Skip to main content

Foundational Trials

SAFE

Image: Shutterstock / sfam_photo

Publication

  • Title: A Comparison of Albumin and Saline for Fluid Resuscitation in the Intensive Care Unit
  • Acronym: SAFE
  • Year: 2004
  • Journal published in: The New England Journal of Medicine
  • Citation: SAFE Study Investigators. A comparison of albumin and saline for fluid resuscitation in the intensive care unit. N Engl J Med. 2004;350(22):2247-2256.

Context & Rationale

  • Background
    • Intravenous fluid resuscitation is a near-universal ICU intervention; small treatment effects could translate to substantial population-level benefit or harm.
    • Physiological arguments for colloids (oncotic pressure, less administered volume) competed with concerns about endothelial leak, interstitial oedema, and context-dependent harm.
    • Pre-trial evidence was dominated by small heterogeneous studies and meta-analyses; a 1998 systematic review reported higher mortality with albumin, amplifying uncertainty and practice variation 1.
    • A large, adequately powered, blinded RCT was required to resolve the “albumin safety” question in contemporary ICU care.
  • Research Question/Hypothesis
    • In adult ICU patients requiring intravascular volume resuscitation, does 4% human albumin (vs 0.9% saline) change 28-day all-cause mortality?
    • Hypothesis: clinically meaningful differences in mortality were possible, but direction was uncertain given conflicting prior evidence.
  • Why This Matters
    • Fluid choice has immediate bedside relevance and system-level implications (cost, supply, stewardship).
    • Establishing safety equivalence (or superiority) would reshape resuscitation practice across heterogeneous ICU case-mix.
    • Definitive evidence could redirect research towards subgroup-specific signals (e.g., sepsis, trauma/brain injury).

Design & Methods

  • Research Question: In critically ill adults requiring fluid resuscitation, does 4% albumin (vs 0.9% saline) alter 28-day all-cause mortality?
  • Study Type:
    • Randomised, double-blind, parallel-group trial.
    • Multicentre (16 ICUs), investigator-initiated; Australia and New Zealand.
    • Randomisation stratified by institution and by trauma diagnosis at ICU admission; central minimisation via secure web-based service.
    • Setting: ICU (administration of study resuscitation fluid in ICU only; fluids outside ICU not controlled).
  • Population:
    • Adults admitted to ICU, where the treating clinician determined intravascular volume resuscitation was required.
    • Baseline illness severity: APACHE II 18.7±7.9 (albumin) vs 19.0±8.0 (saline).
    • Key predefined baseline subgroups (by presence/absence): trauma, severe sepsis, acute respiratory distress syndrome.
    • Exclusions (explicitly reported): ICU admission after cardiac surgery, liver transplantation, or burns; previous adverse reaction to albumin (and related exclusions not further detailed in the published manuscript).
    • Consent approach: delayed consent commonly used; vital status follow-up through day 28.
  • Intervention:
    • 4% human albumin solution (Albumex, CSL) used for all episodes of fluid resuscitation in ICU until ICU discharge/death or day 28.
    • Dose/rate determined by treating clinicians according to clinical status and response.
    • Study fluid supplied in identical 500 mL bottles with masking cartons and bespoke administration sets to preserve blinding.
  • Comparison:
    • 0.9% saline (normal saline) used for all episodes of fluid resuscitation in ICU until ICU discharge/death or day 28.
    • Maintenance fluids, replacement fluids, nutrition, and blood products were permitted at clinician discretion in both groups.
  • Blinding:
    • Double-blind (participants, clinicians, trial staff, outcome assessors).
    • Blinding method included identical bottles, masking cartons, and administration sets; effectiveness of blinding formally assessed before trial initiation.
    • Implication: minimised performance and detection bias for objective primary endpoint (mortality).
  • Statistics:
    • Power calculation (published): a total sample size of ~7000 patients was planned to detect a 3% absolute difference in 28-day mortality (from ~15% to ~12%) with 90% power at a two-sided 5% significance level.
    • Primary analysis: intention-to-treat framework for participants with available 28-day vital status; effect estimates reported as relative risks with 95% CIs; two-sided P values.
    • Interim monitoring: two preplanned interim analyses after 2333 (33%) and 4666 (67%) participants, reviewed by an independent data-monitoring committee.
    • Predefined subgroup analyses: mortality by trauma, severe sepsis, and ARDS status at baseline; heterogeneity testing reported (“test for a common relative risk”).
  • Follow-Up Period:
    • 28 days after randomisation (primary and secondary outcomes).
    • Daily data during ICU stay early (SOFA components daily for 7 days then every 3rd day to ICU discharge/day 28; fluid and physiology recorded daily until ICU discharge/death/day 28).

