Does induction of labour increase the risk of caesarean section? A systematic review and meta-analysis of trials in women with intact membranes
Departments of Obstetrics and Gynaecology, University of Calgary, Calgary, AB, Canada
Community Health Sciences, University of Calgary, Calgary, AB, Canada
Correspondence: Dr S Wood, Department of Obstetrics and Gynaecology, University of Calgary, 4th Floor, North Tower, Foothills Medical Centre, 1441 – 29th Street NW, Calgary, AB, T2N 4J8, Canada. Email Stephen.email@example.com
Recent literature on the effect of induction of labour (compared with expectant management) has provided conflicting results. Reviews of observational studies generally report an increase in the rate of caesarean section, whereas reviews of post-dates and term prelabour rupture of membrane (PROM trials suggest either no difference or a reduction in risk.
To evaluate with a systematic review and meta-analysis of randomised controlled trials (RCTs) whether or not the induction of labour increases the risk of caesarean section in women with intact membranes.
Literature search using electronic databases: MEDLINE, EMBASE, and the Cochrane Database of Clinical Trials.
RCTs comparing a policy of induction of labour with expectant management in women with intact membranes.
Data collection and analysis
A total of 37 trials were identified and reviewed. Quantitative analyses with fixed- and random-effects models were performed with revman 5.1.
Of the 37 RCTs, 27 were trials of uncomplicated pregnancies at 37–42 weeks of gestation. The remaining ten evaluated induction versus expectant management in pregnancies with suspected macrosomia (two), diabetes in pregnancy (one), oligohydramnios (one), twins (two), intrauterine growth restriction (IUGR) (two), mild pregnancy-induced hypertension (PIH) (one), and women with a high-risk score for caesarean section (one). Meta-analysis of 31 trials determined that a policy of induction was associated with a reduction in the risk of caesarean section compared with expectant management (OR 0.83, 95% CI 0.76–0.92).
Induction of labour in women with intact membranes reduces the risk of caesarean section. Review of the trials suggests that this effect may arise from non-treatment effects, and that additional trials are needed.
Rising caesarean delivery rates in North America and Europe continue to be a major concern for health policy experts, administrators, and medical professional associations.[1, 2] A number of factors have been implicated, such as the decline in rates of trial of labour after caesarean section, and increasing rates of obesity, late maternal age, and induced labour.[1, 3, 4] Reducing the frequency of induction is often cited as an approach to reversing the trend in caesarean section rates. The association between induction of labour and caesarean delivery is largely based on the findings of observational studies.[6-11] One clear limitation of the observational literature is that induction is often indicated by complications of pregnancy, which may independently increase the risk of caesarean section. Additionally, some authors have found that using survival analysis and comparing the outcomes of women induced with those who deliver at later gestational ages eliminates the effect of induction; however, this has not been a consistent finding. Therefore, a clearer picture may emerge from clinical trials where women with a specific pregnancy complication are randomised to a policy of induction of labour versus further expectant management. In the setting of prelabour rupture of membranes (PROM) at term, large clinical trials, as well as meta-analysis, have clearly established that a policy of induction does not increase the risk of caesarean section. Additionally, with induction for PROM, the risk of caesarean section appears to be low, perhaps indicating the relative state of cervical maturation, or that the labour process has already begun. A 2006 Cochrane review of trials of induction for post-term pregnancy concluded that there was no difference in the risk of caesarean section between induction and expectant management. Caughey and colleagues. reviewed both observational studies and clinical trials of ‘elective’ induction. They found the observational studies consistently documented a lower rate of caesarean section in women having spontaneous versus induced labour. In contrast, meta-analysis of eight clinical trials, six of which were post-dates trials, concluded that caesarean section was reduced by induction. Of note, the authors excluded all non-English articles.
Our goal was to evaluate whether induction of labour, compared with a policy of expectant management, in women with intact membranes increased the rate of caesarean section. To that end, we undertook a systematic review and meta-analysis.
