PEDIATRICS Vol. 106 No. 2 August 2000, pp. 378-380

The Effects of Blood Transfusion Protocol on Retinopathy of Prematurity

To the Editor.

We wish to comment on the recently published article by Brooks et al evaluating the effect of anemia and transfusion on the incidence and severity of retinopathy of prematurity (ROP). Their manuscript appears to suffer from numerous statistical problems that render their data difficult to interpret. Some of these are listed below:

1. In Table 1, there were 21 infants in group 1 (no transfusion) and 26 infants in group 2 (transfusion), even though only 16 of group 1 and 18 of group 2 appeared to complete the study (Table 3). This partial follow-up introduces a serious potential for bias.
2. In several instances, large differences in the occurrence of ROP were present, but the study did not have a sufficient number of infants to detect a significant difference if one had existed (type II error). Figure 1 shows about twice the incidence of stage 3 ROP in group 1 infants as compared with group 2 infants. Given a total of 8 infants with stage 3 ROP, the statistical power of correctly rejecting the null hypothesis was only 8% at a 95% (P = .05) confidence level. To obtain the conventionally accepted 80% study power, 155 infants are needed in both group 1 and group 2. Furthermore, if only 2 infants had been clinically misassigned with regards to ROP diagnosis (one fewer with ROP in group 2 and one more in group 1), the p value for the overall incidence of ROP would have been .04 rather than .32. This is troubling.
3. Examination of Table 3 shows similar power problems. When stratified by birth weight, group 1 infants weighing <750 g had a 100% incidence of ROP as compared with a 60% incidence in group 2. Our power calculation for this difference was only 3%, although a clinically important observation if it were statistically valid (p = .18). Unfortunately, only 20 patients would have been required in each group to answer this important question.
4. The authors used analysis of variance for examining the number of transfusions, hemoglobin, and hematocrit. They should have used a nonparametric test because of the extremely small sample size and the fact that the number of transfusions may have not been normally distributed.

In view of the above, we would disagree with the authors' conclusion that "Despite a relatively small sample size (italics ours), our data suggest that transfusion policy per se is not a major factor in an infant's risk for developing ROP." Although the authors may in fact be correct, no clinically useful conclusions can be drawn about the effect of blood transfusion or anemia on the incidence or severity of ROP from the data presented. Unfortunately, they are not alone, because statistical inadequacy is a common problem with many clinical studies.¹

Richard A. Saunders
Storm Eye Institute/Medical University of Ophthalmology
Charleston, SC 29425-2236

Dilip Purohit
Professor of Pediatrics

Thomas C. Hulsey

REFERENCE

1. Javitt JC When does the failure to find a difference mean that there is none? Arch Ophthalmol. 1989; 107:1034-1040 [Abstract]

To the Editor.

We have just read the article by Brooks et al¹ in which they discuss the effect of blood transfusion protocol on retinopathy of prematurity (ROP). The authors conclude that "Despite a relatively small sample size, our data suggest that transfusion policy per se is not a major factor in an infants' risk for developing ROP". This affirmation is very important because it counteracts the results of previous paper: Hesse et al² studied 114 preterm infants and found, that blood transfusions are an independent risk factor for ROP. Inder et al³ studied 36 preterm infants weighing <1250 g at birth and showed that an elevation of serum iron and transferrin saturation at 7 days of age is associated with an increased risk of ROP; they could not ascertain an independent role of blood transfusions as a risk factor for ROP, but they found that the iron status and the amount of transfused blood are highly correlated factors.

It is difficult for us to evaluate the results of Brooks et al, who studied 34 preterm infants, in comparison with previous studies. It would be useful if the authors may clarify these points:

1. What is the real power of their study, considering that only 65% (16/25) of expected patients in group 1 and 72% (18/25) of expected patients in group 2 concluded the study? Why did they stop the study before that the recruitment of expected 25 patients, who had completed the study period, was concluded?
2. To compare the number of transfusions per se is insufficient. If we consider that the reported mechanisms by which blood transfusions could contribute to the development of ROP are 1) the increase of oxygen delivery to the retina⁴ and 2) the secondary iron overload (increased amounts of free iron may catalyze Fenton reactions, which produce free hydroxyl radicals),⁵ it is easily demonstrated that the volume of blood transfusions and not their number is important. In fact, Hesse et al found that there is a direct relationship between the volume of blood transfusions and risk of ROP.
3. The authors did not report data about respiratory distress syndrome severity and oxygen therapy requirements of study groups, while these notices is fundamental to draw any conclusions about the risk factors for ROP and should be reported at least to exclude great differences between the groups.
4. The incidence of ROP in the studied population is so high that it seems to be difficult to decrease it with a blood transfusion protocol, because of the multifactorial pathogenesis of ROP. The Italian ROP study group⁶ found that the ROP incidence in infants weighing less than 1250 g at birth was 49% (115/246). It could be useful if the authors try to explain such high ROP frequency.

