Beneficial population
An estimate of the total population is required for the VOI analyses, defined as the total population who could benefit from future research that reduces decision uncertainty. The relevant population is all singleton births to nulliparous women in England, excluding those women opting for elective caesarean section for reasons other than breech presentation.
The NHS Maternity Statistics201 states that there were 636,401 births in England in the financial year 2016–17. Of these, 91.8% were at ≥ 37 weeks’ gestation, 33.6% of which were to nulliparous women.201 The statistics do not disaggregate by reason for elective caesarean section (specifically, whether or not because of suspected breech position). Therefore, this means that there were:
deliveries in England per annum that met our population definition.
Assuming a 10-year time horizon for the VOI analysis (a proxy for the length of time for which the decision question remains relevant before technological development changes it), an approximately stable number of deliveries per annum and a discount rate of 3.5% yields a beneficiary population of 1,689,663.
If our analyses are assumed generalisable to all pregnancies, then the beneficiary population is 636,401 per annum, or 5,477,940 over the 10-year horizon (discounted at 3.5%).
Probabilities
Prevalence of small for gestational age fetuses, large for gestational age fetuses and breech presentation: nodes A1 and A2
LGA and SGA are defined as a birthweight in the highest and lowest decile of the distribution, respectively.202,203 The prevalence of each in the population is therefore 10%.
The prevalence of breech presentation at the third-trimester scan is estimated at 4.6%, based on the POP study, a large prospective cohort study conducted in Cambridge, UK.11
Sensitivity and specificity of ultrasound: nodes B, S_B, L_B, B_B
Estimates of the sensitivity and specificity of ultrasound scanning were based on the POP study.8,11,138 Note that, because of the structure of the model, these figures are not the true sensitivity and specificity of the tests per se, but the probability of detection if everyone is screened (‘universal screening’) compared with the probability of detection with selective screening. The estimates are thus the actual sensitivity and specificities multiplied by the proportion of the population screened. Note that we assume that the sensitivity and specificity of a positioning scan are 100%, as this is an extremely simple procedure requiring solely the identification of the skull and spinal column to determine orientation of the fetus.
Interventions for breech presentation: nodes B_ECV, B_ECVs, B_noECV, B_ECVs_rC and B_ECVf_RC
Data on the proportion of mothers accepting ECV, the success rate and the reversion rates were extracted from the POP study.8 These methods and results have been published separately.11
Delivery modes for true negative (appropriate for gestational age infants): node C1
An otherwise healthy infant (i.e. true negative for SGA, LGA and breech presentation, node C1) can be delivered by emergency caesarean section or vaginally.
A study of 14,100 singleton liveborn and stillborn infants in French maternity units in 2010 found that approximately 19.4% (2504/12881) of non-SGA infants were delivered by emergency caesarean section.22 The POP study found that 19.9% (735/3689) of non-breech position infants were delivered via emergency caesarean section.11 A 2018 Cochrane systematic review16 of IOL compared with expectant management in women at or beyond term found an 18.42% (1056/5734) caesarean section rate in the expectant management arm (see analysis 1.1316).
The most relevant population to this analysis is the POP study.11 Of the 3689 deliveries, 141 were by elective caesarean section. Our defined population excludes elective caesarean sections for indications other than breech presentation; therefore, we assume that 20.7% (735/3548) of AGA deliveries result in emergency caesarean section (95% CI 19.4% to 22.06%), with 79.3% of AGA babies being delivered vaginally.
We chose to use data from the POP study11 (a prospective cohort study) for the risk of emergency caesarean section, rather than those from Monier et al.22 (a population-based setting), because the study design of the former made the validity of the numbers easier to verify. Compared with a network meta-analysis, relying on a single study risks potentially overestimating uncertainty; however, because of time constraints, conducting a network meta-analysis was unfeasible.
