Antenatal magnesium sulphate and adverse neonatal outcomes: A systematic review and meta-analysis

Background There is widespread, increasing use of magnesium sulphate in obstetric practice for pre-eclampsia, eclampsia, and preterm fetal neuroprotection; benefit for preventing preterm labour and birth (tocolysis) is unproven. We conducted a systematic review and meta-analysis to assess whether antenatal magnesium sulphate is associated with unintended adverse neonatal outcomes. Methods and findings CINAHL, Cochrane Library, LILACS, MEDLINE, Embase, TOXLINE, and Web of Science, were searched (inceptions to 3 September 2019). Randomised, quasi-randomised, and non-randomised trials, cohort and case–control studies, and case reports assessing antenatal magnesium sulphate for pre-eclampsia, eclampsia, fetal neuroprotection, or tocolysis, compared with placebo/no treatment or a different magnesium sulphate regimen, were included. The primary outcome was perinatal death. Secondary outcomes included pre-specified and non-pre-specified adverse neonatal outcomes. Two reviewers screened 5,890 articles, extracted data, and assessed risk of bias following Cochrane Handbook and RTI Item Bank guidance. For randomised trials, pooled risk ratios (RRs) or mean differences, with 95% confidence intervals (CIs), were calculated using fixed- or random-effects meta-analysis. Non-randomised data were tabulated and narratively summarised. We included 197 studies (40 randomised trials, 138 non-randomised studies, and 19 case reports), of mixed quality. The 40 trials (randomising 19,265 women and their babies) were conducted from 1987 to 2018 across high- (16 trials) and low/middle-income countries (23 trials) (1 mixed). Indications included pre-eclampsia/eclampsia (24 trials), fetal neuroprotection (7 trials), and tocolysis (9 trials); 18 trials compared magnesium sulphate with placebo/no treatment, and 22 compared different regimens. For perinatal death, no clear difference in randomised trials was observed between magnesium sulphate and placebo/no treatment (RR 1.01; 95% CI 0.92 to 1.10; 8 trials, 13,654 babies), nor between regimens. Eleven of 138 non-randomised studies reported on perinatal death. Only 1 cohort (127 babies; moderate to high risk of bias) observed an increased risk of perinatal death with >48 versus ≤48 grams magnesium sulphate exposure for tocolysis. No clear secondary adverse neonatal outcomes were observed in randomised trials, and a very limited number of possible adverse outcomes warranting further consideration were identified in non-randomised studies. Where non-randomised studies observed possible harms, often no or few confounders were controlled for (moderate to high risk of bias), samples were small (200 babies or fewer), and/or results were from subgroup analyses. Limitations include missing data for important outcomes across most studies, heterogeneity of included studies, and inclusion of published data only. Conclusions Our findings do not support clear associations between antenatal magnesium sulphate for beneficial indications and adverse neonatal outcomes. Further large, high-quality studies (prospective cohorts or individual participant data meta-analyses) assessing specific outcomes, or the impact of regimen, pregnancy, or birth characteristics on these outcomes, would further inform safety recommendations. PROSPERO: CRD42013004451.


Introduction
Antenatal magnesium sulphate is commonly used in obstetric practice. Systematic reviews and clinical practice guidelines support its use when given for maternal neuroprotection in preeclampsia or eclampsia [1][2][3] and for neuroprotection of the fetus in women at risk of preterm birth (for cerebral palsy prevention) [4][5][6][7]. Despite continued use in some countries [8], available evidence does not support its role in preventing preterm birth in women with, or following, threatened preterm labour (for tocolysis) [7,9].
Concerns surrounding possible unintended adverse outcomes for fetuses or neonates following exposure to antenatal magnesium sulphate emerged over 50 years ago [10,11], and uncertainty persists today. While the clinical consequences of hypermagnesemia, related to increased serum concentrations, are well known and documented (such as lethargy, drowsiness, flushing, nausea, vomiting, muscle weakness, loss of deep tendon reflexes, hypotension, apnoea, coma, cardiac arrest, and, ultimately, death [12]), whether neonates are at risk of such adverse outcomes following exposure to antenatal magnesium sulphate is unclear.
We have systematically reviewed the maternal adverse effects of different antenatal magnesium sulphate regimens [13], and a further systematic review has summarised the effects specifically on fetal heart rate [14]. To our knowledge, a broad evaluation of evidence surrounding potential unintended neonatal adverse outcomes, informed by current guidance [15][16][17], has not previously been conducted. Implementation of this treatment may be strengthened, and its safety improved, if guidelines and recommendations for practice can be based on such knowledge.
The aim of our study, therefore, was to conduct a comprehensive systematic review to assess whether antenatal magnesium sulphate is associated with perinatal death and other unintended adverse neonatal outcomes.

