The published literature was searched using strategies designed by a medical librarian (AH) for the concepts of AI and neonatal morbidity, with related synonyms. These strategies were created using a combination of controlled vocabulary terms and keywords, and were executed in Ovid MEDLINE (1946 to present), Embase.com (1946 to present), CINAHL Plus, Cochrane Library (including CENTRAL), Scopus (1823 to present), and Clinicaltrials.gov (1997 to present) from database inception. No limits or filters were applied to the search results. All database searches were completed on 24 November 2021. Details to reproduce the search are available in our Supplementary Material.

Studies were excluded if they received AI for any other indication (e.g., meconium aspiration syndrome, previable and prelabor premature rupture of membranes, intraamniotic infection, oligohydramnios), neonatal morbidity outcomes were not separated by indication for AI, there was no control group (i.e., no AI), contained duplicate data previously in another publication by the same or different author, or if the authors did not report raw data for included neonatal outcomes. In addition, all review/commentary articles identified by our literature search were reviewed to ensure no studies were missing in our literature review. Two studies had multiple indications for AI (including recurrent variable decelerations as the predominant group) [25, 26]. These authors were contacted in attempt to obtain individual patient data; however, the author was deceased and the data no longer available and the other author reported the original data were unavailable as they were more than 40 years ago [25, 26].

We used a predesigned methodology according to guidelines for Preferred Items for Systematic Reviews and Meta-analysis and Meta-analysis of Observational Studies in Epidemiology [27, 28]. The study protocol was registered with the International Prospective Register of Systematic Reviews (#PROSPERO 2022 CRD42022327133 Available from: https://tinyurl.com/2p9238x7) prior to review of articles.

Titles and abstracts were screened by 2 independent reviewers (BEP, KHB) for inclusion and full manuscripts were reviewed by (BEP, JR). Data pertaining to all primary and secondary outcomes were abstracted from the primary literature. Included studies reported on any of the following outcomes: Composite neonatal neurologic morbidity as defined by the primary source authors; composite neonatal morbidity defined by the primary source authors; neonatal death, seizures, HIE, and therapeutic hypothermia; apgar score less than 7 at 1, 5 and 10 minutes of life; umbilical cord gases including umbilical artery lactate, pH, and/or base deficit; neonatal head imaging including brain magnetic resonance imaging, ultrasound, electroencephalogram; neonatal resuscitation at delivery defined by resuscitation greater than 10 minutes at delivery with assisted ventilation, chest compression, or cardiac medications; neonatal physical exam findings consistent with neurologic dysfunction such as hyperalert, irritable, lethargic, obtunded, diminished spontaneous movements, weak/absent cry, respiratory difficultly, feeding difficultly, posturing, abnormal reflexes; and other outcomes related to morbidity such as intensive care unit admission, specialty care nursing, hospital length of stay. In addition, outcomes for this systematic review and meta-analysis included clinical criteria for hypoxic-ischemic encephalopathy utilized when making management decisions surrounding treatment for therapeutic hypothermia: Gestational age $\ge$36 weeks and 6 hours from birth, metabolic acidosis with pH 7.0 or base deficit $\ge$16, 10-minute Apgar score 5, and moderate to severe encephalopathy on clinical examination (Box 1) [12, 24].

Meta-analysis was performed using the meta-analysis statistics in Stata (Stata SE 18.0, StataCorp, College Station, TX, USA). Two-by-two contingency tables were created to compare the presence or absence of composite neonatal morbidity and individual outcomes, stratified by the presence or absence of intrapartum AI for recurrent variable decelerations. As individual patient-level data were not available, each individual neonatal outcomes were summed in the composite neonatal morbidity until the overall cohort was reached for that specific randomized trial. Therefore, neonates may be counted more than once within the composite neonatal morbidity outcome. We calculated pooled risk ratios (RRs) using random effects models to account for the clinical heterogeneity between studies. We estimated heterogeneity across studies and tested its significance using the Higgins I${}^2$ statistics and Cochran’s Q test. I${}^2$ of 50% was considered evidence of significant heterogeneity. Forest plots were created to visually assess both effect size and identify outliers.

