Construcción social de la participación

In our systematic review we made considerable effort to include all RCTs with no language restrictions, which led to the inclusion of>600 trials, with data for>100,000 women and babies. The NMA provided

an opportunity to examine the relative effectiveness of all treatments used for the induction of labour in a coherent and methodologically robust way across important clinical outcomes. Although there are now increasing numbers of NMAs reported in the literature, and some relate to competing treatments in obstetrics,970_{as far as we are aware this NMA includes more trials and participants than any other in this} topic area.

Network meta-analysis is only valid on the assumption that all of the treatments in the network would be suitable for all included women. We were thorough in our evaluation of six important potential treatment effect modifiers (previous CS, parity, membrane status, Bishop score, gestational age and single/multiple pregnancy) and found no clinically important differences in the distribution of these potential effect modifiers across the interventions. We also conducted informal and formal statistical checks of model fit and inconsistency. When lack of fit and/or inconsistency between evidence sources was observed, it was resolved by excluding studies that were assessed as being at high risk of bias.

To our knowledge this is the first attempt to simultaneously compare more than two treatments for the induction of labour in a cost-effectiveness analysis. A study by Petrouet al.14_{suggested that PGE2gel was} more cost-effective than PGE2tablets, and Van Baarenet al._’s study955_{concluded that Foley catheter} induction was more cost-effective than PGE2gel. These results are not directly comparable with the results from this study, as they use different measures of benefit, but it is still worth mentioning that in our cost-effectiveness analysis these interventions were found to be less effective and more expensive than titrated (low-dose) oral misoprostol solution, vaginal PGE2pessary (normal release) and vaginal misoprostol (dose 50 µg).

Limitations

Systematic review and network meta-analysis

Broadly, the aim of induction of labour is to achieve early delivery of the baby with the minimum harm to women and babies, and we selected outcomes to reflect these aims. However, not all of the included trials provided data on all of our key outcomes. The number of women undergoing CS was generally well reported. However, in view of high heterogeneity and apparent inconsistency it was necessary to restrict our analysis to RCTs at low risk of bias for the allocation concealment domain.

The number of women who did not give birth vaginally within 24 hours (our main efficacy outcome) was reported in less than one-quarter of trials.

Key safety outcomes were also reported relatively infrequently. Approximately one-third of trials were included in the NMA for infant admission to NICU (205) and there was considerable heterogeneity between trials (possibly as a result of inconsistent definitions of this outcome). Similarly, uterine hyperstimulation and low infant Apgar score were reported in fewer than one-third of trials.

Overall, maternal mortality and severe morbidity and infant mortality event rates, when reported, were very low. Unfortunately, these outcomes were too infrequently reported to make the pooled analysis possible. We had also intended to report serious infant morbidity but this outcome was poorly reported and inconsistently defined in trials. Consequently, we used admission to NICU as a proxy outcome for infant morbidity. Neonatal mortality was reported in only 21.3% of these trials and the incidence was low at 0.3%. Of course, it should not be assumed that if infant mortality was not reported then it did not occur, but it is probable that death rates were also low in those trials failing to report this important outcome. Very few trials collected data and reported findings relating to women’s views about the induction process. This was surprising, as some methods of induction are likely to be both painful and unpleasant. Again, because of the dearth of data and inconsistency in the way outcomes were measured and reported across trials, we were unable to include findings on maternal preferences and satisfaction in our formal

quantitative analysis. There was also insufficient information from trials evaluating alternative and complementary methods to include them in the analysis of our main efficacy outcome. None of the trials included for the analysis of number of women failing to deliver within 24 hours included an alternative method of induction. For safety outcomes, alternative and complementary methods of induction did not appear to be safer than pharmacological and mechanical methods.

The trials included in the review recruited women with varied clinical characteristics, and it is important to bear this in mind when interpreting results. The indications for induction were not always reported and, when they were, these varied across trials. Many trials excluded women with a history of CS or multiple pregnancies. Predominantly, women recruited to trials were at>37 weeks_’gestation, including post-term

pregnancies and term PROM. Most of the trials were carried out in hospital settings because most

methods of labour induction require constant attendance and monitoring by skilled clinical staff. However, we did include 79 trials examining interventions that were carried out in outpatient, community or

home settings.

