 |
|
 |
|
 |
 |
|
 |
 |
 |
 |
 |
 |
 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Drs Swamy and Embleton express concern that we may have overinterpreted the findings of our recently published study comparing liberal and restrictive transfusion guidelines in preterm infants. Granted, the primary outcome of the study was the number of red blood cell transfusions, and the sample-size estimation (and therefore the stopping point) for the trial was based on the difference in number of transfusions. However, the secondary outcomes of apnea frequency and severity and brain ultrasound findings were planned from the time that the study was designed. One must always exercise caution in the interpretation of differences in secondary outcomes, and we have tried to do so. The excess of serious brain abnormalities by ultrasound in the restrictive-transfusion group is an important finding, and it would have been irresponsible of us not to report and discuss this finding. The possibility of a type I error remains, however, and we were the first to acknowledge this. However, the probability that the association of restrictive transfusion practice is false and occurred purely by chance is only 1.2% based on our data; this finding would be expected to occur by chance only 1 in 83 times.
We will address other specific points raised by Swamy and Embleton. Infants with significant patent ductus arteriosus (PDA) were excluded from enrollment for 2 reasons. First, we were concerned that liberal transfusion of these infants might make management of their PDA more difficult, increasing the likelihood that they would need surgery to close the PDA. Second, the calculation of cardiac output would have been complicated by significant systemic-to-pulmonary shunting through a PDA. Although this exclusion affects the generalizability of the results of our study, as suggested by Swamy and Embleton, it is unlikely that liberal transfusion would have been found advantageous among infants with significant PDA.
Swamy and Embleton correctly point out that stratification by gender would have prevented the slight imbalance in gender between the treatment groups. Prognostic stratification is generally applied to assure balance of factors that are known to independently affect the primary outcome. We stratified by birth weight, which is also a proxy for gestational age, because this is the most important factor known to influence the number of transfusions needed by small preterm infants (our primary outcome). Gender is not known to influence the need for transfusions, nor is it known to influence secondary outcome, apnea, or adverse neurologic events. The slight excess of males (P = .049) in the restrictive-transfusion group occurred by chance. With random allocation of treatment assignment, it is expected that 1 of 20 demographic factors will be different between the treatment groups. We randomized within each of the 3 birth weight strata in blocks of 4, which was done to assure that the total numbers of subjects in the 2 treatment groups would be approximately equal. Stratifying by gender as well as birth weight would have doubled the number of strata to 6 and also would have increased the likelihood that the treatment groups would have been unequal in size, which would have compromised the statistical power of the study. We offer this information to explain our decision not to stratify by gender in addition to birth weight. The issue remains that the slight excess of males in the restrictive-transfusion group may have influenced the results of the study, although gender is not known to affect the risk of intraventricular hemorrhage or periventricular leukomalacia.
Swamy and Embleton claim that our stratification of subjects by birth weight precluded balance by gestational age. On the contrary, birth weight is a very good surrogate marker of gestational age, and the treatment groups were perfectly balanced by gestational age with means of 27.8 and 27.7 weeks.
Swamy and Embleton note that not all of our subjects underwent late head ultrasound examination for periventricular leukomalacia (PVL). Of the infants with late head ultrasound examinations, 4 of 28 infants in the restrictive-transfusion group had PVL, whereas none of the 24 infants in the liberal-transfusion group had PVL. The statistical significance of this difference reported in our article was P = .115. When we recalculate the significance now, without correcting for the interim analysis, which was not performed for secondary outcomes, the significance (Fisher's exact test) is .076. In calculating the statistical significance of this difference, we made no assumption about the incidence of PVL in the infants who did not have late head ultrasound examinations. Had we assumed no PVL in any of the infants in either treatment group without late head ultrasound examinations, the significance level would have been .064. Had we assumed that the incidence was the same in the unexamined infants as it was in those examined (14% vs 0%), the significance level would have been .007. Thus, the statistical approach to examining differences in PVL in our published report provided the most conservative interpretation.
Swamy and Embleton point out that we allowed up to 2 transfusions before enrollment and assignment of treatment group. This was done as a concession to the challenge of obtaining valid informed consent from anxious parents in the first day or two after the birth of a preterm infant. It also reflects the reality that some of these infants need very early transfusions. The mean number of transfusions before enrollment was not significantly different: 0.4 in the liberal-transfusion group and 0.6 in the restrictive-transfusion group. Our inclusion of infants who received 1 or 2 transfusions before enrollment allows the results of the study to be generalized to include such infants.
Swamy and Embleton comment on the fact that the infants in our liberal-transfusion group were more likely to have grade 3 intraventricular hemorrhage but less likely to have grade 4 hemorrhage. We interpret this to mean that the impact of transfusion practice was on the severity but not the incidence of significant (grade 3 or 4) intraventricular hemorrhage.
In their final paragraph, Swamy and Embleton describe our study as having shown that restrictive transfusion practice minimizes transfusion rates and donor exposures. In fact, we found reduced transfusions but no reduction in donor exposures with restrictive transfusion practice. We attribute this finding to our use of a single-donor transfusion program.
We thank Swamy and Embleton for their careful reading of our article and for giving us the opportunity to discuss our findings and their significance further.
Related articles in Pediatrics:
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
 |
||
|
||
Home | My Pediatrics | Journal Information | Current Issue | Past Issues | Subscriptions & Services | Contact Us Click here for faster international access © 2006 American Academy of Pediatrics. All rights reserved. |
||
|
||
