosis and interstitial oedema than are associated with the use of intravenous buffered solutions (Chowdhury 2012; Kellum 1998). Experimental studies have shown that hyperchloraemic acidosis increases the risk of worsening renal function through effects such as renal vasoconstriction, low renal perfusion pressure, and low glomerular filtration rate (Schnermann 1976; Wilcox 1983). Effects of acidosis on the immune system have been described in a rat model (Kellum 2006). Further, changes in systemic inflammation response have been linked to acidosis in humans (Wu 2011). In a before‐and‐after trial, a chloride‐restrictive strategy was associated with significant reduction in the incidence of acute kidney injury and in the need for renal replacement therapy (RRT) in adult intensive care patients (Yunos 2012; Yunos 2014). In a recent observational study, hyperchloraemic acidosis was associated with increased mortality among critically ill patients (Raghunathan 2014a). The effect observed could be independent of the fluid volume administered and may be more closely related to the chloride load (Shaw 2014). These data suggest that buffered solutions, with their lower chloride concentrations than 0.9% saline, may reduce the incidence of hyperchloraemic acidosis, thus decreasing the risk of patient‐centred outcomes such as renal failure, need for RRT, and mortality (McCluskey 2013; Raghunathan 2014a).
Why it is important to do this review
Several observational studies in the critical care setting have reported an association between hyperchloraemic acidosis and relevant outcomes such as acute renal injury and mortality (Raghunathan 2014a; Shaw 2014; Yunos 2014).
A recent Cochrane Review concluded that buffered intravenous fluids reduce the incidence of hyperchloraemia and metabolic acidosis in elective surgery (Bampoe 2017). However, the trials reviewed were not adequately powered to permit conclusions about renal failure or mortality, and did not include critically ill patients. Several recent randomized controlled trials (RCTs) have examined the effects of buffered solutions in critically ill patients. In one such study, faster pH normalization was observed among severely dehydrated participants receiving buffered fluid therapy (Cieza 2013). Two RCTs conducted in trauma patients described a reduction in the incidence of hyperchloraemic acidosis without a rise in intracranial pressure (Roquilly 2013; Young 2014). Other trials have explored the potential benefits of buffered solutions in clinical settings such as diabetes (Van Zyl 2012), acute pancreatitis (Wu 2011; Zhao 2013), dengue (Wills 2005), and doxylamine‐induced rhabdomyolysis (Cho 2007). However, in the recent (0.9% saline vs Plasma‐Lyte 148 for intensive care unit fluid therapy) SPLIT study, which included 2278 critical care participants, no differences between 0.9% saline and Plasma‐Lyte 148 were reported (Young 2015).
The present systematic review identified and synthesized all available evidence on the efficacy and safety of isotonic buffered solutions versus 0.9% saline when used in critically ill patients. On the basis of wide use of this therapy and the amount of solution administered, minor effects of the intervention may have an impact on clinical outcomes. In addition, possible cost differentials may reduce overall costs.
Objectives
To assess the effects of buffered solutions versus 0.9% saline for resuscitation in critically ill adults and children.
Methods
Criteria for considering studies for this review
Types of studies
We included randomized controlled trials (RCTs) with parallel or cross‐over design, regardless of language of publication. We excluded studies performed in persons undergoing elective surgery. We considered unpublished studies and abstracts if adequate information on methods and results was provided.
Types of participants
We included studies on participants with critical illness (including trauma, burns, or emergency surgery during critical illness) who require intravenous fluid therapy. We included studies of adults and children, and we explored the effects of age in a subgroup analysis (Subgroup analysis and investigation of heterogeneity).
Types of interventions
We considered interventions that included the use of intravenous buffered solutions containing bicarbonate or its precursors versus intravenous 0.9% saline as control (Appendix 1). We considered all uses of fluids in a critical care setting of resuscitation or maintenance after enrolment. We required that included fluids were isotonic (osmolarity 250 to 350 mmol/L). Fluids before enrolment were registered and analysed if they had been reported.
To minimize confounding factors, we did not consider studies with multiple interventions (e.g. colloids plus buffered solutions). Therefore, we excluded studies that compared only crystalloids versus colloids, or those that compared different types of colloids, even if buffered colloids were used. However, we included studies with more than two arms if they fulfilled all of our inclusion criteria and compared buffered solutions versus 0.9% saline.
Types of outcome measures
Primary outcomes
-
Overall (in‐hospital) mortality
-
Acute kidney injury (AKI) during hospitalization as defined by risk of renal dysfunction, injury to the kidney, failure of kidney function, loss of kidney function, and end‐stage kidney disease (RIFLE), or by Acute Kidney Injury Network (AKIN) criteria
Secondary outcomes
-
Number of participants with organ system dysfunction (respiratory, haemodynamic, central nervous system, and hepatic) as defined in the included studies during admission
-
Electrolyte disturbances during admission (hyperchloraemic acidosis; serum sodium, potassium, calcium, and chloride concentrations; pH; serum bicarbonate; base excess; strong ion difference) measured as serum levels or defined by study authors (e.g. presence or absence of hyperchloraemic acidosis)
-
Blood loss or transfusion requirement during admission
-
Coagulation disturbances (expressed as thrombocytopenia or coagulopathy) during admission
-
Total volume of intravenous fluids needed during resuscitation
-
Quality of life measured with Short Form (SF)‐36 and the EuroQOL quality of life questionnaire (EQ‐5D) (Angus 2003)
-
Cost
Search methods for identification of studies
Electronic searches
We searched for studies with systematic and sensitive search strategies as described in the Cochrane Handbook for Systematic Reviews of Interventions, Chapter 6 (Higgins 2011). We applied no language, publication year, or publication status restrictions.
We searched the following databases.
-
Cochrane Central Register of Controlled Trials (CENTRAL; 2018 Issue 6).
-
MEDLINE (Ovid SP, 1946 to July 2018).
-
Embase (Embase Elsevier, 1974 to July 2018).
-
CINAHL (Ebsco, 1980 to July 2018).
We developed a subject‐specific search strategy in MEDLINE and modified it appropriately for the other databases. Where appropriate, we used the highly sensitive search strategy designed by Cochrane for identifying RCTs and controlled clinical trials as described in the Cochrane Handbook for Systematic Reviews of Interventions, Chapter 6 (Higgins 2011). Our search strategies can be found in Appendix 2.
Searching other resources
We checked the bibliographic references and citations of relevant studies and reviews for further references to trials. We also searched ClinicalTrials.gov (www.clinicaltrials.gov), as well as the World Health Organization International Clinical Trials Registry Platform (WHO ICTRP) (apps.who.int/trialsearch/), for unpublished and ongoing studies; Open Grey for grey literature (http://opengrey.eu/), and Google Scholar for additional trials (July 2018).
We searched abstracts from the most relevant meetings from 2000 to 2018.
-
Society of Critical Care Medicine (SCCM).
-
European Society of Intensive Care Medicine (ESICM).
-
Spanish Society of Intensive Care Medicine (SEMICYUC).
-
American Society of Critical Care Medicine (ASCCM).
-
European Society of Anaesthesiologists (ESA).
-
American Society of Anesthesiologists (ASA).
-
International Anesthesia Research Society (IARS).
-
Spanish Society of Anaesthesia (SEDAR).
-
American Thoracic Society (ATS).
-
American College of Surgeons (ACS).
-
Society of Thoracic Surgeons (STS).
We contacted experts in the field and main investigators to ask for any ongoing, missed, or unreported studies. We checked the reference lists of all included studies and relevant systematic reviews.
Data collection and analysis
Selection of studies
We selected studies according to the methods of the Cochrane Emergency and Critical Care Group. At least two review authors (AMA and JAB) independently screened all relevant abstracts and titles to determine whether they fulfilled the inclusion criteria. We piloted eligibility criteria on a sample of reports (including studies that the review authors deemed definitely eligible, definitely not eligible, and doubtful) to ensure their performance. We classified studies into three categories: 'exclude', 'uncertain', and 'include', according to pre‐determined criteria for this review (see Criteria for considering studies for this review). At that stage, we excluded only papers classified as 'exclude'. During a second stage of the process, the same two review authors (AMA and JAB) independently examined full‐text reports to check whether the studies complied with the eligibility criteria of this review. We resolved disagreements by discussion and consulted a third review author (MC) if we could not reach consensus. If appropriate, we contacted study authors to clarify the eligibility of a study. Finally, if we were unable to obtain necessary information, we excluded the study in question. We were not blinded to the names and affiliations of study authors, to journals, nor to study results at any stage of the review process. We performed the selection of studies and data extraction using Covidence software.
Data extraction and management
We modified the Cochrane Emergency and Critical Care Group data extraction form and piloted it with five studies to ensure its suitability (Appendix 3). Two review authors (AMA and JAB) independently extracted data. We compared results and resolved discrepancies by discussion or by consultation with a third review author (MC). We extracted the following information from each trial.
