LABORATORIO DE URGENCIAS

Balanced crystalloids versus saline for critically ill patients (BEST-Living): a systematic review and individual patient data meta-analysis

Fernando G Zampieri, Alexandre B Cavalcanti, Gian Luca Di Tanna, Lucas P Damiani, Naomi E Hammond, Flavia R Machado, Sharon Micallef, John Myburgh, Mahesh Ramanan, Balasubramanian Venkatesh, Todd W Rice, Matthew W Semler, Paul J Young.

Lancet Respir Med 2024; 12: 237–46 Published Online November, 2023 https://doi.org/10.1016/ S2213-2600(23)00417-4

Introduction
During an admission to the intensive care unit (ICU), most patients receive intravenous fluid therapy as supportive therapy for the presenting critical illness.1 Crystalloid solutions are the fluids used most commonly to correct symptomatic hypovolaemia due to fluid losses, improve haemodynamic function to optimise vital organ function and as an intravenous vehicle to administer medications.2 Widely used intravenous crystalloid solutions include 0·9% sodium chloride (saline) and solutions characterised by an electrolyte profile similar to extracellular fluid. These buffered, or balanced, crystalloids are constituted with use of alternative anions to reduce the chloride concentration in the fluid, and include compound sodium lactate (Ringer’s lactate, Hartmann’s solution) or acetate-containing solutions (Plasma-Lyte 148, Ringer’s acetate). Potential benefits of balanced crystalloid solutions include the reduction of development of an iatrogenic hyperchloraemic metabolic acidosis.2 Randomised clinical trials have compared the effects of balanced solutions with those of saline on patient-centred outcomes in critically ill patients.3−8 One study reported that the use of balanced solutions was associated with a reduction in a composite outcome measure of death, new renal replacement therapy, or persistent kidney dysfunction.4 No trial reported a statistically significant difference in mortality. A recent trial-level meta-analysis of randomised clinical trials reported a 0·895 probability that the use of balanced solutions was associated with lower mortality than saline.9 To address the remaining uncertainty over the use of balanced solutions in the ICU, we did an individual patient-data meta-analysis of randomised clinical trials analysed within a Bayesian framework to provide estimates of the probability of benefits associated with use of balanced solutions compared with saline in a heterogeneous population of adult patients and in prespecified subgroups. Bayesian methods were used as they allow a more nuanced interpretation of the results, compared with traditional frequentist null hypothesis testing that relies on a dichotomised interpretation of a p value to accept or reject a hypothesis.

Methods - Overview
The BEST-Living Study is a living, individual patient-data meta-analysis with a prespecified protocol and statistical analysis plan designed to compare the effects of using balanced solutions with those of saline in the ICU on patient-centred outcomes in critically ill adult patients.10,11 Given the results of a recent aggregated meta-analysis, which were close to traditional statistical significance, we chose to use a Bayesian framework to provide a more comprehensive analysis of the effect of balanced solutions. Bayesian analyses provide an assessment of the probability of benefit for situations in which benefit is considered probable, but the results are not statistically significant on the basis of traditional frequentist null hypothesis testing.12 They also provide a more nuanced approach to the effect sizes that are compatible with data (and prior information, if available). Reviews of Bayesian methods in critical care can be found elsewhere.13,14 In relation to the use of individual patient data in meta[1]analyses, the Cochrane handbook states, “In most cases participants will not have specifically consented to inclusion in the meta-analysis. However, as the meta[1]analysis is posing the same research question, and is essentially updating the trial they did consent to, the usual view is that separate consent is not required. However, it is advisable that data received are anonymised.” In keeping with this principle, our meta[1]analysis addressed the same question as the trials for which we have data, and all data were anonymised before transfer. This systematic review is reported in accordance with PRISMA-IPD Guidelines
Data collection and data items
The review was sponsored by the HCor Research Institute (São Paulo, Brazil) with formalised collaborative research agreements between the sponsor and respective author investigators and their associated institutions. A data skeleton (appendix p 4) was sent to corresponding authors for populating respective trial information, which was housed on a secure server at the HCor Research Institute. Complete cases were included in the analysis, with no imputation for missing data. Data quality was assured by replicating the primary analysis of each trial, the results of which were compared with the published results of each trial; trialists were informed about any discrepancies and allowed to respond or update their uploaded results.