We assessed study quality using three key factors considered most likely to threaten the validity of study results: Valid randomization method, completeness of follow-up (loss less than 10%), use of intention-to-treat analysis [29]. Valid randomization included use of random number tables, computer-generated random sequences, and other widely accepted random allocation. Studies complying with all these criteria were considered to be of high quality. Studies were scored 0 to 4 with higher scores suggestive of high quality. Each study received a single point if it was randomized, had completed follow-up, performed an intention-to-treat analysis, and specified the primary outcome as neonatal neurologic morbidity.

The risk of bias in each included study was assessed using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions using the revised Cochrane risk-of-bias tool for randomized trials [30]. Five domains (selection, performance, attrition, reporting, and other) related to risk of bias were assessed in each included trial: Random sequence generation, allocation concealment, selective reporting, other sources of bias, blinding (participants and personnel), blinding (outcome assessment), and incomplete outcome data. These were categorized as high, low or unclear risk of bias.

We could not summarize data for our primary outcome—a composite of neonatal neurologic morbidity based on individual neonatal outcomes used clinically with suspected HIE and/or clinical criteria utilized when deciding to treat neonates with therapeutic hypothermia (Box 1)—as no included studies provided neurological outcome data [12, 24].

Composite neonatal morbidity, defined as neonatal intensive care unit (ICU) admission, umbilical artery pH 7.20, and Apgar scores 7 at 1 and 5 minutes. This composite neonatal morbidity was not statistically different on pooled analysis (3 studies; 28.7% vs. 59.1%, pooled risk ratio, –0.30; 95% CI (95% confidence interval), –0.99–0.40; I${}^2$ = 94.51%; p = 0.40; Fig. 2, Ref. [6, 31, 32]). When individual outcomes within the composite neonatal morbidity were examined independently, ICU admission occurred less frequently among those with recurrent variable decelerations randomized to intrapartum AI compared to no intrapartum AI (ICU: 1 study; 6.8% vs. 16.5%; risk ratio 0.45; 95% CI 0.25–0.83) while there were no differences in umbilical artery pH 7.20 (pH 7.20: 1 study; 19% vs. 8%; p = 0.62). There was no difference in Apgar score 7 at 1 minute (2 studies; pooled risk ratio –0.46; 95% CI –1.57–0.65; Fig. 3, Ref. [31, 32]) or at 5 minutes (3 studies; pooled risk ratio –0.46, 95% CI –1.64–0.71; Fig. 4, Ref. [6, 31, 32]).

Fig. 2.

Composite neonatal morbidity by amnioinfusion intrapartum for recurrent variable decelerations during electronic fetal monitoring. 95% CI, 95% confidence interval; REML, restricted maximum likelihood.

Fig. 3.

Apgar score 7 at 1 minute by amnioinfusion intrapartum for recurrent variable decelerations during electronic fetal monitoring.

Fig. 4.

Apgar score 7 at 5 minutes by amnioinfusion intrapartum for recurrent variable decelerations during electronic fetal monitoring.

Intrapartum AI for recurrent variable decelerations is commonly utilized during labor, yet the impact of intrapartum AI on neonatal morbidity, particularly neonatal neurologic morbidity, has not demonstrated conclusive evidence of benefit [16, 17, 18, 19, 20]. Results of this analysis suggest that there is insufficient high-quality evidence surrounding the effect of intrapartum AI for recurrent variable decelerations on neonatal morbidity, yet individual studies suggest a potential benefit in reducing short term surrogate markers of morbidity such as neonatal ICU admissions. We could not provide insight as to whether this intervention is associated with differences in neonatal neurologic morbidity due to a lack of primary data.