-
Study authors, journal and year of publication.
-
Study design.
-
Study hypothesis.
-
Statistical information: estimation of sample size, statistical power, and analysis methods.
-
Participant characteristics: demographics, previous diseases, fulfilment of inclusion criteria, and baseline comparability.
-
Intervention and control groups, outcome measures.
-
Results.
-
Risk of bias domains.
-
Conflicts of interest and funding.
Two review authors (JAB and AMA) entered the data into Review Manager 5 and checked them for accuracy (Review Manager 2014).
Dealing with duplicate publications
If we found several publications that referred to the same trial, we included the primary version of the study and referenced all secondary reports. We selected the most complete data among all publications found.
Assessment of risk of bias in included studies
Two review authors (JAB and AMA), independently assessed the methodological quality of each trial that fulfilled the inclusion requirements of our review. We extracted information related to risk of bias from each study according to the following domains: random sequence generation; allocation concealment; blinding of participants/investigators; outcome assessment; incomplete outcome data; and selective reporting, in accordance with the recommendations of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). If any of the above data were not available for the publication of interest, or if it was unclear whether criteria were met, we contacted the study author by email to request further information. If needed, we considered other sources of bias related to the design of the studies, as specified in theCochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). In accordance with recent publications, we considered industry funding as a potential source of bias (Bero 2013; Lundh 2012). We considered funding when work force, materials, or grants were provided for the study. We established the risk of funding bias as defined in Appendix 4. We did not exclude trials on the basis of risk of bias but conducted subgroup analyses to explore the effects of risk of bias in the meta‐analysis (see Subgroup analysis and investigation of heterogeneity). We performed the subgroup analysis to determine if the effect on primary outcome changed when we assessed only studies with low risk of bias for randomization, allocation concealment, and study blinding.
Measures of treatment effect
We performed all analyses according to standards specified in the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2011). We selected measures of treatment effects according to how data were expressed in studies as follows.
-
Dichotomous data (e.g. need for RRT): odds ratio (OR).
-
Continuous data (e.g. potassium levels): mean difference (MD), standard deviation (SD), or standardized mean difference (SMD) when original studies used different scales.
We reported outcomes with 95% confidence intervals (CIs).
Unit of analysis issues
We dealt with trials with cluster or cross‐over design and multiple arms as specified in the Cochrane Handbook for Systematic Reviews of Interventions (Elbourne 2002; Higgins 2011). First, for cluster design, if adjusted data were available, we incorporated them into analyses. Adjusted data for blood products outcomes in Semler 2018 were not available. For this outcome, we reduced the size of the trial to its ‘effective sample size’, considering an intracluster correlation coefficient of 0.01 (Hannan 1994). Second, for multiple‐arms trials, we included each buffered pair‐wise comparison separately and divided up the sample size of the 0.9% saline group among the comparisons.
Dealing with missing data
We conducted analyses on an intention‐to‐treat basis, meaning that:
-
we analysed participants in the intervention groups to which they were randomized, regardless of the intervention they actually received; and
-
we considered outcome data for all participants and included all randomized participants in the analysis.
For missing data, we used this strategy.
-
Contact study authors by email or by letter to request those data.
-
Describe missing outcomes for studies as proportions of participants for whom we had no data with reasons.
We performed sensitivity analysis by using two alternative scenarios for participants with incomplete or missing data for the two primary outcomes considered (overall mortality and acute renal injury during hospitalization) (Jakobsen 2014).
-
'Best‐worst’ case scenario analyses: we considered participants with missing outcome data to be successes in the experimental group and failures in the control group. The denominator included all participants in the trial.
-
'Worst‐best’ case scenario analyses: we considered participants with missing outcome data to be failures in the experimental group and successes in the control group. The denominator included all participants in the trial.
Assessment of heterogeneity
We checked for heterogeneity by considering:
-
clinical and methodological characteristics of studies;
-
forest plots of study results to visually check for overlaps in confidence intervals; and
-
results of the Chi² test for statistical heterogeneity (we considered trial results as heterogeneous with P < 0.10) and results of the I² statistic for quantification of heterogeneity. We judged the importance of the observed value of I² according to the magnitude and direction of effects and the strength of evidence of heterogeneity (from 0% to 40% heterogeneity ‐ might not consider important; from 30 to 60% heterogeneity ‐ might be moderate; from 50% to 90% heterogeneity ‐ might be substantial; and from 75% to 100% heterogeneity ‐ might be considerable) (Deeks 2011). We explored the reasons behind substantial or considerable heterogeneity by performing subgroup analyses.
Assessment of reporting biases
We minimized reporting bias by including both published and unpublished studies. We developed a strategy to search for unpublished studies (Searching other resources), and we looked for publication bias in every outcome reported. The pre‐defined strategy for assessment of reporting bias consisted of the following.
-
Graphically, if more than 10 studies were included for the considered outcome, we created a funnel plot (a scatter plot of the intervention effect against a measure of study size).
-
We assessed funnel plot asymmetry statistically if 10 or more studies were meta‐analysed.
We interpreted results while considering all causes of asymmetry (not only publication bias).
Data synthesis
We performed statistical tests according to the recommendations of the Cochrane Emergency and Critical Care Group using the Review Manager 5 package (Review Manager 2014), provided by Cochrane for data synthesis and analysis.
Assessment of significance
We assessed our intervention effects by using a random‐effects meta‐analytical model. If one or two trials contributed to 75% of randomized participants, we used a fixed‐effect model (Jakobsen 2014). We compared results of both models in a sensitivity analysis (see Sensitivity analysis). We did not calculate trial sequential analyses (TSA) because of the recent position statement published by Cochrane (Schmid 2018).
We calculated Bayes factor (ratio between the probability that the meta‐analysis result given the null hypothesis is true divided by the probability that the meta‐analysis result given the alternative hypothesis is true) for primary outcomes based on a relative risk reduction (RRR) of 20%. We chose a Bayes factor less than 0.1 as the threshold for significance.
Subgroup analysis and investigation of heterogeneity
If the number of studies was appropriate, we conducted subgroup analyses to examine the following.
Participant subsets
-
Neurocritical participants
-
Septic participants
-
Burn or trauma participants
-
Surgical critically ill participants
-
Participants with primary electrolyte imbalance (dehydration or diabetic ketoacidosis)
Intravenous buffered fluid received
-
Fluids containing bicarbonate as buffer
-
Fluids containing a bicarbonate precursor as buffer
Age group
-
Adults ≥ 16 years old
-
Children < 16 years old
Risk of bias
-
Trials with low risk of bias
-
Trials with unclear or high risk of bias
We assessed differences between subgroups by performing the test of interaction (Altman 2003).
Summary of findings and GRADE
We designed a 'Summary of findings' table using GRADE profiler software to summarize the key results of our review (GRADEpro GDT). We indicated in this table the population, intervention, and comparison, along with relevant outcomes. We used the GRADE system to assess the quality of evidence associated with the following outcomes (Guyatt 2008).
-
Overall (in‐hospital) mortality.
-
Acute renal injury.
-
Organ system dysfunction.
-
Electrolyte disturbances.
The GRADE approach appraises the quality of evidence according to the extent to which one can be confident that an estimate of effect or association reflects the item assessed. Two review authors (JAB and AMA) independently assessed the quality of the body of evidence for the design of summary of findings Table for the main comparison. Our confidence in the estimate of effect was evaluated in terms of study limitations, inconsistency of effect or unexplained heterogeneity, imprecision of results, indirectness, and publication bias. We used trial sequential analyses as a supplement for a more thorough assessment of imprecision (Guyatt 2011).
Sensitivity analysis
We undertook the following sensitivity analyses.
-
We evaluated the impact of risk of bias from individual trials in the magnitude or direction of overall effect. We excluded studies with high or unclear risk of bias in the following domains: allocation features, levels of missing data, and blinding of outcome assessment.
-
We explored potential differences when the fixed‐effect model versus the random‐effects model was used.
Results
Description of studies
See Characteristics of included studies,Characteristics of excluded studies, Characteristics of studies awaiting classification, and Characteristics of ongoing studies.
Results of the search
We conducted a search on 7 July 2018 (Figure 1). We retrieved 10,286 records. After reading the abstracts, we excluded 10,251 articles for one of the following reasons: irrelevance, overlap, or design. We identified 35 papers for further examination. We included 21 studies in the review. We excluded 14 studies because they did not fulfil our inclusion criteria. We found three ongoing trials in registries of clinical trials.
Flow diagram.
Included studies
We included 21 studies, with a total of 20,213 participants (Aditianingsih 2017; Allen 2016; Choosakul 2018; Dung 1999; Hasman 2012; Hassan 2017; Mahajan 2012; Mahler 2011; Ngo 2001; Ratanarat 2017; Reddy 2014; Semler 2016; Semler 2018; Van Zyl 2012; Vasu 2013; Verma 2016; Viaene 2014; Wu 2011; Young 2014; Young 2015; de‐Madaira 2018). Three RCTs with 19,054 participants (94.2%) contributed.
