• In adults with severe sepsis or septic shock, does 20% albumin solution with crystalloid fluid compared to crystalloid fluid alone reduce death with 28 days?

• Randomised, controlled clinical trial
• Stratified according to participating ICU and time between onset of sepsis and randomisation (<6 hours vs >6 hours)
• Open-label (non-blinded)

• 100 intensive care units in Italy
• August 2008 to February 2012

• Inclusion: adults who met the clinical criteria for severe sepsis (SIRS + proven or suspected infection + at least one organ dysfunction as measured by SOFA score)
• Exclusions: Head injury, congestive heart failure, over 24 hours since inclusion criteria met, objection to blood products, albumin indicated for specific clinical reason.
• 1818 participants

• 20% albumin and crystalloid solution
  • 300ml of 20% albumin daily to maintain serum albumin at 30 g/l or more
  • Crystalloid as clinically indicated at physician’s discretion
  • 20% albumin accounted for median daily average of 4.3% of the total administered fluid
  • Median total daily fluid was 3,738ml for first 7 days (compared with control this was a non-statistically significant difference [p=0.10])

• Crystalloid as clinically indicated at physician’s discretion
• Median total daily fluid was 3,825ml for first 7 days

• Primary outcome: mortality at 28 days was not different between groups
  • 31.8% in albumin group versus 32.0% in crystalloid group; RR 1.00 (CI 0.87–1.14; p=0.94).
• Secondary outcome: mortality at 90 days was not different between groups

• The addition of albumin to crystalloids during the first 28 days of treatment to maintain a serum albumin level of 30 g/l or more is safe, but does not provide a survival advantage over crystalloids alone over a follow-up period of 90 days

• Pragmatic study directly relevant to clinical practice.
• Only 29 participants lost-to-follow-up (1.6%), which is well below primary outcome rate (minimal attrition bias).

• Open-label, non-blinded methodology may have influenced clinicians’ decisions, which may have biased toward either intervention or control.
• Lower incidence of death compared to anticipated rate (45% used in power calculation) may have lead to type 2 error (failure to detect a true positive outcome).

The Bottom Line

Links

Full text pdf not available without subscription / abstract / doi: 10.1056/NEJMoa1305727

Editorials, Commentaries or Blogs

• None listed

Metadata

Summary author: @DuncanChambler
Summary date: May 2014
Peer-review editor: @stevemathieu75