Balanced fluid versus saline in children with septic shock: a discourse graph

How to read these graphs

A discourse graph breaks a piece of research into typed one-sentence nodes and draws how they bear on one another. The grammar is that of Discourse Graphs (Chan and Akamatsu). Read each graph left to right: a question, the experiments that addressed it, the results they produced, and the hypothesis every result is drawn against.

QUE Question: What the work set out to answer.
HYP Hypothesis: The answer the authors expected. Every result earns one edge against it.
EXP Experiment: One attempt at the question: a trial arm, an assay, a survey, an analysis.
RES Result: One finding from an experiment, with where in the source it comes from.
ISS Issue: An open question or untried branch the work leaves for someone else.

addresses: An experiment addresses the question.
produces: An experiment produces a result.
supports: A positive result that backs the hypothesis.
opposes: A negative result that counts as evidence against the hypothesis, or a refutation of an earlier claim.
informs: A partial or mixed result that bears on the hypothesis without settling it.
consistent with: A negative or inconclusive result that does not count either way: no positive control passed, or there is no sensitivity or power statement, so the absence of an effect cannot be told from a measurement that failed.
relation: A thin dotted line: a link the source itself states between two experiments, such as one being derived from another.

A dashed box marks a result the rule above counts as uninformative, or an issue. In the drawing, click a node to jump to its text below.

Balanced fluid versus 0.9% saline in children with septic shock (NEJM, 2026)

Null trial

Read left to right. The question (QUE) is addressed by 4 experiments (EXP), each producing results (RES), and each result bears on the hypothesis (HYP). 0 support it, 1 oppose it, 8 inform it. Issues (ISS) hang below: what the paper leaves open.

HYP, in the authors’ words (Introduction): “We hypothesized that resuscitation and maintenance hydration with balanced fluid would decrease the rate of MAKE30 compared with 0.9% saline.”

supports positive finding
opposes a negative with a control and a stated sensitivity, or a failed replication
informs bears on the question without deciding the hypothesis
thin grey a relation between two experiments

Nodes, in reading order. Sources: Balamuth et al. 2026, 10.1056/NEJMoa2601969; CC BY author manuscript PMC13134814, read in full on 20 October 2026.
Node	Text	Type	Outcome	Edge
EXP	PRoMPT BOLUS: pragmatic open-label randomised trial in 47 emergency departments in five countries; 9,041 children enrolled, 4,235 balanced fluid and 4,247 saline analysed for MAKE30 and secondary effectiveness outcomes Balanced fluid did not reduce MAKE30, its components, mortality or hospital-free days compared with 0.9% saline.	`clinical-trial`	`negative-no-effect`
RES	MAKE30 occurred in 137 of 4,073 (3.4%) with balanced fluid and 124 of 4,068 (3.0%) with saline: absolute difference 0.4 percentage points (95% CI -0.5 to 1.3), risk ratio 1.10 (0.88 to 1.40), P=0.85. MAKE30 = 137/4,073 (3.4%) vs 124/4,068 (3.0%); absolute difference 0.4 percentage points, 95% CI -0.5 to 1.3; risk ratio 1.10, 95% CI 0.88 to 1.40; P=0.85. Power: 8,800 participants gave 95% power to detect a reduction from 6.0% to 4.3% at alpha 0.05; significance threshold at final analysis P<0.044. Results, Primary Outcome; Table 2 · “absolute difference, 0.4 percentage points; 95% CI, -0.5 to 1.3; risk ratio, 1.10; 95% CI, 0.88 to 1.40, P=0.85”		`negative-no-effect` informative	opposes HYP
RES	No component of MAKE30 differed: death within 30 days 41/4,214 (0.97%) vs 39/4,226 (0.92%); new RRT 26/4,214 (0.62%) vs 31/4,226 (0.73%); persistent kidney dysfunction 93/4,085 (2.3%) vs 78/4,079 (1.9%). Death within 30 days RR 1.10 (0.70 to 1.60); new renal replacement therapy RR 0.84 (0.50 to 1.40); persistent kidney dysfunction at discharge RR 1.10 (0.88 to 1.50); balanced fluid vs saline Table 2, Components of the primary outcome · “Components of the primary outcome were also not different between groups.”		`negative-no-effect` informative	informs HYP
RES	Hospital-free days out of 28 were 23 (IQR 19 to 25) in both groups; death before discharge 48/4,216 (1.1%) vs 47/4,230 (1.1%); death within 90 days 90/3,857 (2.3%) vs 83/3,867 (2.1%), RR 1.10 (0.80 to 1.40). Hospital-free days median 23 [19, 25] vs 23 [19, 25]; hospital length of stay 5.0 [3.0, 9.0] days in both; death before discharge RR 1.0 (0.69 to 1.50); death within 90 days RR 1.10 (0.80 to 1.40) Results, Secondary Outcomes; Table 2 · “Mortality was not different between groups at hospital discharge or within 90 days”		`negative-no-effect` informative	informs HYP
EXP	Fluid administration check in the same 8,482 analysed children: volumes of total, bolus and maintenance crystalloid by group and the share of each group receiving at least 75% of crystalloid as the assigned fluid Total volumes were similar between groups and the assigned fluid dominated in each arm, so treatment separation was adequate.	`clinical-trial`	`positive`
RES	Median total crystalloid was 85 mL/kg (IQR 55 to 119) with balanced fluid and 88 mL/kg (57 to 123) with saline; 80% and 88% respectively received at least 75% of crystalloid as the assigned fluid. Total crystalloid median 85 mL/kg (IQR 55-119) vs 88 mL/kg (IQR 57-123), difference in medians -2.3 mL/kg (95% CI -4.3 to -0.3); 0.9% saline received 20 mL/kg (IQR 5.5-32) in balanced group vs 79 mL/kg (IQR 49-113) in saline group; at least 75% assigned fluid in 80% vs 88% Results, Fluid Administration; Figure 1; Table S10 · “80% of patients in the balanced fluid and 88% in the 0.9% saline groups received at least 75% of total crystalloid as their randomized fluid type”		`positive` informative	informs HYP
EXP	Safety outcomes in the same trial: electrolyte abnormalities within three days of enrollment among children with follow-up laboratory values (about 2,750 to 2,860 per group for sodium and chloride), plus thrombosis and cerebral edema in 4,216 vs 4,230 Balanced fluid produced less hyperchloremia and hypernatremia but more hyperlactatemia, with no difference in thrombosis, cerebral edema or other adverse events.	`clinical-trial`	`partial`
RES	Chloride above 110 mEq/L occurred in 866 of 2,751 (31%) with balanced fluid and 1,378 of 2,813 (49%) with saline: risk difference -0.18 (95% CI -0.20 to -0.15), RR 0.64 (0.60 to 0.69), P<0.001. Chloride >110 mEq/L = 866/2,751 (31%) vs 1,378/2,813 (49%); risk difference -0.18 (-0.20 to -0.15); RR 0.64 (0.60 to 0.69); P<0.001; remained significant across sensitivity analyses (Table S14) Table 2, Safety outcomes; Results, Secondary Outcomes · “balanced fluid resulted in less hyperchloremia and hypernatremia, but these biochemical effects did not translate to improved patient-centered outcomes”		`positive` informative	informs HYP
RES	Sodium above 155 mEq/L occurred in 52 of 2,805 (1.9%) with balanced fluid and 89 of 2,862 (3.1%) with saline: risk difference -0.013 (95% CI -0.021 to -0.005), RR 0.60 (0.43 to 0.84), P=0.003. Sodium >155 mEq/L = 52/2,805 (1.9%) vs 89/2,862 (3.1%); risk difference -0.013 (-0.021 to -0.005); RR 0.60 (0.43 to 0.84); P=0.003 Table 2, Safety outcomes · “hyperchloremia and hypernatremia occurred less often, and hyperlactatemia more often, with balanced fluid”		`positive` informative	informs HYP
RES	Lactate above 4 mmol/L occurred in 259 of 1,296 (20%) with balanced fluid and 228 of 1,348 (17%) with saline: risk difference 0.03 (95% CI 0.001 to 0.06), RR 1.20 (1.00 to 1.40), P=0.04. Lactate >4 mmol/L = 259/1,296 (20%) vs 228/1,348 (17%); risk difference 0.03 (0.001 to 0.06); RR 1.20 (1.00 to 1.40); P=0.04; unadjusted P-value, lactate measured in a subset only Table 2, Safety outcomes · “Hyperlactatemia was also more frequent with balanced fluids”		`partial` informative	informs HYP
RES	Thrombosis occurred in 55 of 4,216 (1.3%) vs 55 of 4,230 (1.3%), RR 1.00 (0.69 to 1.50), and cerebral edema in 18 of 4,216 (0.43%) vs 17 of 4,230 (0.40%), RR 1.10 (0.55 to 2.10). Thrombosis RR 1.00 (0.69 to 1.50), P=0.91; cerebral edema RR 1.10 (0.55 to 2.10), P=0.57; other adverse events similar (Table S15) Table 2, Safety outcomes; Results, Secondary Outcomes · “Thrombosis and cerebral edema did not differ between groups (Table 2), nor did other adverse events”		`negative-no-effect` informative	informs HYP
EXP	Pre-specified subgroup analyses (age, sex, cancer, AKI at presentation, fluid volume, country), post-hoc analyses (measured vs imputed creatinine, acidosis, lactate, total volume) and sensitivity analyses (complete-case, per-protocol, tipping-point) of MAKE30 No subgroup showed a difference and the null result was robust to sensitivity analyses, but point estimates favoured balanced fluid in the sickest children and the authors cannot exclude a benefit there.	`secondary-analysis`	`negative-no-effect`
RES	MAKE30 did not differ between arms in any pre-specified or post-hoc subgroup, and the primary result was unchanged in complete-case and per-protocol analyses and robust to all but extreme departures from missing-at-random. No numbers given in the text for subgroup risk ratios (Figure 2, Figures S14 to S16 and Tables S11 to S12 are not reproduced in the manuscript text) Results, Primary Outcome and Subgroup Analyses; Figure 2 · “MAKE30 did not differ between balanced fluid and 0.9% saline within pre-specified subgroups (Fig. 2) or post-hoc analyses”		`negative-no-effect` informative	informs HYP
RES	Point estimates favoured balanced fluid among children who received the highest fluid volumes and who presented with more extreme acidosis and hyperlactatemia, without formal heterogeneity, so a benefit in the sickest children is not excluded. No point estimates or confidence intervals given in the text; the authors state only the direction of the estimates Discussion, limitations; Figure 2; Figures S15 to S16 · “point estimates favoured balanced fluid among patients who received the highest fluid volumes and presented with more extreme acidosis and hyperlactatemia”		`inconclusive-underpowered` uninformative	consistent-with HYP