In summary, we think that the good message of this study is that blood transfusion policy must be regulated by accurate protocol, while it does not report conclusive data about the role of blood transfusion as a risk factor for ROP.

Carlo Dani
Firmino F. Rubaltelli
Division of Neonatology
Careggi University Hospital
Viale Morgani 85
I-50134 Florence, Italy

REFERENCES

1. Brooks SE, Marcus DM, Gillis D, Pirie E, Johnson MH, Bhatia J The effect of blood transfusion protocol on retinopathy of prematurity: a prospective randomized study. Pediatrics. 1999; 104:514-518 [Abstract/Free Full Text]
2. Hesse L, Eberl M, Schlaud C, Poets F Blood transfusion. Iron load and retinopathy of prematurity. Eur J Pediatr. 1997; 156:465-470 [CrossRef][Medline]
3. Inder TE, Clemett RS, Austin NC, Graham P, Darlow BA High iron status in very low birth weight infants is associated with an increased risk of retinopathy of prematurity. J Pediatr. 1997; 131:541-544 [CrossRef][Medline]
4. James L, Greenough A, Naik S The effect of blood transfusion on oxygenation in premature ventilated neonates. Eur J Pediatr. 1997; 156:139-141 [CrossRef][Medline]
5. Gutteridge JMC Fate of oxygen free radicals in extracellular fluids. Biochem Soc Trans. 1982; 10:72-73 [Medline]
6. The Italian ROP Study Group Italian multicentre study on retinopathy of prematurity. Eur J Pediatr. 1997; 156:939-943 [CrossRef][Medline]

In Reply.

We appreciate Drs Saunders, Purohit, and Hulsey's interest in our article. We would like to respond to their comments, however, and hopefully clarify some of the issues they have raised. We will address each of the comments in order.

1. Table 1 shows data from 24 infants in group 1 and 26 infants in group 2. The loss of infants to follow-up was indeed a significant problem, and may be a problem for any prospective study involving neonatal intensive care unit patients, because deaths, transfers, and discharges from the hospital are difficult factors to control. However, the magnitude of the problem was remarkably similar for the 2 groups: 8 infants in group 1 (0 in birth weight (BW) <751 g, 4 in BW 751-1000 g, and 4 in BW 1001-1250 g) and 8 in group 2 (1 in BW <751 g, 3 in BW 751-1000 g, and 4 in BW 1001-1250 g) did not complete the full 6-week study period. Given this similarity, bias from this problem is likely to be minimal.
2. The authors of the letter express concern regarding the statistical power of our study, and the possibility of type II error. Because the sample size of our study was relatively small, it is clear that we would not have been likely to detect small differences between groups, let alone subgroups. However, our prospective, randomized, masked study was not designed to detect to small differences, but rather a large difference in ROP incidence between the 2 primary groups, as stated clearly in the "Methods" section. Fisher's exact test in Figure 1 tested the overall association between treatment group and stage of ROP, and was not applied to subgroups in that instance. We do agree that the small sample size of the BW subgroups might make the application of statistical methods misleading, including posthoc analysis, and therefore we specifically did not do this.
3. As stated in the preceding comments, the primary goal of the study was to assess the incidence of ROP in 2 cohorts of premature infants managed using 2 different transfusion protocols. The goal was not to provide subgroup analysis of the type Saunders et al mentions. The a priori power analysis and sample size determination was therefore based on the presence of any ROP versus no ROP in the 2 groups, regardless of BW. Stratification by BW during the randomization process was done to ensure that this important factor would be equally represented in the 2 groups. It is not surprising, therefore, that there is little power in the subgroup analyses, especially in the lowest BW category. Additional studies would be helpful to specifically clarify any potential interaction between transfusion protocol, BW, and ROP that may exist. The final sentence of the article clearly acknowledges the need for further investigation of this and other relevant variables.
4. If sampling is from a normal distribution, then analysis of variance (ANOVA) is an appropriate test regardless of the size of the sample. The data did not contain any extreme outliers that would unduly influence an ANOVA analysis.

Finally, we are disappointed that Saunders et al could draw no clinically useful conclusions from this study. On the contrary, we believe our study clearly demonstrates that the overall incidence of ROP in infants weighing <1251 g at birth is not significantly different in those who receive replacement blood transfusions compared with those who receive them primarily for symptomatic anemia. We would welcome additional studies to confirm these findings, as well as address some the other important variables that may influence the development of ROP.

