Room Air for Initiating Term Newborn Resuscitation: A Systematic Review With Meta-analysis

This systematic review and meta-analysis was conducted in accordance with the Cochrane Handbook for Systematic Reviews of Interventions and reported following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement for meta-analysis in health care interventions.^8,9 The protocol was registered in advance of article selection with the Prospective Register of Systematic Reviews (registered January 8, 2018; CRD42018084902; see Supplemental Information). The protocol included term and preterm newborns as predetermined subgroups, and these were separated into different analyses after initial article selection. Studies were included in this systematic review if >75% of the newborns were ≥35 weeks’ gestation. The cutoff at 35 weeks’ gestation was chosen on the basis of the experience of our clinical experts that late preterm infants were more similar physiologically to term infants than the population of very premature infants enrolled in the preterm randomized controlled trials (RCTs) and pragmatically based on the knowledge of the enrollment criteria for previous studies.

The selection and importance rating of patient-oriented outcomes for term and late preterm newborns ≥35 weeks’ gestation were determined in advance through discussion and consensus with the ILCOR NLS task force.¹⁰ The outcomes centered on all-cause mortality and neurologic impairment.¹⁰ When available, we collected all-cause mortality at 2 time intervals: short-term (primary outcome, in-hospital or up to 30 days postnatal) and long-term (1–3 years); and neurologic impairment at 2 time intervals: short-term HIE (Sarnat Stage II–III), and long-term NDI (moderate to severe at 1–3 years).¹¹ NDI is commonly defined as having at least 1 of cerebral palsy, cognitive impairment, visual impairment, or hearing impairment and is categorized by severity. Where available, we extracted data for moderate to severe NDI at 1 to 3 years based on the Gross Motor Function Classification System and Bayley Scales of Infant and Toddler Development, Third Edition.^12,13 

Ovid Medline, Embase, all Evidence Based Medicine Reviews (including Cochrane CENTRAL and others), and EBSCOhost Cumulative Index to Nursing and Allied Health Literature (CINAHL) were searched for relevant neonatal literature between January 1, 1980, and December 11, 2017, without language restrictions. The search was updated from December 1, 2017, to August 10, 2018, before publication (Supplemental Tables 6 and 7). The searches were limited to the last 4 decades because no pertinent studies were expected before this. An iterative approach was used to ensure that key articles (identified by content experts and in previous systematic review articles) were found. Additionally, we searched the first 200 hits on Google Scholar, references of systematic reviews on the topic, references of the ILCOR 2010 and 2015 CoSTRs, and trial registries (clinicaltrials.gov; the International Standard Randomized Controlled Trial Number Registry, isrctn.com; and the EU Clinical Trials Register, clinicaltrialsregister.eu; last searched August 10, 2018).

Covidence software was used for study selection in 2 steps (Covidence systematic review software; Veritas Health Innovation, Melbourne, Australia; www.covidence.org). Pairs of independent reviewers screened titles and abstracts. In the event of a disagreement at abstract screening, the full text was reviewed. Independent reviewers subsequently completed full-text review for eligibility in duplicate. A third reviewer was involved for disagreements at the full-text stage, and final decisions were determined by consensus. The first reason for exclusion was captured according to a predetermined, ordered list of exclusions. Interrater agreement for article selection was assessed by using Cohen’s κ coefficient at the abstract and full-text stages.

RCTs, quasi randomized controlled trials (qRCTs), and nonrandomized (observational) studies were eligible if they included comparison of low with high initial oxygen concentration for respiratory support at birth. Review articles, editorials, comments, case reports, and small case series (≤10 patients) were excluded. We excluded studies that were focused on oxygen use beyond the initial stabilization in the delivery room or studies that were focused on oxygen saturation targeting and not initial oxygen concentration. To avoid publication bias, the protocol was amended to include data from conference abstracts (not otherwise published) in a sensitivity analysis if the authors provided enough information to confirm the methods, key patient characteristics, and outcomes.

For each study, pairs of authors independently extracted predetermined study characteristics and study outcomes and then achieved consensus. Pairs of independent authors evaluated risk of bias (RoB) in individual studies using the Cochrane Risk of Bias Tool for RCTs and the Risk of Bias in Non-Randomized Studies of Interventions tool for observational studies.^8,14 Similarly, 2 authors assessed the certainty of evidence (confidence in the estimate of effect) for each outcome on the basis of the GRADE framework, including calculating the optimal information size to assess imprecision (GRADEpro Guideline Development Tool, McMaster University, Ontario, Canada; available at gradepro.org).¹⁵ The RoB and GRADE assessments were then reviewed by ILCOR content experts who are also authors to achieve consistency and consensus.