The details of the included trials are provided in the Characteristics of included studies table. All publications were written in English. All 21 included trials were published between 1999 and 2018.
Participants
Sixteen included trials were conducted in adults, four were conducted in the paediatric population, and one trial limited neither minimum nor maximum age as an inclusion criterion (Young 2015). Most participants were males, and the proportion of females ranged between 12% and 63%, in the 19 trials that reported this information. All studies enrolled critically ill participants ‐ five specifically acute pancreatitis (Choosakul 2018; de‐Madaira 2018; Reddy 2014; Vasu 2013; Wu 2011); three severe dehydration (Allen 2016; Hasman 2012; Mahajan 2012); three diabetic ketoacidosis (Aditianingsih 2017; Mahler 2011; Van Zyl 2012); two dengue shock syndrome (Dung 1999; Ngo 2001); two severe trauma (Hassan 2017; Young 2014); and one shock (Ratanarat 2017).
The intervention
-
Nine studies administered Ringer's lactate (Choosakul 2018; de‐Madaira 2018; Dung 1999; Mahajan 2012; Ngo 2001; Reddy 2014; Vasu 2013; Van Zyl 2012; Wu 2011)
-
Five trials used Plasma‐Lyte A (Allen 2016; Mahler 2011; Verma 2016; Young 2014; Young 2015)
-
Three studies used Sterofundin (Hassan 2017; Ratanarat 2017; Viaene 2014)
-
Two studies used either Ringer's lactate or Plasma‐Lyte A in an intervention arm (Semler 2016; Semler 2018)
-
One trial administered both Ringer's lactate and Plasma‐Lyte A in different arms of multiple interventions (Hasman 2012)
-
One used Ringerfundin (Aditianingsih 2017)
Setting
There was geographical variability across the included studies. Trials were performed in Belgium (Viaene 2014), Canada (Allen 2016), India (Mahajan 2012; Reddy 2014; Vasu 2013), Indonesia (Aditianingsih 2017), Malaysia (Hassan 2017), New Zealand and Australia (Verma 2016; Young 2015), South Africa (Van Zyl 2012), Spain (de‐Madaira 2018), Thailand (Choosakul 2018; Ratanarat 2017), Turkey (Hasman 2012), USA ( Allen 2016; Mahler 2011; Semler 2016; Semler 2018; Wu 2011; Young 2014), and Vietnam (Dung 1999; Ngo 2001).
Type of study
Five trials were multi‐centric studies (Allen 2016; Semler 2018; Verma 2016; Wu 2011; Young 2015). Most of the included trials had medium sample sizes, enrolling between 22 and 230 participants. The largest study enrolled 15,904 participants (Semler 2018). Most of the included trials were parallel studies, and three were cluster multiple‐cross‐over (Semler 2016; Semler 2018; Young 2015). Three studies compared multiple interventions, and we considered only their buffered solution and 0.9% saline arms (Hasman 2012; Dung 1999; Ngo 2001). One study used a four‐arm (2×2) factorial design (Wu 2011).
Financial sponsorship
Financial sponsorship was provided by non‐industry sources in the majority of included studies. One trial was supported by industry (Allen 2016), and two were supported by mixed funding (Verma 2016; Young 2015). Six studies did not report the sponsorship source in their publications (Aditianingsih 2017; Hasman 2012; Mahler 2011; Reddy 2014; Vasu 2013; Viaene 2014).
Contact with study authors
We contacted study authors for further information. Three authors of included studies responded. Dr. de‐Madaria provided a recent full‐text publication (de‐Madaira 2018), Dr. Semler provided additional methodological details regarding the characteristics of clusters, and Dr. Aditianingsih confirmed the data related to in‐hospital mortality. We contacted 13 study authors regarding methodology and outcomes, but they did not respond.
Excluded studies
We excluded 14 studies because they did not fulfil our inclusion criteria (Benoit 2016; Chen 2004; Cho 2007; Crivits 2016; Dybvik 1995; Fang 2008; Galas 2009; Kartha 2017; Martin 2018; Omar 2018; Rainier‐Pope 1962; Roquilly 2013; Rowell 2016; Yung 2017). The characteristics of these studies are summarized in the Characteristics of excluded studies table.
Studies awaiting classification
We identified no studies awaiting classification.
Ongoing studies
We found three ongoing trials in registries of clinical trials (NCT02721654; NCT02835157; NCT02875873). We have summarized the characteristics of these studies in the Characteristics of ongoing studies table.
Risk of bias in included studies
Details of individual studies are available in the Characteristics of included studies, the risk of bias summary (Figure 2), and the risk of bias graph (Figure 3).
Risk of bias summary: review authors' judgements about each risk of bias item for each included study.
Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.
Allocation
All trials were randomized. A majority of studies explained their randomization and allocation concealment methods. The most frequently used method was web‐based randomization. Because of this, we rated the majority as having low risk of bias. Five studies did not provide any information about the method used for randomization nor allocation concealment (Aditianingsih 2017; Ratanarat 2017; Reddy 2014; Vasu 2013; Viaene 2014) . We asked for this information by mail, but we received no information. We rated those studies as having unclear risk for selection bias. The number of participants in most trials was low (fewer than 50 participants in each arm). Five trials randomized more than 100 participants (Allen 2016; Ngo 2001; Semler 2016; Semler 2018; Young 2015).
Blinding
We included four open‐label trials and eight participant‐, personnel‐, and assessor‐blinded studies. We could not assess the blinding status of participants, personnel, or assessors in nine trials. We requested this information by emailing the study authors, but we received no information. We distinguished between objective outcomes (e.g. mortality) and subjective outcomes (e.g. AKI defined by study authors) in evaluating detection bias. Regarding mortality (considered objective), we rated 15 studies as having low risk of performance and detection bias (Allen 2016; de‐Madaira 2018; Dung 1999; Hasman 2012; Mahajan 2012; Mahler 2011; Ngo 2001; Ratanarat 2017; Semler 2016; Semler 2018; Van Zyl 2012; Verma 2016; Young 2014; Young 2015; Wu 2011). Concerning subjective outcomes, we assessed eight studies as having low risk of bias (Allen 2016; de‐Madaira 2018; Dung 1999; Mahajan 2012; Van Zyl 2012; Verma 2016; Young 2014; Young 2015); nine trials did not detail the methods used in blinding assessors, and we rated them as having unclear risk; four studies were open‐label and we categorized them as high risk.
Incomplete outcome data
Nine studies performed their analyses in an intention‐to‐treat way (Aditianingsih 2017; Choosakul 2018; Dung 1999; Hasman 2012; Mahajan 2012; Semler 2016; Semler 2018; Wu 2011; Young 2015). We rated two trials as having high risk of attrition bias because they used a modified intention‐to‐treat analysis or a per‐protocol analysis not prespecified in the protocol. Five studies reported incomplete outcome data or changes in the analysis plan or in the inclusion criteria; we rated them as high risk. The remaining five trials did not provide enough information, and we classified them as unclear risk.
Selective reporting
To assess reporting bias, we obtained information on published protocols or on clinical trials registers if it was available. Four studies had published protocols, and we identified no reporting bias (Semler 2016; Semler 2018; Verma 2016; Young 2015). Five studies had public information in clinical trials registers related to prespecified outcomes with no evidence of selective reporting (Allen 2016; Choosakul 2018; Mahajan 2012; Wu 2011; Young 2014). The remaining 12 studies did not provide information; we rated them as having unclear risk of reporting bias.
Other potential sources of bias
In our protocol (Barea‐Mendoza 2016), we specified the criteria used to assess funding bias according to the other authors (Appendix 4). Most of the studies provided information about funding, and we rated them as having low risk of bias. Two studies received mixed funding through unrestricted grants from the companies involved (Verma 2016; Young 2015). We rated both trials as low risk of bias. One study stated no information about sponsorship source, and its authors declared no conflicts of interest (Hasman 2012). We rated this study as having low risk of bias. One study reported a company source of funding without providing information about the role of the company (e.g. design or analysis) (Allen 2016). We rated it as having high risk of bias with pre‐specified criteria (Appendix 4). Five studies did not provide information about study funding, and we rated them as having unclear risk of bias (Aditianingsih 2017; Mahler 2011; Reddy 2014; Vasu 2013; Viaene 2014). We included three cluster‐randomized, multiple‐cross‐over trials (Semler 2016; Semler 2018; Young 2015). We rated all of them as having low risk of recruitment bias, baseline imbalance between groups, loss to follow‐up of clusters, non‐comparability, and carry‐over effect. With regard to the risk of bias for incorrect analyses, we judged these three studies to be at low risk because adjustments of analyses for cluster were performed via individual‐level analyses based on a multi‐level model and generalized estimating equations (Semler 2016; Semler 2018; Young 2015).