We systematically searched Ovid MEDLINE (1950–2012, week 23, 2010) and EMBASE (1980–2012, week 23). The MEDLINE search for labour induction included the Medical Subject Headings (MeSH) term ‘labour, induced’ and the keywords ‘induced labour’, ‘induction of labour’, and ‘labour induced’. EMBASE followed a similar strategy with the MeSH term ‘labour induction’ and keywords ‘labour induced’, ‘induction of labour’, and ‘induced labour’. All results were then limited to ‘clinical trial’. A review of the Cochrane Database of Clinical Trials was also performed. All citations were then independently reviewed by two of the authors, S.W. and S.C. Titles and abstracts were reviewed and full texts were obtained for all articles that the reviewers felt were relevant, or also if they were uncertain of their relevance. The references of these studies were also searched for additional articles. Studies were included if a policy of induction of labour, for indications other than preterm rupture of membranes (PROM), was compared with expectant management and outcome data on route of delivery were provided. All three of the authors independently assessed studies for potential selection, performance, and attrition bias, as recommended for assessing the quality of clinical trials in meta-analysis. The compliance with allocated treatment, as well as the degree of crossover, were also assessed.
All studies were graded as high or low quality based on three key quality indicators: adequate randomisation and allocation concealment; limited losses to follow-up (<20%); and intention-to-treat analysis. Studies that were deficient in any of these areas were graded as low quality. Studies were not included in the quantitative summary analysis, unless intention-to-treat analysis was presented or could be calculated from the available data. Quantitative analyses with fixed- and random-effects models were performed with revman 5.1 (The Nordic Center, The Cochrane Collaboration, Copenhagen). Statistical assessment for heterogeneity was performed and considered statistically significant if P < 0.05. A subgroup analysis limited to high-quality trials was planned a priori.
The literature search identified 2794 citations. A review of the abstracts determined that there were 1368 unique citations. Details of the study selection are outlined in Figure 1. Review of the abstracts identified 42 possible trials for which the full articles were reviewed by all of the authors. Three of the publications were not in English: one was in German, one was in French, and one was in Spanish. These were reviewed with the assistance of two obstetrician-gynaecologists and an Internal Medicine-OB specialist at the authors' institution. Five of the trials were determined to be cervical ripening trials and were excluded. A citation search identified two additional papers. One of these, an abstract, did not clearly report the frequency of caesarean section in the two study groups, and was therefore excluded. The other was a conference abstract that did supply sufficient data to be included. Another article was included as the trial protocol had been identified in our literature search, and the publication of the final results was later identified. Therefore, a total of 38 trials were included in the review. Further review identified that one of the trials was a duplicate publication of a previously published German article[20, 27]; therefore, only the English version was included in the review. One trial, which randomised women with twins to early term delivery versus standard care included women who planned to have a caesarean section. The primary author of the article was contacted and the details of the outcomes of the planned vaginal delivery groups were kindly provided.
Of the remaining 37 studies, 27 were induction trials of uncomplicated pregnancies at 37–42 weeks of gestation.[21, 22, 27-51] The remaining ten studies evaluated induction versus expectant management in pregnancies with suspected macrosomia,[24, 52] diabetes in pregnancy, oligohydramnios, twins,[26, 55] intrauterine growth restriction (IUGR),[56, 57] mild pregnancy-induced hypertension (PIH), and women with a high-risk score for caesarean section. The characteristics of the 37 trials are outlined in Table 1(and are further detailed in Table S1). The methods of induction were not identical in all studies. The early investigations generally used artificial rupture of membranes with oxytocin to achieve induction. Later studies used combinations of cervical ripening with prostaglandins or mechanical means, and subsequent oxytocin. Two studies used only prostaglandins for induction. In six trials, women were only included if they had an unfavourable cervix, and in five trials only women with a favourable cervix were enrolled. The remaining trials did not use cervical status as an inclusion criterion. The reported overall caesarean section rates varied between studies, with a range of 1–47%. Compliance with treatment in the induction groups was not uniformly high, with reported rates of <70% in six of the trials. In almost all of the studies induction of labour became necessary in many of the expectantly managed patients. The reported rates varied from 4 to a high of 50%. Most trials reported that the time to delivery increased in the expectantly managed versus the induced groups, but this was not reported in a sufficiently uniform manner to allow a quantitative summary. Typically, trials reported latency to delivery of about 1 week in the expectant management group. One trial reported no difference in average gestational age at delivery between those induced and those managed expectantly.