Covidence, GRADEpro, and Review Manager software (RevMan 5.3; The Nordic Cochrane Centre, Copenhagen, Denmark) were used to abstract, summarize, and analyze the data, respectively.

Meta-analyses were performed if ≥2 studies were available. Heterogeneity was measured by using the I² statistic.¹⁶ Because multiple small studies (<250 patients) were anticipated, a random effects model was used for analysis. We report pooled unadjusted risk ratios (RRs) and corresponding 95% confidence intervals (CIs) using the Mantel-Haenszel method for dichotomous variables. Forest plots were used for graphical representation of RRs. To assess for publication bias, we visually inspected funnel plots where >8 studies were available. The absolute risk difference and number needed to treat (NNT) were calculated where the pooled estimate from RCTs revealed a statistically significant benefit by using the method recommended by the Cochrane Collaboration.¹⁷ 

Sensitivity analyses were completed where inclusion of 1 or more studies was uncertain because of a high RoB, incongruent allocation, mixture of adjusted and nonadjusted analyses, or significant heterogeneity.

Prespecified subgroup analyses were planned if >2 studies were available with relevant outcome information related to gestational ages (≥35 and ≥37 weeks’ gestation), specific Fio₂ ranges, or oxygen saturation targeting as a cointervention.

Using this search strategy, we identified a total of 2366 records; after removing 967 duplicates, 1399 records were screened by title and abstract. Five additional studies (abstracts) were found via reference searches and added to full-text screening. A total of 59 full-text articles were assessed for eligibility, and 12 publications on term or late preterm newborns were included.^{2,–4,18,–26} The Cohen’s κ coefficient, a measure of interrater agreement, was 0.81 (excellent) at the abstract stage and 1.0 (full agreement) at the full-text stage. See Fig 1 for the PRISMA study selection diagram including the reasons for article exclusion.

Of the additional studies found via reference searches, 1 study published in Romanian with an English abstract was originally excluded because it did not include outcomes of interest.²² However, the publication included a full description of methods and the first author provided additional relevant outcome data; therefore, the full-text of the study was translated and included. Three additional titles by the same first author were identified but were only published as conference abstracts.^23,–25 The author provided additional information regarding methodology, patient characteristics, and outcomes for all 3 of these studies and thus they were included in sensitivity analysis.

Another article of note included both randomized and observational data for 830 term newborns over a 6-year time period (1994–1999) at a single center.¹⁹ Some of the randomly assigned patients overlapped with other included publications.^3,20 The first author provided additional information and outcomes for a subset of 537 patients who were randomly assigned but not published in other included studies. Thus, the study was included in the systematic review and considered for the meta-analysis.

Lastly, a search of clinical trial registries (www.clinicaltrials.gov, www.isrctn.com, and www.clinicaltrialsregister.eu) revealed no additional published or unpublished term or late preterm studies.

In Table 1, we summarize the characteristics of the included studies and highlight these differences. Of the 12 included articles, there were 10 original studies and 2 publications of follow-up data. Five were quasi randomized on the basis of alternating days and 5 were RCTs, but only 2 were fully randomized with blinding of allocation and intervention. No eligible observational cohort studies were identified. Ten studies contained reports of short-term mortality (in hospital or up to 30 days postnatal), 7 contained reports of HIE (Sarnat Stage II–III), 1 contained a report of long-term mortality (1–3 years), and 3 contained reports of NDI (moderate to severe, 1–3 years).

A total of 2164 patients were included in studies ranging from 44 to 609 patients. Most studies were from India and Europe; they were published between 1993 and 2007, with patient recruitment from 1990 to 2007. In 4 of 10 studies, the authors reported multicenter data. The authors of all the included studies compared initiating Fio₂ 0.21 with Fio₂ 1.0.

The Resair-2 study was a large, nonblinded, multicenter qRCT (allocated by even and odd days). There were 10 sites, which were predominantly low-resource settings with high event rates.³ The authors of the original study indicated that the researchers enrolled 609 patients but subsequently determined data from 18 neonates had been duplicated.⁴ The authors reported that the corrected demographics and outcomes revealed no difference. As the specific outcome and demographic data for the corrected numbers was not available, the original data were used. The follow-up data at 18 to 24 months used the corrected numbers, but only some of the original centers were included, and there was significant loss to follow-up such that only 66% had follow-up data.