Effects of interventions
See summary of findings Table for the main comparison for the main comparison buffered solutions versus 0.9% saline for resuscitation in non‐surgical critically ill adults and children.
All included studies reported at least one outcome of interest, as specified previously in Types of outcome measures. For primary outcomes, each outcome section is subdivided into six main subheadings: 'Overall effect', ‘Bayes factor’, 'Sensitivity analysis', and 'Subgroup analysis.'
We collected and analysed data from the 21 included studies (20,213 participants). Given that one trial dominated, with 75% of randomized participants, we used a fixed‐effect model in the primary outcomes analysis (Semler 2018). For primary outcomes (overall mortality and acute renal injury), we presented pooled results as ORs with 95% CIs. We performed additional Bayes factor analyses for primary outcomes. We performed formal assessment of funnel plot asymmetry only for overall (in‐hospital) mortality outcomes because more than 10 studies were combined. We presented secondary outcomes according to the protocol.
Primary outcomes
1. Overall (in‐hospital) mortality
Fourteen trials including 19,664 participants reported results on mortality (Aditianingsih 2017; Allen 2016; Choosakul 2018; de‐Madaira 2018; Dung 1999; Mahajan 2012; Ngo 2001; Semler 2016; Semler 2018; Van Zyl 2012; Verma 2016; Young 2014; Young 2015; Wu 2011). A total of 1102/9927 (11.1%) participants in the buffered solutions group died versus a total of 1154/9737 (11.9%) participants in the 0.9% saline group. Meta‐analysis showed no evidence of a difference between the two intravenous fluid therapies (odds ratio (OR) 0.91, 95% confidence interval (CI) 0.83 to 1.01; P = 0.06; I² = 0%; 14 trials; high‐certainty evidence). We did not observe any statistical heterogeneity among the studies (P = 0.94; I² = 0%).
Bayes factor
We calculated the Bayes factor based on the intervention effect shown by the meta‐analysis result (OR 0.91, 95% CI 0.83 to 1.01) and the intervention effect hypothesized in the estimation of the required information size (RRR 0.2). The Bayes factor was 3.9, which is above the Bayes factor threshold for significance of 0.1, seeming to provide more evidence for the null hypothesis compared to the evidence for an intervention effect of 20% RRR.
Sensitivity analysis
We performed sensitivity analysis as specified in the protocol (Barea‐Mendoza 2016). A random‐effects model revealed no differences between meta‐analysis results showing analysis under the fixed‐effect model because heterogeneity was low (OR 0.91, 95% CI 0.83 to 1.01; P = 0.06). We performed sensitivity analysis after excluding six studies with high risk of bias (Allen 2016; de‐Madaira 2018; Ngo 2001; Van Zyl 2012; Verma 2016; Young 2014). The analysis showed no difference for mortality when only studies with low risk of bias were included (OR 0.91, 95% CI 0.83 to 1.00; P = 0.06).
Subgroup analysis
We planned to perform subgroup analysis for mortality outcomes by disease subsets, type of fluid received, age group, and risk of bias when data were available. The included trials did not report outcomes by disease subsets.
Intravenous buffered fluid received
Four studies used Plasma‐Lyte (Allen 2016; Verma 2016; Young 2014; Young 2015); seven studies used Ringer's lactate (Choosakul 2018; de‐Madaira 2018; Dung 1999; Mahajan 2012; Ngo 2001; Van Zyl 2012; Wu 2011); and one study used Ringerfundin (Aditianingsih 2017). Two studies used Plasma‐Lyte or Ringer's lactate indistinctly (Semler 2016; Semler 2018). Subgroup analysis comparing trials using different types of buffered solutions showed no difference in mortality (Analysis 1.1).
Age group
Only four studies with a total of 258 participants involved children during the resuscitation (Allen 2016; Dung 1999; Mahajan 2012; Ngo 2001). Ten studies with a total of 19,406 participants were performed in the adult population (Aditianingsih 2017; Choosakul 2018; de‐Madaira 2018; Semler 2016; Semler 2018; Van Zyl 2012; Verma 2016; Wu 2011; Young 2014; Young 2015). Subgroup analysis showed no difference in mortality (Analysis 1.2).
See Figure 4.
Forest plot of comparison: 1 Buffered solutions vs SS 0.9% (saline solution), outcome: 1.1 Mortality.
2. Acute renal injury during hospitalization as defined by risk of renal dysfunction, injury to the kidney, failure of kidney function, loss of kidney function, and end‐stage kidney disease (RIFLE), or by Acute Kidney Injury Network (AKIN) criteria
This outcome was reported in nine studies including 18,701 participants (Aditianingsih 2017; Choosakul 2018; Ratanarat 2017; Semler 2016; Semler 2018; Verma 2016; Wu 2011; Young 2014; Young 2015). A total of 1077/9452 (11.39%) participants in the buffered solutions group developed an acute renal injury versus a total of 1118/9249 (12.09%) participants in the 0.9% saline group. Meta‐analysis showed no evidence of a difference between the two intravenous fluid therapies (OR 0.92, 95% CI 0.84 to 1.00; P = 0.06; I² = 0%; 9 trials; low‐certainty evidence). We did not observe any statistical heterogeneity among the studies (P = 0.93; I² = 0%).
Bayes factor
We calculated the Bayes factor based on the intervention effect shown by the meta‐analysis result (OR 0.92, 95% CI 0.84 to 1.00) and the intervention effect hypothesized in the estimation of the required information size (RRR 0.2). Bayes factor was 5.3, which is above the Bayes factor threshold for significance of 0.1, seeming to provide more evidence for the null hypothesis compared to the evidence for an intervention effect of 20% RRR.
Sensitivity analysis
We performed a sensitivity analysis using a random‐effects model. We found no differences between results obtained with a fixed‐effect model and those obtained with a random‐effects model (OR 0.92, 95% CI 0.84 to 1.00; P = 0.06). We performed sensitivity analysis, excluding studies with high risk of bias for the specified domains in the protocol (allocation features, blinding of outcome data, and incomplete outcome data) (Barea‐Mendoza 2016). There were no differences between results after exclusion of one study (Ratanarat 2017), and risk of bias for allocation features and blinding was unclear (OR 0.91, 95% CI 0.83 to 1.00; P = 0.06). We performed the analysis after excluding three studies with high risk of bias for incomplete outcome data (Ratanarat 2017; Verma 2016; Young 2014). The results were very similar (OR 0.91, 95% CI 0.83 to 1.00; P = 0.06).
Subgroup analysis
We planned to perform subgroup analyses for acute renal injury outcome by disease subsets, type of fluid received, and age group when data were available. However, the included trials did not report outcomes by disease subsets, and no trials reported renal injury in the paediatric population.
Intravenous buffered fluid received
Three studies used Plasma‐Lyte as buffer solution (Verma 2016; Young 2014; Young 2015); two studies used Ringer's lactate (Choosakul 2018; Wu 2011); and two studies used Plasma‐Lyte or Ringer's lactate indistinctly (decided by the treating clinician) (Semler 2016; Semler 2018). Both studies by Semler did not report individualized outcomes according to the type of fluid. Finally, two studies used Ringerfundin and Sterofundin as buffered solutions (Aditianingsih 2017; Ratanarat 2017). Subgroup analysis comparing trials using different types of buffered solutions showed no difference in acute renal injury Analysis 1.3.
See Figure 5.
Forest plot of comparison: 1 Buffered solutions vs SS 0.9% (saline solution), outcome: 1.3 Acute renal injury.
Secondary outcomes
1. Number of participants with organ system dysfunction as defined in the included studies
Six studies reported on organ system dysfunction. Five studies with a total of 266 participants were included in the analysis (Analysis 1.4) (Choosakul 2018; De Madaira 2018; Dung 1999; Ngo 2001; Wu 2011). None of the studies found a significant effect of Ringer's lactate solutions in preventing organ system dysfunction compared with control at an alpha level of 0.05. Overall, we found no clear evidence of an effect of buffered solutions on the occurrence of organ system dysfunction in critically ill patients, as evidenced by a fixed‐effect model (OR 0.80, 95% CI 0.40 to 1.61; P = 0.53; I² = 0%; 5 trials; very low‐certainty evidence). We could not include Vasu 2013, which examined the incidence of organ failure, in the analysis because total sample and event rates were not reported properly, and we were unable to get further information from the trial author.