Key Results

This trial was not stopped early. Two preplanned interim analyses were performed (at 2333 and 4666 participants), and recruitment proceeded to completion.

Outcome 4% Albumin 0.9% Saline Effect p value / 95% CI Notes
28-day all-cause mortality 726/3473 (20.9%) 729/3460 (21.1%) RR 0.99 95% CI 0.91 to 1.09; P=0.87 Primary outcome (vital status available for 6933 participants; 67 missing)
Alive in ICU at day 28 111/3473 (3.2%) 87/3460 (2.5%) RR 1.27 95% CI 0.96 to 1.68; P=0.09 Secondary outcome
Alive in hospital at day 28 793/3473 (22.8%) 848/3460 (24.5%) RR 0.93 95% CI 0.86 to 1.01; P=0.10 Secondary outcome
ICU length of stay (days) 6.5 ± 6.6 6.2 ± 6.2 Mean diff 0.24 95% CI −0.06 to 0.54; P=0.44 Secondary outcome
Hospital length of stay (days) 15.3 ± 9.6 15.6 ± 9.6 Mean diff −0.24 95% CI −0.70 to 0.21; P=0.30 Secondary outcome
Duration of mechanical ventilation (days) 4.5 ± 6.1 4.3 ± 5.7 Mean diff 0.19 95% CI −0.08 to 0.47; P=0.74 Secondary outcome
Duration of renal-replacement therapy (days) 0.48 ± 2.28 0.39 ± 2.0 Mean diff 0.09 95% CI −0.0 to 0.19; P=0.41 Secondary outcome
New organ failure (0–5 organs; distribution) 0: 1397 (52.7%)
1: 795 (30.0%)
2: 369 (13.9%)
3: 68 (2.6%)
4: 18 (0.7%)
5: 2 (0.1%)		 0: 1424 (53.3%)
1: 796 (29.8%)
2: 361 (13.5%)
3: 75 (2.8%)
4: 17 (0.6%)
5: 0		 Not reported P=0.85 Assessed by SOFA; data available for 2649 (albumin) and 2673 (saline)
28-day mortality (predefined subgroup: trauma) 81/596 (13.6%) 59/590 (10.0%) RR 1.36 95% CI 0.99 to 1.86; P=0.06 Heterogeneity vs non-trauma: P=0.04 (“test for a common relative risk”)
28-day mortality (predefined subgroup: severe sepsis) 185/603 (30.7%) 217/615 (35.3%) RR 0.87 95% CI 0.74 to 1.02; P=0.09 Heterogeneity vs no severe sepsis: P=0.06 (“test for a common relative risk”)
28-day mortality (predefined subgroup: ARDS) 24/61 (39.3%) 28/66 (42.4%) RR 0.93 95% CI 0.61 to 1.41; P=0.72 Heterogeneity vs no ARDS: P=0.74 (“test for a common relative risk”)
28-day mortality (trauma with traumatic brain injury definition) 59/241 (24.5%) 38/251 (15.1%) RR 1.62 95% CI 1.12 to 2.34; P=0.009 Post-hoc refinement of trauma subgroup; excess deaths in albumin group = 21
28-day mortality (trauma without brain injury) 22 (6.2%) 21 (6.2%) RR 1.00 95% CI 0.56 to 1.79; P=1.00 Denominators not explicitly stated in the manuscript text for this comparison
  • Overall 28-day mortality was essentially identical: 20.9% with albumin vs 21.1% with saline (RR 0.99; 95% CI 0.91 to 1.09; P=0.87).
  • Albumin achieved higher serum albumin over days 1–4 (e.g., day 2: 30.8±6.4 g/L vs 24.5±5.9 g/L; P<0.001) and higher CVP (e.g., day 2: 11.6±4.9 mm Hg vs 10.4±4.3 mm Hg; P<0.001) while requiring less study fluid (day 2: 602.7±892.7 mL vs 954.0±1484.4 mL; P<0.001).
  • A clinically important subgroup signal emerged in trauma with brain injury (RR 1.62; 95% CI 1.12 to 2.34; P=0.009), contrasted with no signal in trauma without brain injury (RR 1.00; 95% CI 0.56 to 1.79; P=1.00).