Although the reporting of mortality was similar, the definition of NDI was different among the 3 studies that contained reports of it. In the Resair-2 follow-up publication, they evaluated whether the neonates had developed normally using an unvalidated simple assessment tool.⁴ In Bajaj et al,²¹ NDI assessment was performed by using the Baroda development screening test (modified Bayley), and <97% score on the development screening test was defined as neurologically abnormal. In Toma et al,²² they assessed NDI using the Bayley Scales of Infant and Toddler Development, Third Edition and divided neonates into 3 categories: high, moderate, and low according to a total score. Where available, we extracted information on moderate to high NDI. Additionally, HIE reported in Ramji et al²⁶ required clarification, and the data used in our analysis were confirmed with the authors.

In Table 2, we outline the patient characteristics of the included studies. The intervention and comparator groups were similar in most of the key prognostic variables (although patient characteristics was not available for 1 study).¹⁹ Eight of the 10 studies included either term or late preterm newborns (≥35 weeks’ gestation), 1 study included newborns ≥34 weeks’ gestation, and 1 study (including follow-up publication) included some preterm patients but <25% were <35 weeks’ gestation (median 38, interquartile range [IQR] 31–42 weeks’ gestation).^3,4,22 The gestational age mean and median ranged from 35.3 to 40.5 weeks with well-matched intervention and comparator groups. The birth weight mean and median ranged from 2319 to 3536 g and were also well matched. Researchers did not routinely report sex, but those that did reported a slight male sex predominance (49%–60%). Intubation and mechanical ventilation were not routinely reported but ranged from 10% to 51%. Apgar scores were well matched between the 2 comparators.

The RoB assessment for each study is summarized in Table 3. Only 2 studies were fully randomized and blinded for participants, personnel, and assessors; the majority of the studies were determined to have a high RoB.

The 3 articles that were published as conference abstracts were determined to have a critical RoB and thus were included in the sensitivity analysis but not the primary meta-analysis.^11,24,25 Additionally, although outcome data were provided for the randomly assigned subset of patients from Vento et al,²⁰ the demographic information for this subset was not available, and the imbalance in the numbers randomly assigned to each group was unexplained (300 compared with 237). For these reasons, this study was found to have a critical RoB and only included in the sensitivity analysis and not the primary meta-analysis.

Results of the meta-analysis are detailed below and shown in the Figs 2–5 forest plots.

Seven RCTs and qRCTS involving 1469 term and late preterm newborns (≥35 weeks’ gestation) containing reports on short-term mortality (in-hospital or up to 30 days postnatal) were included in the meta-analysis.^{2,3,19,21,–23} The pooled estimate demonstrated a statistically significant reduction in mortality for Fio₂ 0.21 compared with 1.0 (RR = 0.73; 95% CI: 0.57 to 0.94; I² = 0%). The absolute survival benefit (absolute risk difference) is 4.6% (95% CI: 1.0% to 7.3%) and the relative survival benefit is 27%. Based on an assumed control risk of 17.0% (from the mortality rate of the comparison group), the NNT with room air to have 1 additional survivor (short-term) is 22 (95% CI: 14 to 99). The forest plot is presented in Fig 2A. Heterogeneity was low as evidenced by the low I² = 0%, visual inspection of the forest plot, and similarities in the included populations. Although the CIs from the included studies crossed the null effect line, all studies trended in the same direction (favoring room air), and the summary estimate from the meta-analysis provides confirmation that initial room air improves survival.

To explore the reasons for heterogeneity, a sensitivity analysis was conducted for the primary outcome of short-term mortality to compare studies that were blinded, unblinded, and those at a critical RoB (Fig 3B).^19,23,25 Inclusion of data from the studies at a critical RoB to the meta-analysis (additional 647 neonates from 3 studies) made no appreciable change in the outcome estimate for short-term mortality, favoring an Fio₂ 0.21 compared with 1.0 (RR = 0.71; 95% CI: 0.56 to 0.91; I² = 0%).