Risk of bias assessment and certainty of evidence
Risk of bias in the included studies and certainty of the review evidence was assessed by independent experts using the Cochrane Risk of bias and tool and the GRADE (Grading of Recommendations Assessment, Development and Evaluation) approach for assessing certainty.17,18 Outcomes The primary outcome was all-cause in-hospital mortality censored at 90 days. Secondary outcomes were survival at last follow-up, treatment with renal replacement therapy commenced during the index ICU admission, and days alive and out of hospital and out of ICU within 28 days. Prespecified subgroup analyses were done for the primary outcome and for treatment with renal replacement therapy in subgroups of patients identified by baseline characteristics of sex (male or female), presence or absence of sepsis (as defined by each trial), presence or absence of traumatic brain injury, baseline serum chloride concentration (categorised as low [<100 mmol/L], normal [100–110 mmol/L], or high [>110 mmol/L]), acid-base status as defined by baseline pH (severe acidaemia [pH <7·20], mild acidaemia [pH 7·20 to ><7·35], normal [pH 7·35 to 7·45], or alkalaemia [pH >7·45]), volumes of intravenous saline received before randomisation (none, 1–999 mL, or ≥1000 mL of saline received), and study design (cluster versus individual randomisation). Hypotheses for each prespecified subgroup are presented in the appendix (p 7) and in the statistical analysis plan.11 One planned subgroup analysis estimating effects in patients with non-traumatic acute brain injury was planned but could not be done due to an absence of prospective classification of patients in the included trials (appendix p 8). A post-hoc analysis based on serum sodium con[1]centration at baseline (<135, 135–145, or >145 mmol/L) was also added during the review process.

Statistical analysis
The statistical analysis plan was finalised before data were merged or analysed.11 The primary analysis was based on a one-step meta-analysis using a Bayesian hierarchical model with the intervention of interest (balanced solution vs saline) as a fixed-effect covariate and two layers: the ICU (or cluster) nested within trial as hierarchical effects. The primary model used for this and future iterations is a non-informative prior for the effect size of the intervention (prior for the log odds ratio [OR] of the intervention defined as normal with mean 0 and SD 0·355).19 Alternative analyses for the primary endpoint included an adjusted model (adjusted for age, sex, surgical vs non-surgical admission, and sepsis), different prior analyses (including optimistic and pessimistic priors), and a frequentist mixed model.
Survival analyses were done with time-to-event models following a Bayesian semiparametric survival time with a non-informative prior for the intervention mean of 0 and SD of 4. The proportion of patients treated with new renal replacement therapy initiated during the index ICU admission was assessed using the same model as the primary outcome. Length of hospital stay was analysed with a cumulative logistic model for days alive and free of hospital or ICU admission (both truncated at 28 days, with patients who died given a score of –1 using non-informative priors).
Exploratory analyses were assessed with frequentist generalised mixed models with interaction between time and intervention and with patients as a random intercept. We did a two-stage individual patient data meta[1]analysis for the primary outcome: effect sizes obtained from each trial were aggregated using a Bayesian meta[1]analysis. In addition, we did a Bayesian meta-analysis of aggregated trial-level data for hospital mortality and use of renal replacement therapy.
Results for the primary outcome are presented as median OR (95% credible intervals [CrI]) and the probability of direction, defined as the posterior probability of the intervention (use of balanced solutions) being associated with an OR less than 1·0. We also present relative risks and absolute risk differences obtained from posterior expected probabilities.
As described in the statistical analysis plan, a region of practical equivalence (ROPE) for the effect of the intervention on the primary outcome as an OR was defined as 0·955–1·046,11 shown as the percentage of the posterior probability distribution contained in the ROPE.
Results for the survival outcome are presented as hazard ratio (HR) and corresponding 95% CrI. Results for secondary binary outcomes are presented similarly to the primary endpoint, with exception of ROPE, which was not defined for all secondary endpoints. Results for subgroup analyses are presented as forest plots for the OR, probability of benefit, and probability of direction of treduction are also shown. A list of deviations from original plans to performed analysis is summarised in appendix (p 8). All analyses were done with R version 4.2.2 with packages brms and bayesmeta.20,21 This study was registered with PROSPERO (CRD42022299282). Role of the funding source The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report.