Steven E. Brooks
Dennis M. Marcus
Jatinder Bhatia
Maribeth H. Johnson
Elizabeth Pirie
Delores Gillis

In Reply.

We appreciate the interest in our articles¹ by Drs Dani and Rubaltelli and would like to address their comments.

1. Of the 16 infants (8 in each group) who did not complete the entire 6-week transfusion protocol period, only 3 in group 1 (low hematocrit) and 2 in group 2 (high hematocrit) showed no signs of ROP at the time they left the study. Although it is true that some or all of these infants may have developed ROP, the impact this would have on the overall outcome would have been minimal and would not have approached the large difference (50% reduction in ROP) that we set out to find.

All but 6 infants completed at least 4 full weeks of the transfusion protocol. The incidence of ROP for these 44 infants was 90% for group 1 and 78% for group 2. The power to detect a 50% reduction in ROP incidence (from 80% to 40%) at  = .05 and a sample size of 40 is .75. We feel that the sample size of this study was sufficient to provide adequate power to detect a large difference between groups if one were present.2. We agree with Drs Dani and Rubaltelli that the number of transfusions alone does not fully characterize the iron status or oxidative stress to which a premature infant is exposed. However, the purpose of the study was to evaluate the effect of 2 different transfusion protocols, administered during a defined window of time in the perinatal period, on the incidence of ROP. The study was not designed to specifically produce iron overload in one group and anemia in another. The fact that ROP incidence and severity were not significantly different between the groups, despite significant group differences in hematocrit and hemoglobin throughout the study, argues against a major effect of blood transfusion protocol as performed in our study. In fact, the incidence of ROP in group 1 (the group receiving transfusions based on symptoms) was actually higher than in group 2 (the group receiving replacement transfusions), though the difference was not statistically significant.

3. Drs Dani and Rubaltelli wonder whether there may have been an undetected or unreported "great difference" in respiratory distress syndrome (RDS) severity or oxygen requirements between groups that presumably may have masked an effect of the blood transfusions. Although we did not report data about RDS severity, we did report that the incidence of bronchopulmonary dysplasia (as defined in the "Methods" section of our paper) was not significantly different between the groups. Because of the randomization process used and the remarkable similarity in baseline demographic characteristics and comorbidities between groups, we believe it is highly unlikely that "great differences" existed between groups regarding oxygen therapy requirements. Furthermore, supplemental oxygen was administered to all infants according to specific guidelines. Although we do not dispute the importance of oxygen as a risk factor for ROP, precise quantitation of exposure/dosage in individual infants is problematic. The uncertainty is further complicated by the fact that oxygen levels in the retinal circulation may not directly correlate with levels of inspired oxygen or oxygen tensions measured peripherally.

4. The incidence of ROP (any stage, either eye) in our study was not much higher than the 65.8% rate for infants weighing <1251 g at birth, and the 81.6% rate found for infants weighing <1000 g at birth found in the multicenter CRYO-ROP study.² However, because our study was not designed to compare the incidence of ROP in our neonatal intensive care unit to that found in other centers, we cannot offer an ad hoc explanation as to why our rates were higher than those reported by the Italian ROP Study Group.

Finally, we believe that ROP is very likely a multifactorial disease process. As such, defining the exact role of individual variables can be very difficult, particularly in the clinically dynamic setting of an intensive care unit. We agree with Drs Dani and Rubaltelli that our study does not conclusively rule out the possibility that blood transfusions or iron overload may play some role in ROP pathogenesis. However, under the conditions and time frame of our study, the influence of different transfusion protocols (one based on replacement and the other based on symptoms of anemia) did not appear to have a significant effect on ROP incidence or severity. Furthermore, our results do not support Drs Dani and Rubaltelli's conclusion that "blood transfusion policy must be regulated by accurate protocol," because both groups of infants in our study had similar short-term outcomes. We are glad that our work has stimulated thought about ROP, and hope that additional studies will be done to help understand this serious complication of prematurity.

Steven E. Brooks
Dennis M. Marcus
Jatinder Bhatia
Maribeth H. Johnson
Elizabeth Pirie
Delores Gillis

REFERENCES

1. Brooks SE, Marcus DM, Gillis D, Pirie E, Johnson MH, Bhatia J The effect of blood transfusion protocol on retinopathy of prematurity: a prospective, randomized study. Pediatrics. 1999; 104:514-518
2. Palmer EA, Flynn JT, Hardy RJ, Incidence and early course of retinopathy of prematurity. Ophthalmology. 1991; 98:1628-1640 [Medline]