Vaginal misoprostol versus vaginal dinoprostone for cervical ripening and induction of labour: An individual participant data meta‐analysis of randomised controlled trials

Induction of labour (IOL) is common practice and different methods carry different effectiveness and safety profiles.


| I N TRODUC TION
Induction of labour (IOL) is carried out when the risks of advancing gestation outweigh maternal and fetal benefits.It is one of the most frequent obstetric interventions, accounting for nearly one-quarter of all births in high-income countries, but with generally lower rates in low-and middleincome countries. 1][4] Cervical ripening is usually the first step of IOL in the case of unripe cervices.In contemporary practice, misoprostol and dinoprostone are the two most common pharmacological methods used for cervical ripening. 57][8][9] It is important to find an optimal method of IOL that maximises both effectiveness and safety, as one-fifth of inductions for unripe cervices require a caesarean delivery because of failed induction. 5Although oral misoprostol is increasingly recommended, 10 the vaginal route for misoprostol remains more popular.Vaginal misoprostol has been investigated for various dosing regimens, and there is convincing evidence regarding superior safety with low-dose compared with highdose misoprostol, demonstrating similar effectiveness in labour induction. 11,12Although a more common clinical scenario is to choose between oral misoprostol and vaginal dinoprostone, studying vaginal misoprostol is still valid to distinguish between the different routes of misoprostol.
There have been conventional data meta-analyses comparing the effectiveness and safety outcomes between vaginal misoprostol and vaginal dinoprostone. 13,14However, these data meta-analyses carry limitations owing to the poor reporting of trials and differences in the definitions of outcomes.6][17] Thus, performing a conventional data meta-analysis is not the best way to address this question.Individual participant data (IPD) meta-analysis generates a harmonised database of eligible studies to be re-analysed according to a predefined plan, before which data integrity and replicability need to be assured. 18For this reason, IPD metaanalysis yields higher-quality data to guide clinical practice. 18,19n addition, IPD methodology offers the ability to standardise exposure and outcome definitions, construct composite outcomes of rare adverse events, improve the trustworthiness of the evidence and accommodate advanced statistical analysis with satisfactory power.Thus, comparing the effectiveness and safety of vaginal misoprostol with vaginal dinoprostone for labour induction in an IPD database provides all these benefits, leading to more robust evidence on the topic.
Here, we report an IPD meta-analysis comparing the effectiveness and safety of vaginal misoprostol with vaginal dinoprostone for labour induction.We aim to compare the effectiveness, and perinatal and maternal safety outcomes when labour is induced with vaginal misoprostol versus vaginal dinoprostone.

| M ET HODS
This international collaborative IPD meta-analysis followed a prospectively registered protocol (PROSPERO: CRD42022321378) and a statistical analysis plan produced in advance of data lock and analysis (Appendix S2).The findings are reported following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses checklist for IPD (PRISMA-IPD). 20

| Search strategy
The initial search was conducted in March 2022, and the search was updated in September 2022 and again in March 2023.Databases including CINAHL Plus, ClinicalTrials.gov,Cochrane Pregnancy and Childbirth Group Trial Register, Ovid Embase, Ovid Emcare, Ovid MEDLINE, Scopus and the World Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP), and the reference lists of retrieved studies, were searched to identify further published randomised controlled trials (search strategy outlined in Table S1).There were no language restrictions, and all published and unpublished data were eligible.Two investigators (MP and FC) independently reviewed the identified titles and abstracts followed by the full texts for eligibility, with disagreements solved by a third reviewer (WL).

| Eligibility criteria
Randomised controlled trials (RCTs) of vaginal misoprostol versus vaginal dinoprostone for IOL, irrespective of gestational age, were included.Women with prelabour rupture of membranes and trial of scar were also included.We did not exclude RCTs according to the dose of vaginal misoprostol.However, we used the definition in the Cochrane Review by de Vaan for low-dose misoprostol (≤50 μg every ≥4 h) to categorise the dose. 21Trials with any other methods for IOL, including intracervical dinoprostone, cluster randomised trials and quasi-experimental studies, were excluded.Potentially suitable clinical trials were identified from inception.