For all outcomes we observed moderate heterogeneity between study effects. This is not surprising, given the clinical heterogeneity described above in settings and women who present for induction of labour. Heterogeneity may also be attributable to the varied quality of included studies. Overall, approximately half of the studies were assessed as being at high or unclear risk of bias. Consequently, we conducted REs NMA for all outcomes to allow for this heterogeneity. We report the mean from the REs distribution of study effects, although this assumes that our focus is on the effects observed in an‘average study’, whereas other summaries might be more appropriate, such as the shrunken estimate for the UK trials971 _or a prediction for a new study population.

Although NMA offers the opportunity to rank treatments in terms of relative effects for each outcome, for many results there was considerable uncertainty around effect estimates. Particularly for the analysis of safety outcomes, the findings were not clear-cut (i.e. there were no clear_‘best_’or_‘worst_’treatments for most of these outcomes). This uncertainty did not apply just to results for CS. This uncertainty is not necessarily surprising, as a large number of interventions were examined in the network. Although some interventions were examined in a large number of trials, data for other interventions were sparse, event rates for some outcomes were very low, and some outcomes were also inconsistently defined (e.g. hyperstimulation syndrome). This means that we were not able to use all of the available data in our analysis. In particular, the low event rates for NICU admission meant that in some arms of trials no events were reported, which led to problems in estimation of relative effects and also increased uncertainty in the economic analyses.

Heterogeneity in the analyses may also have been caused by the fact that trials were carried out over a long time period during which induction and CS rates in particular have increased steadily. These temporal changes could have contributed to heterogeneity and increased uncertainty of findings. More intensive surveillance may also have led to apparent increases in some outcomes (e.g. hyperstimulation).

Cost-effectiveness analysis

Our cost-effectiveness analysis was confined to short-term outcomes up until discharge from hospital, although we are aware that some outcomes may have a longer-term impact on women and their families, and also on NHS resources. The analysis was complicated by the fact that outcomes related to both women and their babies, and the two are interlinked. Women may be profoundly affected by any adverse outcomes in their newborn and, conversely, the baby may be affected by adverse outcomes for the mother. In our analysis the well-being of women and babies were combined in a single utility value for each outcome. The evidence sources informing utilities for method of delivery were assumed to represent the mothers_’well-being. However it was not clear whether the utilities for NICU admission and intensity of care required represented utility for the mother, baby or both (and, if so, the relative weight given to mother and baby: women (and even society) may value the health of the baby above their own).

We needed to distinguish those women who had a CS, those who had a VD within 24 hours, and those who had a VD after 24 hours. We found that results from trials were not always reported in a way that allowed us to estimate the outcomes together in this way. There was sufficient information to estimate effects for only 18 interventions and our conclusions on cost-effectiveness are therefore limited to this data subset.

The RCTs identified in the systematic review did not provide any evidence on the proportion of NICU admissions following births by CSs, nor on the proportion of babies admitted to different intensities of NICU care (intensive care, high-dependency care and transitional care). We have, therefore, used routinely collected hospital activity data from Liverpool Women_’s Hospital to inform these inputs to our model. We identified only four studies960–963_{reporting preference-based measures of utility relevant to the} outcomes in our model, none of which reported EQ-5D, our preferred measure. The health states did not correspond directly with those in our model, and so assumptions were necessary. It was also not clear in these studies960–963_{whether or not the utility was for the mother, baby or both (and if so the relative} weight given to mother and baby). Furthermore, measures of uncertainty were not reported alongside the utility estimates. In an attempt to address these limitations, we used our own small-scale survey to put uncertainty limits on the literature-based utilities and to define sensitivity analyses. However, note that our survey is severely limited due to being restricted to the project steering group and also limitations with the VAS instrument that it used.972_{Although the scores are bias prone and may not be comparable to utilities} elicited through other measures, the resulting ordinal preferences we obtained were found to have some face validity (the patterns seen across respondents were broadly comparable and in line with intuition) and can be considered as a first step towards defining utilities for mother/baby pairs. A large-scale study measuring utilities (preferably using EQ-5D) on antenatal and postnatal women, reporting results (together with uncertainty estimates) from both the mother and baby perspective, including time post discharge, would be of great value in addressing the limitations described above.