Table 1. Study characteristics of randomised controlled trials of induction in subjects with intact membranes
A review of trial quality rated 19 trials as high quality and 17 trials as low quality (Tables 1 and S2). One trial could not be graded as sufficient information was not available from the abstract, and the authors did not respond to requests for additional information. The most common reason for a low-quality score was unclear or inadequate randomisation and allocation concealment. Of the low-quality studies, six could not be included in the meta-analysis either because of large post-randomisation exclusions (>20%) or because of insufficient data to perform an intention-to-treat analysis. Only three trials documented a statistically significant difference in caesarean section rates between a policy of induction and expectant management. Two of these trials reported a reduction in the rate of caesarean section with induction,[34, 36] and one reported an increase in risk. The remaining trials reported non-significant differences in the rate of caesarean section, with most point estimates favouring a reduction in risk with induction. A quantitative summary analysis combined the results of 31 trials, with 6248 women randomised to induction and 5918 women randomised to expectant management (Figure 2). The trials were ordered by year in order to display any potential chronological effect, and were subdivided by post-dates trials (≥41 weeks of gestation) and induction for other indications. In this analysis, a policy of induction was associated with a reduction in the risk of caesarean section compared with expectant management (fixed-effects model, OR 0.83, 95% CI 0.76–0.92, P = 0.0002). There was no significant statistical heterogeneity: χ² = 27.11, df = 30, P = 0.62, and I2 = 0%. The same results were obtained with a random-effects model (OR 0.83, 95% CI 0.76–0.92). Visual inspection of the funnel plot did not suggest any significant publication bias. The reduced risk of caesarean section was seen in both the subgroups of the post-dates trials (OR 0.85, 95% CI 0.76–0.95) and in the trials of induction for other indications (OR 0.81, 95% CI 0.69–0.95). Subgroup analysis of only the 19 high-quality trials revealed a similar result (fixed-effects model, OR 0.82 95% CI 0.73–0.91; random-effects model, OR 0.82, 95% CI 0.73–0.91; Figure 3).
Details on the indication for caesarean section were reported in 16 of the trials; however, in the majority the number of outcomes was too few for any meaningful comparisons. Of the seven trials with sufficient numbers, three reported an increase in the number of caesarean sections for fetal distress in the expectant group, but this difference was statistically significant in only one study.[36, 40, 56] One trial reported a non-significant increase in the risk of abdominal delivery for dystocia with expectant management. The remaining trials did not report any differences in indications for caesarean section between those randomised to induction or those randomised to expectant management.
Meta-analysis of the risk of caesarean section for fetal distress with induction versus expectant management did not document a statistically significant difference (Figure 4). A meta-analysis was also performed for the outcomes of postpartum haemorrhage and operative vaginal delivery (Figure 4): neither was increased with induction of labour. Neonatal outcomes were also examined. A composite neonatal morbidity score was described in seven trials, but only one reported a statistically significant difference between the study groups, which, in that study, favoured the induction arm. Unfortunately, the morbidity scores were not sufficiently similar to allow for a pooled analysis. Some assessment of cord pH was reported in 13 trials. One trial reported an increase in the rate of cord pH < 7.05 with expectant management 6% (19) versus 3% (9), P = 0.043. The other trials reported non-significant differences in mean pH or in the rate of low cord pH. As a result of variable reporting between the trials, meta-analysis was not feasible. Meta-analysis was performed for the following neonatal outcomes: an Apgar score at 5 minutes < 7; admission to the neonatal intensive care unit (NICU); and perinatal death, excluding anomaly (Figure 4). No statistically significant differences between induction and expectant management were observed. Perinatal death was reduced with induction of labour, but the result was not statistically significant P = 0.05.
Our meta-analysis of clinical trials of induction versus expectant management in women with intact membranes found that the induction of labour is associated with a moderate but statistically significant reduction in the risk of caesarean section. This effect was evident in the subgroups of post-date trials, trials of induction for other indications, and in the analysis restricted to high-quality studies. Differences in other maternal and neonatal outcomes were not evident, but a non-statistically significant reduction in perinatal death was observed in the induction group.
Strengths and limitations
Our review's main strength is that it included all relevant randomised clinical trials that we identified, and so added to previous reviews by including reports in all languages as well as those published since 2007. As our results are based solely on randomised clinical trials, selection bias and confounding should be limited. This is in contrast to observational studies where selection bias and confounding can be problematic. Admittedly, recent well-designed retrospective studies in electively induced women, have also demonstrated a reduction in caesarean section with induction, compared with expectant management.[60, 61]
Our results were based on an intention-to-treat analysis. Although this is a valid standard, it could also have affected our findings. In most of the trials there was less than perfect compliance with induction in the treatment groups, and there were high rates of induction in the expectant groups. These factors would tend to bias the results towards the null, and therefore could have obscured a true difference in the risk of caesarean section with induction.