Results
The initial search was concluded on March 1, 2022. Of the 5219 records screened, 81 were assessed for eligibility, and six met the eligibility criteria and were selected for analysis (figure 1).3–8 All investigators agreed to share data. Characteristics of the included studies, including the type of balanced fluid used and volumes received after randomisation, are shown in table 1. Of the six included studies, four were cluster-randomised trials and two were individually randomised. Two studies were conducted in the USA, one in Australia, one in New Zealand, one in Australia and New Zealand combined, and one in Brazil. In all trials, patients received balanced solutions or saline for resuscitation, and all received compatible intravenous crystalloid therapy during their ICU stay. The search was repeated on Sept 1, 2023. We identified 691 new records; 37 were duplicates, 634 were ineligible on initial review, the remaining 20 were retrieved and underwent full-text review to assess for eligibility. None met our study’s eligibility criteria. A total of 34685 patients were included in the meta[1]analysis: 17407 assigned to receive balanced crystalloids and 17278 to receive saline. The mean age was 58·8 yearshe interaction for each subgroup; other measurements of effect size including relative risk and absolute risk (SD 17·5). Among patients with available data, 20074 (57·9%) of 34653 patients were male, 14579 (42·1%) were female, 20919 (60·6%) of 34520 were non-surgical admissions, and 6585 (19·1%) of 34427 had an admission diagnosis of sepsis (table 2; appendix pp 10–11). For analyses adjusted for age, sex, and admission type (surgical vs non-surgical) and sepsis, complete case data were available for 17293 assigned to receive balanced crystalloids and 17153 assigned to receive saline. Consent to report in-hospital mortality was obtained for 17 313 participants assigned balanced crystalloids and 17 166 assigned saline. Among these patients, 2907 (16·8%) assigned balanced crystalloids and 2975 (17·3%) assigned saline died in hospital (OR 0·962 [95% CrI 0·909 to 1·019], absolute difference –0·4 percentage points [–1·5 to 0·2]). The posterior probability that the use of balanced crystalloid solutions decreased the risk of in-hospital death compared with saline use was 0·895; 58·8% of the probability mass was concentrated in the defined ROPE (table 3; figure 2A). Using alternative priors, the posterior probability of benefit for balanced solutions could be as low as 0·821 (pessimistic prior) and as high as 0·932 (optimistic prior; table 3). From the frequentist model, the OR for in[1]hospital death with balanced crystalloids compared with saline was 0·962 (95% CI 0·908 to 1·020; absolute difference –0·4 percentage points [95% CI –1·1 to 0·2]; p=0·192; table 3). The two-stage, individual patient data meta-analysis yielded an OR for in-hospital death for balanced solutions versus saline of 0·956 (95% CrI 0·878 to 1·040; probability of benefit of 0·870; appendix p 16). The HR for survival to longest available follow-up for patients assigned balanced crystalloid solutions versus saline was 0·964 (95% CrI 0·918 to 1·010; table 3, figure 2B). New use of renal replacement therapy occurred in 934 (5·6%) of 16628 participants assigned balanced solutions and 993 (5·9%) of 16 803 assigned saline (OR 0·931 [95% CrI 0·849 to 1·020], absolute risk difference –0·4 percentage points)
In the trial-level meta-analysis using non-informative priors for heterogeneity and a neutral prior for the effect of the intervention, the OR for mortality with use of balanced solutions versus saline was 0·94 (95% CrI 0·86–1·03; probability of benefit of 0·92). The corresponding OR for renal replacement therapy was 0 ·94 (0 ·82 to 1 ·09; probability of benefit of 0 ·82; appendix pp 17–18). Subgroup results for in-hospital mortality are shown in figure 3 and in the appendix (p 12), and those for use of renal replacement therapy are shown in the appendix (pp 13, 15). Denominators are reported for each subgroup as the baseline data needed to classify participants into subgroups was not available for all participants. The probability that balanced solutions were associated with lower mortality exceeded 0·90 in the following subgroups: patients without traumatic brain injury (0·975), patients who did not receive 0·9% saline before enrolment (0·988), and female patients (0·950). In patients with traumatic brain injury, 191 (19·1%) of 999 assigned balanced solutions died and 141 (14·7%) of 962 assigned saline died (OR 1·424 [95% CrI 1·100–1·818], absolute difference 3·2 percentage points [0·7–8·7]). the probability that balanced solutions increased mortality in patients with traumatic brain injury was 0·975. Comparing patients with and without traumatic brain injury, the interaction probability of direction was greater than 0·99; for all other comparisons between subgroups, including those based on baseline chloride concentration and pH, the interaction probability of direction was less than 0·80. The probability that balanced solutions were associated with reduced risk of new treatment with renal replacement therapy exceeded 0·90 in the following subgroups: patients with sepsis (97·5), patients without traumatic brain injury (93·3), patients with baseline blood pH of 7·20–7·35 (91·0), patients who did not receive saline before enrolment (90·4), female patients (92·0), patients enrolled in cluster-randomised trials (90·7), and patients with baseline serum sodium concentration less than 135 mmol/L (98·5). The independent risk of bias assessment for included trials is shown in the appendix (p 9). Two trials were deemed to be at low risk of bias6,7 and four at high risk of bias.3–5,8 The evidence relating to in-hospital mortality and use of renal replacement therapy was assessed to be of moderate certainty due to risk of bias (for in-hospital mortality) or inconsistency (for use of renal replacement therapy; appendix p 14).