ISS Whether balanced fluid benefits the most severely ill children (highest fluid volumes, extreme acidosis and hyperlactatemia), where point estimates favoured it but the trial could not exclude a benefit (open-question · Discussion, limitations)check and design
ISS Why this trial found no kidney benefit while Sankar et al. found a 38% relative reduction in new or progressive AKI in 708 children: outcome definition (7-day creatinine rise of 0.3 mg/dL vs MAKE30) or population (open-question · Discussion, paragraph 2)check and design
ISS Whether the null result generalises to low-resource settings and to hospital-acquired sepsis, neither of which the trial enrolled (open-question · Discussion, limitations)check and design
ISS Whether unmeasured fluid given after the 24 to 48 hour intervention phase altered outcomes, since fluid type was not controlled or recorded beyond 11:59 PM of the day after randomisation (open-question · Discussion, limitations; Methods, Intervention)check and design

Extraction notes: Year taken from the manuscript header (issue date 2026 Sep 3) and the DOI; the volume and page string in the header ('24420263959870870-881') is garbled and I did not resolve it. The primary finding is classed negative-no-effect and opposes because the paper states its power calculation (8,800 for 95% power, 6.0% to 4.3%), but note two things that a reviewer may weigh differently: the observed MAKE30 rate (3.0% to 3.4%) was about half the 6.0% assumed, and the authors themselves write that 'the low event rate diminished statistical power to detect planned differences'; also 8,482 children were analysed against a planned 8,800 after 559 withdrawals. An inconclusive-underpowered reading is defensible; I kept negative-no-effect because the risk-ratio CI (0.88 to 1.40) is narrow and the authors' conclusion is 'no significant difference' and 'do not show benefit'. Splitting one trial into four experiments (primary, fluid separation, safety, subgroups) is my reading of the paper's result sections; the paper presents one trial. The outcomeClass 'partial' for the safety experiment is my label for 'biochemical effect present, patient outcome absent, one adverse signal'; 'positive' for hyperchloremia and hypernatremia findings means the expected effect of balanced fluid was observed, not that it is a benefit of the hypothesis. The hyperlactatemia P=0.04 is an unadjusted safety P-value in a subset with lactate measured (1,296 vs 1,348). Subgroup risk ratios exist only in Figure 2 and supplementary figures, which the text file does not reproduce, so the two subgroup findings carry no numbers. The Sankar et al. figures (38% relative reduction, 708 children, 7-day window, 0.3 mg/dL creatinine rise) are as reported by this paper's Discussion, not checked against the source. Baseline AKI counts in the prose (1,172, 415, 432) and in Table 1 (560+612, 201+214, 204+228) agree. The trial registration is NCT04102371; the protocol and statistical analysis plan are referenced as published (Trials 2021;22:776) and at NEJM.org but are not in the text file.

Each open issue: checked, designed, and sent to a lab in mock

These pages are written for the next researcher who finds an open issue, not for the authors. Opening an issue shows, in order: whether anyone has already addressed it, what a registry of negative results holds on it, an experiment designed to close it with a control and a sensitivity statement, and what sending that experiment to a laboratory service would look like. Under each is a place to say what is wrong with it.

ISS Whether balanced fluid benefits the most severely ill children (highest fluid volumes, extreme acidosis and hyperlactatemia), where point estimates favoured it but the trial could not exclude a benefit (open-question · follow-up: partly addressed)

The trial's overall event rate was 3 percent, half of what it was powered for, and the only strata where balanced fluid looked better were the sickest children; the question is whether that is a real severity-dependent effect or the noise expected from several correlated subgroups in a null trial.