Internal Validity

  • Randomisation and allocation
    • Central web-based randomisation with minimisation; stratified by ICU and trauma status.
    • Allocation concealment was robust due to central randomisation and masked identical bottles/sets.
  • Dropout / exclusions (post-randomisation)
    • Vital status at 28 days unavailable for 67/7000 (1.0%): 26 (albumin) vs 41 (saline); 56/67 due to withheld/withdrawn consent.
    • Study fluid administered to all but 197 patients (2.8%): 90 (albumin) vs 107 (saline).
  • Performance / detection bias
    • Double blinding and objective primary outcome (mortality) substantially limit bias.
    • Secondary outcomes (length of stay, ventilation, RRT days) are relatively objective but could still be influenced by local practice; randomisation stratified by site reduces systematic imbalance.
  • Protocol adherence / separation of the variable of interest
    • Resuscitation with additional nonstudy fluid occurred in 309 (8.8%) albumin vs 375 (10.7%) saline; clinician preference accounted for 68 (1.9%) vs 103 (2.9%).
    • Study fluid volumes differed early (demonstrating biological separation): day 1: 1183.9±973.6 mL vs 1565.3±1536.1 mL (P<0.001); day 2: 602.7±892.7 mL vs 954.0±1484.4 mL (P<0.001); day 3: 268.0±554.5 mL vs 348.3±753.5 mL (P=0.03).
    • Net positive fluid balance was lower with albumin early: day 1: 1543.6±1619.7 mL vs 1990.5±2061.7 mL (P<0.001); day 2: 1015.3±1826.9 mL vs 1505.1±2215.9 mL (P<0.001); day 3: 422.1±1633.3 mL vs 553.0±1732.3 mL (P=0.007).
    • Physiological separation supported mechanistic plausibility: serum albumin day 2: 30.8±6.4 g/L vs 24.5±5.9 g/L (P<0.001); CVP day 2: 11.6±4.9 mm Hg vs 10.4±4.3 mm Hg (P<0.001).
    • Haemodynamic endpoints were broadly similar: MAP day 2: 84.4±15.1 mm Hg vs 84.2±15.7 mm Hg (P=0.49).
  • Baseline characteristics / illness severity
    • Groups were closely balanced for age (58.6±19.1 vs 58.5±18.7), APACHE II (18.7±7.9 vs 19.0±8.0), mechanical ventilation at baseline (63.8% vs 64.8%), and baseline serum albumin (27.4±7.8 vs 27.7±7.9 g/L).
    • A small baseline imbalance: CVP was higher in the albumin group at baseline (9.0±4.7 vs 8.6±4.6 mm Hg; P=0.03), with unclear clinical relevance.
  • Heterogeneity
    • Pragmatic enrolment across multiple ICUs with broad case-mix increases clinical heterogeneity but improves estimate stability for average treatment effect.
    • Predefined subgroup analyses were performed, but the trial was not powered for modest subgroup-specific effects.
  • Timing and dose
    • Timing of initiation was aligned with clinician-recognised need for resuscitation; study fluids used for ICU resuscitation episodes for up to 28 days.
    • Albumin and saline are not equipotent volume expanders; albumin required less administered volume (overall first 4 days albumin:saline ≈1:1.4), consistent with greater oncotic effect.
  • Adjunctive therapies
    • Packed red cells were higher in the albumin group in the first two days: day 1: 97.8±360.7 mL vs 71.7±296.8 mL (P<0.001); day 2: 106.5±321.4 mL vs 61.1±235.2 mL (P<0.001); average excess over 4 days 71.0 mL per patient.
    • Nonstudy fluid was slightly higher in the saline group on day 2: 2707.3±1435.7 mL vs 2615.9±1372.5 mL (P=0.009).
  • Outcome assessment and statistical rigour
    • Mortality is objective; survival curves were similar (log-rank P=0.96).
    • Preplanned interim analyses and predefined subgroups were reported; multiple comparisons for subgroups were not adjusted, consistent with hypothesis-generating intent.