Inclusion criteria
We included randomised and quasi-randomised controlled trials as well as non-randomised controlled studies (non-randomised trials, cohort studies, and case-control studies), and case reports. We excluded cross-sectional studies and case series. We included studies available as abstracts only, along with full-text publications.
We included neonates who were exposed to antenatal magnesium sulphate, regardless of their gestational age at exposure or birth. We included studies where antenatal magnesium sulphate was given for pre-eclampsia or eclampsia, for neuroprotection of the fetus, or for tocolysis. We excluded studies where magnesium sulphate was given as an adjuvant during obstetric anaesthesia. We included intervention studies in which magnesium sulphate was compared with no treatment, placebo, or a different magnesium sulphate regimen. We included observational studies where magnesium sulphate was assessed as an 'exposure'. We excluded studies where magnesium sulphate was compared with another therapy (for example, diazepam for pre-eclampsia or eclampsia, or nifedipine for tocolysis).
We included studies that reported on adverse outcomes for neonates, however defined. The primary outcome was perinatal death. Secondary outcomes included pre-specified adverse outcomes (based on non-systematic literature review: stillbirth, neonatal death or death up to hospital discharge, low Apgar scores at 1 and 5 minutes, need for active resuscitation at birth, respiratory depression, spontaneous intestinal perforation, patent ductus arteriosus, hypotension, lethargy, hypotonia or hyporeflexia, osteopenia or bone fractures, neonatal intensive care unit admission, and duration of neonatal care unit admission), along with other non-pre-specified adverse neonatal outcomes.

Study selection
After screening all titles and abstracts, we obtained full-text articles for studies that appeared to meet the inclusion criteria. All full-text articles were assessed for inclusion. Each stage was carried out by 2 reviewers, and we resolved any discrepancies through discussion, or, if required, we consulted a third reviewer.

Data extraction and management
For included studies, data were extracted using a standardised form, including information regarding design, participants, the magnesium sulphate regimen(s), the control/comparison if applicable, neonatal adverse outcomes reported, results relevant to the review, and the risk of bias. For all randomised trials, all case reports, and 60% of non-randomised studies, extraction was carried out by 2 reviewers (for 40% of non-randomised studies, extractions were checked by a second reviewer), and we resolved discrepancies through discussion, or, if required, we consulted a third reviewer.

Assessment of risk of bias
Quality appraisal of intervention studies was undertaken utilising established guidelines provided in the Cochrane Handbook for Systematic Reviews of Interventions [19]. The quality assessment of observational studies was guided by the RTI Item Bank [20].

Data synthesis and analysis
Data analyses were undertaken by study design. Statistical analyses for randomised trials were performed using Review Manager, version 5.3 [21]. We present quantitative data from individual studies as risk ratios (RRs) for dichotomous outcomes and mean differences (MDs) for continuous outcomes, with 95% confidence intervals (CIs). For all outcomes, we carried out analyses as far as possible on an intention-to-treat basis. Pooled estimates were calculated using fixed-effects meta-analysis (Mantel-Haenszel method) where there was a sufficient quantity of data, with clinical homogeneity. Where there was substantial statistical heterogeneity (where I 2 was greater than 30% and either T 2 was greater than 0 or there was a low P value [less than 0.10] in the χ 2 test), summary estimates were calculated using random-effects meta-analysis. Where there were 10 or more trials in a meta-analysis, we investigated reporting biases (such as publication bias) using funnel plots, which we assessed visually.
Separate comparisons were performed for those studies assessing magnesium sulphate versus no treatment/placebo and those comparing different magnesium sulphate regimens. For our primary review outcome (perinatal death) and other mortality outcomes, we conducted subgroup analyses based on indication for use and characteristics of the magnesium sulphate loading and maintenance dose regimens, as these factors were considered likely to influence outcomes. It was not possible to conduct subgroup analyses based on other pregnancy or birth characteristics (gestational age at magnesium sulphate administration, birthweight, mode of birth, and concomitant maternal treatments) due to paucity of data. We assessed subgroup differences by interaction tests available within Review Manager, and, where applicable, we have quoted the χ 2 statistic and P value, and the interaction test I 2 value.
For observational studies (non-randomised trials, cohort studies, and case-control studies), we present effect estimates where possible as adjusted RRs or odds ratios if reported with 95% CIs, unadjusted RRs or odds ratios with 95% CIs, P values only, or percentages (rates), in tabular format; we used narrative synthesis to summarise the studies. Data from case reports were grouped according to common adverse outcomes, tabulated, and summarised narratively.