Long-term mortality (1–3 years) was reported in 1 abstract with a critical RoB, involving a total of 54 patients with no deaths in either group.²⁴ 

NDI (moderate to severe, 1–3 years) was reported in 2 qRCTs involving 360 term and late preterm newborns (≥35 weeks’ gestation, Table 2). The summary estimate from meta-analysis revealed no difference in initiating respiratory support with Fio₂ 0.21 compared with 1.0 (RR = 1.41; 95% CI: 0.77 to 2.60; I² = 0%).^4,21 

The sensitivity analysis for NDI involving 1 additional study that was determined to be at a critical RoB (414 patients from 3 studies), still demonstrated no statistically significant difference (RR = 1.24; 95% CI: 0.73 to 2.10; I² = 0%) for Fio₂ 0.21 compared with 1.0.²⁴ 

HIE (Sarnat Stage II–III) was reported in 5 RCT and/or qRCTs involving 1315 term and late preterm newborns (≥35 weeks’ gestation) receiving respiratory support at birth included in the meta-analysis and revealed no statistically significant difference in Fio₂ 0.21 compared with 1.0 (RR = 0.90; 95% CI: 0.71 to 1.14; I² = 8%) (Fig 5D).^{3,18,21,22,26} 

Sensitivity analysis for HIE (Sarnat Stage II–III) was performed with 3 additional studies that were determined to be at a critical RoB. Data from these 8 studies and 2006 neonates did not change the RR appreciably and still demonstrated no statistically significant difference (RR = 0.89; 95% CI: 0.73 to 1.10; I² = 0%) for Fio₂ 0.21 compared with 1.0.^19,23,25 

The authors of all studies compared Fio₂ 0.21 with 1.0, and thus subgroup analysis according to different initial oxygen concentrations was not possible. There were no data for the planned subgroups analyses related to different gestational ages and whether oxygen saturation targeting was included as a cointervention.

The GRADE summary of quality evaluation for the primary outcomes is presented in Table 4. RCTs start at high certainty. Because of concerns with RoB, inconsistency, and imprecision, the certainty of the results was downgraded. The ILCOR NLS Task Force provided the expert opinion that it was unlikely that there were any additional unpublished studies on this topic given the intense clinical interest in this topic, the international reach and involvement of the committee, and the extensive search including uncovering abstracts and conference proceedings. Therefore, the outcomes were not downgraded for publication bias. The GRADE certainty was determined to be low for short-term mortality and HIE and very low for NDI because of serious concerns with RoB and imprecision. The ratings of the importance of outcomes for the GRADE analysis were all “critical” and ranged from 7 to 9 on the 9-point scale.

In this systematic review and meta-analysis involving 2164 neonates ≥35 weeks’ gestation, we demonstrate a 27% relative survival benefit and 4.6% absolute survival benefit (short-term) when initial room air is compared with Fio₂ 1.0 for neonates receiving respiratory support at birth. This corresponds to an NNT with room air to have 1 additional survivor (short-term) of 22. There were no statistically significant differences for HIE and NDI. The GRADE certainty of evidence was low for short-term mortality and HIE and very low for NDI.

The last ILCOR analysis of initial oxygen use for term neonatal resuscitation was completed in 2010 before the adoption of the GRADE methodology for ILCOR reviews. The ILCOR 2010 NLS CoSTR stated, “In term infants receiving resuscitation at birth with positive pressure ventilation, it is best to begin with air rather than 100% oxygen. If despite effective ventilation there is no increase in heart rate or if oxygenation (guided by oximetry) remains unacceptable, use of a higher concentration of oxygen should be considered.”⁶ After the release of this CoSTR, there was widespread adoption of initial room-air resuscitation worldwide. The Neonatal Resuscitation 2015 International Consensus on Cardiopulmonary Resuscitation and Emergency Cardiovascular Care Science with Treatment Recommendations did not update the CoSTR.⁷ 

The ILCOR NLS Task Force prioritized this research question in 2017 because of the strong desire to have the existing evidence subjected to a methodologically rigorous review including GRADE analysis of confidence in effect. In Table 5, we compare this analysis to key previously published meta-analyses.^5,27,–30 The results are fairly consistent across the meta-analyses, with the largest number of patients included in the current publication’s sensitivity analysis and the Saugstad et al²⁷ publication.

There have been no new publications of term studies on this topic since 2007; therefore, we are left with these few older studies from a different era, with their inherent differences to current state-of-the-art neonatal resuscitation to reflect on the evidence for current care. This previous evidence, despite low and very low confidence in the point effect estimates, has low statistical heterogeneity and consistent short-term mortality excess with Fio₂ 1.0. The findings of the current systematic review and meta-analysis are similar to the 2010 ILCOR review and indicate benefit in short-term mortality by initiating respiratory support at birth for term and late preterm newborns (≥35 weeks’ gestation), with room air compared with Fio₂ 1.0. There are no identified term or late preterm studies comparing outcomes at any intermediate oxygen concentration between these 2 extremes. However, given that practice has changed to initiating resuscitation with room air and using oxygen saturation monitoring to adjust oxygen administration, the precise initial Fio₂ may have become less important, and it is unlikely that there will be any future studies on this topic.