Conclusion on Internal Validity: Overall, internal validity appears strong for the primary outcome given concealed central randomisation, effective blinding, objective endpoint, and minimal missing outcome data; interpretability is more limited for subgroup signals due to multiplicity and lower power.

External Validity

  • Population representativeness
    • Broad adult ICU case-mix in closed multidisciplinary ICUs; baseline APACHE II ~19 suggests moderate-to-severe critical illness.
    • Included meaningful proportions of trauma (~17%) and severe sepsis (~18%), and a smaller ARDS subgroup (~2%).
    • Excluded key populations (explicitly): burns, post-cardiac surgery, post-liver transplantation; paediatric patients were not included.
  • Applicability
    • Findings generalise best to adult ICU resuscitation decisions where the clinician considers either fluid acceptable.
    • Intervention tested was 4% albumin; results do not directly address 20–25% albumin strategies or balanced crystalloids as the crystalloid comparator.
    • Because fluids outside ICU were not controlled, translation to ED/pre-hospital settings or peri-operative pathways requires caution.

Conclusion on External Validity: External validity is high for adult ICU resuscitation in comparable high-income settings, but limited for excluded populations (burns, post-cardiac surgery/transplant, paediatrics) and for other albumin concentrations or crystalloid formulations.

Strengths & Limitations

  • Strengths:
    • Large sample size (~7000 planned; 6933 with known 28-day vital status) enabling precise estimation of average treatment effect.
    • Double-blind design with robust masking procedures; objective primary outcome.
    • Pragmatic protocol embedded in routine ICU practice, enhancing applicability.
    • Clear separation in biological/physiological markers (serum albumin, CVP) and administered volume/net balance.
  • Limitations:
    • Not powered to detect modest but clinically relevant effects within predefined subgroups; multiple subgroup comparisons increase false-positive risk.
    • Only 4% albumin tested; generalisability to hyperoncotic albumin or albumin-targeting strategies is uncertain.
    • Resuscitation fluids outside ICU were not controlled.
    • No formal cost-effectiveness evaluation within the trial.
    • Follow-up limited to 28 days; longer-term functional outcomes (particularly for brain injury) not assessed.

Interpretation & Why It Matters

  • Practice-changing message
    • In a heterogeneous adult ICU population, 4% albumin and 0.9% saline produced clinically equivalent 28-day mortality (RR 0.99; 95% CI 0.91 to 1.09).
    • This supports crystalloids as an acceptable default resuscitation fluid when albumin is being considered primarily for volume expansion.
  • Physiology vs outcomes
    • Albumin achieved similar haemodynamic endpoints with less administered study volume and a lower early net positive balance, but without measurable improvement in hard outcomes (mortality, organ failure, length of stay).
  • Subgroup signals
    • Trend towards benefit in severe sepsis (RR 0.87; 95% CI 0.74 to 1.02; P=0.09) was hypothesis-generating.
    • Signal towards harm in trauma with brain injury (RR 1.62; 95% CI 1.12 to 2.34; P=0.009) prompted focused follow-up and ultimately influenced guideline recommendations.