Study selection
The results of the search strategy, including the sources of the studies, their assessment, and final inclusion are shown in Fig 1. The database searching identified 5,890 records, and other searching identified a further 11 records. Review of the titles and abstracts and exclusion of irrelevant and duplicate records yielded 777. Of these, we excluded 572 for the documented reasons (see S3 Text for list of records excluded due to absence of an English translation). We included a total of 205 articles, relating to 197 studies. See S4 Text for references for all included studies. In the case of multiple publications from the same study, we included the report with the most relevant data as the primary reference, and only included further publications as secondary references if they provided additional relevant data. The methodological quality of the 40 trials varied considerably. Considering selection bias, 18 trials were at low risk, reporting adequate methods for sequence generation and allocation concealment. Twelve and 8 trials received an unclear judgement for 1 and 2 of the selection bias domains, respectively. Two trials appeared to be quasi-randomised and thus were at high risk of selection bias. Twelve trials were at low risk of performance and detection bias (with blinding of participants, personnel, and outcome assessors); 21 were at high risk of both performance and detection bias (all trials of different magnesium sulphate regimens, with no reported blinding), and the remaining 7 trials had an unclear judgement for 1 or 2 of the blinding domains. The majority of trials (31) were at low risk of attrition bias for neonatal outcome data, though for 9 trials, this was unclear. Only 5 trials were at low risk of reporting bias, 1 was at high risk of reporting bias, and for the remaining 34 trials, selective reporting was unclear. Magnesium sulphate regimens assessed varied considerably: 4-gram IV loading dose only (2 trials), 4-gram IV loading dose and 1-gram-per-hour IV maintenance dose (4 trials), 4-gram IV loading dose and 2-gram-per-hour IV maintenance (4 trials), 6-gram IV loading dose and 2-gram-per-hour IV maintenance dose (6 trials), and 4-gram IV and 10-gram IM loading dose and 5-gram IM maintenance dose every 4 hours (2 trials), with duration of treatment generally ranging from 12 to 24 hours. Fourteen trials compared magnesium sulphate with a placebo, while 4 trials had a no-treatment comparison (see Table 1 and S1 Appendix for effect estimates, forest plots, and funnel plots).