2. Electrolyte disturbances (hyperchloraemic acidosis; serum sodium, potassium, calcium, and chloride concentrations; pH; serum bicarbonate; base excess; strong ion difference) measured as serum levels or defined by study authors (e.g. presence or absence of hyperchloraemic acidosis)
2.1 Sodium
Seven studies reported sodium levels (Hasman 2012; Hassan 2017; Mahajan 2012; Semler 2016; Semler 2018; Van Zyl 2012; Young 2014). We excluded three studies from the analysis. We excluded the studies by Semler and colleagues because they reported only the highest sodium levels during hospitalization. We excluded Van Zyl 2012 because investigators did not report the standard deviation (SD). We included four studies with a total of 222 participants in the analysis (Analysis 1.5). Quantitative analysis showed no evidence of an effect (mean difference (MD) ‐0.48, 95% CI ‐1.67 to 0.70; I² = 0%; 4 trials; very low‐certainty evidence).
2.2 Potassium
Seven studies reported potassium levels (Hasman 2012; Mahajan 2012; Semler 2016; Semler 2018; Van Zyl 2012; Young 2015; Young 2014). We excluded four studies from the analysis. We excluded the two studies by Semler and colleagues because they reported only the highest potassium levels during hospitalization. We excluded Young 2015 because study authors reported no difference in the incidence of hyperkalaemia (risk ratio (RR) 1.93, 95% CI 0.35 to 10.50). We excluded Van Zyl 2012 because the SD was not reported. We included three studies with a total of 158 participants in the quantitative analysis (Analysis 1.6). Data showed no differences (MD 0.09, 95% CI ‐0.10 to 0.27; I² = 0%; 3 trials; very low‐certainty evidence).
2.3 Chloride
Nine studies reported chloride levels (Allen 2016; Hasman 2012; Hassan 2017; Mahajan 2012; Mahler 2011; Semler 2016; Semler 2018; Van Zyl 2012; Young 2014). We excluded three studies from the analysis. We excluded the two studies by Semler and colleagues because they reported only the highest chloride levels during hospitalization. We excluded Van Zyl 2012 because the SD was not reported. We included six studies with a total of 351 participants from the quantitative analysis (Analysis 1.7). Data showed lower levels of chloride in the buffer solution group (MD ‐3.02, 95% CI ‐5.24 to ‐0.80; 6 trials; very low‐certainty evidence). Heterogeneity was significant (I² = 89%); therefore, we used the random‐effects model. The subgroup analysis showed significant differences between groups according to the type of buffer solution used (P = 0.02; I² = 73.3%): Plasma‐Lyte subgroup (MD ‐4.56, 95% CI ‐6.68 to ‐2.44); Ringer's lactate (MD ‐0.56, 95% CI ‐2.58 to 1.47); Sterofundin (MD ‐1.55, 95% CI ‐4.39 to 1.29).
2.4 pH
Four studies reported on this outcome (Hasman 2012; Hassan 2017; Van Zyl 2012; Young 2014). We excluded one study because the SD was not reported (Van Zyl 2012). We included three studies with a total of 200 participants in the quantitative analysis (Analysis 1.8). Data showed higher pH in the buffer solution group (MD 0.04, 95% CI 0.02 to 0.06; I² = 59%; 3 trials; very low‐certainty evidence).
2.5 Bicarbonate
Eight studies reported the bicarbonate levels (Allen 2016; Hasman 2012; Hassan 2017; Mahajan 2012; Mahler 2011; Semler 2016; Semler 2018; Young 2014). We excluded the two studies by Semler and colleagues because they reported only the highest bicarbonate levels during hospitalization. We included six studies with a total of 344 participants in the quantitative analysis (Analysis 1.9). Data show higher levels of bicarbonate in the buffer solution group (MD 2.26, 95% CI 1.25 to 3.27; 6 trials; very low‐certainty evidence). Heterogeneity was significant (I² = 72%); therefore we used the random‐effects model.
3. Blood loss or transfusion requirement
3.1 Blood loss
None of the included trials reported on this outcome.
3.2 Transfusion requirement
Two studies with a total of 16,776 participants reported on blood products (Analysis 1.10) (Semler 2016; Semler 2018). Blood products included the total of red blood units, fresh frozen plasma, platelets, and cryoprecipitate transfused. Neither of the two studies found a significant effect of buffered solutions reducing the volume of blood products compared with the control at an alpha level of 0.05. Overall, we found no clear evidence of an effect of buffered solutions on the requirement of blood products in critically ill patients, as evidenced by a fixed‐effect model (MD ‐17.53, 95% CI ‐44.52 to 9.46; P = 0.20; I² = 0%; moderate‐certainty evidence).
Two studies examined the transfusion requirement detailing the total units of red blood cells, plasma, and platelet (Young 2014; Young 2015). We did not include them in the analyses because the mean and the SD were not reported. They are listed as orphan outcomes in Table 1.
| Study | Outcome | Participants | Results |
| Total volume of intravenous fluids needed during resuscitation | 30 | Mean in the 0.9% saline group was 6.23 (4.3 to 8.2) litres and in the Rigerfundin group was 6.23 (4.1 to 8.4) litres | |
| Total volume of intravenous fluids needed during resuscitation | 100 | There were no significant differences in volume, duration of fluid administration, or maintenance fluid between groups | |
| Total volume of intravenous fluids needed during resuscitation | 40 | Median from 0 to 72 hours after randomization was 6904 (6400 to 8600) mL for 0.9% saline and 5900 (4930 to 7002) mL for Ringer's lactate solution; P = 0.045 | |
| Total volume of intravenous fluids needed during resuscitation | 22 | Median total fluids (including intravenous and oral rehydration solution) requirement was less in the Ringer's lactate group (310 mL/kg (IQR 230 to 365)) as compared to the 0.9% saline group (530 mL/kg (IQR 324 to 750)); P = 0.01 | |
| Total volume of intravenous fluids needed during resuscitation | 111 | Mean in the Ringer's lactate group was 134.2 (± 19.9) mL/kg and in the 0.9% saline group was 132.9 (± 16.6) mL/kg; P = 0.954 | |
| Total volume of intravenous fluids needed during resuscitation | 974 | Patients in the saline and balanced groups received a similar total volume of intravenous crystalloid by 30 days (1424 (500 to 3377) mL vs 1617 (500 to 3628) mL); P = 0.40 | |
| Total volume of intravenous fluids needed during resuscitation | 15,802 | Patients in the saline group received a mean of 2171 (± 3942) mL of saline solution and 216 (± 1394) mL of buffered solution. Patients in the buffered group received a mean of 492 (± 2303) mL of saline solution and 2083 (± 3310) mL of buffered solution | |
| Total volume of intravenous fluids needed during resuscitation | 65 | There was no significant difference in the amount of study fluid administered | |
| Total volume of intravenous fluids needed during resuscitation | 2,278 | Patients in the saline group received a mean of 2554 (± 2120) mL of saline solution and 1.8 (± 60) mL of buffered solution. Patients in the Plasma‐Lyte group received a mean of 0.5 (± 15) mL of saline solution and (2655 ± 3052) mL of Plasma‐Lyte solution | |
| Transfusion requirement | 65 | 16 (67%) patients in the saline group and 11 (50%) patients in the buffered solution group received pRBC transfusion (mean difference of 0.75, 95% CI 0.5 to 1.2) | |
| Transfusion requirement | 2,278 | Packed red blood cells: 26 (9%) patients in the saline group received a mean of 45 (± 277) mL, and 29 (9%) patients in the buffered group received a mean of 39 (± 199) mL of packed red blood cells at study day 3 | |
| Cost | 65 | The cost‐minimization analysis reported a 24‐hour cost differential of USD 12.35 in favour of Plasma‐Lyte A compared with non‐buffered solutions |
Data potentially of interest to this review but reported with variation in outcome measures or reported in single study, and therefore not suitable for numerical analysis. IQR: interquartile range; mL: millilitres; mL/kg: millilitres per kilogram; pRBC:packed red blood cells; vs: versus; USD: USA dollars.
4. Coagulation disturbances (expressed as thrombocytopenia or coagulopathy)
None of the included studies reported on this outcome.
5. Total volume of intravenous fluids needed during resuscitation
Nine studies with a total of 19,422 participants reported on the total volume of fluids needed during resuscitation (Aditianingsih 2017; Allen 2016; de‐Madaira 2018; Mahajan 2012; Ngo 2001; Semler 2016; Semler 2018; Young 2014; Young 2015). The data were heterogeneous in the types of measures of effect used, identification of study fluid infused per arm of comparison, and time scales, and we were unable to get further information from the investigators; therefore, we did not combine these for meta‐analysis. Two trials found a significant difference in the total fluid requirement in favour of the intervention arm (de‐Madaira 2018; Mahajan 2012). Four studies reported a similar cumulative volume of fluids administered between groups (Aditianingsih 2017; Allen 2016; Ngo 2001; Semler 2016; Young 2014).
We did not subject these results to any further analysis, and this outcome is listed as orphan outcome in Table 1.
6. Quality of life measured with Short Form (SF)‐36 and the EuroQOL quality of life questionnaire (EQ‐5D) (Angus 2003)
None of the included trials reported on quality of life.