1. Has anyone done this since?

Status

partly addressed

Works since

Balanced crystalloids versus 0.9% saline for fluid resuscitation in children with septic shock: an updated systematic review and meta-analysis of randomized controlled trials (Raheem et al. 2026, Irish Journal of Medical Science, doi:10.1007/s11845-026-04639-x)
Pools six trials and 10,308 children including this one: no mortality or acute kidney injury reduction, but renal replacement therapy reduced (risk ratio 0.67, 95% CI 0.48 to 0.92); it reports no analysis by illness severity, so it leaves the severity question open while adding a signal on the outcome most tied to severe illness.
Balanced Solutions Versus Normal Saline in Pediatric Septic Shock: A GRADE-Assessed Systematic Review and Meta-Analysis of Randomized Controlled Trials (Elbahloul et al. 2026, Journal of Intensive Care Medicine, doi:10.1177/08850666261498038)
Six trials, 9,743 children: mortality unchanged (RR 0.97, 0.81 to 1.16), renal replacement therapy reduced (RR 0.68, 0.47 to 0.97); again no severity stratification, and the two meta-analyses include different trial sets despite both claiming six.
Fluid Resuscitation in Patients with Sepsis: For Whom the Balance Tolls (O'Connor and Bubeck-Wardenburg 2026, New England Journal of Medicine, doi:10.1056/NEJMe2609856)
The accompanying editorial; it has no abstract in Europe PMC and the version of record is not CC BY, so I did not read it and cannot say whether it takes a position on the severity subgroup.

Searched

All ten titles in the citing-works file, with Europe PMC abstracts for the three relevant ones; Europe PMC for 'balanced saline sepsis lactate acidosis subgroup treatment effect' and 'PRoMPT BOLUS'; web search for a secondary analysis or editorial discussing the severity subgroup.

Still open

No published analysis stratifies the pediatric randomised evidence by severity; the pooled renal replacement therapy signal is the only severity-adjacent result and it comes from two meta-analyses that disagree on which six trials to include.

Registry query

0 of 49 records in the open node mention any of: balanced crystalloid, 0.9% saline, pediatric septic shock, make30, hyperlactatemia, heterogeneity of treatment effect. The open node holds biology and chemistry so far, so this is the expected answer; a private node of a clinical group would be asked the same question.

2. A proposed experiment

Question

In children with septic shock and initial lactate above 4 mmol/L or vasoactive support, does balanced fluid reduce MAKE30 compared with saline, when pooled across the randomised trials?

Hypothesis

In the pre-specified high-severity stratum, the risk ratio for MAKE30 with balanced fluid is below the overall PRoMPT BOLUS estimate of 1.10 and its credible interval excludes 1.0; outside that stratum it does not.

Attempt type

secondary-analysis

Design

A pre-registered individual patient data (IPD) reanalysis, modelled on the adult BEST-Living IPD meta-analysis, of PRoMPT BOLUS (8,482 children analysed, 261 MAKE30 events) and Sankar 2023 (708 children, Indian PICUs, ICU mortality 33 percent), with any further pediatric trial the two 2026 meta-analyses include that will share data. Severity is defined before unblinding by three markers the PRoMPT BOLUS record already holds: initial lactate above 4 mmol/L (the paper's own hyperlactatemia threshold), vasoactive medication during the intervention phase (received by 14.2 percent), and KDIGO stage 2 or 3 acute kidney injury at presentation (4.9 and 5.1 percent). The outcome is MAKE30, a composite of death, new renal replacement therapy or persistent kidney dysfunction within 30 days, rebuilt from each trial's creatinine and dialysis records. A Bayesian hierarchical logistic model with a sceptical prior centred on the overall risk ratio estimates the fluid-by-severity interaction; the analysis plan names the single primary interaction (lactate) and treats the others as secondary. If the interaction credible interval excludes the overall estimate, the next step is an enrichment trial in that stratum; if it does not, the issue closes.

Controls

reference-case Before any stratification, the pipeline must recover each trial's published primary estimate from the shared data: PRoMPT BOLUS MAKE30 risk ratio 1.10 (95% CI 0.88 to 1.40) and Sankar's new or progressive AKI risk ratio 0.62 (0.49 to 0.80).
Rules out: Errors in data harmonisation or outcome reconstruction masquerading as a subgroup effect.
permutation-null Treatment labels are shuffled within trial and site many times to produce the distribution of the interaction statistic under no effect, given the actual correlation among the severity markers.
Rules out: An interaction that looks strong only because three correlated severity markers were tested on one set of 261 events.
background A stratum with no plausible mechanism for effect modification (sex, which the trial pre-specified) is analysed with the same model and must show no interaction.
Rules out: A model that generates interactions wherever it is pointed.
Vocabulary note: No control kind in the vocabulary names a negative-control stratum; 'background' is the closest.

Sensitivity

It cannot be set from the published record. The number of MAKE30 events inside the high-lactate stratum is in Figure S15, which the text file does not reproduce; lactate was measured in only 1,296 and 1,348 children per arm, so the stratum is a fraction of the 261 events. What is needed is the per-stratum event count from the data coordinating centre; with it, the width of the interaction credible interval can be stated before the analysis is run.

Would count as

Supported → positive · opposed → negative-no-effect
Inconclusive if: The high-severity stratum holds too few MAKE30 events for the interaction credible interval to exclude the overall estimate in either direction, or Sankar's data cannot be rebuilt into MAKE30 because 30-day creatinine and dialysis were not collected.

Known before

In adults with sepsis in SMART, balanced crystalloids were associated with lower 30-day in-hospital mortality (26.3 vs 31.2 percent, adjusted OR 0.74, 0.59 to 0.93) and lower MAKE30 (35.4 vs 40.1 percent). (Brown et al. 2019, American Journal of Respiratory and Critical Care Medicine, doi:10.1164/rccm.201903-0557OC)
So: The adult sepsis signal exists at event rates ten times those in PRoMPT BOLUS, which is why severity is the natural axis to test.
In BaSICS, the effect of balanced solution on 90-day mortality interacted with fluid volume, with lower mortality as fluid use increased in septic patients; at high fluid-use sites the absolute risk reduction was 0.13 (credible interval -0.27 to -0.01, probability of benefit 0.98). (Zampieri et al. 2024, Intensive Care Medicine, doi:10.1007/s00134-023-07264-9)
So: The PRoMPT BOLUS highest-volume signal has an adult precedent, and this paper supplies the Bayesian interaction method the design copies.
In BaSICS, benefit was concentrated in patients who had received only balanced fluid before enrollment, most of all unplanned sepsis admissions (OR 0.70, 89% CrI 0.50 to 0.97). (Zampieri et al. 2022, American Journal of Respiratory and Critical Care Medicine, doi:10.1164/rccm.202111-2484OC)
So: Pre-enrollment saline (median 17 to 18 mL/kg in PRoMPT BOLUS) is a candidate effect modifier the IPD model should include.
The adult IPD meta-analysis of six trials and 34,685 patients found in-hospital mortality OR 0.962 (95% CrI 0.909 to 1.019), probability of benefit 0.895, with harm in traumatic brain injury. (Zampieri et al. 2024, Lancet Respiratory Medicine (BEST-Living), doi:10.1016/S2213-2600(23)00417-4)
So: The template: subgroup findings in an IPD framework can be decisive in both directions, as the brain-injury harm shows.
In a secondary analysis of CLOVERS, balanced crystalloids were associated with lower mortality than saline under liberal resuscitation (12.6 vs 16.8 percent) and higher under restrictive resuscitation (16.3 vs 13.2 percent), a fluid-by-strategy interaction. (Khan et al. 2026, CHEST Critical Care, doi:10.1016/j.chstcc.2025.100226)
So: Fluid type was not randomised there, but it is a second adult result pointing at volume as the modifier, and a warning that the direction could reverse in low-volume patients.
Sankar's Indian PICU trial of 708 children, ICU mortality 33 to 34 percent, found new or progressive AKI reduced (21 vs 33 percent, RR 0.62) with no mortality difference. (Sankar et al. 2023, Critical Care Medicine, doi:10.1097/CCM.0000000000005952)
So: It is the only pediatric trial in a high-severity population, so it carries most of the severity information in any pooled analysis and dominates the result.