| Data access
We approached the investigators of all potentially eligible RCTs to further confirm eligibility and requested individual participant-level data from RCTs confirmed as eligible.The contact details of trial investigators were obtained through the published articles or the websites of their associated institutions, and invitations to participate were sent by e-mail at least four times if there was no response.Where the contact details for the primary or corresponding authors were unavailable or no response was obtained, attempts were made to contact other authors involved in the RCTs and lead/co-authors were copied in at least every month until March 2023.In addition to e-mailing the authors, in the absence of a response attempts were made to contact the authors through social media platforms or their institutions or clinics by e-mail or phone call.Our academic contacts from particular countries were called to help contact any authors/ institutions did not respond to the initial enquiries.Journal editors or funding bodies were also contacted as a last resort for some of the studies.Once RCT investigators responded indicating that that they were interested in participating, regular e-mails were sent every month coordinating the data-sharing process and clarifying the details.
The RCT investigators supplied de-identified participantlevel data, which were harmonised and recoded to the predefined IPD meta-analysis definitions.Data were requested for all women randomised, even if excluded from original trial analyses.Where possible, the received data were examined for missing data, error, internal consistency, consistency with the publication, and pattern of treatment allocation and data presentation.Identified issues were communicated with RCT investigators for a solution before acceptance in the IPD meta-analysis data set.Studies with insolvable inconsistencies or trustworthiness concerns were excluded from the analysis.

| Outcomes and effect modifiers
The primary outcomes (Table 1) included vaginal birth rate, a composite measure of adverse perinatal outcomes and a composite measure of adverse maternal outcomes.The composite adverse perinatal outcome included at least one of stillbirth, neonatal death, neonatal Apgar score of <7 at 5 min, arterial umbilical cord pH of <7.1, suspected or proven neonatal infection, neonatal seizures, hypoxic ischaemic encephalopathy, neonatal intensive care unit (NICU) admission, need for mechanical ventilation, meconium aspiration syndrome and severe neonatal respiratory compromise.The composite adverse maternal outcome included admission to ICU, maternal infection (suspected or confirmed), postpartum haemorrhage of ≥1000 mL, maternal death and uterine rupture.When calculating composite outcomes, the presence of one or more of the predefined components of the composite outcome led to a positive composite maternal or perinatal outcome.
Secondary outcomes (Table 1) further explored various aspects of effectiveness and safety.Delivery outcomes included the mode of birth (unassisted/spontaneous vaginal, instrumental vaginal or caesarean birth), indication for instrumental vaginal birth, indication for caesarean birth and the cumulative rate of vaginal birth.Labour progression outcomes included uterine tachysystole, uterine hyperstimulation, the need for oxytocin augmentation and the presence of meconium-stained amniotic fluid.Secondary neonatal safety outcomes assessed individually included an Apgar score of <7 at 5 min, an arterial umbilical cord pH of <7.10 and NICU admission.Secondary maternal safety outcomes assessed individually included maternal antibiotic use, maternal fever (≥38°C) and severe postpartum haemorrhage (≥1000 mL).Potential effect modifiers of interest for vaginal birth included parity, maternal age, body mass index (BMI), initial Bishop score and indications for labour induction (Table 1).

| Risk of bias of included studies and certainty of evidence
The risk of bias (RoB) was independently assessed by two researchers (MP and FC) using the Cochrane RoB 2 tool. 22ifferences were resolved by consensus; if the information was insufficient, clarification was sought from trialists.We used the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach to assess the overall certainty of the evidence for the primary outcomes. 23