Magnesium sulphate versus placebo or no treatment
No clear difference was seen between magnesium sulphate and placebo/no treatment for the primary review outcome perinatal death (RR 1.01; 95% CI 0.92 to 1.10; 8 trials, 13,654 babies; analysis 1.1), nor for stillbirth, neonatal death (no obvious asymmetry observed on visual assessment of funnel plot), death later than 28 days but before discharge, early neonatal death, or late neonatal death (Table 1; S1 Appendix). When considering indication for use, the tocolysis subgroup showed an increase in perinatal death (RR 7.99; 95% CI 1.00 to 63.49; 2 trials, 257 babies; analysis 1.1.1) that was not observed in the pre-eclampsia or fetal neuroprotection subgroups. The subgroup interaction test, however, did not indicate a differential effect according to treatment indication (χ 2 = 4.07, P = 0.13, I 2 = 50.8%). For the remaining mortality outcomes, subgroup interaction tests did not indicate differential treatment effects according to indication for administration (see Tables 2 and 3 for effect estimates for individual subgroups and results from subgroup interaction tests).
Babies exposed to antenatal magnesium sulphate had a 67% relative increase in the risk of having an Apgar score less than 7 at 1 minute (RR 1.67; 95% CI 1.02 to 2.73; 2 trials, 199 babies; analysis 1.7), and over 2 times the risk of need for volume expansion compared with babies not exposed (RR 2.03; 95% CI 1.01 to 4.10; 1 trial, 87 babies; analysis 1.28). A subgroup of babies born less than 32 weeks gestation exposed to antenatal magnesium sulphate had a 62% relative reduction in the risk of intracerebral echodensity, compared with babies not exposed (RR 0.38; 95% CI 0.19 to 0.79; 1 trial, 1,613 babies; analysis 1.46). While a difference in intracerebral echolucency was not observed in all babies, a 39% relative reduction was seen for babies born less than 32 weeks exposed to antenatal magnesium sulphate (RR 0.61; 95% CI 0.38 to 0.97; 1 trial, 1,613 babies; analysis 1.47.2) ( Table 1; S1 Appendix).
There were no clear differences between magnesium sulphate and placebo/no treatment for all remaining secondary outcomes reported (Table 1; S1 Appendix).

Lower versus higher dose regimens
This comparison included 8 trials, with 6 assessing magnesium sulphate for treatment of eclampsia or severe pre-eclampsia [22,23,44,52,56,59], and 2 for the prevention of preterm birth (tocolysis) [26,60]. Regimens assessed varied: lower dose regimens included a 4-to 10-gram loading dose with a 0.625-to 2-gram-per-hour maintenance dose; higher dose regimens assessed included a 4-to 14-gram loading dose with a 1.25-to 5-gram-per-hour maintenance dose (see Table 4 and S1 Appendix for effect estimates and forest plots).
No clear differences between the lower and higher dose regimens of magnesium sulphate were seen for the primary review outcome perinatal death (RR 1.01; 95% CI 0.75 to 1.36; 6 trials, 543 babies; analysis 2.1), nor for stillbirth or neonatal death (Table 4; S1 Appendix). For all mortality outcomes, subgroup interaction tests did not indicate differential treatment effects according to indication for administration of antenatal magnesium sulphate (see Table 5 for effect estimates for individual subgroups and results from subgroup interaction tests).
Babies exposed to the lower dose versus higher dose regimens of magnesium sulphate had an increased risk of neonatal intensive care unit admission (RR 1.75; 95% CI 1.06 to 2.88; 5 trials, 409 babies; analysis 2.17). On average, babies exposed to the lower dose regimen had a longer duration of stay in the neonatal intensive care unit compared with those exposed to the higher dose regimen (MD 3.10 days; 95% CI 0.78 to 5.42; 1 trial, 104 babies; analysis 2.18) ( Table 4; S1 Appendix). Antenatal magnesium sulphate and adverse neonatal outcomes No clear differences were seen between the lower and higher dose regimens of magnesium sulphate for the remaining secondary outcomes reported (Table 4; S1 Appendix).

Loading dose versus loading and maintenance doses
This comparison included 5 trials, all assessing magnesium sulphate for the prevention or treatment of eclampsia [25,41,50,53,57]. Trials compared a cumulative 8-, 10-, or 14-gram loading dose (4 grams IV and 4 to 10 grams IM) with Dhaka or Pritchard's regimen (see descriptions above; and see Table 6 and S1 Appendix for effect estimates and forest plots).
No clear differences for loading dose only versus loading and maintenance dose regimens were seen for the primary review outcome perinatal death (average RR 0.94; 95% CI 0.33 to 2.72; 3 trials, 632 babies; analysis 4.1), nor for stillbirth, neonatal death, or neonatal death at less than 7 days (Table 6; S1 Appendix).
No clear differences for loading dose only versus loading and maintenance dose regimens were seen for the remaining secondary outcomes reported (Table 6; S1 Appendix).