Searched: Europe PMC: 'balanced crystalloid saline sepsis secondary analysis SMART subgroup', 'BaSICS secondary analysis', author search Zampieri FG 2021 to 2024, 'balanced saline sepsis lactate acidosis subgroup'; abstracts fetched by DOI for SMART, PLUS, FLUID, Hammond 2022 and BEST-Living. I looked for a pediatric severity-stratified analysis of fluid type and found none.

Feasibility

The PRoMPT BOLUS data coordinating centre at Children's Hospital of Philadelphia with the PECARN, PERC and PREDICT networks, jointly with the AIIMS New Delhi group that ran Sankar 2023 · 6 to 12 months once data-sharing agreements are signed · low, since the data exist and the analysis is one biostatistician's project
Blockers: Data-sharing agreements across five countries plus India; Consent obtained under exception from informed consent may restrict secondary use in some jurisdictions; Sankar's trial may lack 30-day creatinine to build MAKE30

Strongest objection

A trialist would say that a favourable point estimate in two correlated strata of a null trial with six pre-specified and several post-hoc subgroups is exactly what chance produces, that no reanalysis creates events that did not happen, and that pooling with Sankar, whose population had ten times the mortality and a different outcome, will simply restate Sankar's result with PRoMPT BOLUS's name attached.

Not derivable from what is published

The fluid-volume cutoffs used for the Figure 2 subgroup
The MAKE30 event count within the high-lactate and high-volume strata (Figures S15 and S16)
The bicarbonate threshold the authors used for 'extreme acidosis'
Whether Sankar 2023 collected 30-day creatinine and dialysis

3. Send it to a lab

Send to a lab: This design is a reanalysis of existing data. There is nothing for a bench to run; it needs the data holder.
No bench job exists: this is a reanalysis of individual patient data from completed randomised trials.

Tell us what is wrong with it

Nothing here is sent anywhere by itself. The copy button puts your answers on the clipboard, and the email link opens your own mail client.

I would run this I would run it with changes I would not run this

Is initial lactate above 4 mmol/L the right primary severity marker, or would you pre-specify vasoactive support within the first 6 hours instead?

How sceptical should the prior on the interaction be: centred on the overall risk ratio of 1.10 with what width?

Would you accept Sankar's 7-day new or progressive AKI as a stand-in for MAKE30 in the pooled model, or exclude that trial and accept a smaller analysis?

Anything else, including what the design got wrong about your field

Email it instead

ISS Why this trial found no kidney benefit while Sankar et al. found a 38% relative reduction in new or progressive AKI in 708 children: outcome definition (7-day creatinine rise of 0.3 mg/dL vs MAKE30) or population (open-question · follow-up: partly addressed)

The authors offer outcome definition as the explanation for the conflict with Sankar 2023, but the two trials also differ in population (Indian PICU, ICU mortality 33 percent, versus emergency departments in five high-income countries, mortality about 1 percent), blinding, and whether maintenance fluid was included; the question is which difference carries the discrepancy.

1. Has anyone done this since?

Status

partly addressed

Works since

Balanced crystalloids versus 0.9% saline for fluid resuscitation in children with septic shock: an updated systematic review and meta-analysis of randomized controlled trials (Raheem et al. 2026, Irish Journal of Medical Science, doi:10.1007/s11845-026-04639-x)
Pooling both trials, acute kidney injury was not reduced (RR 0.76, 0.46 to 1.27, a wide interval that reflects the disagreement) while renal replacement therapy was (RR 0.67, 0.48 to 0.92); it pools across definitions rather than testing whether definition explains the gap.
Balanced Solutions Versus Normal Saline in Pediatric Septic Shock: A GRADE-Assessed Systematic Review and Meta-Analysis of Randomized Controlled Trials (Elbahloul et al. 2026, Journal of Intensive Care Medicine, doi:10.1177/08850666261498038)
Same pattern, renal replacement therapy RR 0.68 (0.47 to 0.97) with mortality unchanged; neither meta-analysis re-applies one AKI definition across trials.

Searched

The citing-works file; Europe PMC for 'acute kidney injury definition MAKE pediatric sepsis', 'MAKE30 major adverse kidney events pediatric sepsis' and 'Sankar multiple electrolytes solution saline bolus'; web search for any cross-definition reanalysis.

Still open

Nobody has applied Sankar's outcome to the PRoMPT BOLUS data or MAKE30 to Sankar's, so the authors' explanation remains a conjecture.

Registry query

4 of 49 records in the open node mention any of: balanced crystalloid, saline, pediatric septic shock, acute kidney injury definition, kdigo, make30. arcadia-neuroimaging-capsaicin: Intradermal capsaicin produces widespread oscillatory cortical activation absent from paired vehicle trials; arcadia-neuroimaging-histamine: Histamine-induced itch produces no cortical signal separable from saline injection (iced for a technical gap); arcadia-neuroimaging-path: Label-free flavoprotein imaging of the mouse cortex as a behaviour-independent assay for touch, pain and itch (iced)

2. A proposed experiment

Question

Does applying Sankar's outcome (new or progressive KDIGO AKI within 7 days) to PRoMPT BOLUS reproduce a balanced-fluid benefit, and does applying MAKE30 to Sankar's data remove it?

Hypothesis

Outcome definition explains the discrepancy: in PRoMPT BOLUS, balanced fluid reduces 7-day new or progressive KDIGO AKI (risk ratio below 1 with an interval excluding 1) while MAKE30 stays at 1.10.

Attempt type

secondary-analysis

Design

A pre-registered cross-application of outcome definitions in both datasets. In PRoMPT BOLUS (4,235 balanced, 4,247 saline), each child's creatinine values through day 7, which the trial extracted from the medical record for its follow-up laboratory analyses, are scored against the Kidney Disease Improving Global Outcomes (KDIGO) stages using the same baseline creatinine rule the trial used (lowest value 12 months to 24 hours before enrollment, or imputed by age and sex in 40.5 percent). The outcome is new or progressive AKI within 7 days, as Sankar defined it, and the between-arm risk ratio is estimated with the trial's own site-stratified Cochran-Mantel-Haenszel method. In Sankar's data (351 vs 357), the reverse is done: death, dialysis and final creatinine at least 200 percent of baseline with a rise of at least 0.3 mg/dL are combined into MAKE30 as far as the follow-up allows. A second analysis restricts PRoMPT BOLUS to children with KDIGO stage 2 or 3 AKI at presentation or vasoactive support, to see whether a Sankar-like population inside PRoMPT BOLUS shows a Sankar-like effect. Three outcomes decide the issue: definition reproduces the benefit, population reproduces it, or neither.