Discussion
In this individual patient data meta-analysis incorporating data from randomised clinical trials, the use of balanced solutions for intravenous fluid therapy in adults in the ICU was associated with a high probability of reduced in-hospital mortality overall. The main Bayesian analysis suggests the OR for in-hospital death for balanced solutions compared with saline is 0·962 (ie, approximately a 4% reduction in the odds of dying) with a 95% CrI of 0·909 to 1·019 (ie, around a 9% reduction or a 2% increase in the odds of dying, respectively). These values are consistent with those from the frequentist analysis (OR 0·962 [95% CI 0·908–1·020]). The point estimates and the CrIs and CIs are consistent with the posterior probability of 89·5% that the use of balanced solutions reduces mortality by some amount, the converse of which is a 10·5% probability that it does not reduce mortality. While the results of the Bayesian and frequentist analyses are consistent, the inferences that can be drawn from them might differ. A purely frequentist view would be that there was no difference between balanced solutions and saline, as the upper limit of the 95% CI is 1·020. By contrast, the Bayesian analysis implies a high probability that balanced solutions reduce mortality. We defined a margin of practical equivalence, which can be viewed as the range of ORs indicating that a treatment should not be selected based on assessment of efficacy alone, as an OR of 0·955–1·046. Thus, although our analysis suggests a high probability that balanced solutions reduce mortality by a small amount, other factors such as acquisition cost might influence whether clinicians choose to use them. When acquisition costs are similar, clinicians might consider a 4% or less reduction in risk of death sufficient evidence to choose balanced solutions. However, the estimate for the absolute reduction in mortality was 0·4 percentage points, meaning that 250 patients would have to be treated with balanced solutions to prevent one death. In situations where balanced solutions are more expensive and health-care resources are limited, this possible benefit might not be considered compelling. The use of balanced solutions was associated with increased in-hospital mortality in patients with traumatic brain injury and decreased in-hospital risk of death in patients without traumatic brain injury. These findings support the use of balanced solutions in patients without traumatic brain injury. There was also a high probability that balanced solutions reduced treatment with renal replacement therapy.
The direction of treatment effect replicates the findings of a recent trial-level systematic review and meta[1]analysis,9 but the ability of our analysis to separate patients with and without traumatic brain injury provides stronger evidence for both of those populations. In addition, a probabilistic interpretation of the effect of balanced solutions on treatment with renal replacement therapy is also provided.
The use of individual patient[1]level data from randomised clinical trials allowed for estimation of treatment effects in a range of subgroups of potential interest to clinicians. Apart from patients with traumatic brain injury, subgroups in which saline use was associated with improved outcomes were not identified. In particular, neither blood chloride concentration nor blood pH at enrolment could be used to identify patients in whom saline use was associated with reduced mortality. These data do not support the hypothesis that blood pH or chloride concentration are the mechanism by which balanced solutions and saline are associated with differences in outcomes. We found no heterogeneity of treatment effect on the basis of baseline serum sodium concentration. Balanced solutions and saline are widely used in clinical practice. This study provides clinicians with information that could inform choices about which fluid to use for patients treated in an ICU.22 The combined trials provide evidence that balanced solutions should be favoured in critically ill patients without traumatic brain injury. This individual patient data meta-analysis has limitations:
* First, in an independent assessment, some trials included in the analysis were judged to have a high risk of bias, predominantly due to cluster randomisation, resulting in concerns over allocation concealment.
* Second, primary outcome data were not available for some participants because, in critical care trials of urgent interventions, institutional review boards and ethics committees might allow participant enrolment without previous informed consent; however, participants or their legal surrogates can subsequently withhold consent to the use of their data, leading to loss of outcome data. This loss resulted in 15 patients from the BaSICS trial6 and 191 patients from the PLUS trial7 being excluded from the primary analysis, and four patients from the SCOPE-DKA trial8 being excluded from the survival analysis due to missing follow-up data.
* Third, in the cluster-randomised trials, patients might have received both balanced solution and saline if present in the ICU when the units changed from using one fluid to the other, and this carryover effect might have affected the results. SMART5 was the largest cluster-randomised trial (representing 15 802 of the 19 147 cluster-randomised participants included in this meta-analysis) and had multiple crossovers, and therefore the highest risk of carryover effects. As a secondary analysis to adjust for carryover effects did not affect the SMART study results, we did not repeat that analysis.
* Fourth, we planned to evaluate the subgroup of patients with non-traumatic acute brain injuries; however, extracting data on this subgroup of patients was not possible due to lack of prospective subclassification of patients in the included trials.Thus, the relative risks and benefits of balanced solutions versus saline in patients with non-traumatic acute brain injury remains unclear, and further research in such patients is warranted.
* Fifth, this study included data from randomised clinical trials that compared a range of balanced solutions with saline, but did not allow for the effects of different balanced solutions to be compared.
Our report is the first iteration of a living review in which the search will be repeated each year and the analysis updated when new eligible trial data become available. In conclusion, overall, there is a high probability that use of balanced solutions compared with saline in the ICU is associated with reduced in-hospital mortality and reduced treatment with renal replacement therapy, with the evidence being of moderate certainty. However, in patients with traumatic brain injury, balanced solutions probably increase mortality.

NOTE: tables, graphs and references in the journal cited at the beginning.