A limitation of our review is that we did not include a formal systematic review of cost effectiveness. Two of the trials we reviewed, the Canadian Post Term Pregnancy Trial and Hypertension and Pre-eclampsia Intervention Trial at Term (HYPITAT), have published secondary economic analyses.[62, 63] Both of these analyses found that the increase in costs associated with induction of labour were more than offset by a reduction in the frequency of fetal testing. These results may not apply to term elective inductions, but may be similar for induction for indications such as small for gestational age (SGA), multiple pregnancy, and maternal diabetes.
Some caution should be sounded before unreservedly accepting the conclusion that induction of labour reduces the risk of caesarean section. In our meta-analysis, this result, especially in the post-date trials, is heavily influenced by the largest study in our review, the Canadian Post Dates Trial, where cervical ripening with intracervical prostaglandin E was only planned for women in the immediate induction group. Similar treatment was discouraged for inductions in women who were expectantly managed. Eventually 32% of the expectant group were induced, but only 9% received prostaglandins for ripening, compared with 51% of the immediate induction group. In a recent Cochrane review, cervical ripening with intracervical prostaglandins was reported to reduce the risk of caesarean section compared with oxytocin alone. Therefore, this co-intervention in the induction group may have biased their results. Additionally, a disproportionate number of women in the expectant group had induction for suspected fetal compromise. A similar trend was noted in our meta-analysis. Although many of these were likely to be false-positive results, the higher gestational ages in women who were expectantly managed may have increased the anxiety of doctors, and led to a greater number of caesarean sections. Similar labelling has been documented as a factor in increasing the rate of caesarean sections in women with gestational diabetes and suspected macrosomia.[66, 67] Such concern about poor outcomes may affect delivery decisions for expectantly managed women with PIH, SGA, oligohydramnios, and multiple pregnancy. It may be that the results of our review reflect doctors' discomfort with delayed delivery in high-risk women that, once they are in labour, manifests as more frequent caesarean sections: an example of research confirming the biases of the health care community.
It is difficult to reconcile the results of this meta-analysis with the consistent finding that lack of cervical maturation prior to induction is strongly associated with subsequent caesarean section.[69, 70] Clinicians have generally pursued a policy of expectant management, not only to allow spontaneous labour, but also to provide time for cervical ripening should induction become necessary later. Therefore, a policy of delay would be expected to produce an improvement in cervical state and related reduction in the risk of caesarean section. A few possible explanations can be postulated to resolve this paradox. First, although ripening would have eventually occurred in many of expectantly managed women, the actual delay to delivery was too short: most of the included trials documented only modest differences between treatment groups in gestational age at delivery. A second explanation is that cervical state is a characteristic of the patient that is not modifiable by expectant management. Support for this supposition comes from Smith and colleagues's. fascinating cohort study, which found that greater cervical length at mid-gestation in primiparous women was associated with an increased risk of caesarean section at term, and that this was largely because of poor progress in labour. Ultimately, either one of these possibilities could explain both why the trials that included only women with unripe cervixes had similar findings as those restricted to women with mature cervixes, and the result of a subgroup analysis of Koopman and colleagues's. trial, which found no differences in the risk of caesarean section stratified by Bishop score.
Some readers may feel that the results of the trials and this meta-analysis are sufficient to justify a policy of elective induction. Before concluding this, it should also be carefully considered that only one of the trials was designed to evaluate the impact of induction on the risk of caesarean section. Ideally, we feel that a trial primarily evaluating the risk of caesarean section would have a number of features that were not prominent in the trials where it was a secondary outcome. First, cervical state should not be known by the recruiting clinicians, or even by the women themselves. This would reduce the significant risk of pre-randomisation exclusions of women with the most adverse cervical findings. Secondly, the latency to delivery in the expectant group should be sufficient both for a meaningful improvement in cervical state and for a significant number of women to go into labour spontaneously. This can only be achieved if the treatment group is induced when fetal maturation is assured (at 38 weeks of gestation?) and the women are of sufficiently low risk to anticipate a delay in induction until at least 41 weeks of gestation. Furthermore, there needs to be a strong commitment at study sites to perform what might be perceived as ‘elective’ inductions in a timely fashion to reduce crossover. Finally, an economic evaluation conducted alongside the trial will be important. Until such rigorous trials are performed we think it is premature to conclude that induction does not affect the risk of caesarean section.
