Does high-dose erythropoietin decrease the risk of death or severe neurodevelopmental impairment in preterm infants?

 Does high-dose erythropoietin decrease the risk of death or severe neurodevelopmental impairment in preterm infants?

MANUSCRIPT CITATION: Juul SE, Comstock BA, Wadhawan R, et al. A Randomized Trial of Erythropoietin for Neuroprotection in Preterm Infants. N Engl J Med. 2020;382(3):233-243. doi:10.1056/NEJMoa1907423

REVIEWED BY

 Molly C. Easterlin, MD, MS¹ and Rangasamy Ramanathan, MD¹

AFFILIATION: ¹Division of Neonatology, Department of Pediatrics, LAC+USC Medical Center, Keck School of Medicine, University of Southern California

CONFLICTS OF INTEREST: The authors have no conflicts of interest to declare.

MANUSCRIPT LENGTH: 6 double spaced pages / 6 double spaced pages

KEYWORDS: Erythropoietin, neuroprotection, extremely preterm infants, missing data, multiple imputation

CORRESPONDING AUTHOR

Molly C. Easterlin, MD, MS

Neonatal-Perinatal Medicine Fellow

Department of Pediatrics

Division of Neonatology

LAC+USC Medical Center, Keck School of Medicine,

University of Southern California, Los Angeles, CA

1200 North State Street, IRD Building-Room 820,

Los Angeles, CA 90033-1029

Phone: (626) 372-5686

Email: [email protected] 

TYPE OF INVESTIGATION: Treatment safety and efficacy

QUESTION: In extremely preterm infants born at 24 weeks 0 days to 27 weeks 6 days, does high-dose erythropoietin (EPO) administered from 24 hours after birth through 32 weeks postmenstrual age, decrease the risk of death or severe neurodevelopmental impairment at 22 to 26 months postmenstrual age, compared to placebo?

METHODS

Design: Multicenter, double-blind, randomized, placebo-controlled phase 3 trial.

Allocation: Block randomization within recruitment sites stratified by single or multiple births and gestational age (24+0 to 25+6, or 26+0 to 27+6), with blocks of 4, 6, 8, and 10 infants.

Blinding: Double blind - all trial personnel (aside from the staff performing randomization, site pharmacist, and staff administering injections) and outcome assessors were unaware of group assignment.

Follow-up period: From birth until 22 to 26 months postmenstrual age.

Setting: Multicenter study including 19 sites and 30 hospitals.

Participants: Inclusion Criteria - Extremely preterm infants born at 24 weeks 0 days to 27 weeks 6 days between December 2013-September 2016. Exclusion Criteria - Known life-threatening anomalies, chromosomal anomalies, disseminated intravascular coagulopathy, twin-to-twin transfusion, hematocrit >65%, hydrops fetalis, or known congenital infection.

Intervention & Control: Subjects randomized to the intervention arm received high-dose EPO (1000 units/kg) intravenous within 24 hours of birth and every 48 hours for a total of 6 doses, followed by EPO 400 units/kg subcutaneous 3 times per week through 32 weeks 6 days. Subjects randomized to the control received saline intravenous infusions and then sham injections according to the same schedule.

For all subjects, iron supplementation was standardized and adjusted based on serum ferritin levels. Standard iron-containing formula was used (if no breastmilk). Transfusion criteria were not standardized.

Outcomes - Primary outcome: Death or severe neurodevelopmental at 22 to 26 months postmenstrual age, defined as 1) severe cerebral palsy based on Gross Motor Function Classification System (GMFCS) level>2, or 2) composite motor or cognitive score <70 on the Bayley Scales of Infant and Toddler Development, 3rd edition (Bayley-III), which corresponds to 2 standard deviations below the mean.  

Secondary outcomes:

1) Death or moderate-to-severe neurodevelopmental impairment at 22 to 26 months postmenstrual age. 2) Adverse Events - Included events most likely related to the sequelae of prematurity (including a severe subset) and events that may have been related to administration of EPO. 3) Adverse events, or death, or severe neurodevelopmental impairment by sex. Post hoc analyses examined transfusions (average cumulative volume, number of transfusions, unique donors per infant).