We used strict inclusion criteria from the primary literature. Unlike prior meta-analysis analyzing neonatal outcomes after AI [33], our meta-analysis was limited to intrapartum AI utilized for in utero resuscitation of recurrent variable decelerations in labor and the results demonstrated that intrapartum AI may impact short-term neonatal morbidity such as intensive care unit admission but did not affect overall composite neonatal morbidity. More importantly, however, this review highlights the need for novel high-quality research aimed at examining whether intrapartum AI for recurrent variable decelerations affects neonatal outcomes, including neonatal neurologic morbidity. A main limitation of our systematic review and meta-analysis is the lack of patient-level data on neonatal morbidity outcomes. In particular, the generation of neonatal neurologic morbidity data after intrapartum AI should be a research priority given the absence of current data and the potential for intrapartum AI as a tool to begin therapeutic cooling in utero [12, 24]. Lastly, our composite neonatal morbidity is largely built on surrogate markers and may overestimate the effect of intrapartum AI for recurrent variable decelerations as we were unable to rank which outcome occurred in an individual more than once and thus may over represent the morbidity.

A similar systematic review was performed by Hofmeyr et al. [33] examining AI for potential or suspected umbilical cord compression. They concluded AI may be of considerable benefit resulting in a higher mean umbilical cord artery pH (mean difference 0.03, 95% CI 0.00–0.06) and fewer neonates with low cord arterial pH (average risk reduction 0.58, 95% CI 0.29 to 1.14) but these data are not statistically significant, demonstrate a small effect size, and are not clinically meaningful [33]. Our results demonstrate a similar trend in surrogate markers but represents a purely defined cohort of intrapartum AI for recurrent variable decelerations suggesting a role in reducing morbidity.

Data from this systematic review and meta-analysis suggests that intrapartum AI for the resuscitation of fetal tracings with recurrent variable decelerations is insufficiently studied, yet individual studies suggest intrapartum AI may plausibly reduce the risk of short-term neonatal morbidity such as ICU admission [34]. We speculate this reduction in neonatal morbidity occurs through a reduction in the total deceleration area—a feature of electronic fetal monitoring found to be the most predictive feature for umbilical artery acidemia in the two hours before delivery [21]. However, the clinical significance of intrapartum AI on neonatal morbidity remains insufficiently studied. Clinicians should consider these limited data when deciding to use an amnioinfusion during childbirth.

The results also demonstrate the need to critically assess in utero resuscitation techniques. For example, intermittent versus continuous AI techniques demonstrated no difference in outcomes except for lower amount of AI fluid and decreased costs [35]. However, the indication for intrapartum AI (e.g., recurrent variable deceleration versus reducing risk of neurologic injury) is important when considering administration techniques. For example, to decrease in utero and fetal temperatures, a continuous AI may counteract the blood flow from the gravid uterus and fetus which are being continuously warmed. The review also highlights varying descriptions for intrapartum AI techniques including the use of prophylactic antibiotics or antiseptics; type, temperature, amount, rate and specific fluid content of the AI; definitions of “recurrent” variable decelerations and when an intrapartum AI is classified as failed. While intrapartum AI for recurrent variable decelerations holds promise to reduce neonatal morbidity, we also highlight the rise and fall of intrapartum AI for meconium aspiration syndrome as a cautionary tale of changing clinical practice before intrapartum AI for reducing neonatal morbidity is rooted in evidence-based, high-quality data [16]. Overall, our systematic review and meta-analysis highlight there are insufficient high quality, randomized trials that examine the effect of amnioinfusion on neonatal morbidity outcomes, in particularly neonatal neurologic morbidity. Future research should identify the optimal amnioinfusion technique (e.g., type, temperature, amount, rate and specific fluid content), indication for amnioinfusion (e.g., reducing recurrent variables, intrauterine neuroprotection, thermoregulation), and examine clinically meaningful neonatal and maternal outcomes as they relate to the use of intrapartum AI.