Ultimately, induction of labour is one of the common treatments available to clinicians. It could also be argued that, based on our results, it is one of the only interventions that has been shown in a clinical trial to reduce the risk of caesarean section. Therefore, it may be that some women at high risk of caesarean section, such as women of late maternal age or women who are obese, could benefit from an elective induction. Ideally, this will be the subject of further clinical trials.
Disclosure of interests
The authors have no interests to disclose.
Contribution to authorship
All authors contributed to the design of the study. SW and SC reviewed all of the abstracts. All of the authors critically appraised the selected articles. SW performed the statistical analysis, and all the authors contributed to the interpretation of the results. SW drafted the article, which was reviewed by SC and RS.
Details of ethics approval
The study was approved by the local ethics review board (ref. no. E–23622).
The authors are members of the Partnership for Research and Education in Mothers and Infants, which has been funded by Ross Products, a division of Abbott.
The authors wish to acknowledge the financial support of their research group by Ross Products, a division of Abbott. The authors also wish to thank Dr Magali Robert, Dr Vreni Kuret, and Dr Eliana Castillo for their assistance with reviewing the French, German, and Spanish articles. The authors would also like to thank Ms Selphee Tang for assistance with the figures.
Mini commentary on ‘Does induction of labour increase the risk of caesarean section? A systematic review and meta-analysis of trials in women with intact membranes’
Research in Childbirth and Health Group, University of Central Lancashire, Preston, UK
There is global interest in reducing caesarean section rates. As Wood and colleagues note, studies of induction of labour have variously reported both increased and decreased caesarean rates. The authors have undertaken a careful and thorough meta-analysis, and have shown consistent results in favour of labour induction; however, they are justifiably cautious about the policy implications of their findings, on the basis that the results may arise from non-treatment effects. They propose more better-designed studies. Very importantly, they note the need for future research to include cost-effectiveness analyses, and an assessment of women's views and experiences.
Apart from these factors, future studies should also assess unanticipated long-term outcomes for mother and baby. For example, there is growing evidence that so-called ‘late prematurity’ (37–38 weeks of gestation) may be linked to a range of adverse outcomes for infants in the longer term (Boyle et al. BMJ 2012;344:e896). This is important, given the known margin of error in dating of any particular pregnancy. Factors that may independently affect labour management decisions (and, consequently, policy implementation decisions) should also be taken into account. For example, a study of associations between induction of labour and rates of caesarean section in the USA found that the results varied with sociodemographics (parity, age, race, and education), maternal behaviours (number of prenatal visits), and, intriguingly, service design factors, such as hospital teaching status and ownership (Wilson et al. J Nurs Scholarship 2010;42:130–138).
Wood et al. comment that induction of labour seems to be the only technique that has successfully reduced caesarean section in the context of a randomised controlled trial (RCT). This is not, in fact, the case. The current Cochrane review of continuous labour support (Hodnett et al. Cochrane Database Syst Rev 2012: art. no.: CD000012. DOI 10.1002/14651858.CD000012.pub4) shows a reduction in caesarean section rate that is very similar to that found by Wood et al. (OR 0.78, 95% CI 0.67–0.91). RCTs of place of birth, and, particularly, home birth, are notoriously hard to set up. In the next-best approach, the population-based birthplace study found very significant reductions in caesarean section for planned birth centres or home births, when compared with hospital births [ranging from an adjusted OR of 0.31 (95% CI 0.23–0.41) for a planned homebirth to 0.39 (95% CI 0.29–0.53) for a planned birth in a birth centres], with no increase in adverse outcomes, except among neonates born at home to primiparous women. These data suggest that the mechanism that underlies high and rising caesarean section rates is likely to include local childbirth philosophies and management approaches, as much as maternal and fetal physiology.
In conclusion, in addition to the factors suggested by Wood et al., future studies of the impact of induction of labour on caesarean section rates should take into account the contextual factors and longer-term outcome measures identified above. Indeed, it may be premature to undertake such studies at all, until there is a more precise understanding of the mechanisms that underpin high and rising caesarean section rates, and the consequent identification of factors that could plausibly be addressed by additional interventions such as continuous labour support, place of birth, and labour induction.
Disclosure of interests
I am a co-author on the current Cochrane review of alternative versus conventional institutional settings for birth, and of a structured review of the outcomes of free-standing, midwife-led birth centres. I was the founding Chair of the Royal College of Midwives Campaign for Normal Birth. I have no financial interests in any of the issues discussed in the commentary above.