Controls

known-positive-task Hyperchloremia, the one effect both trials agree on, must reproduce under the harmonised pipeline: PRoMPT BOLUS RR 0.64 (0.60 to 0.69) and Sankar's lower hyperchloremia at 24, 48 and 72 hours.
Rules out: A harmonisation that destroys real between-arm differences, which would make a null cross-definition result uninterpretable.
reference-case Each trial's published primary estimate must be recovered first from the shared data before any redefinition.
Rules out: Data transfer or coding errors.
permutation-null Allocation labels shuffled within site to give the distribution of the 7-day AKI risk ratio under no effect, with creatinine sampling frequency carried along with each child.
Rules out: An apparent AKI difference produced by differential creatinine measurement between arms in an open-label trial.

Sensitivity

PRoMPT BOLUS has 8,482 children; its 7-day AKI rate is not reported, but 13.8, 4.9 and 5.1 percent already had stage 1, 2 and 3 AKI at presentation and Sankar's saline arm had 33 percent new or progressive AKI, so a 7-day AKI outcome would yield far more events than the 261 MAKE30 events. A 38 percent relative reduction, Sankar's effect, would be detectable; what cannot be stated without the data is the exact event count, which depends on how often creatinine was drawn after discharge from the emergency department.

Would count as

Supported → partial · opposed → negative-not-replicated
Inconclusive if: Creatinine was measured too infrequently or too differentially after the first 72 hours in the pragmatic record (children discharged early have no day-7 value) for 7-day AKI to be ascertained without bias, or Sankar's follow-up cannot support a 30-day composite.

Known before

Sankar 2023: blinded multicentre trial in four Indian PICUs, bolus fluid only, outcome new or progressive AKI within 7 days, RR 0.62 (0.49 to 0.80), ICU mortality 33 vs 34 percent, hyperchloremia lower at 24, 48 and 72 hours. (Sankar et al. 2023, Critical Care Medicine, doi:10.1097/CCM.0000000000005952)
So: Fixes the exact outcome and population to be cross-applied; note it was blinded and bolus-only, two further differences from PRoMPT BOLUS.
MAKE30 in pediatric sepsis had an incidence of 9.6 percent (death 4.5, kidney replacement 1.7, persistent dysfunction 5.8) in 1,685 admissions and tracked mortality, cost and later eGFR below 60. (Weiss et al. 2019, Clinical Journal of the American Society of Nephrology, doi:10.2215/CJN.12201018)
So: The 6.0 percent base rate PRoMPT BOLUS was powered on came from this cohort; the trial observed half that, which is itself part of why a creatinine-rise outcome has more events.
In adults, SMART's MAKE30 reduction (14.3 vs 15.4 percent) came with no difference in persistent renal dysfunction (6.4 vs 6.6 percent); the composite moved on death and dialysis. (Semler et al. 2018, New England Journal of Medicine, doi:10.1056/NEJMoa1711584)
So: Warns that creatinine-defined and composite-defined kidney outcomes can move independently even in a positive trial.
SPLYT-P in 516 PICU children: balanced solutions halved the odds of a chloride rise of 5 mEq/L or more, with new-onset AKI 6.1, 3.7 and 3.2 percent across arms and no difference. (Raman et al. 2023, JAMA Pediatrics, doi:10.1001/jamapediatrics.2022.4912)
So: A pediatric trial where chloride separated and KDIGO AKI did not, the opposite pattern to Sankar, so definition alone cannot be assumed to favour balanced fluid.
Five pediatric RCTs with 992 children before PRoMPT BOLUS: mortality RR 0.82 (0.45 to 1.50), AKI RR 0.64 (0.50 to 0.82), renal replacement RR 0.52 (0.35 to 0.76), hyperchloremia RR 0.74. (Chandelia et al. 2026, Journal of Intensive Care Medicine, doi:10.1177/08850666251315705)
So: The pre-PRoMPT BOLUS evidence base was small and driven by Sankar; the design should weight it accordingly.
A 42-child Thai trial of saline, Ringer's lactate and Sterofundin boluses found no outcome differences but a larger urinary NGAL fall with large-dose Ringer's. (Trepatchayakorn et al. 2021, Indian Journal of Pediatrics, doi:10.1007/s12098-021-03808-3)
So: A biomarker outcome (urinary NGAL) exists as a third definition that could be added if the two creatinine-based ones disagree.

Searched: Europe PMC by DOI for Sankar 2023, Weiss 2019, SMART, SPLYT-P, the 2025 BMC Pediatrics and 2026 Chandelia meta-analyses and the Thai trial; title searches for pediatric balanced-versus-saline sepsis RCTs (two hits) and for AKI definition papers in pediatric sepsis (none specific to fluid trials).

Feasibility

The PRoMPT BOLUS data coordinating centre at Children's Hospital of Philadelphia together with the AIIMS New Delhi group; a pediatric nephrologist should own the KDIGO scoring · 6 to 12 months · low, a reanalysis of two existing datasets
Blockers: Access to Sankar's patient-level data; Creatinine sampling after 72 hours in a pragmatic emergency-department trial may be sparse; Urine output, part of KDIGO, was probably not captured in either record

Strongest objection

A pediatric nephrologist would object that the premise is wrong: in a child whose baseline creatinine is 0.30 mg/dL (the trial's median), a rise of 0.3 mg/dL is a doubling, so Sankar's definition is not more lenient than MAKE30's 200 percent and redefinition will not bridge the gap; the real difference is a PICU population with 33 percent mortality against an emergency-department population with 1 percent, and the only honest test of that is a trial in the sicker population, not a reanalysis.

Not derivable from what is published

How often creatinine was measured in PRoMPT BOLUS after day 3 and whether that differed by arm
Whether Sankar's data include 30-day vital status, dialysis and final creatinine
Sankar's exact KDIGO staging rule for 'progressive' AKI
Whether either trial recorded urine output

3. Send it to a lab

Send to a lab: This design is a reanalysis of existing data. There is nothing for a bench to run; it needs the data holder.
No bench job exists: the work is outcome redefinition and reanalysis of two completed trials' patient-level data.

Tell us what is wrong with it

Nothing here is sent anywhere by itself. The copy button puts your answers on the clipboard, and the email link opens your own mail client.

I would run this I would run it with changes I would not run this

Would you count a child discharged before day 7 with no further creatinine as no AKI, or as missing, and does the answer differ by arm in an open-label trial?

Should the Sankar-like subpopulation inside PRoMPT BOLUS be defined by vasoactive use, by KDIGO stage at presentation, or by both?

If definition reproduces the benefit but MAKE30 does not move, would you call that a real but clinically unimportant effect, or a measurement artefact of creatinine dilution by saline?