Analysis and Sample Size: Enrollment of 940 infants was planned using power calculations assuming 1) 18% of infants would die irrespective of group assignment, 2) 80% power to detect a 25% relative difference in neurodevelopmental impairment between the two groups (30% in EPO vs. 40% in placebo) with a two-sided significance level of 0.05, and 3) accounting for correlation among multiple gestations (with a variance inflation factor) and attrition (estimated 12.5%). A per-protocol efficacy analysis evaluated the primary outcome of death or severe neurodevelopmental impairment at age 2 using generalized estimating equations (GEE) to account for potential correlation between twins and with adjustment for gestational age at birth and recruitment site as a fixed effect (Table 1). A modified intention-to-treat safety analysis compared the percent of serious adverse events in the EPO group vs. placebo from the time of the first dose to hospital discharge. GEE with robust standard errors were used to account for potential correlation between twins. For analysis of total number of serious adverse events a Poisson regression with Wald test was used. For analysis of individual events a logistic-regression was used with adjustment for gestational age at birth and recruitment site as fixed effects.

MAIN RESULTS:

941 infants were randomized. 936 were included in the safety analysis (476 received EPO, 460 received placebo). 741 were included in the per-protocol efficacy analysis (376/476 who received EPO (79%), 365/460 who received placebo (79%)). Maternal, pregnancy, and infant characteristics at enrollment were similar between the two groups. The number of infants who completed the 6 intravenous injections, all subcutaneous injections, and who were discharged alive (as well as time to discharge) was similar in the two groups.

Efficacy Analysis - Primary outcome: There was no significant difference in rates of death or severe neurodevelopmental impairment between EPO and placebo (26% in each, RR 1.03, 95% CI (0.81-1.32)). There was also no difference in the individual components of the primary outcome between the two groups (death occurred in 13% of EPO vs. 11% of placebo, RR 1.27, 95% CI (0.91-1.79); severe neuro-developmental impairment occurred in 11% of EPO vs. 14% of placebo, RR 0.79, 95% CI (0.51-1.22)).

Secondary outcomes: There was no significant difference in rates of death or moderate-to-severe neurodevelopmental impairment between EPO and placebo (frequency 48% in EPO vs. 47% in placebo, RR 1.03, 95% CI (0.88-1.21)) or the components of the secondary outcome (moderate-to-severe neurodevelopmental impairment occurred in 38% of both groups, RR 0.97, 85% CI (0.79-1.18)).

Overall, the frequency of death or severe neurodevelopmental impairment was lower at older gestational ages and for female infants, but did not differ by treatment group.

Safety Analysis - There was no significant difference in the number of total serious adverse events between the two groups (59% in EPO vs. 62% in placebo, RR 1.01, 95% CI (0.83-1.22)), number of infants with at least one serious adverse event (37% in EPO vs. 40% in placebo, RR 0.97, 95% CI (0.84-1.13)), rates of individual complications of prematurity, or the rates of individual serious adverse events.

STUDY CONCLUSION: In extremely preterm infants born at 24 weeks 0 days to 27 weeks 6 days, high-dose EPO administered from 24 hours after birth through 32 weeks postmenstrual age, did not decrease the risk of death or severe neurodevelopmental impairment at 22-26 months postmenstrual age.

COMMENTARY:

In this study, high-dose erythropoietin administered to extremely preterm infants did not decrease the risk of death or severe neurodevelopmental impairment at 2 years post-menstrual age; but it also was not associated with more adverse events or complications of prematurity.[1] This trial provides important evidence to inform a question that has plagued the neonatal-perinatal community for years: what can we do to improve neurodevelopmental outcomes in this vulnerable population? This is an important issue given the significant rates of preterm birth, survival (50-70%), and neurodevelopmental impairment (20-50%).[2]