In this systematic review and meta-analysis, we include a prespecified protocol, broad search strategy, additional unpublished data from authors, unpublished studies (abstracts) included in sensitivity analyses, use of GRADE to describe certainty in effect estimate, a team of expert systematic reviewers coupled with international multidisciplinary experts in neonatology, and adherence to PRISMA reporting.

However, several limitations are worth noting. First, out of 10 trials, all but 2 had high RoB, 5 studies used a quasi-randomized design (alternating days), and 8 did not have allocation concealment or personnel blinding. This serious RoB, as well as imprecision, make the certainty of the evidence low or very low. In addition, the mortality incidence was vastly different between studies. Short-term mortality ranged from 1.5% in Vento et al¹⁹ (Spain) to 15% in Ramji et al²⁶ (India), and 18% in Saugstad et al³ (patients were mostly from India, Egypt, and Philippines). Studies with high mortality are more heavily weighted in the meta-analyses, and thus caution is needed to ensure similar results are found in countries with lower newborn mortality. Furthermore, all of the studies were conducted >10 years ago and before continuous oxygen saturation monitoring and oxygen titration during delivery room resuscitation were routine. Therefore, it is not clear whether the same results would be found in the setting of current clinical practice in which the inspired oxygen concentration is titrated to achieve targeted levels of oxygen saturation.⁷ The evidence in which long-term NDI was evaluated was limited because of the small sample size included in the 2 eligible trials and 1 abstract. We assessed HIE as a short-term neurologic outcome as oxygen administration was historically investigated as a treatment; however, the pathogenesis of HIE starts in utero and HIE may not accurately predict important long-term neurologic outcomes.³¹ Lastly, all the trials included in this review compared initial room air with Fio₂ 1.0; therefore, whether room air is superior to other low or intermediate Fio₂ (eg, Fio₂ 0.30) is not known.

With this systematic review and meta-analysis, we confirm a statistically significant reduction in short-term mortality (without statistically significant differences in short- and long-term neurologic outcomes) by using initial room air compared with Fio₂ 1.0 (100% oxygen) for term and late preterm newborns (≥35 weeks’ gestation) receiving respiratory support at birth. The certainty of effect is low (short-term mortality); however, the results are consistent across studies with no evidence of statistical heterogeneity and represent the best available evidence to answer this important question.

CI

confidence interval

CINAHL

Cumulative Index to Nursing and Allied Health Literature

CoSTR

consensus on the science with treatment recommendation

Fio₂

fraction of inspired oxygen

GRADE

Grading of Recommendations Assessment, Development and Evaluation

HIE

hypoxic-ischemic encephalopathy

ILCOR

International Liaison Committee on Resuscitation

IQR

interquartile range

NDI

neurodevelopmental impairment

NLS

Neonatal Life Support

NNT

number needed to treat

PRISMA

Preferred Reporting Items for Systematic Reviews and Meta-Analyses

qRCT

quasi randomized controlled trial

RCT

randomized controlled trial

RoB

risk of bias

RR

risk ratio

The authors would like to express their appreciation to the following for contributing valuable support to improve this review: Laurie J. Morrison (ILCOR Continuous Evidence Evaluation Working Group Liaison), and Carolyn Ziegler (St. Michael's Hospital Information Specialist) and Andrei Harabor (article translation). The authors would also like to highlight the following researchers who kindly contributed valuable information and data from their studies to improve this review: Naveen Bajaj, Ola Saugstad, Adrian Toma, and Maximo Vento.

Besides the authors Tetsuya Isayama, Charles Christoph Roehr, Myra H. Wyckoff, and Yacov Rabi, members of the International Liaison Committee on Resuscitation Neonatal Life Support Task Force include: Jonathan Wyllie, Jeffrey M. Perlman, Khalid Aziz, Ruth Guinsburg, Maria Fernanda de Almeida, Vishal Kapadia, Daniele Trevisanuto, Sithembiso Velaphi, Lindsay Mildenhall, Helen Liley, Shigeharu Hosono, Han-Suk Kim, and Edgardo Szyld.