Anything else, including what the design got wrong about your field

Email it instead

ISS Whether the null result generalises to low-resource settings and to hospital-acquired sepsis, neither of which the trial enrolled (open-question · follow-up: partly addressed)

The authors flag two populations outside their trial; this design takes the first, low-resource settings, where mortality is many times higher and the one existing trial (Sankar, India) was positive on kidney injury, and leaves hospital-acquired sepsis as a separate question noted below.

1. Has anyone done this since?

Status

partly addressed

Works since

Balanced crystalloids versus 0.9% saline for fluid resuscitation in children with septic shock: an updated systematic review and meta-analysis of randomized controlled trials (Raheem et al. 2026, Irish Journal of Medical Science, doi:10.1007/s11845-026-04639-x)
Pools PRoMPT BOLUS with the smaller trials from India and Thailand and finds no mortality effect in any window; it does not report results by country income level, so it shows consistency only in aggregate.
Comparison of acetate-based, lactate-based and multiple electrolyte solution versus saline (MeSLAc trial) (Sankar, AIIMS New Delhi, listed 2023 to 2026 on the India Science and Technology research portal; not in the citing-works file, found by web search, https://www.indiascienceandtechnology.gov.in/research/comparison-acetate-based-lactate-based-and-multiple-electrolyte-solution-versus-saline-meslac-trial)
An ongoing multicentre pediatric septic shock trial in India comparing three balanced solutions with saline; if it reports, it is the next low-resource datum, but I could not find a registry number, sample size or primary outcome, and the portal page could not be fetched (certificate error).

Searched

The citing-works file; Europe PMC for pediatric balanced-versus-saline trials in Africa, South Asia or resource-limited settings (only diabetic ketoacidosis and diarrhoea trials surfaced); web search for ClinicalTrials.gov or CTRI registrations of pediatric fluid-type trials in Africa, India or Pakistan from 2024 onward.

Still open

No trial of fluid type in children with septic shock has run in sub-Saharan Africa, none in any low-resource setting has used MAKE30 or a 30-day outcome, and the hospital-acquired sepsis population has no trial at all.

Registry query

0 of 49 records in the open node mention any of: balanced crystalloid, ringer's lactate, 0.9% saline, pediatric septic shock, low-resource setting, cluster crossover. The open node holds biology and chemistry so far, so this is the expected answer; a private node of a clinical group would be asked the same question.

2. A proposed experiment

Question

In children with septic shock treated in low- and middle-income country hospitals, does assigning balanced fluid rather than saline for all crystalloid reduce death or MAKE30 within 30 days?

Hypothesis

Where baseline mortality is far higher than PRoMPT BOLUS's 1 percent, balanced fluid reduces 30-day MAKE30 compared with saline, with the chloride separation PRoMPT BOLUS observed reproduced as the mechanism check.

Attempt type

clinical-trial

Design

A pragmatic cluster-randomised crossover trial, in the pattern of the FLUID trial in Ontario, in which whole hospitals (paediatric emergency department plus PICU) in India and East Africa switch their stocked isotonic crystalloid between Ringer's lactate and 0.9% saline for alternating periods, so that no per-patient consent or allocation step is needed at the bedside. Participants are children 2 months to under 18 years treated with fluid resuscitation for suspected septic shock with abnormal perfusion, the PRoMPT BOLUS definition, with bolus volume left to the local protocol; in African sites that protocol must be the post-FEAST conservative one, so the trial compares fluid composition inside a bolus-restrictive policy rather than reintroducing boluses. The primary outcome is MAKE30 rebuilt from death, dialysis and a creatinine at discharge or day 30, with 28-day all-cause mortality as the co-primary because 30-day creatinine will be missing for many discharged children. Chloride at 24 to 72 hours is the mechanism outcome. Period length, number of hospitals and washout are set from a pilot that measures the local MAKE30 base rate.

Controls

internal-positive Serum chloride above 110 mEq/L at 24 to 72 hours must be less frequent in balanced-fluid periods, as in PRoMPT BOLUS (31 vs 49 percent) and Sankar.
Rules out: A null result caused by fluid substitution or stock-outs, which are the main threat to a supply-level intervention.
permutation-null Period and hospital labels are permuted to give the null distribution of the cluster-level effect, as cluster-crossover designs require.
Rules out: Secular trends in sepsis mortality and seasonal epidemics (malaria, dengue) masquerading as a fluid effect.
blinding-check Outcome adjudicators are blind to period allocation and are asked to guess it; guesses must be no better than chance.
Rules out: Assessor bias in classifying deaths and dialysis decisions in an open-label design.
Vocabulary note: Fluid supply switching cannot be blinded; the check is of outcome assessors only.

Sensitivity

The base rate is the missing number. PRoMPT BOLUS was powered for a fall from 6.0 to 4.3 percent and observed 3.0 percent; Sankar's Indian PICUs had ICU mortality of 33 to 34 percent, so MAKE30 in those settings is plausibly an order of magnitude above 3 percent, and a trial a fraction of 8,800 could detect the same relative effect. The exact size needs a pilot cohort's MAKE30 rate and the between-hospital variance for the cluster design; neither is published.

Would count as

Supported → positive · opposed → negative-no-effect
Inconclusive if: Chloride does not separate between periods (fluids were substituted), or loss to follow-up after discharge leaves MAKE30 unascertained in more than the tipping-point analysis can absorb, or an epidemic season falls into one period only.

Known before

Sankar 2023, four Indian PICUs, 708 children, blinded: new or progressive AKI 21 vs 33 percent (RR 0.62), ICU mortality 33 vs 34 percent. (Sankar et al. 2023, Critical Care Medicine, doi:10.1097/CCM.0000000000005952)
So: The only completed low-resource trial; it supplies the base rates and shows the kidney signal appears where mortality is high, which is the design's premise.
In African children with severe febrile illness, fluid boluses increased 48-hour mortality compared with no bolus (FEAST). (Maitland et al. 2011, New England Journal of Medicine, doi:10.1056/NEJMoa1101549)
So: Any African fluid-type trial must sit inside a bolus-restrictive protocol; it cannot copy the PRoMPT BOLUS bolus practice, so the result would answer a narrower question.
A hospital-wide cluster-crossover of lactated Ringer's versus saline in seven Ontario hospitals, 43,626 patients, found death or readmission within 90 days 20.3 vs 21.4 percent (adjusted difference -0.53 points, -1.85 to 0.79). (McIntyre et al. 2025, New England Journal of Medicine, doi:10.1056/NEJMoa2416761)
So: The design template; it also shows how small the hospital-wide effect is in a high-income setting, so the low-resource trial must justify a larger expected effect.
Five pediatric RCTs, 992 children, mostly from South Asia and Southeast Asia, before PRoMPT BOLUS: mortality RR 0.82 (0.45 to 1.50), AKI RR 0.64. (Chandelia et al. 2026, Journal of Intensive Care Medicine, doi:10.1177/08850666251315705)
So: The low-resource evidence is small and kidney-positive, mortality-null; the design should expect the same pattern and power for MAKE30, not mortality alone.
A double-blind trial of 0.9% saline versus Plasma-Lyte as initial fluid in Indian children with diabetic ketoacidosis (SPinK) was completed in a tertiary PICU. (Williams et al. 2020, Critical Care, doi:10.1186/s13054-019-2683-3)
So: Shows a blinded individually randomised fluid-type trial is feasible in an Indian PICU, which argues against my cluster design if blinding is judged essential.
A 42-child Thai trial of saline, Ringer's lactate and Sterofundin boluses found no difference in complications. (Trepatchayakorn et al. 2021, Indian Journal of Pediatrics, doi:10.1007/s12098-021-03808-3)
So: Illustrates the size of trials the region has run so far; a network, not a single centre, is needed.