7. Cost
A single study examined cost inputs (Smith 2014 used data on participants enrolled in Young 2014). This study enrolled 65 participants and detailed an analysis that included costs for intravenous fluid acquisition, electrolyte acquisition, and nurse labour. The cost‐minimization analysis reported a 24‐hour cost differential of USD 12.35 in favour of Plasma‐Lyte A compared with non‐buffered solutions. We did not subject this outcome to any further analysis, and we listed this outcome as an orphan outcome in Table 1.
Discussion
Summary of main results
We included in this review the data from 21 studies with 20,213 participants. Primary outcomes analyses showed that use of buffered solutions for resuscitation in non‐surgical critically ill adults did not improve mortality or the incidence of acute renal injury. Fourteen trials including 19,664 participants contributed to the mortality analyses. Data showed no effect on mortality (odds ratio (OR) 0.91, 95% confidence interval (CI) 0.83 to 1.01; P = 0.06; Figure 4). We graded the certainty of evidence as high. We performed a sensitivity analysis for the analysis model (fixed‐effect vs random‐effects) and for risk of bias without changes in the results. Only four studies involved a paediatric population. Subgroup analysis by age showed no differences in mortality. Nine trials including 18,701 participants were included in the acute renal injury outcome. Data showed no effect on the incidence of acute renal injury (OR 0.92, 95% CI 0.84 to 1.00; P = 0.06; Figure 5). We graded the certainty of evidence as low because several studies were not blinded and publication bias was not excluded. The sensitivity analysis for the analysis model and for the risk of bias did not show any difference. Analyses of all secondary outcomes suggest that patients treated with buffer solutions have lower levels of chloride (mean difference (MD) ‐3.21, 95% CI ‐5.52 to ‐0.89), higher pH (MD 0.05, 95% CI 0.04 to 0.07), and higher levels of bicarbonate (MD 2.48, 95% CI 1.37 to 3.59). However, the clinical significance of these results for the incidence of hyperchloraemic acidosis in the critical care setting is doubtful. We have very low‐certainty evidence for these outcomes. The analyses did not show differences between groups in terms of organ dysfunction or levels of sodium or potassium. The included trials did not report results on blood loss, coagulopathy, or quality of life.
Overall completeness and applicability of evidence
This current systematic review includes published trials comparing buffered solutions versus 0.9% saline for resuscitation in critically ill adults and children. We excluded trials carried out in the setting of major surgery because a recently updated Cochrane Review has already assessed this question (Bampoe 2017). Fluids are used in a wide range of situations in critically ill patients such as sepsis, trauma, or different shock profiles. These populations are well represented in the current review. We planned to carry out subgroup analysis according to clinical conditions (burns, neurocritical, septic, etc.). We did not perform these analyses because most of the trials included case mixed intensive care patients and reported pulled results. We think that most of our conclusions are applicable to a general critical care setting. However, some subgroups of patients could be not adequately represented in this systematic review, and they deserve additional comments. Only four trials with 258 participants were included in the subset of the paediatric population (Allen 2016; Dung 1999; Mahajan 2012; Ngo 2001). The subgroup analysis did not show differences (P = 0.96; I² = 0%). No trial reported on acute renal injury in the paediatric population. Furthermore, participants included in these trials were less sick than participants included in the adult trials. For these reasons, we should be very cautious in interpreting results from the paediatric population. In addition, we consider that neurocritical patients need special attention. Buffered solutions trend to have lower osmolarity than 0.9% saline (Appendix 1). For this reason, several studies excluded neurocritical patients. In our review, we included only one study, which included 66 participants with traumatic brain injury (Hassan 2017). The authors of this trial did not report any signal of harm. One other study carried out on neurocritical patients was excluded because it included colloid solutions as the co‐intervention (Roquilly 2013). In this trial, intracranial pressure was not different between study groups (MD 4 mmHg, 95% CI ‐1 to 8; P = 0.088). A recent consensus for neurointensive care patients recommends the use of buffered solutions in this population based on lower rates of hyperchloraemia (Oddo 2018). Safety data, or patient‐centred outcomes, were not considered in this consensus. The authors of this consensus graded the quality of these recommendations as low and very low. Due to the paucity of data in this population, we consider that our conclusions should be translated with caution for these patients. The 21 trials included in this current Cochrane Review included a total of 20,213 participants. The included trials reported many outcomes but inconsistently. Many trials did not report on our primary outcomes of mortality and acute renal injury. Reporting of secondary outcomes was highly variable in terms of time points, units, and measures. A potential confounder is related to volumes used due to range of lack of reporting data prior randomization (type and/or volume) or ability to combine them. Although analysing a dose‐response relationship could have been interesting, the paucity of data related to doses of the intervention precluded this kind of analysis. Closely linked with the previous comment, we should highlight the wide range of enrolment timing (emergency criteria, intensive care unit (ICU) admission, within 60 minutes of hospital arrival, etc.). Furthermore, important outcomes such as costs and quality of life were never reported. Expected adverse events are related to electrolyte disturbances. They were broadly but inconsistently reported in the included studies. In spite of this, no signal of harm is related to the buffered solution.
Quality of the evidence
The certainty of evidence range in the included trials was between very low and high. For the two primary outcomes, certainty was high and low for mortality and acute renal injury, respectively (summary of findings Table for the main comparison). Six out of 14 trials included in the mortality outcome were graded as high risk of bias, most because we could not rule out selective reporting bias nor attrition bias (Figure 2). Sensitivity analysis did not modify the results after high‐risk studies were excluded; they represented only 1% of the overall weight. Relevant heterogeneity was not detected. Seven trials did not report mortality data. Reporting or publication bias was not suspected upon visual assessment of the funnel plot (Figure 6). For these reasons, we considered the certainty of evidence as high (no downgrading).
In contrast with mortality, eight out of nine trials included in the acute renal injury outcome were graded as high risk of bias (Figure 2). The high risk of bias was related to detection bias. We considered acute renal injury as a subjective outcome, and several studies were not blinded (or unclear). Sensitivity analysis did not modify the results after exclusion of these studies, but they represented more than 90% of the overall weight. Only 9 out of 21 included studies reported acute renal injury data. We did not prepare a funnel plot because we included fewer than 10 studies in the analysis. We did not suspect inconsistency, indirectness, or imprecision in the acute renal injury certainty assessment. We therefore classified the certainty of evidence as low (risk of bias: ‐1; publication bias: ‐1).
For secondary outcomes, the certainty of evidence was very low (summary of findings Table for the main comparison). Reasons for this included risk of bias (high in several trials), indirectness (surrogate outcomes), inconsistency (different time points across studies), and imprecision (few participants and events in the trial that reported these outcomes). The certainty of evidence was very low for the secondary outcomes, which means that further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. In the future, trial authors should improve the consistency of measures and reports of secondary outcomes.
Potential biases in the review process
Despite the fact that we used a broad search strategy, we may have missed published studies not listed in the resources searched for this review. We imposed no language restrictions on the search. We requested additional data from authors of results published only in abstract form, and we did not receive an answer from three authors (Reddy 2014; Vasu 2013; Viaene 2014). All trials were independently evaluated by two review authors. We assessed the risk of bias of included studies by using published data and clinical trials registers. When the study's protocol was not available, we assessed reporting bias according to the methods detailed in the study's report. This may have introduced bias if study authors removed outcomes from their methods section. We attempted to make contact with several authors of included studies, and four investigators provided further information (Aditianingsih 2017; de‐Madaira 2018; Semler 2018; Young 2015). The lack of universal definitions for concepts such as critical illness or resuscitation is a potential source of bias in the present review. As a result, we chose broad inclusion criteria for participants and decided to carry out a subgroup analysis by participant subsets. However, this was not conducted because the included studies did not state data by disease subsets. Moreover, we included inconsistent trials with regards to the time points. This is particularly relevant in studies with few participants in each arm, such as those with electrolyte disturbances. Chloride concentrations, pH, and bicarbonate level analyses showed moderate heterogeneity and may have undermined the confidence of the estimated effect (Analysis 1.7, Analysis 1.8 and Analysis 1.9, respectively). We considered a clinical cause as a possible explanation for this heterogeneity because the gradient of levels of electrolytes showed the same tendency as the chloride concentration and a strong ion difference among different buffered preparations (Sterofundin > Ringer's lactate > Plasma‐Lyte A). We used a random‐effects model as defined in our protocol (Barea‐Mendoza 2016). Moreover, we included studies with more than two intervention groups (Dung 1999; Hasman 2012; Ngo 2001). However, we feel confident about including them in meta‐analyses because data were reported for each of the arms to which participants were randomized and contributed several independent comparisons (without interventions groups in common) in fixed‐effect model analyses (Analysis 1.1; Analysis 1.4 (Higgins 2011). We also included in the review some studies in which participants in the crystalloid and buffered groups received additional buffered or crystalloid fluids, respectively (Semler 2016; Semler 2018; Young 2015). This may have introduced clinical differences or bias, and we did not explore this in the review. Some trial outcomes did not contribute to the meta‐analyses because they were reported in a different form than that defined by our protocol (Barea‐Mendoza 2016), and we found methodological challenges in converting them to the desired format. Data on total volume of intravenous fluids needed during resuscitation were not combined in our analyses owing to the fact that the outcomes distribution was skewed, and it was not possible to estimate means and standard deviations from medians and interquartile ranges. In particular, the lack of meta‐analysis of the total volume requirement hampers the general understanding of our findings because we ignored the dose of the intervention. The Review includes three ongoing studies that could be included in an updated version (NCT02721654; NCT02835157; NCT02875873).
