Balanced crystalloids versus saline and mortality in hospitalized patients: a hierarchical Bayesian meta-analysis

Bottom line

In this Bayesian meta-analysis of 6 cluster-randomized trials (42,223 patients), balanced crystalloids carried an 83% posterior probability of reducing mortality versus saline but no statistically "clear" benefit, and essentially no AKI-benefit signal (40% posterior probability).

PICO at a glance

• Population: Hospitalized adults across perioperative, emergency, and critical care settings enrolled in 6 cluster-randomized or cluster-allocated fluid-policy trials (42,223 patients pooled).
• Intervention or exposure: A unit- or hospital-level balanced-crystalloid fluid policy.
• Comparator: A unit- or hospital-level 0.9% saline fluid policy.
• Outcomes: Primary: all-cause mortality; secondary: acute kidney injury (AKI).
• Study design: Hierarchical Bayesian meta-analysis of cluster-randomized or cluster-allocated trials (MEDLINE, Embase, and CENTRAL search through September 2025; prospectively PROSPERO-registered, CRD420251146439).

Clinical question and design

Fluid resuscitation is among the most common interventions ordered in hospitalized patients, and the choice between balanced crystalloids (e.g., lactated Ringer's, Plasma-Lyte) and 0.9% saline has been debated for over a decade, driven by saline's association with hyperchloremic metabolic acidosis and theoretical renal harm. Individually-randomized trials have produced mixed, modest signals on mortality and acute kidney injury (AKI). This meta-analysis takes a different angle: rather than pooling individual-patient-randomized data, it focuses on cluster-randomized and cluster-allocated trials - where whole units, hospitals, or time periods are randomized to a fluid policy - mirroring how fluid choice is actually rolled out in routine practice.

This is a policy-level comparison across six trials, not one fixed fluid prescription. Protocols varied: examples include Plasma-Lyte 148, Hartmann’s solution or Plasma-Lyte 148 with 20% albumin, and lactated Ringer’s or Plasma-Lyte A. Across 41,163 mortality participants, the Bayesian RR was 0.92 (95% credible interval 0.77-1.18), compatible with benefit and harm; the frequentist estimate was 0.91 (95% CI 0.80-1.02; I² 0.01%). For AKI (42,223 participants), the RR was 1.02 (95% credible interval 0.86-1.23; frequentist I² 63.7%). GRADE certainty was moderate for mortality and low for AKI because of imprecision and heterogeneity. These data inform a fluid-selection policy, not a universal formulation, volume, or duration.

Intervention and outcomes measured

The intervention was a unit- or hospital-level balanced-crystalloid fluid policy compared against a unit- or hospital-level 0.9% saline policy. The primary outcome was all-cause mortality; acute kidney injury (AKI) was the secondary outcome.

Results

Across the 6 trials (42,223 patients), the pooled Bayesian relative effect for mortality was 0.92 (95% CrI, 0.77-1.18) - an 83% posterior probability of benefit with balanced crystalloids. Both 95% credible intervals in this analysis cross 1, so neither outcome clears a strict frequentist-significance threshold; the value of the Bayesian framing is in the posterior probability itself, which is a favorable directional signal even though it does not amount to a "clear" or decisive benefit by the authors' own characterization.

Additional findings

For AKI, the pooled Bayesian relative effect was 1.02 (95% CrI, 0.86-1.23) - only a 40% posterior probability of benefit, essentially a coin-flip with no meaningful directional effect on renal outcomes at the routine-implementation level. Sensitivity analyses yielded findings similar to the primary analysis, per the abstract, though specific sensitivity-analysis point estimates are not itemized there.

Uncertainty and limitations

Only six trials were pooled, and settings, formulations, and implementation protocols varied. The full review reports moderate certainty for mortality and low certainty for acute kidney injury. Posterior probabilities depend on the data, priors, and model; credible intervals that include benefit and harm do not establish equivalence. A pooled fluid-policy estimate does not specify one product, volume, or duration for every patient.

Literature review and evidence synthesis

This cluster-RCT-focused Bayesian analysis sits within a dense literature comparing balanced crystalloids to saline using individually-randomized designs: a 2022 NEJM Evidence systematic review and meta-analysis (PMID 38319180) and a 2024 Lancet Respiratory Medicine individual patient data meta-analysis, the BEST-Living collaboration (PMID 38043564), both pooled individually-randomized trials - the evidence stream this cluster-level analysis complements, not replaces.

A 2019 Critical Care physiologic-lens review (PMID 31752973), a 2019 American Journal of Emergency Medicine review with trial sequential analysis (PMID 30852043), and a 2021 Medicine PRISMA meta-analysis (PMID 34559108) round out that individually-randomized evidence base.

How this fits with the broader evidence

Together these sources position the current analysis as answering a distinct question - real-world policy effectiveness rather than idealized patient-level efficacy. Pharmacists on P&T committees, order-set teams, or ICU/ED rounds are routinely asked whether a formulary default should favor balanced crystalloids over saline. This analysis supports a calibrated answer: the signal favors balanced crystalloids for mortality (83% posterior probability of benefit), but the effect is modest and not "clear," with essentially no AKI-benefit signal. Because the abstract does not specify which balanced-crystalloid product(s) were used, this pooled result should not be extrapolated to endorse one specific solution - it is a policy-level signal, not a product-specific one.

ACPE UAN: 0683-0000-26-036-H01-P

---

Study source: PMID:42361768