The recent RCT comparing transfusion thresholds of 25,000 per mm3 to 50,000 in very preterm babies has generated a great deal of discussion. The result was somewhat unexpected, I think. I would not have been surprised to find that there was no impact on hemorrhage, but the increase in mortality surprised me.
Curley A, et al. Randomized Trial of Platelet-Transfusion Thresholds in Neonates. NEJM. 2018. The PlaNet2 MATISSE trial.
We should always be circumspect when evaluating trials that have a composite primary outcome, in this case “death or serious bleeding’. Does it make sense to put the different parts of the outcome together? Are they likely to be of approximately similar frequency, and of approximately similar importance to families? Are they likely to change in the same direction, and if they do not, then how will we interpret the results?
I think this composite outcome has some validity, especially if the definition of serious bleeding is restricted to bleeds that are potentially life-threatening.
When we examine the results there were fewer babies with the primary outcome in the low threshold group, 19%, compared to the high threshold group, 26%. This outcome was made up of death (low threshold 33/330 (10%), versus higher threshold 48/326 (15%) OR, 1.56 (95% CI 0.95–2.55)) and major bleeding episodes (low threshold 35/330 (11%) versus higher threshold 45/328 (14%) Hazard Ratio, 1.32 (95% CI 1.00–1.74))
Remember, however, that that is only mortality up to 28 days after study enrolment, we don’t know if mortality by the time of discharge was different between groups. We do know from the data in table 2 that by 36 weeks, when the outcome of BPD was determined, there were 50 deaths in the low threshold group, and 60 deaths in the high transfusion group, which is probably consistent with a chance effect. It shows also that between 28 days of age, and 36 weeks postmenstrual age (if we assume that the deaths were among babies less than 32 weeks gestation) there were 17 deaths in the low threshold, 25,000, group, and 12 deaths in the higher threshold, 50,000, group.
There are one or two other things about the trial that are a little confusing, only 90% of the babies in the high threshold group actually had a transfusion, whereas to be in the trial the babies had to have a platelet count under 50,000, so surely 100% of them should have had a transfusion, in the article it is noted that the platelet counts increased to above 50,000 while they were waiting for the platelets, but, once I have decided to give platelets we almost always have them in the NICU to be given within an hour, so it is very unusual to have another platelet count between ordering the platelets and giving them, so I find that a little confusing. In the article it is noted that there were 94 platelet transfusions in the high threshold group that should have been given and were not (and 30 in the low threshold group). There is a per-protocol analysis in the online supplementary materials, which shows almost identical differences between the 2 groups when only those babies who actually followed the protocol are included: mortality 10% vs 14%, and major bleed 9% vs 14%.
Another question to ask is whether or not this is biologically plausible. I think it is, platelet transfusions are pro-inflammatory, they have been associated with TRALI (transfusion related lung injury) in adults, and platelets prepared for transfusion release CD40 ligand (whatever that is) which activates cyclo-oxygenase. So there are potential pathways for adverse effects, which could be diverse and could potentially increase lung injury and other complications. In adults platelet transfusions for thrombocytopenia have also been associated with excess thrombotic complications, again a potential risk for our preterm patients.
My final question is the following : what is MATISSE? I understand the acronym PLaNEt2, even though I can’t remember which letters are capitalised, but what on earth is MATISSE doing there?
In the supplementary data we can see that infants who were IUGR did not have any difference in their incidence of the primary outcome (18% vs 20%) between thresholds, but beware, the p-value for the interaction was 0.3, so we shouldn’t make too much about that as a subgroup analysis, also the overall incidence of the primary outcome was not much different from the babies who were not growth restricted. Of course IUGR babies can get sepsis and NEC also, so those babies weren’t necessarily in the trial because of early onset thrombocytopenia associated with IUGR. Those babies with IUGR, maternal hypertension, and/or placental insufficiency are often thrombocytopenic in the first days of life, and have a reduced proportion of immature platelets. Those with consumptive coagulopathy and an increased proportion of immature platelets often have other derangements of coagulation, and I think are probably at higher risk of bleeding. If the authors have the data it would be fascinating to see whether babies with low production of platelets, due to placental insufficiency, have a different incidence of complications, and a different response to platelet transfusions, to those with platelet consumption.
I think we should be careful making excessive claims for a study with a result like this, especially when the 2 parts of a composite outcome are both barely at the threshold of what is classically considered to be statistically “significant”. On the other hand, as I mentioned before, these are the only reliable data we have, and there is no sign at all of a benefit to transfusion at a higher threshold. So for preterm babies who are not actively bleeding, I for one will no longer transfuse unless they fall below 25,000.
Great thoughts, particularly on the use of the composite outcome. In supplemental Table 3, authors provide serious adverse event rate (including death) till discharge. Interestingly, rate of death to discharge in both groups seem similar (54/331 deaths in low threshold and 58/329 in high threshold). This adds some difficulty in interpreting the death outcome. Why would PLT transfusion increase your risk of death within 28 days and not at day 29-30 (or after).
Also, median age at enrollment was 7 days (IQR 4-20) (and based on data of median GA at birth vs median CGA at randomization, one assumes that the more preterm infants <26 wks likely entered the trial at higher postnatal age) – so not quite sure what to do with the 4days after delivery – and avoiding non-beneficial interventions is better.