| Data synthesis
For each outcome, an intention-to-treat analysis was performed using all available data comparing vaginal misoprostol with vaginal dinoprostone (reference group).In this IPD meta-analysis, vaginal dinoprostone was considered as the reference group for all outcomes.
Our primary analysis strategy involved a two-stage meta-analysis method to synthesise the IPD.However, for outcomes that had no events in any intervention group of any included RCT, a one-stage method was used as the primary analysis.In the first step of the two-stage method, we compared outcomes between vaginal misoprostol and vaginal dinoprostone for each included study, adjusting for maternal age and parity.For binary outcomes, we calculated odds ratios (ORs) along with 95% confidence intervals (95% CIs) using logistic regression.For timeto-event outcomes like the cumulative rate of vaginal births, we estimated hazard ratios (HRs) and 95% CIs using the sub-distribution hazard model, which considered caesarean birth as a competing risk.In the second step, we combined the generated relative estimates using random-effects models (restricted maximum-likelihood estimator with Hartung-Knapp-Sidik-Jonkman variance correction). 24,25For the one-stage method, we used multilevel mixed-effects logistic regression (with a stratified intercept by study and with a random treatment effect, covariates as fixed effects and a maximum-likelihood estimator), adjusting for maternal age and parity. 25We tested treatment-covariate interactions for vaginal birth using interaction terms between treatment and potential effect modifiers, adjusting for maternal age and parity.To avoid ecological bias, we only considered within-trial interactions.All variables besides the identification variable were checked for missing values and entries that are outside the expected ranges.The pattern of missing data was analysed separately for each data set using the pattern charts to look for monotonicity and the likely mechanism of missingness. 26In the event of missing values for covariates or potential effect modifiers in any RCT, we performed multiple imputations using chained equations (20 imputed data sets) within the RCT before the meta-analysis.
We extracted summary data for vaginal birth from RCTs that did not contribute IPD and performed a summary data meta-analysis using the same random-effects model to assess the risk of data unavailability bias of the IPD meta-analysis.This was not possible for the two composite outcomes of maternal and neonatal safety.Sensitivity analysis for primary outcomes was performed as a onestage meta-analysis to investigate the impact of the metaanalytical method on the pooled estimates, as well as an as-treated analysis to compare with the intention-to-treat analysis.We intended to examine complications requiring maternal hospital assistance; however, this was not readily available in the trials included.
Statistical analyses were performed in R (R Foundation for Statistical Computing, Vienna, Austria): the 'meta' package was used for the two-stage meta-analysis and aggregate data meta-analysis; the 'lme4' package was used for the one-stage meta-analysis; and the 'survival' package was used for competing risk analysis. 27SPSS 28.0 (IBM, Armonk, NY, USA) was used for summary statistics.Neither the public nor the patients were involved in planning this IPD meta-analysis.

| Study selection and participants
Our systematic search identified 1744 unique references (for the PRISMA IPD flow diagram, see Figure 1).After the screening, 52 RCTs (both high-and low-dose vaginal misoprostol vs vaginal dinoprostone) were deemed potentially eligible for inclusion and enquiries were sent to the trial investigators; of these, 42 RCTs did not participate in this IPD meta-analysis.The 42 RCTs did not share data for the following reasons: 12 RCTs had incorrect or no contact details, nine author groups never responded to multiple rounds of contact, three author groups initially responded but then discontinued contact and seven author groups declined to share the data.The requested data were unavailable for 11 RCTs.Table S4 shows the characteristics of the RCTs excluded after enquiry or not contributing to the analysis.

| Study characteristics
Of the ten studies that shared IPD (4326 women) for this meta-analysis, two RCTs had to be excluded following data checks.One was excluded because of imbalance in baseline characteristics and this trial used coin tossing for randomisation, without allocation of concealment. 28The authors of the other RCT excluded shared only a part of the data set (n = 30/72) without any ability to reproduce the published data. 291][32][33][34][35][36] For the misoprostol group, all eight RCTs that shared IPD used low-dose vaginal misoprostol (≤50 μg and ≥4 hourly).Four were three-arm RCTs, where only the arms regarding vaginal misoprostol and vaginal dinoprostone were included in this IPD meta-analysis. 31,33,35,36

| Risk of bias of included studies
On screening for risk of bias, RCTs were mostly identified as having 'some concerns' , largely owing to the inability to blind the participants and/or investigators because of the nature of the intervention, the unavailability of a prespecified analysis plan or the lack of prospective trial registration (Figures 2 and  S1).Of note, seven out of eight RCTs were performed before the introduction of mandatory trial registration in 2010.

| Descriptive analysis of participants
Of the 4180 participants included, 2077 women were allocated to receive vaginal misoprostol and 2103 women were allocated to receive vaginal dinoprostone.There were 2506 (60.0%) nulliparous women and 1674 (40.0%) multiparous women.The baseline participant characteristics of each included trial are available in Table S2.

| Publication bias
The funnel plot of all RCTs (participating and nonparticipating) shows that all RCTs fall around the middle, representing a low risk of publication bias (Figure S17).The datasheet we received only had 640 participants after post-randomisation exclusions, and the published article included 650 participants.This was discussed with the trial investigators, and 640 was the agreed final sample; comparing baseline variables and results did not yield any reason to exclude this trial from this meta-analysis.