This multicenter, double blind, randomized phase III controlled trial was thoughtfully designed and conducted. The trial’s strengths include large size, double blinding, equal allocation of participants, appropriate analysis methods for outcomes, and multiple sensitivity analyses to enhance internal validity of results. Participants were allocated using block randomization within recruitment sites stratified by single or multiple birth and gestational age. Randomization helps to ensure that the comparison groups are similar in measured and unmeasured characteristics and block randomization with stratification ensures that characteristics that are strongly associated with the primary outcome, are equally distributed between the groups.[3] This allocation scheme resulted in intervention and control groups that were similar, ensuring unbiased comparisons. Additionally, they used appropriate analysis methods for various outcomes, employing GEE to account for potential correlation between twins[4] and count models (Poisson regression) when analyzing the number of serious adverse outcomes. Multiple sensitivity analyses[5] and multiple imputation ensured that decisions regarding missing or incomplete data did not affect results (see EBM lesson).

As the authors acknowledge, the limitations of this study include the use of neurodevelopmental testing at 2 years of age as compared to older ages when the Bayley-III may be more sensitive and specific.[6] Additionally, the sensitivity and specificity of the Bayley-III is variable and not precise, creating some difficulty in using Bayley-III scores as a primary outcome.[7] That being said, multiple prior studies have used Bayley-III scores, and used them at age 2, making it a useful primary outcome.[8] Additionally, the study intends to follow the cohort to identify problems that may present later in life.

The authors point out that their conclusions are in contrast to those of a meta-analysis of four randomized trials where EPO appeared to reduce the risk of a Bayley-II Mental Developmental Index (MDI) score <70 at 18-24 months postmenstrual age.⁸ The studies included in the meta-analysis used different EPO dosing regimens and lengths of treatment than the current study, which could account for the difference. However, significant prior research, including animal studies, and studies examining erythropoietic agents and brain vulnerability, supported the dosing used in this RCT.  Additionally, the studies in the meta-analysis included infants of older gestational ages (3 of the 4 included infants up to 31+6 or 32+0 weeks, including the largest which had the most effect on the meta-analysis primary outcome, MDI<70 at 18-24 months). Importantly, in the meta-analysis there was no significant effect of EPO when the analysis was restricted to <28 weeks gestational age (the population in this RCT), although this was based on 2 studies and the power may have been limited by smaller sample size; there was also no effect when the largest study was excluded which included infants up to 32+0 weeks. Thus, the findings of the reviewed RCT do not seem entirely contrasting and do seem plausible.

Interestingly, the rate of the primary outcome, death or severe neurodevelopmental impairment (26%), was significantly lower than the expected rate (40%). This was also true for the secondary outcome (47-48% vs expected rate of 60%). This raises the question of why these infants did better than expected, regardless of group assignment. Are we learning to care for these infants better as a specialty? Is it because they received closer monitoring due to being in an RCT (Hawthorne effect)? Or is it something specific to the care provided at the centers in this trial? Ultimately EPO did not appear to be harmful and it remains unclear whether a different dosing regimen or administration at different gestational ages would have affected the outcome. Given the non-significant effect of EPO on neurodevelopmental outcomes in this RCT the results are unlikely the change current management; however, the finding that EPO was not harmful, and the fact that the optimal dosing regimen, timing of administration, and follow-up period is unknown, warrants further study of EPO and longer-term follow-up of the participants in this RCT.

EBM LESSON: Missing Data and Multiple Imputation

Missing data occurs commonly in research studies. Data may be missing for various reasons. How researchers deal with missing data is important in interpreting the results of a study. “Inadequate handling of the missing data in a statistical analysis can lead to biased and/or inefficient estimates of parameters such as means or regression coefficients, and biased standard errors resulting in incorrect confidence intervals and significance tests” (which may lead to inaccurate conclusions).[9] In this study, the efficacy analysis for the primary outcome was based only on infants with complete data collected according to protocol, and included 79% of the EPO group and 79% of placebo. The researchers used multiple imputation by chained equations (MICE) to impute, or fill in, missing data on neuro-developmental outcomes.