Serial IV boluses versus continuous IV maintenance dose
This comparison included 1 trial, assessing magnesium sulphate for the treatment of severe pre-eclampsia, and compared a serial intravenous bolus regimen (6-gram IV loading dose, and 2-gram IV bolus over 10 minutes every 2 hours as maintenance) with a continuous infusion (4-gram IV loading dose, and 1-gram-per-hour continuous IV maintenance dose) [37] (see Table 6 and S1 Appendix for effect estimates and forest plots).
No clear differences between the serial bolus and continuous infusion regimens were seen for the primary review outcome perinatal death (RR 0.44; 95% CI 0.08 to 2.34; 1 trial, 197 babies; analysis 5.1), nor for stillbirth or neonatal death (Table 6; S1 Appendix).
No clear differences between the serial bolus and continuous infusion regimens were seen for the remaining secondary outcomes reported (Table 6; S1 Appendix).

Short versus standard maintenance course
This comparison included 1 trial, assessing magnesium sulphate for the treatment of eclampsia. Both groups received a 4-gram IV and 10-gram IM loading dose, followed by either a short maintenance course (2 doses of 5 grams IM 4 hours apart after birth or last seizure) or a standard maintenance course (5 grams IM every 4 hours for 24 hours after birth or last seizure) [29].
No clear difference between the short and standard maintenance course regimens was seen for stillbirth. No clear difference between the short and standard maintenance course regimens was seen for the only other secondary outcome reported, birth asphyxia (see Table 6 and S1 Appendix for effect estimates and forest plots).

Slower versus standard rate of loading dose
This comparison included 1 trial, assessing magnesium sulphate for fetal neuroprotection, and compared a slower (over 60 minutes) versus standard (over 20 minutes) rate of administering a 4-gram IV loading dose of magnesium sulphate (all women received a 1-gram-per-hour maintenance dose for 24 hours or until birth) [24].
No clear difference between a slower and standard rate of loading dose administration was seen for stillbirth (see Table 6 and S1 Appendix for effect estimate and forest plot).

Weaning versus no weaning
This comparison included 1 trial, assessing magnesium sulphate for the prevention of preterm birth (tocolysis), and compared weaning (by 1 gram IV per 4 hours) versus not weaning magnesium sulphate (all women received a 6-gram IV loading dose, and 2-to 3.5-gram IV maintenance dose per hour until tocolysis was achieved) [42].
No clear difference between weaning and no weaning was seen for Apgar score less than 7 at 5 minutes (see Table 6 and S1 Appendix for effect estimate and forest plot). Antenatal magnesium sulphate and adverse neonatal outcomes