Agreements and disagreements with other studies or reviews
Our findings are in line with those of other reviews, whilst making an additional contribution to the current knowledge base with some trials that were not included in any of these previous reviews. We identified four systematic reviews that focused on critically ill patients who received buffered solution compared to 0.9% saline fluids (Kawano‐Dourado 2018; Rochwerg 2014; Serpa 2017; Zayed 2018). Kawano‐Dourado and colleagues examined 15 randomized trials with 4067 critically ill and perioperative adult patients who received intravenous high‐ versus low‐chloride content solutions (colloids and crystalloids) for intravascular volume expansion or maintenance. The pooled effect did not differ between groups for all‐cause mortality (odds ratio (OR) 0.90, 95% confidence interval (CI) 0.69 to 1.17; P = 0.44; I² = 0%; 11 trials; 3710 participants; low‐certainty evidence), AKI (risk ratio (RR) 0.99, 95% CI 0.78 to 1.25; I² = 0%; 5 trials; 3245 participants), and allogenic blood transfusion rates (RR 0.69, 95% CI 0.44 to 1.09; I² = 0%; 4 trials; 137 participants), highlighting that trial sequential analyses (TSA) carried out for mortality and AKI estimated an optimal sample size of 9517 and 12,000 participants, respectively (Kawano‐Dourado 2018). Rochwerg and colleagues examined 14 parallel‐group randomized trials (including factorial designs) involving 18,916 adults with severe sepsis or septic shock who required fluid resuscitation. The indirect low‐quality evidence from a network meta‐analysis suggested no evidence of a significant effect of balanced fluids preventing mortality compared with saline solutions (OR 0.78, 95% CrI 0.58 to 1.05; low‐certainty evidence) (Rochwerg 2014).
The other two systematic reviews collected a subgroup of the population included in the present review (Serpa 2017Zayed 2018). Serpa and colleagues evaluated three randomized clinical trials involving 2348 adults admitted to the ICU and randomized to receive either buffered or 0.9% saline solutions. Study authors did not find any significant differences in in‐hospital mortality (OR 0.87, 95% CI 0.65 to 1.17; P = 0.36; I² = 0%) and AKI during hospital stay (OR 1.00, 95% CI 0.75 to 1.34; P = 0.97; I² = 0%) (Serpa 2017). Zayed and colleagues evaluated six randomized clinical trials involving 19,332 participants who received either balanced crystalloids or 0.9% saline fluids in the ICU setting. Study authors did not find any significant differences in in‐hospital mortality (OR 0.92, 95% CI 0.85 to 1.01; P = 0.09; I² = 0%) or incidence of AKI (OR 0.92, 95% CI 0.84 to 1.01; P = 0.1; I² = 0%) (Zayed 2018). We consider that despite differences in the types of trials included, data comparisons, and analyses, the data from our review probably support these conclusions. The European Society of Intense Care Medicine consensus focused on fluid resuscitation in neurointensive care patients (Oddo 2018). The Society recommended buffered compared to isotonic crystalloids for subarachnoid haemorrhage (SAH) (GRADE low‐quality evidence) (Oddo 2018). This recommendation was based on two small multi‐intervention trials in patients with SAH ‐ Lehmann 2013 ‐ and with severe traumatic brain injury ‐ Roquilly 2013, and investigators reported a statically significant reduction in hyperchloraemia rate in favour of buffered fluids when compared to 0.9% saline solutions (RR 0.57, 95% CI 0.37 to 0.75; P < 0.001). Although the results of our review show a small difference, this trend is consistent with this observation (MD ‐2.9, 95% CI ‐5.07 to 0.90; P < 0.005). Nevertheless, we ruled out the Lehmann 2013 and Roquilly 2013 trials from our review owing to their study design and intervention because Lehmann and colleagues randomized participants to receive 0.9% saline and hydroxyethyl starch dissolved in 0.9% saline, or balanced crystalloid and colloid solutions. Roquilly and colleagues randomized participants to isotonic balanced solutions (crystalloid and hydroxyethyl starch; balanced group) or isotonic sodium chloride solutions (crystalloid and hydroxyethyl starch; saline group). We are not aware of any other systematic reviews or good quality studies investigating the effects of buffered versus 0.9% saline solutions in non‐surgical critically ill patients.
Risk of bias summary: review authors' judgements about each risk of bias item for each included study.
Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.
Forest plot of comparison: 1 Buffered solutions vs SS 0.9% (saline solution), outcome: 1.1 Mortality.
Forest plot of comparison: 1 Buffered solutions vs SS 0.9% (saline solution), outcome: 1.3 Acute renal injury.
Comparison 1 Buffered solutions vs SS 0.9% (saline solution), Outcome 1 Mortality.
Comparison 1 Buffered solutions vs SS 0.9% (saline solution), Outcome 2 Mortality (by age).
Comparison 1 Buffered solutions vs SS 0.9% (saline solution), Outcome 3 Acute kidney injury.
Comparison 1 Buffered solutions vs SS 0.9% (saline solution), Outcome 4 Organ system dysfunction.
Comparison 1 Buffered solutions vs SS 0.9% (saline solution), Outcome 5 Sodium.
Comparison 1 Buffered solutions vs SS 0.9% (saline solution), Outcome 6 Potassium.
Comparison 1 Buffered solutions vs SS 0.9% (saline solution), Outcome 7 Chloride.
Comparison 1 Buffered solutions vs SS 0.9% (saline solution), Outcome 8 pH.
Comparison 1 Buffered solutions vs SS 0.9% (saline solution), Outcome 9 Bicarbonate.
Comparison 1 Buffered solutions vs SS 0.9% (saline solution), Outcome 10 Blood products.