| IPD unavailability bias
Out of the 42 RCTs that did not share data (Table S4) with this IPD meta-analysis, we could extract summary data for vaginal birth from 35 RCTs.Aggregate data meta-analysis combining all the RCTs found a comparable rate of vaginal delivery between the vaginal misoprostol and vaginal dinoprostone groups (45 RCTs, 12 733 women, OR 1.10, 95% CI 0.92-1.31,P = 0.29, I 2 = 62%) (Figure S18), suggesting a low risk of data unavailability bias.Aggregate data meta-analysis combining all the RCTs with low-dose misoprostol regimens also found comparable vaginal delivery rates between the vaginal misoprostol and vaginal dinoprostone groups (34 RCTs, 9792 women, OR 1.10, 95% CI 0.87-1.32,P = 0.52, I 2 = 67%) (Figure S19).After applying the Cochrane risk-of-bias tool, eight RCTs were excluded owing to 'some concerns' in the randomisation domain (Figure S20).Aggregate data meta-analysis with the remaining 26 RCTs showed that the rate of vaginal delivery was comparable between the vaginal misoprostol and vaginal dinoprostone groups (26 RCTs, 8429 women, OR 0.99, 95% CI 0.80-1.23,P = 0.93, I 2 = 67%) (Figure S21).Results of these conventional aggregate data meta-analysis for vaginal birth in both shared and non-shared RCTs have been summarised in Table S5.

| Sensitivity analysis
Sensitivity analysis for primary outcomes using the one-stage method produced similar findings (Table S6).An 'as treated' analysis of the primary outcomes (Figures S22-S24 and Table S7) for six of the included RCTs did not lead to different conclusions from the main analyses for vaginal delivery and perinatal composite outcomes of the primary outcomes.The maternal composite outcome was available only from four studies and the result was not significant in the 'as treated' analysis (four RCTs, 2773 women, aOR 0.77, 95% CI 0.54-1.12,P = 0.11, I 2 = 0.0%).This is likely to be associated with the reduced sample size.

| Post hoc analysis
Post hoc analysis for primary outcomes comparing two misoprostol doses (50 and 25 μg) used in the included RCTs produced similar findings (Table S8) for vaginal delivery and perinatal composite outcomes.The maternal composite outcome was available only from one trial in the misoprostol 50 μg group (369 women, aOR 0.59, 95% CI 0.34-1.02,P = 0.06) and five RCTs in the misoprostol 25 μg group (3312 women, aOR 0.85, 95% CI 0.72-1.01,P = 0.06, I 2 = 0.0%).The trend is the same as that found in the main analysis, irrespective of the misoprostol dosage.

| Main findings
Based on an analysis of eight RCTs involving 4180 women undergoing IOL, it was found that low-dose vaginal misoprostol and vaginal dinoprostone are equally effective in achieving vaginal birth and offer similar perinatal safety profiles.Although there was a lower incidence of composite adverse maternal outcomes with low-dose vaginal misoprostol, none of the individual components reached statistical significance.

| Strengths and limitations
The key strength of this IPD meta-analysis is the predefined analysis and a larger sample size of 4180 randomised women.The largest study contributed less than 40% of the total data, 7 and therefore, the findings are unlikely to be driven by a single study.The IPD meta-analysis study design improved the statistical power, and enabled outcome harmonisation and treatment-covariate interaction analysis.The central coordinating team continuously collaborated with trial investigators to ensure data consistency and quality.Previous meta-analyses were underpowered for maternal and perinatal adverse effects, but the composite outcomes in this IPD metaanalysis had sufficient statistical power.Seven of the included RCTs had an overall low risk of bias, except for the lack of blinding owing to the nature of the intervention.Although there were 52 eligible RCTs, only eight RCTs were included in the final analysis.IPD were unavailable from 32 RCTs that used low-dose vaginal misoprostol combinations.The authors acknowledge this as a main limitation of this study.However, most of these RCTs were published before 2010 and most data were unlikely to be available anymore.Second, even with a sample size of 4180, we could not clearly demonstrate which component(s) is contributing towards better maternal safety with low-dose vaginal misoprostol, despite the significant finding for the composite outcome.Although data sharing for reanalysis is increasingly recognised as a major necessity in clinical research, in reality, IPD are not always available. 37The International Committee of Medical Journal Editors (ICMJE) highlighted a data-sharing statement from manuscript submissions reporting clinical trials effective from 1 July 2018 as an ethical obligation, because participants have put themselves at risk for these trials. 38 survey conducted by Tan et al., contacting principal investigators of all interventional trials (n = 1517) registered on the Australian New Zealand Clinical Trials Registry from 1 December 2018 to 30 November 2019, has found that despite a high willingness for sharing IPD, the actual commitment to share data is low (with 77% willingness vs 40% actual sharing).39 Unfortunately, another reason for not sharing data is that it may challenge the trustworthiness of the study or reveal that there are no data underlying the study findings at all.40 In addition to the generation of new knowledge through reanalysis and novel subgroup analyses, data sharing has an important ethical and economic dimension.41 Data sharing can, therefore, increase data validity, but it also offers more value from the original research investment and helps to avoid the unnecessary repetition of studies.
Constructing a five-item adverse maternal composite outcome improves the statistical power, but more serious complications such as maternal death or uterine rupture could be overshadowed by more common outcomes, such as maternal infection.Two RCTs did not report any of the maternal outcomes and were excluded from the analysis. 33,35Despite this, we believe that the use of composite outcomes is a practical solution for measuring safety in labour induction studies.Adverse maternal outcomes are usually rare, and therefore planning a sufficiently powered RCT to investigate any of the individual components would require a very large sample size and a huge budget, making it is not feasible.Third, seven out of eight included studies were conducted in high-resource settings.Hence, there could be potential clinical heterogeneity when the results are applied in low-and middle-income settings.Lastly, one trial was found to have a high risk for two domains because of 93 post-randomisation exclusions without any details of the distribution among the three trial arms. 35We could not get any more information from the authors as the trial was performed more than 20 years ago.Evidence with moderate certainty for maternal safety outcomes should be acknowledged.