Evidence from non-randomised comparative studies
One hundred thirty-eight non-randomised studies were included: 5 non-randomised trials, 35 prospective cohort studies (7 with nested case-control analyses), 82 retrospective cohort studies (16 with nested case-control analyses), 8 non-concurrent cohort studies, and 8 case-control studies . The characteristics of the studies and risk of bias assessments are detailed in S1 and S2 Tables. There was substantial variation in the characteristics of these studies regarding participants, indications for use of magnesium sulphate (prevention or treatment of eclampsia: 30 studies; prevention of preterm birth (tocolysis): 28 studies; fetal neuroprotection: 25 studies; combination of aforementioned indications: 38 studies; unclear: 17 studies), comparison groups, outcomes assessed (and their definitions), and analysis methods employed. Methodological quality (specifically in relation to the risk of bias for reported review outcomes of interest) also differed across the studies, with overall judgements of unclear (specifically when only abstracts were available), high, moderate to high, and moderate risk of bias assigned to 43 studies, 49 studies, 35 studies, and 11 studies, respectively. The most common concerns across studies related to the potential for confounding (with no attempt to balance allocation between groups or match groups, and/or important confounding variables not taken into account in relevant outcome analyses), detection bias (with the consistent implementation of valid and reliable measures being unclear, and/or the absence of blinding of exposure or outcome assessors), and performance bias (with protocols not available to assess important variations).
The primary review outcome, perinatal death, was reported by 11 of the 138 non-randomised studies, 10 of which showed either a possible reduction (or lower rate) or no clear difference (or similar rate) in perinatal death among babies exposed to magnesium sulphate compared with no magnesium sulphate or a different magnesium sulphate regimen. A possible increase in perinatal death, specifically among babies exposed to >48 versus �48 grams of magnesium sulphate for tocolysis was shown in in the 11th study (retrospective cohort of 127 babies, moderate to high risk of bias) [183]. See Table 7 and S3 Table for summaries of individual study results.
For the majority of secondary pre-specified and non-pre-specified adverse neonatal outcomes reported, the results from non-randomised studies were consistent with those observed in randomised controlled trials, with no clear differences (and in some cases, possible benefits of magnesium sulphate) observed. The direction of the findings (no clear difference, possible benefit, possible harm, or mixed) from the non-randomised studies are summarised in Table 8, with the detailed individual study results provided in S3 Table. Seventeen of the 138 non-randomised studies (14 at moderate or moderate to high risk of bias, and 3 at unclear risk of bias [abstracts only]) observed a possible increase in the risk of adverse neonatal outcomes with antenatal magnesium sulphate, with some consideration of important confounding variables in relevant outcome analyses [68,73,84,90,106,108,109, 129,132,134,140,152,153,173,180,184,191]. Potential increased risks with antenatal magnesium sulphate of 4 outcomes (discussed below) were observed by more than 1 study. For the remaining outcomes (nosocomial infection [184], enteral feeding intolerance [68], respiratory disease composite [153], pulmonary interstitial emphysema [191], early germinal matrix/intraventricular haemorrhage [180], thalamostriate or mineralising vasculopathy [152], and a composite neonatal adverse outcome [73]), single studies reported possible harms.
Potential increased risk of neonatal death before intensive care unit discharge, and of a composite outcome of neonatal death before intensive care unit discharge and/or necrotising enterocolitis, was shown among a subgroup of babies born less than 26 weeks gestation with the use of antenatal magnesium sulphate for fetal neuroprotection (293 babies, in a retrospective cohort of 697 babies born less than 28 weeks gestation) [129]. A possible increased risk of the composite outcome spontaneous intestinal perforation or neonatal death was also shown among a subgroup of babies born less than 25 weeks gestation with higher cumulative   Antenatal magnesium sulphate and adverse neonatal outcomes Table 8. Summary of outcomes from non-randomised studies.    antenatal magnesium sulphate doses for fetal neuroprotection (non-concurrent cohort of 155 babies born less than 1,000 grams) [173]. A possible increased risk of patent ductus arteriosus was observed with magnesium sulphate given for pre-eclampsia (retrospective cohort of 81 'very low birthweight' babies; abstract only) [140] or for pre-eclampsia or tocolysis (retrospective cohort of 941 babies born 500 to 1,000 grams) [90]. Further, a possible increased risk of patent ductus arteriosus in babies born 26 weeks gestation or later was shown with cumulative antenatal magnesium sulphate doses for pre-eclampsia or tocolysis of at least 50 grams (retrospective cohort of 941 babies born 500 to 1,000 grams) [90]. Potential increased risk of symptomatic patent ductus arteriosus, and of failure of early closure of the ductus arteriosus, was also observed with the use of antenatal magnesium sulphate for tocolysis (retrospective cohort of 160 babies born less than 28 weeks gestation, who all received indomethacin prophylaxis) [132]. A potential increased risk of intraventricular haemorrhage grade 3 or 4 was observed with antenatal magnesium sulphate for tocolysis (case-control study of 121 babies born less than 1,500 grams) [134], and with a higher dose regimen of antenatal magnesium sulphate for fetal neuroprotection (6-gram IV loading dose and 2-gram-per-hour IV maintenance dose for 12 hours versus 4-gram IV loading dose only) (retrospective cohort, including 54 babies exposed to magnesium sulphate within 12 hours of birth) [84].

