Uncertain benefits, certain harms

Abstract

The AABB recently published a guideline conditionally recommending a liberal transfusion strategy in acute myocardial infarction (AMI) using grading of recommendations assessment, development, and evaluation (GRADE) methodology.1 This recommendation will increase the number of transfusions in AMI and accordingly increase transfusion-associated harms. With no statistically significant effects observed on primary outcomes in trials or the meta-analysis of transfusion thresholds in AMI, the compensatory benefit to doing so is unclear. There have been two large trials of restrictive and liberal transfusion in acute myocardial infarction, restrictive and liberal transfusion strategies in patients with acute myocardial infarction (REALITY) and myocardial ischemia and transfusion (MINT), both of which had composite primary endpoints that did not reach commonly accepted significance thresholds.2, 3 In the meta-analysis of these two trials and two additional smaller trials, the AABB guideline panel identified that the point estimate of the effect of a liberal transfusion threshold on 30-day mortality was a decrease of 1.2%.1 This was greater than the panel's consensus of 1% as the minimum important difference (MID) in mortality that an individual patient would consider meaningful. While the point estimate of the confidence interval is the most likely effect, the confidence interval of the odds ratio from the meta-analysis includes 1.0, which would typically be interpreted as a statistically nonsignificant result. This is similarly described in the meta-analysis's evaluation of the overall effect on 30-day mortality, which had a p-value of .16.1 For a recommendation with the potential to radically change transfusion practice, there remains a high level of statistical uncertainty regarding any potential benefit. Requiring a p-value of .05 or less for statistical significance is a historical standard originally selected to reduce the risk of misidentifying chance associations as real, otherwise known as type I error. Utilizing alternative, more permissive criteria for statistical significance, including substituting our judgment of potential clinical significance in its stead, will increase the rate of type I error. The risk of type I error also increases in proportion with the number of statistical tests on a population, also known as the problem of multiple comparisons. The most recent AABB guidelines on red blood cell transfusion identified 30 trials of transfusion thresholds with mortality data,4 not including MINT or several other recent transfusion threshold trials in intracranial hemorrhage. This extended family of inferences in transfusion threshold research is large, with multiple statistical tests repeated on similar populations. From this perspective, a p-value of .05 is more likely to be inappropriately permissive, rather than too strict. Another way to assess the risk of type I error for this recommendation is simulation. If we instead presume that transfusion threshold has no effect on mortality in AMI, consistent with the established prior for other patients,4 a Monte Carlo approach using total patient numbers and event rates mirroring the meta-analysis simulating a 4311 patient trial with 8.7% mortality occurring at random produces a 1% or greater decrease in mortality in one arm in 24,387 of 100,000 iterations (Figure 1). If there is no actual difference in mortality between liberal and restrictive transfusion arms in AMI, utilizing GRADE methodology would still result in a recommendation for a specific transfusion strategy in AMI purely by chance about 24% of the time. There is already evidence of an increase in transfusion-associated harms with a liberal red cell transfusion strategy in this population. The MINT trial reported that 4 patients in the liberal arm had an acute hemolytic transfusion reaction, 6 had transfusion-related acute lung injury (TRALI), and 1 had transfusion-associated sepsis, versus 0 patients in the restrictive arm reported having any of these potentially fatal transfusion-associated events, as well as 2 patients in each arm with a reported anaphylactic transfusion reaction. This is a statistically significant increase in these rare, serious transfusion-associated harms affecting 13 out of 1754 eligible patients in the liberal arm versus 2 out of 1748 in the restrictive arm (p = .0073, Fisher's exact test), as well as a significant increase in transfusion-associated circulatory overload (TACO) in the liberal arm identified by the authors of MINT.3 In the meta-analysis, the AABB guideline panel defined transfusion-associated serious adverse events (SAEs) as TACO, TRALI, or anaphylaxis. For the purposes of the recommendation, the AABB guideline panel considered a 2.5% absolute difference in SAEs to be clinically significant as the MID. While there was a statistically significant increase in SAEs seen with more liberal transfusion, the point estimate of that increase was 1.4%, below the MID.1 Selecting an MID for SAEs at this level has led the AABB to publish a guideline that includes the statement “A liberal transfusion threshold possibly has no important effect on serious adverse events.” While the internal logic of this recommendation is sound in the context of the MID, it is also obviously externally inconsistent with other statements of the AABB as well as the entire thrust of the patient blood management movement on reducing the harms of transfusion. Transfusion medicine providers are not unfamiliar with clinical situations where we know a transfusion can hurt and are uncertain about whether it may help. We are routinely involved in individualized decisions surrounding transfusion in the context of guidelines that have historically trended toward more restrictive use of blood products.4 Having these conversations in the context of this new guideline promoting more liberal transfusions for certain patient populations will be a novel experience and may be perceived as a medical reversal by our clinical colleagues, damaging their trust in our ability to make recommendations that limit transfusion-related harms in other contexts. This new AABB guideline1 goes beyond commonly accepted thresholds of statistical significance in order to give a clinical recommendation in favor of a more liberal transfusion threshold in AMI. This is a radical change to transfusion practice based on statistically nonsignificant results. Without additional information on transfusion in AMI, we will most likely remain uncertain about who benefits from this recommendation. But as reports of rare events accrue over time, we will eventually be certain about those harmed by it. All data analysis for this letter was performed using the R statistical computing language.5 The author has disclosed no conflicts of interest. Data sharing not applicable to this article as no datasets were generated or analysed during the current study.

Ähnliche Arbeiten