Searched: Europe PMC: 'balanced crystalloid versus saline children low-income OR Africa OR India randomized', title search for pediatric balanced-versus-saline sepsis RCTs, 'FEAST trial fluid bolus African children', 'hospital-acquired sepsis balanced crystalloid saline' (nothing on fluid type in hospital-onset pediatric sepsis); web search for registered trials in Africa, India and Pakistan. The FEAST figures are from memory of its title result only, since I did not fetch its abstract; the claim is stated without numbers for that reason.

Feasibility

An Indian PICU trial network such as the four-centre group behind Sankar 2023, with the East African paediatric trial groups that ran FEAST; a funder such as the Medical Research Future Fund or the UK Medical Research Council global health stream · 3 to 4 years including a 6-month pilot for base rates · moderate, since the intervention is a change of stocked fluid but 30-day follow-up after discharge in these settings is the expensive part
Blockers: Reliable supply of Ringer's lactate in all sites for whole periods; Creatinine measurement at 30 days after discharge; Ethics review of cluster-level allocation without individual consent in each country; Bolus policy differences between Asian and African sites that make a single protocol hard to write

Strongest objection

A global-health trialist would say the live question in low-resource paediatric sepsis is bolus volume, settled badly by FEAST, not bolus composition; that a 30-day creatinine outcome is unobtainable where children leave hospital on day 3 and never return; and that where Ringer's lactate costs more or is intermittently available, a positive result would not change what is in the cupboard.

Not derivable from what is published

The MAKE30 base rate in any low- or middle-income paediatric setting
Between-hospital variance needed for the cluster design
Whether MeSLAc is registered, its size and its primary outcome
Any data at all on fluid type in hospital-acquired paediatric sepsis, the second half of the authors' issue, which this design does not cover

3. Send it to a lab

Send to a lab: This design is a study in people. It cannot be sent to a laboratory service; it needs a trial network, ethics approval and a sponsor.
No bench job exists: this is a cluster-randomised crossover trial of stocked fluid type in hospitals, with outcomes drawn from patient records.

Tell us what is wrong with it

Nothing here is sent anywhere by itself. The copy button puts your answers on the clipboard, and the email link opens your own mail client.

I would run this I would run it with changes I would not run this

Would you insist on individual randomisation and blinding, as Sankar did, over the cheaper hospital-level crossover, and why?

Is MAKE30 the right primary outcome where day-30 creatinine is rarely available, or should 28-day mortality stand alone?

Should African sites be included at all given FEAST, or should the first trial be South Asian only?

Anything else, including what the design got wrong about your field

Email it instead

ISS Whether unmeasured fluid given after the 24 to 48 hour intervention phase altered outcomes, since fluid type was not controlled or recorded beyond 11:59 PM of the day after randomisation (open-question · follow-up: nothing found)

The authors wonder whether fluid of either type given after the intervention window, which they neither assigned nor recorded, was large enough to dilute the contrast between arms before MAKE30 was assessed; the first step is simply to measure it.

1. Has anyone done this since?

Status

nothing found

Works since

IV bolus, maintenance and medication carrier fluid in children with community-acquired sepsis: a multicentre cohort study (Long et al. 2026, Archives of Disease in Childhood, doi:10.1136/archdischild-2025-330189)
In 5,352 children in 11 Australian and New Zealand emergency departments, maintenance fluid was the largest share of the first 24 hours (27.1 of 40.0 mL/kg median); it cites this paper and comes from the same PREDICT network but stops at 24 hours, so it does not measure post-window fluid.

Searched

The citing-works file; Europe PMC for 'fluid type after 48 hours crystalloid maintenance fluid chloride burden children sepsis' and 'fluid creep OR maintenance fluid chloride children critically ill 2018 to 2026'.

Still open

Nobody has quantified the volume or type of crystalloid given to PRoMPT BOLUS children after the intervention phase, so the size of the possible dilution is unknown.

Registry query

5 of 49 records in the open node mention any of: balanced crystalloid, saline, maintenance fluid, fluid creep, pediatric septic shock, chloride. arcadia-neuroimaging-capsaicin: Intradermal capsaicin produces widespread oscillatory cortical activation absent from paired vehicle trials; arcadia-neuroimaging-histamine: Histamine-induced itch produces no cortical signal separable from saline injection (iced for a technical gap); arcadia-neuroimaging-path: Label-free flavoprotein imaging of the mouse cortex as a behaviour-independent assay for touch, pain and itch (iced)

2. A proposed experiment

Question

How much crystalloid, and of which type, did PRoMPT BOLUS children receive between the end of the intervention phase and hospital discharge or day 30, and does it attenuate the between-arm chloride difference?

Hypothesis

Post-window crystalloid of the non-assigned type was large enough that the between-arm difference in serum chloride had closed by day 7 in children still in hospital, which would mean the MAKE30 contrast was diluted.

Attempt type

secondary-analysis

Design

A record-level extension of the PRoMPT BOLUS data: for each of the 8,482 analysed children, all intravenous crystalloid (bolus, maintenance and medication carrier fluid) from the end of the intervention phase to discharge or day 30 is pulled from the medical record at the 47 sites, typed as balanced, saline or other, and expressed in mL/kg, as the trial did within the window (Table S10). The primary descriptive measure is the proportion of total hospital crystalloid that was the assigned type, compared with the 80 and 88 percent within-window figures. The analytic test is a dose-response: serum chloride, bicarbonate and creatinine at days 3 to 7 against post-window volume of the non-assigned fluid, by arm (the trial found chemistry differences widen with volume, Figures S11 to S13). A second analysis re-runs MAKE30 restricted to children in whom at least 75 percent of all hospital crystalloid, not just intervention-phase crystalloid, was the assigned type, mirroring the published per-protocol analysis. If the assigned-type share stays near the within-window figure the issue closes; if it falls and chemistry converges, a whole-stay trial, as in the adult trials, is the follow-on.

Controls

reference-case The extraction must first reproduce the published within-window figures: saline 20 mL/kg in the balanced arm versus 79 in the saline arm, and balanced fluid 58 versus 0 mL/kg.
Rules out: A record-extraction method that cannot recover known values, which would make the post-window numbers untrustworthy.
instrument-noise-floor In a sample of children at each site, fluid volumes from the medical record are checked against pharmacy or infusion-pump logs to measure how much fluid the record misses.
Rules out: Apparent post-window crossover or absence of it that is an artefact of charting gaps rather than fluid actually given.
Vocabulary note: Closest kind for a record-completeness check.
permutation-null Arm labels are shuffled within site to give the null distribution of the chemistry-by-volume gradient.
Rules out: A dose-response that is really confounded by illness severity, since sicker children get more fluid of every type and stay longer.