| Patient or population: critically ill adults and children | |||||
| Outcomes | Anticipated absolute effects* (95% CI) | Odds ratio | № of participants | Certainty of the evidence | |
| Risk with saline 0.9% | Risk with buffered solutions | ||||
| Overall mortality (in‐hospital death) | Study populationa | OR 0.91 | 19,664 | ⊕⊕⊕⊕ | |
| 119 per 1000 | 108 per 1000 | ||||
| Acute renal Injury (defined as RIFLE or AKIN criteria by study authors) | Study population | OR 0.92 | 18,701 | ⊕⊕⊝⊝ | |
| 121 per 1000 | 111 per 1000 | ||||
| Organ system dysfunction (during admission) | Study population | OR 0.80 | 266 | ⊕⊝⊝⊝ | |
| 156 per 1000 | 137 per 1000 | ||||
| Electrolyte disturbances | |||||
| Plasma potassium, mmol/L (during admission) | Median potassium in control groups was 3.55 | MD 0.09 | ‐ | 158 | ⊕⊝⊝⊝ |
| Plasma chloride, mmol/L (during admission) | Median chloride in control groups was 109.75 | MD ‐3.02 | ‐ | 351 | ⊕⊝⊝⊝ |
| Plasma pH (during admission) | Median pH in control groups was 7.36 | MD 0.04 | ‐ | 200 | ⊕⊝⊝⊝ |
| Plasma bicarbonate, mmol/L (during admission) | Median bicarbonate in control groups was 19.75 | MD 2.26 | ‐ | 344 | ⊕⊝⊝⊝ |
| *The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). * The basis for the basal risk assumption in continuous outcomes is the median control group response across studies. | |||||
| GRADE Working Group grades of evidence. | |||||
| aFew included studies were rated as high risk of bias, and the weight of them was 1%. Relevant heterogeneity was not detected. Several studies did not report mortality but publication bias was not suspected upon the funnel plot. Certainty of the outcome was not downgraded. bWe estimated a serious limitation of risk of bias because several studies were not blinded. Certainty of the outcome was downgraded two levels because of risk of bias and publication bias. cThe possibility of publication bias is not excluded because several studies did not report this outcome. dSeveral studies included in this outcome have high or unclear risk of bias for blinding or incomplete reporting. eStudies included few participants and few events. fCertainty was downgraded by three levels because risk of bias, imprecision, and publication bias. gOnly four studies with a total of 258 participants involved children. | |||||
| Study | Outcome | Participants | Results |
| Total volume of intravenous fluids needed during resuscitation | 30 | Mean in the 0.9% saline group was 6.23 (4.3 to 8.2) litres and in the Rigerfundin group was 6.23 (4.1 to 8.4) litres | |
| Total volume of intravenous fluids needed during resuscitation | 100 | There were no significant differences in volume, duration of fluid administration, or maintenance fluid between groups | |
| Total volume of intravenous fluids needed during resuscitation | 40 | Median from 0 to 72 hours after randomization was 6904 (6400 to 8600) mL for 0.9% saline and 5900 (4930 to 7002) mL for Ringer's lactate solution; P = 0.045 | |
| Total volume of intravenous fluids needed during resuscitation | 22 | Median total fluids (including intravenous and oral rehydration solution) requirement was less in the Ringer's lactate group (310 mL/kg (IQR 230 to 365)) as compared to the 0.9% saline group (530 mL/kg (IQR 324 to 750)); P = 0.01 | |
| Total volume of intravenous fluids needed during resuscitation | 111 | Mean in the Ringer's lactate group was 134.2 (± 19.9) mL/kg and in the 0.9% saline group was 132.9 (± 16.6) mL/kg; P = 0.954 | |
| Total volume of intravenous fluids needed during resuscitation | 974 | Patients in the saline and balanced groups received a similar total volume of intravenous crystalloid by 30 days (1424 (500 to 3377) mL vs 1617 (500 to 3628) mL); P = 0.40 | |
| Total volume of intravenous fluids needed during resuscitation | 15,802 | Patients in the saline group received a mean of 2171 (± 3942) mL of saline solution and 216 (± 1394) mL of buffered solution. Patients in the buffered group received a mean of 492 (± 2303) mL of saline solution and 2083 (± 3310) mL of buffered solution | |
| Total volume of intravenous fluids needed during resuscitation | 65 | There was no significant difference in the amount of study fluid administered | |
| Total volume of intravenous fluids needed during resuscitation | 2,278 | Patients in the saline group received a mean of 2554 (± 2120) mL of saline solution and 1.8 (± 60) mL of buffered solution. Patients in the Plasma‐Lyte group received a mean of 0.5 (± 15) mL of saline solution and (2655 ± 3052) mL of Plasma‐Lyte solution | |
| Transfusion requirement | 65 | 16 (67%) patients in the saline group and 11 (50%) patients in the buffered solution group received pRBC transfusion (mean difference of 0.75, 95% CI 0.5 to 1.2) | |
| Transfusion requirement | 2,278 | Packed red blood cells: 26 (9%) patients in the saline group received a mean of 45 (± 277) mL, and 29 (9%) patients in the buffered group received a mean of 39 (± 199) mL of packed red blood cells at study day 3 | |
| Cost | 65 | The cost‐minimization analysis reported a 24‐hour cost differential of USD 12.35 in favour of Plasma‐Lyte A compared with non‐buffered solutions | |
| Data potentially of interest to this review but reported with variation in outcome measures or reported in single study, and therefore not suitable for numerical analysis. IQR: interquartile range; mL: millilitres; mL/kg: millilitres per kilogram; pRBC:packed red blood cells; vs: versus; USD: USA dollars. | |||
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 1 Mortality Show forest plot | 14 | 19664 | Odds Ratio (Fixed, 95% CI) | 0.91 [0.83, 1.01] |
| 1.1 Buffered vs saline 0.9% | 2 | 16776 | Odds Ratio (Fixed, 95% CI) | 0.92 [0.83, 1.01] |
| 1.2 Plasma‐Lyte vs saline 0.9% | 4 | 2513 | Odds Ratio (Fixed, 95% CI) | 0.90 [0.65, 1.25] |
| 1.3 Ringer lactate vs saline 0.9% | 7 | 345 | Odds Ratio (Fixed, 95% CI) | 0.32 [0.05, 2.14] |
| 1.4 Ringerfundin vs saline 0.9% | 1 | 30 | Odds Ratio (Fixed, 95% CI) | 0.62 [0.09, 4.34] |
| 2 Mortality (by age) Show forest plot | 14 | 19797 | Odds Ratio (Fixed, 95% CI) | 0.91 [0.83, 1.00] |
| 2.1 Adults | 12 | 19539 | Odds Ratio (Fixed, 95% CI) | 0.91 [0.83, 1.00] |
| 2.2 Children | 4 | 258 | Odds Ratio (Fixed, 95% CI) | 0.97 [0.10, 9.80] |
| 3 Acute kidney injury Show forest plot | 9 | 18701 | Odds Ratio (Fixed, 95% CI) | 0.92 [0.84, 1.00] |
| 3.1 Plasmalyte vs saline 0.9% | 3 | 2413 | Odds Ratio (Fixed, 95% CI) | 1.01 [0.75, 1.36] |
| 3.2 Ringer lactate vs saline 0.9% | 2 | 87 | Odds Ratio (Fixed, 95% CI) | 0.45 [0.06, 3.21] |
| 3.3 Buffered vs saline 0.9% | 2 | 15990 | Odds Ratio (Fixed, 95% CI) | 0.91 [0.82, 1.00] |
| 3.4 Ringerfundin vs saline 0.9% | 1 | 30 | Odds Ratio (Fixed, 95% CI) | 0.62 [0.09, 4.34] |
| 3.5 Sterofundin vs saline 0.9% | 1 | 181 | Odds Ratio (Fixed, 95% CI) | 1.00 [0.49, 2.06] |
| 4 Organ system dysfunction Show forest plot | 5 | 266 | Odds Ratio (M‐H, Fixed, 95% CI) | 0.80 [0.40, 1.61] |
| 5 Sodium Show forest plot | 4 | 222 | Mean Difference (IV, Random, 95% CI) | ‐0.48 [‐1.67, 0.70] |
| 5.1 Plasma‐Lyte vs saline 0.9% | 2 | 91 | Mean Difference (IV, Random, 95% CI) | ‐0.67 [‐2.90, 1.55] |
| 5.2 Ringer's lactate vs saline 0.9% | 2 | 67 | Mean Difference (IV, Random, 95% CI) | 0.43 [‐3.53, 4.38] |
| 5.3 Sterofundin vs saline 0.9% | 1 | 64 | Mean Difference (IV, Random, 95% CI) | ‐0.66 [‐4.95, 3.63] |
| 6 Potassium Show forest plot | 3 | 158 | Mean Difference (IV, Random, 95% CI) | 0.09 [‐0.10, 0.27] |
| 6.1 Plasma‐Lyte vs saline 0.9% | 2 | 91 | Mean Difference (IV, Random, 95% CI) | 0.10 [‐0.13, 0.33] |
| 6.2 Ringer's lactate vs saline 0.9% | 2 | 67 | Mean Difference (IV, Random, 95% CI) | 0.07 [‐0.24, 0.38] |
| 7 Chloride Show forest plot | 6 | 351 | Mean Difference (IV, Random, 95% CI) | ‐3.02 [‐5.24, ‐0.80] |
| 7.1 Plasma‐Lyte vs saline 0.9% | 4 | 220 | Mean Difference (IV, Random, 95% CI) | ‐4.56 [‐6.68, ‐2.44] |
| 7.2 Ringer's lactate vs saline 0.9% | 2 | 67 | Mean Difference (IV, Random, 95% CI) | ‐0.56 [‐2.58, 1.47] |
| 7.3 Sterofundin vs saline 0.9% | 1 | 64 | Mean Difference (IV, Random, 95% CI) | ‐1.55 [‐4.39, 1.29] |
| 8 pH Show forest plot | 3 | 200 | Mean Difference (IV, Random, 95% CI) | 0.04 [0.02, 0.06] |
| 8.1 Plasma‐Lyte vs saline 0.9% | 2 | 91 | Mean Difference (IV, Random, 95% CI) | 0.06 [0.03, 0.08] |
| 8.2 Ringer's lactate vs saline 0.9% | 1 | 45 | Mean Difference (IV, Random, 95% CI) | 0.04 [0.01, 0.07] |
| 8.3 Sterofundin vs saline 0.9% | 1 | 64 | Mean Difference (IV, Random, 95% CI) | 0.02 [0.00, 0.04] |
| 9 Bicarbonate Show forest plot | 6 | 344 | Mean Difference (IV, Random, 95% CI) | 2.26 [1.25, 3.27] |
| 9.1 Plasma‐Lyte vs saline 0.9% | 4 | 213 | Mean Difference (IV, Random, 95% CI) | 2.58 [1.33, 3.84] |
| 9.2 Ringer's lactate vs saline 0.9% | 2 | 67 | Mean Difference (IV, Random, 95% CI) | 2.42 [‐0.71, 5.55] |
| 9.3 Sterofundin vs saline 0.9% | 1 | 64 | Mean Difference (IV, Random, 95% CI) | 1.15 [0.21, 2.09] |
| 10 Blood products Show forest plot | 2 | 1453 | Mean Difference (IV, Fixed, 95% CI) | ‐17.53 [‐44.52, 9.46] |