Studies showing no clear difference
A possible increased risk of neonatal intensive care unit admission with the use of antenatal magnesium sulphate for pre-eclampsia was observed (retrospective cohorts of 264 babies and 2,166 babies born at 37 weeks gestation or later) [106,109]. Further increased risks of neonatal intensive care unit and special care unit admission were observed with higher total hours, higher total doses, more than 12 hours, and more than 30 grams of antenatal magnesium sulphate for pre-eclampsia (retrospective cohort of 242 babies born at 35 weeks gestation or later) [108].
Clinical features of neonatal hypermagnesemia, magnesium 'toxicity', or magnesium 'intoxication' at birth were the focus of 5 reports (35 neonates), in which antenatal magnesium sulphate was given for pre-eclampsia/eclampsia (with exposure durations and doses ranging from 3.5 to 46 hours and 11 to 60.4 grams prior to birth, respectively) [202,205,206,213,218], and were variably described throughout the remaining reports. These included low Apgar scores, apnoea, cyanosis, hypotonia, and/or hyporeflexia, with or without the need for active resuscitation and calcium gluconate administration. Two reports described neonatal death following antenatal magnesium sulphate exposure. In the first report, death was considered to be related to 'the toxic effects of magnesium on the myocardium' when given for tocolysis (4-gram IV loading dose followed by 2.5-gram-perhour IV maintenance dose for approximately 1 day: 51.4 grams total) [204], and in the second, 1 death (of 7) was attributed to 'overdose' (unclear dose/regimen) of magnesium sulphate when given for pre-eclampsia/eclampsia [210].
In the context of hypermagnesemia at birth (following exposure to a total of 24 to 28 grams of magnesium sulphate for pre-eclampsia), neonatal gentamicin administration for suspected sepsis was associated with respiratory arrest and cardiac arrest in 2 reports [211,215]. Other specific adverse outcomes attributed to antenatal magnesium sulphate exposure included • Microcolon or meconium-plug syndrome (3 reports, 14 neonates, when given for preeclampsia/eclampsia: 25 to 41 grams in the day prior to birth in 1 report; regimen not described in 2 reports) [201,209,216]; • Nonoliguric hyperkalaemia (1 report, 1 neonate, when given for pre-eclampsia: 0.1 gram to 2 grams per hour IV for 12 days) [217]; • Bone abnormalities (commonly metaphyseal osteopenia, in some cases leading to fracture) (6 reports, 35 neonates, when given for tocolysis: ranging from 1 to 4 grams per hour for between 8 and 13 weeks) [200,203,207,208,212,214].

Discussion
Overall, no clear difference in our primary review outcome, perinatal death, was shown in the randomised trials comparing antenatal magnesium sulphate with placebo/no treatment, nor in regimen comparisons in randomised trials. While 11 of the 138 non-randomised studies reported on perinatal death, only 1 cohort study (at moderate to high risk of bias) observed a possible increased risk of perinatal death, with high-dose (more than 48 grams) antenatal magnesium sulphate exposure for tocolysis [183]. Results for secondary adverse neonatal outcomes were reassuring, with very few clear differences observed between antenatal magnesium sulphate and placebo/no treatment or between different magnesium sulphate regimens. Where possible harms of magnesium sulphate were seen, commonly no confounders were taken into account (and studies were judged to be at high risk of bias), study samples were small (less than 200 babies), and/or differences were observed in (non-formal) subgroup analyses only. Non-randomised studies identified a limited number of outcomes justifying further evaluation, such as from large, high-quality studies (prospective cohorts, individual participant data meta-analyses, or randomised trials of regimen comparisons). These included neonatal death and intestinal morbidity in very preterm neonates with exposure for fetal neuroprotection, patent ductus arteriosus in very preterm or very low birthweight neonates with exposure for pre-eclampsia or tocolysis, and intensive care unit admission in term neonates with exposure for pre-eclampsia. Case reports suggested an association between neonatal bone abnormalities and long-term, high-dose exposure to antenatal magnesium sulphate for tocolysis.
We are not aware of any other published systematic reviews with a focus on potential adverse outcomes for neonates following antenatal magnesium sulphate exposure, though we identified 3 review registrations focused specifically on magnesium sulphate for tocolysis and the outcomes neonatal respiratory depression (CRD42017058912), patent ductus arteriosus (CRD42017060049), and bone abnormalities (CRD42017062550) [18]. Three previous systematic reviews have assessed 'adverse events' [13], 'side effects' [219], and 'safety' [220] for women, and a further systematic review has assessed the effects of antenatal magnesium sulphate specifically on fetal heart rate parameters, finding a small negative effect on rate, variability, and accelerative pattern, 'not sufficient clinically to warrant medical intervention' [14].
Our review findings are consistent with those from the relevant Cochrane reviews comparing antenatal magnesium sulphate with placebo/no treatment, or different magnesium sulphate regimens [1,2,4,9,221], though our review includes a wider range of outcomes. As was observed in our review's tocolysis subgroup for perinatal death, the Cochrane review assessing magnesium sulphate for tocolysis demonstrated a borderline increased risk of fetal, neonatal, or infant death with antenatal magnesium sulphate [9].
A recent non-systematic narrative review evaluated 'whether antenatal MgSO 4 is beneficial or harmful' in extremely and very preterm neonates [222]. Relevant systematic reviews, metaanalyses, randomised controlled trials, and observational studies were retrieved, with a broad search strategy focused on neuroprotection and cerebral palsy, necrotising enterocolitis, and spontaneous intestinal perforation. The narrative review suggested that current evidence supports the neuroprotective role of antenatal magnesium sulphate for preterm neonates, and that, while the effects are 'controversial' and 'not well established', a 'high index of suspicion of gastrointestinal complications in extremely preterms, particularly < 26 weeks of gestation' is recommended [222]. While our review similarly identified a possibility of harm, the relevant studies were of questionable methodological quality, and our systematic review included additional reports (of higher quality, and involving much larger cohorts) that indicated no increased risk of intestinal morbidity.
The findings of this review are reassuring and can be considered in conjunction with those from relevant reviews demonstrating a clear benefit of antenatal magnesium sulphate [1,2,4], and current international clinical practice guideline recommendations [3,7]. Our review findings of possible adverse outcomes with long-term, high-dose use for tocolysis have implications for settings with continued use for this indication [8] in spite of the absence of benefit shown in systematic reviews and international guidance [7][8][9].