| Interpretation
A meta-analysis of aggregate data performed by Austin et al. in 2010 found that vaginal misoprostol resulted in a higher incidence of vaginal birth with comparable safety to vaginal dinoprostone. 14However, they only included a combination of 11 RCTs with high-and low-dose vaginal misoprostol.According to a meta-analysis of eight RCTs performed by Wang et al. in 2016, 13 vaginal misoprostol requires less oxytocin augmentation and is similarly effective and as safe as dinoprostone.However, this meta-analysis missed several eligible RCTs, included two quasi-experimental RCTs, and did not distinguish the use of intracervical and vaginal dinoprostone.To the best our knowledge, the present IPD meta-analysis is the most comprehensive review of the literature on this topic and used individual-level data to generate robust evidence.
The occurrence of uterine hyperstimulation with misoprostol has been a significant concern, which might result from the use of high doses.None of the RCTs included in this IPD meta-analysis and none in the non-shared group have reported significant uterine hyperactivity with the use of low-dose misoprostol.A 2015 network metaanalysis of 611 RCTs assessing different methods used for IOL found unclear safety profiles of various methods, as most RCTs did not report rare safety events. 42High-dose vaginal misoprostol (≥50 μg) and vaginal dinoprostone have been ranked as two of the top methods in achieving vaginal birth within 24 h.However, increasing odds of uterine hyperstimulation have been reported with highdose vaginal misoprostol (≥50 μg).Two Cochrane reviews found that low-dose and high-dose misoprostol have similar effectiveness, but that low-dose misoprostol has superior safety profiles. 11,12Our findings further add to the body of evidence that low-dose misoprostol is a better option than high-dose misoprostol.
With regards to the maternal safety outcomes included in this study, there seems to be a favourable effect for low-dose vaginal misoprostol.Low-dose misoprostol may reduce the risk of maternal infection and maternal ICU admission, although these results were not statistically significant, likely associated with the limited statistical power for rare adverse outcomes.According to the molecular and immunological basis of prostaglandins, prostaglandin E1 and E2 have separate pathways of action. 43Prostaglandin E1 mainly exerts its effects on myometrial contractility and to soften the cervix. 43Whereas prostaglandin E2 plays a pivotal role in the inflammatory processes (pro-inflammatory) that trigger cervical ripening and the initiation of labour. 43his inflammatory response is characterised by an influx of inflammatory cells into the cervix and an increased expression of inflammatory cytokines, leading to a potential increased risk of infections.In post hoc analysis, the magnitude of the safety effect of misoprostol stays the same across 25 and 50 μg doses of low-dose vaginal misoprostol, compared with vaginal dinoprostone.Low-dose vaginal misoprostol provides better maternal safety, lower costs and stable storage at room temperature without compromising effectiveness or perinatal safety, making it a more attractive option for low-resource settings where cost-effectiveness is paramount.This is especially important in low-resource settings with higher maternal morbidity and mortality rates and poor refrigeration facilities.
At present, the UK National Institute for Health and Care Excellence (NICE) guideline recommends only three primary methods for cervical ripening: vaginal dinoprostone, mechanical methods (balloon catheter and osmotic dilator) and oral misoprostol. 5The American College of Obstetricians and Gynaecologists recommends similar options. 44A recent expert consensus article from the Royal College of Obstetricians and Gynaecologists has highlighted the same guidance. 45Our study suggests that low-dose vaginal misoprostol can be a useful addition to contemporary obstetric practice.This expands the range of available methods and allows for a greater degree of personalisation.The physical properties and low cost of low-dose vaginal misoprostol make it highly applicable in low-resource settings.In terms of wider implications, a large IPD meta-analysis on balloon catheters versus vaginal prostaglandins recently showed that balloon catheters had favourable perinatal safety and comparable maternal safety and mode of delivery. 46Another IPD meta-analysis has shown that oral misoprostol has better effectiveness compared with balloon catheters, with comparable maternal outcomes but with potentially worse perinatal safety. 47Further, an IPD meta-analysis comparing single balloon versus double balloon in IOL has shown comparable vaginal birth rates and maternal and perinatal outcomes. 48Based on the available high-quality evidence and the present IPD meta-analysis, it is recommended to consider low-dose vaginal misoprostol along with oral misoprostol and balloon catheters as the preferred methods for cervical ripening.However, a large network meta-analysis using IPD would be the best possible solution to summarise and draw wider and more practical implications into clinical practice for all these comparisons under one study.