Controversies & Subsequent Evidence

  • Pre-trial controversy: a 1998 systematic review reported increased mortality with albumin, motivating the need for a definitive blinded RCT; SAFE directly addressed this uncertainty in contemporary ICU care 1.
  • The accompanying editorial highlighted SAFE’s methodological strengths and interpreted the overall effect as no evidence of benefit or harm of 4% albumin compared with saline in general ICU resuscitation 2.
  • Correspondence raised three recurring critiques: (i) heterogeneous case-mix could mask subgroup effects; (ii) baseline haemodynamics (e.g., CVP ~9 mm Hg) might indicate earlier or less severe volume depletion than implied by “resuscitation”; (iii) the sepsis subgroup trend warranted dedicated trials; the authors replied that contemporary practice resuscitates early, and that subgroup hypotheses required appropriately powered RCTs 3.
  • The trauma/brain injury signal was debated as a chance subgroup finding within SAFE; a subsequent detailed analysis in traumatic brain injury reported higher mortality with albumin, shifting practice away from albumin in this population 4.
  • Follow-up sepsis-specific evidence: ALBIOS tested albumin replacement in severe sepsis/septic shock and did not demonstrate an overall mortality benefit, providing counterweight to SAFE’s sepsis trend signal 5.
  • Broader fluid-choice evidence: CRISTAL (colloids vs crystalloids) did not show 28-day mortality benefit but suggested differences at longer horizons, reinforcing that “fluid class” effects can be outcome- and time-dependent 6.
  • Guidelines incorporating SAFE-era evidence commonly recommend crystalloids as first-line resuscitation fluids and reserve albumin for selected contexts (e.g., persistent shock after large crystalloid volumes in sepsis), while avoiding albumin in traumatic brain injury populations 7891011.

Summary

  • SAFE compared 4% albumin vs 0.9% saline for ICU fluid resuscitation in a large, double-blind, pragmatic adult ICU RCT.
  • There was no difference in 28-day mortality: 20.9% vs 21.1% (RR 0.99; 95% CI 0.91 to 1.09; P=0.87).
  • Albumin achieved similar haemodynamic endpoints with less study fluid volume and lower early net positive fluid balance, without improvements in organ failure or length-of-stay outcomes.
  • Subgroup signals were hypothesis-generating: trend towards benefit in severe sepsis (RR 0.87; P=0.09) and signal towards harm in trauma with brain injury (RR 1.62; P=0.009).
  • The trial reset ICU fluid practice: crystalloids remained default; albumin became a selective adjunct (notably avoided in traumatic brain injury).

Further Reading

Other Trials

Systematic Review & Meta Analysis

Observational Studies

Guidelines

Notes

  • SAFE tested 4% albumin versus 0.9% saline (not balanced crystalloids) and did not address hyperoncotic albumin strategies.
  • Subgroup findings were not powered for definitive inference; later dedicated evidence influenced practice most strongly for traumatic brain injury and sepsis-specific questions.

Overall Takeaway

SAFE established that 4% albumin and 0.9% saline yield equivalent 28-day mortality when used as ICU resuscitation fluids in a broad adult ICU population, decisively moving the field beyond conflicting small-study meta-analyses. The trial’s enduring impact is twofold: it normalised crystalloids as the default resuscitation strategy while simultaneously identifying subgroup hypotheses—particularly potential harm in traumatic brain injury and potential benefit in sepsis—that shaped subsequent trials and guideline recommendations.

Overall Summary

  • In adult ICU resuscitation, 4% albumin was neither superior nor inferior to 0.9% saline for 28-day mortality, but the trauma-with-brain-injury signal changed practice and research priorities.

Bibliography