Strengths and limitations
The main limitations of our review relate to missing data for important outcomes across most studies, the inclusion of published data only, and the heterogeneity of included studies.
Of the 40 randomised trials included in this review, our primary outcome (perinatal death) was reported by 22 (55%); it was reported by only 11 (8%) of the 138 included non-randomised studies. Aside from related mortality outcomes (stillbirth and neonatal death), all other adverse outcomes were reported sparsely, by less than a third of trials, with many outcomes reported by single trials only. While a broader range of adverse neonatal outcomes were reported by the non-randomised studies, apart from neonatal death (50 studies), intraventricular haemorrhage (39 studies), and necrotising enterocolitis (36 studies), all other outcomes were reported by less than a fifth of studies, again, with many reported by single studies only.
In addition to missing data for important outcomes across most studies, a further limitation includes the number of studies with relatively small sample sizes comparing different antenatal magnesium sulphate regimens. Moreover, many studies assessing antenatal magnesium sulphate for relevant indications were not included due to lack of reporting of adverse neonatal outcomes.
We searched extensively across multiple databases and reviewed reference lists for additional reports; however, we did not seek unpublished data. Recent evidence has suggested that while much adverse event information remains unpublished, inclusion of such data generally does not change the direction or statistical significance of pooled risk estimates [223]. While we were not able to fully evaluate non-English publications without available translations for inclusion, we have provided a list of these for readers to consider (S3 Text).
As the aim of the review was to provide a comprehensive, general view of potential unintended adverse outcomes, we designed this review with broad scope [15,16]. This presented challenges, including the number of diverse outcomes, inconsistent reporting, and the vast quantities of heterogeneous data. The study characteristics (including designs, settings, participating women and neonates, and antenatal magnesium sulphate regimens) varied greatly, and reporting was commonly incomplete. The ability to conduct subgroup analyses for the randomised trials was limited, and we did not inappropriately pool data from non-randomised studies. We conducted this review in accordance with recommendations for systematic reviews of adverse events [15][16][17], and of randomised and non-randomised studies more generally [224]. Our evaluation thus provides a firm basis for any further, narrowly focused studies of specific outcomes and characteristics.

Conclusions
In conclusion, our findings do not support any clear associations between perinatal death or other adverse neonatal outcomes and antenatal magnesium sulphate exposure when given for the beneficial indications of maternal neuroprotection in pre-eclampsia/eclampsia and fetal neuroprotection in cerebral palsy prevention. To further inform safety recommendations of this widely used treatment in pregnancy, future research should be directed towards identified research gaps surrounding specific adverse neonatal outcomes and the impact of particular regimen, pregnancy, and/or birth characteristics.