| CONCLUSION
This large IPD meta-analysis shows that low-dose vaginal misoprostol and vaginal dinoprostone for IOL are comparable in terms of effectiveness and perinatal safety.However, low-dose vaginal misoprostol is likely to have a better maternal safety profile than vaginal dinoprostone.These findings are important for clinical decision-making in IOL, and current guidelines and patient information should be updated accordingly.
Further study into the effectiveness of low-dose vaginal misoprostol versus vaginal dinoprostone cannot be justified as this has been widely studied.However, further research into safety remains a potential area for further investigation.Three RCTs included maternal satisfaction, using three different questionnaire formats. 7,32,36Thus, satisfaction and long-term effects of IOL on neonates represent a research gap.

T A B L E 1
Definition of outcomes and potential effect modifiers.Primary outcomes Vaginal delivery rate: rate of vaginal delivery following randomisation per woman randomised Composite of adverse perinatal outcomes: Apgar score of <7 at 5 min, arterial umbilical cord pH of <7.1, admission to neonatal intensive care unit, severe respiratory compromise a , neonatal seizures, neonatal infection, need for mechanical ventilation b , hypoxic ischaemic encephalopathy, meconium aspiration syndrome c , neonatal death or stillbirth Composite of adverse maternal outcomes: maternal infection (temperature of ≥38°C at any time during labour or delivery, antibiotic use or clinically diagnosed infection, such as endometritis), maternal admission to intensive care unit, severe postpartum haemorrhage (>1000 mL estimated blood loss), uterine rupture or maternal death Secondary outcomes Delivery outcome: • Mode of delivery (caesarean delivery, unassisted vaginal birth, instrumental vaginal birth) • Caesarean delivery, including analysis for fetal compromise or failure to progress (if both fetal compromise and failure to progress apply as indications, fetal compromise prevails) • Indication for instrumental vaginal birth • Cumulative rate of vaginal birth (time-to-event analysis) Labour progression outcomes: • Uterine tachysystole • Uterine hyperstimulation • Oxytocin augmentation • Meconium-stained amniotic fluid Neonatal safety outcomes: • Apgar score of <7 at 5 min • Arterial umbilical cord pH of <7.10 • Neonatal intensive care unit admission Maternal safety outcomes: • Antibiotic administration d • Maternal fever of >38°C • Severe postpartum haemorrhage (>1000 mL estimated blood loss) • Maternal admission to intensive care unit • Other complications requiring hospital assistance e Potential effect modifiers

F
I G U R E 1 Trial identification (PRISMA-IPD flow diagram).RCTs, randomised controlled trials.

F I G U R E 2
Risk-of-bias assessment of included trials in the intention-to-treat population.

F I G U R E 3
Forest plots of IPD meta-analyses of the primary outcomes.95% CI, 95% confidence interval; HK, Hartung-Knapp-Sidik-Jonkman variance correction; OR, adjusted odds ratio.T A B L E 3 Vaginal misoprostol compared to vaginal dinoprostone for secondary outcomes.
Table 2 summarises the characteristics of the eight RCTs included.

Table 3 and
Figures Characteristics of included trials.