TITRe2: Liberal or Restrictive Transfusion after Cardiac Surgery

Murphy. NEJM 2015;372(11):997-1008. doi:10.1056/NEJMoa1403612

Clinical Question

• In adults undergoing cardiac surgery, does a restrictive transfusion strategy (Hb > 75 g/l) compared to a liberal transfusion strategy (Hb > 90 g/l) lead to fewer infections and ischaemic events within 3 months?

Design

• Multi-centre, randomised, controlled trial
• Secure online allocation process with complete concealment
• Allocation minimisation to balance assignment by centre and surgery
• Clinicians not blinded, but patients and investigators were
  • Patients were asked if they thought they knew their allocation at:
    • discharge:  15.1% believed they did and 75.6% were correct
    • 3 months: 27.5% believed they did and 56.6% were correct
• Intention-to-treat analysis
• Power analysis concluded 1468 patients needed to have 90% power with two-sided 5% level of significance
  • Increased to 2000 to allow for non-adherence
  • Based upon estimate of primary outcome of:
    • restrictive group 11%
    • liberal group 17%

Setting

• 17 cardiac surgery centres across the UK
• July 2009 to February 2013

Population

• Inclusion: over 16 years undergoing non-emergency cardiac surgery
  • Patients recruited and consented prior to surgery but only randomised and included if their Hb was recorded at below 90 g/l at any post-operative time
• Exclusion: patients unwilling to receive blood products (e.g. Jehovah’s Witnesses), congenital or acquired platelet, red cell or clotting disorders, sepsis, critical limb ischaemia, emergency surgery (occurring on same day as admission)
• 3565 consented, 2007 became anaemic and were randomised, 2003 analysed

Intervention

• Restrictive transfusion-threshold
  • Threshold of 75 g/l
  • 1 unit packed red cells transfused if threshold breached
  • Repeated as needed if remained or dropped below 75 g/l again

Control

• Liberal transfusion-threshold
  • Threshold of 90 g/dl
  • 1 unit packed red cells transfused immediately after randomisation
  • Repeated as needed if remained or dropped below 90 g/l again

Treatment common to both groups

Outcome

• Primary outcome: there was no statistically significant difference in the composite outcome of serious infections or ischaemic events within 3 months
  • This outcome required the event to occur within the index hospital admission, or to have resulted in re-admission, or through detection by postal or phone survey of patients at home
  • Restrictive transfusion group 35.1% vs Liberal transfusion group 33.0%
    • Odds ratio 1.11 (95% CI 0.91 to 1.34); p-value 0.30
    • Absolute risk increase 2.11% (95% CI -2.14% to 6.36%)
    • Number needed to harm (by a restrictive transfusion practice) is 48
• Secondary outcome: the only secondary outcome (other than the dependent variable of red cell transfusions) to demonstrate a statistically significant difference was death at 90 days (according to a Cox proportional regression analysis) which favoured a liberal transfusion strategy
  • Red cells transfused: liberal 95% vs restrictive 64%
    • Number needed to prevent transfusion (by a restrictive transfusion practice) is 4
  • Occurrence of infection: liberal 25% vs restrictive 25%
  • Occurrence of an ischaemic event: liberal 14% vs restrictive 16%
  • Median duration of ICU stay, HDU stay and hospital stay (see table)
  • All cause mortality at 90 days: liberal 1.9% vs restrictive 2.6%
    • p-value 0.045
    • Number need to harm (by a restrictive transfusion practice) is 63
  • Significant pulmonary complications: liberal 12% vs 13%
  • EQ-5D health status: no statistical difference between groups or before and after surgery!
• Tertiary data:
  • Severe non-adherence to protocol: liberal 6.2% vs restrictive 9.7%
  • Proportion of patients that suspected they knew their allocation at 3 months: liberal 26% vs restrictive 29%
    • The proportion of these that were correct: liberal 45% vs restrictive 68%

Authors’ Conclusions

• The authors concluded that a restrictive transfusion strategy did not demonstrate clinical (or financial – data not presented here) superiority over a liberal transfusion strategy.

Strengths

• Important clinical question in a population not previous investigated through studies like TRICC, TRISS or Villanueva
  • Patients in this trial had cardiovascular disease (although they were ‘treated’ for this during the index admission)
  • Therefore, it might be reasonable to extrapolate these findings to unwell patients with cardiovascular disease (such as acute coronary syndrome)
• Very well planned and executed clinical research
• Attempts to blind pragmatically, and effort to investigate how successful this was
• Intention-to-treat, allowing clinicians to contravene allocation on clinical grounds
  • Sensitivity analysis (data not presented here) investigated impact of this and found no significant effect on conclusion
• Considerable effort made to follow-up and detect outcome events in patients at home (leading to a rate double that of the anticipated rate)
• Transparent data available in detailed supplementary data

Weaknesses

• Composite outcome – can lead to clinically less significant but common outcomes being very influential on the conclusion, whilst clinically more significant but rarer outcomes are over-looked
  • The secondary breakdown of events suggests this isn’t the case in this trial, since all sub-groups demonstrate little difference between the groups
• Un-blinding of patients – the difference in practice (give blood early vs give blood later) appears to have led to un-blinding of patients
  • More patients in the restrictive group correctly identifying their allocation
  • The authors attempted to collect objective outcomes using blinded assessors, limiting the bias that could occur if patients believed they knew (and correctly knew) their allocation
  • The published supplementary data suggests that patients believed a liberal threshold was beneficial, thus the un-blinded restrictive-threshold patients may have biased their reporting of symptoms
  • It is unclear exactly what impact, if any, this had on the conclusion
  • It is likely that other trials have had this problem, but it has not been investigated or reported as succinctly as this trial has done so
• Number needed to prevent transfusion of 4 – although no clinical or financial differences were found in this trial, there are unmeasured consequences of transfusion
  • If all measured factors are equal, a number needed to prevent transfusion of 4 may be very relevant to individuals who wish to avoid transfusion
• The only statistically significant difference was death at 90 days – if all other measured outcomes are almost equal, what is causing the difference in death at 90 days?
  • More commonly, trials find a difference in morbidity but not in mortality, so this is an oddity
  • Statistically, there is ~5% chance it is a false positive
  • The authors suggest that the small numbers that died in this trial prevent a cause-and-effect conclusion being drawn, but that the trend is supportive of a theory that warrants further investigation

The Bottom Line