Multiple imputation is a statistical technique to deal with missing data where, plausible values for each missing observation are generated (imputed) a number of times to create multiple “complete” data sets. The imputed values are generated using variables relevant to the analysis or missingness in the outcome.  The prediction is done multiple times, resulting in multiple datasets, to account for the statistical uncertainty in the generated values.[10] Each complete data set is then analyzed individually and the estimates are combined to obtain the final multiply-imputed estimates (i.e. variances, confidence intervals). The chained equations approach is commonly used for large datasets because it can handle thousands of observations, and hundreds of variables. In addition, it can manage variables of different types, such as continuous, binary, and categorical, because each missing variable is generated using its own regression model.^9,[11] Multiple imputation is now achievable using statistical programs such as R and STATA. In this study, analysis using multiple imputation for missing primary outcome data showed similar results to the primary efficacy analysis giving further validity to the results presented (117 children, RR 1.00, 95% CI (0.80-1.25)).

Table 1. Summary of Analyses and Results

³N=117 refers to the number of subjects lost to follow-up only. It is not clear if the imputed values were analyzed alone or with the larger dataset.

Acknowledgment:

The Journal club is a collaboration between the American Academy of Pediatrics- Section of Neonatal Perinatal medicine and the International Society of Evidence- based neonatology (EBNEO.org)

REFERENCES

[1] Includes severe BPD, medically or surgically treated PDA, all grades of NEC, all grades of IVH, PVL, cerebral hemorrhage, all grades of ROP

[2] Includes hypertension, polycythemia, major thrombosis, severe pulmonary hemorrhage, SEC stage 2b/3, Severe sepsis, grade III/IV IVH, severe ROP, nonfatal cardiac arrest, other life threatening events, SIP, death

[1] Juul SE, Comstock BA, Wadhawan R, et al. A Randomized Trial of Erythropoietin for Neuroprotection in Preterm Infants. N Engl J Med. 2020;382(3):233-243. doi:10.1056/NEJMoa1907423

[2] Glass HC, Costarino AT, Stayer SA, Brett CM, Cladis F, Davis PJ. Outcomes for extremely premature infants. Anesth Analg. 2015;120(6):1337-1351. doi:10.1213/ANE.0000000000000705

[3] Broglio K. Randomization in Clinical Trials: Permuted Blocks and Stratification. JAMA. 2018;319(21):2223–2224. doi:10.1001/jama.2018.6360

[4] Nair J, Kumar V. Is glucose variability associated with worse brain function and seizures in neonatal encephalopathy?. J Perinatol. 2020;40(5):827-830. doi:10.1038/s41372-020-0651-z

[5] Anchan JC, Carr NR, Ahmad KA. Neonatal hypoglycemia: is there a neurodevelopmental impact in early childhood?. J Perinatol. 2019;39(1):4-7. doi:10.1038/s41372-018-0244-2

[6] Marlow N, Wolke D, Bracewell MA, Samara M; EPICure Study Group. Neurologic and developmental disability at six years of age after extremely preterm birth. N Engl J Med. 2005;352(1):9-19. doi:10.1056/NEJMoa041367

[7] Wong HS, Santhakumaran S, Cowan FM, Modi N; Medicines for Neonates Investigator Group. Developmental Assessments in Preterm Children: A Meta-analysis. Pediatrics. 2016;138(2):e20160251. doi:10.1542/peds.2016-0251

[8] Fischer HS, Reibel NJ, Bührer C, Dame C. Prophylactic Early Erythropoietin for Neuroprotection in Preterm Infants: A Meta-analysis. Pediatrics. 2017;139(5):e20164317. doi:10.1542/peds.2016-4317

[9] White, I.R., Royston, P. and Wood, A.M. (2011), Multiple imputation using chained equations: Issues and guidance for practice. Statist. Med., 30: 377-399. doi:10.1002/sim.4067

[10]   Pigott T.D. (2001), “A Review of Methods for Missing Data,” Educational Research and Evaluation, 7 (4), 353-383.

[11] Azur MJ, Stuart EA, Frangakis C, Leaf PJ. Multiple imputation by chained equations: what is it and how does it work?. Int J Methods Psychiatr Res. 2011;20(1):40-49. doi:10.1002/mpr.329