Sensitivity

It cannot be set from the published record, because the quantity to be measured, post-window fluid volume, is the unknown. What is known is that chemistry differences in the trial widened with volume, so the gradient to be detected has already been seen within the window; what is needed is the distribution of post-window volumes, after which the chloride difference the design could detect follows from the within-window variance in Figures S11 to S13.

Would count as

Supported → partial · opposed → negative-no-effect
Inconclusive if: Post-window fluid data are not retrievable at enough sites (the trial monitored data quality only for the collected variables), or most children were discharged within the window so that post-window exposure is confined to the sickest and cannot be separated from severity.

Known before

In a 45-bed adult ICU over 103,098 patient-days, maintenance and replacement fluids were 24.7 percent and fluid creep 32.6 percent of daily volume against 6.5 percent for resuscitation; a hypotonic maintenance strategy cut chloride burden by 30.8 mmol/day versus 3.0 mmol/day for balanced resuscitation. (Van Regenmortel et al. 2018, Intensive Care Medicine, doi:10.1007/s00134-018-5147-3)
So: Non-resuscitation fluid, which continues after any 48-hour window, carries most of the chloride; the authors cite this paper as the reason maintenance fluid was included.
In 43 ventilated children under 3 years, fluid creep supplied 56 to 58 percent of sodium and chloride load in the first 48 hours. (Langer et al. 2020, BMC Pediatrics, doi:10.1186/s12887-020-02322-3)
So: Medication carrier fluid, which the trial did not assign, is a large chloride source and must be typed in the extraction.
PLUS assigned Plasma-Lyte 148 or saline for all ICU fluid therapy for up to 90 days in 5,037 adults and found death within 90 days 21.8 vs 22.0 percent. (Finfer et al. 2022, New England Journal of Medicine, doi:10.1056/NEJMoa2114464)
So: Extending exposure to the whole ICU stay did not produce an effect in adults, which lowers the prior that post-window dilution hid one here.
SMART assigned fluid type for the whole ICU stay in 15,802 adults and found MAKE30 14.3 vs 15.4 percent. (Semler et al. 2018, New England Journal of Medicine, doi:10.1056/NEJMoa1711584)
So: The one positive adult trial also used whole-stay exposure, so a whole-stay pediatric trial is the natural follow-on if post-window crossover proves large.
The ESPNIC guideline on maintenance fluids in acutely and critically ill children systematically reviewed tonicity and composition of maintenance fluids. (Brossier et al. 2022, Intensive Care Medicine, doi:10.1007/s00134-022-06882-z)
So: Sets what maintenance fluid children in these hospitals are likely to have received after the window, which the extraction should code against.

Searched: Europe PMC: 'fluid creep OR maintenance fluid AND chloride AND children OR critically ill, 2018 to 2026' and the earlier query on fluid type after 48 hours; abstracts by DOI for Van Regenmortel 2018, Langer 2020, PLUS and SMART. I searched for a pediatric study of crystalloid type beyond 48 hours of sepsis treatment and found none.

Feasibility

The PRoMPT BOLUS data coordinating centre at Children's Hospital of Philadelphia and the site investigators who hold medical-record access · 3 to 6 months if records are accessible, longer if each of 47 sites must re-abstract · low to moderate, dominated by record abstraction at 47 sites under five countries' regulatory oversight
Blockers: Re-abstraction permissions under exception-from-consent enrollment; Medication carrier fluid is often not charted by type; Sites in five countries chart fluids differently

Strongest objection

The authors already state, with four citations, that most crystalloid is given within 48 hours, and their per-protocol analysis matched the primary result; adult trials with whole-stay exposure were also null or nearly so; so the likely outcome is a confirmation that post-window fluid was small, which changes nothing, and the effort would be better spent on the severity question.

Not derivable from what is published

The distribution of post-window crystalloid volumes and types in PRoMPT BOLUS
Whether post-window fluid orders are retrievable at all 47 sites
The within-window variance of chloride against volume (Figures S11 to S13, not in the text)

3. Send it to a lab

Send to a lab: This design is a reanalysis of existing data. There is nothing for a bench to run; it needs the data holder.
No bench job exists: this is medical-record abstraction and reanalysis of a completed trial.

Tell us what is wrong with it

Nothing here is sent anywhere by itself. The copy button puts your answers on the clipboard, and the email link opens your own mail client.

I would run this I would run it with changes I would not run this

Is day 7 the right time point to test chloride convergence, given how many children are discharged by then?

Would you count medication carrier fluid as assigned-type exposure, or only bolus and maintenance?

If the assigned-type share after the window turns out to be, say, half the within-window share, is that enough to justify a whole-stay trial, or would you want the chemistry to have converged as well?

Anything else, including what the design got wrong about your field

Email it instead

Designed on 21 October 2026 against outreach/nejm-experiments/brief.md. Grounded on: record, full-text, citing-works, prior-art. Full text: Europe PMC PMC13134814 fullTextXML, CC BY author manuscript. Designer notes: Unresolved: the fluid-volume cutoffs for the Figure 2 subgroup, the event counts in the severity strata, the creatinine sampling schedule and all supplementary figures are not in the CC BY text file, so three of the four sensitivity statements say the number cannot be set. The NEJM editorial (doi:10.1056/NEJMe2609856) was not read because the version of record is not CC BY and Europe PMC holds no abstract. The MeSLAc listing could not be fetched (self-signed certificate on the Indian government portal); what I report about it comes from a web-search summary and should be verified against CTRI before it is cited. The FEAST claim in design iss-3 is stated without numbers because I did not fetch its abstract. The citing-works file holds only ten works, all from 2026, and I used zero OpenAlex queries beyond that file; Europe PMC was the main search tool, with two web searches. Inferred rather than stated: that the two 2026 pediatric meta-analyses include different trial sets (both say six trials but report 9,743 versus 10,308 children); that PRoMPT BOLUS creatinine values exist through day 7 (the text says follow-up chloride, bicarbonate and creatinine were analysed, with electrolyte safety outcomes defined within three days, so day 7 coverage is a guess). Design iss-3 covers only the low-resource half of the authors' issue; hospital-acquired sepsis has no fluid-type evidence at all and is listed under notDerivable rather than designed for. Vocabulary fit: the control kinds are built for benches and trials, not for reanalyses; I used 'reference-case' for pipeline checks that recover published estimates, 'permutation-null' for shuffled-label nulls, 'background' for a negative-control stratum and 'instrument-noise-floor' for a record-completeness check, each with a kindNote where the fit is poor. 'blinding-check' was applied to outcome assessors in an unblindable cluster design, which is a stretch. The attempt types fitted well; three designs are secondary-analysis and one is clinical-trial, and no design is a bench job, which is what this paper's issues demand.

The other five

Licence: the author manuscript of this paper is CC BY 4.0 in PubMed Central, so a graph of it is published here with attribution, exactly as the pilot’s records are. The NEJM version of record is under the publisher’s terms and nothing was taken from it; where the manuscript and the version of record differ, the manuscript is what is drawn. Method: one extraction pass against outreach/nejm-graphs/brief.md, every finding with a locator and a verbatim quote; the extraction notes above say what was inferred.