Delivery modes for false negatives for small for gestational age fetuses and large for gestational age fetuses: nodes S_C2, L_C2
If an infant is SGA and this is not spotted (i.e. is a false negative, node S_C2), the relative risk of emergency caesarean section is taken from the French cohort study, which reported an adjusted relative risk of ‘caesarean after onset of labour’ (assumed to meet the definition of emergency caesarean section) in low-risk pregnancies of 1.9 (95% CI 1.4 to 2.5; table 3, Monier et al.,22 figures reported to only one decimal place).
If a baby is LGA and this is not spotted (i.e. is a false negative, node L_C2), the odds ratio of emergency caesarean section compared with that for an AGA infant is assumed to be 1.792 (95% CI 0.718 to 4.471). This probability was obtained from a retrospective analysis carried out in the USA in 2005 that included 241 nulliparous women whose pregnancies were induced and who were delivered at term.146 Breech position, stillbirth and pregnancies with other abnormalities were excluded. All women underwent estimation of fetal weight with ultrasound prior to labour. In total, 23 out of 241 (9.5%) overestimated the EFW by ≥ 15%. Caesarean section delivery rates for labour arrest (assumed to be emergency caesarean section) were 34.8% in the overestimation group and 13.3% in the no-overestimation group. This equates to 8 out of 23 and 29 out of 218 in each group, respectively, yielding an odds ratio of 1.792 with a standard error of the log of the odds ratio of 0.466.
Delivery modes for true positives for small for gestational age fetuses and large for gestational age fetuses: nodes S_C3, L_C3
The relative risk of ‘caesarean after onset of labour’ (assumed to meet the definition of emergency caesarean section) in true-positive SGA infants following induction compared with true-negative infants (i.e. AGA infants) is assumed to be 2.9 (node S_C3). This may be an overestimate as according to the data source22 this is the relative risk of emergency caesarean section for true-positive SGA fetuses, whether or not labour was induced, and only 27.1% (36/133) were induced at < 39 weeks’ gestation.
We could not identify data on how early IOL would affect the risk of emergency caesarean section among true-positive LGA pregnancies. For this reason, we used data from Middleton et al.,16 implicitly assuming the same relative risk reduction for LGA pregnancies as for non-LGA pregnancies. The relative risk for induced versus non-induced LGA pregnancies was 0.92 (95% CI 0.85 to 0.99) and was modelled using log-normal distribution (mean –0.08, standard error 0.037).
If the policy for LGA infants is expectant management (node L_C2), then the emergency caesarean section rate is assumed the same as for a false-negative diagnosis.
Delivery modes for false positives for small for gestational age and macrosomia: nodes S_C4, L_C4, L_C1
False positives for SGA will be induced. False positives for LGA will be handled depending on the selected management strategy: expectant management or IOL.
A prospective RCT (n = 6106) of IOL at 39 weeks’ gestation in low-risk nulliparous women yielded a relative risk of (emergency) caesarean section of 0.84 (95% CI 0.76, 0.93) associated with induction.154 Note that the Monier et al.22 study described above reported a relative risk of emergency caesarean section in false positives for SGA of 1.0 (95% CI 0.5 to 2.2). However, as a RCT is generally considered at a lower risk of bias than an observational study, we opted for the RCT results154 and applied these to nodes S_C4 and L_C4, representing the probabilities of emergency caesarean section following IOL for false-positive diagnoses of SGA and LGA, respectively.
Where the selected management strategy for LGA is expectant management, the risk of emergency caesarean section after a false-positive diagnosis (node L_C1) is logically assumed to be the same as that for an AGA infant (node C1).
Delivery modes for breech presentation: false negative and true positive – nodes B_C2, B_C3a–B_C3f
If an infant is breech and is a false negative for this (i.e. undetected breech, node B_C2), we assume that the probability of an emergency caesarean section is 57.7% (95% CI 38.67% to 75.62%). No comparative data were identified for the risk of emergency caesarean section with unidentified breech compared with that with cephalic presentation. However, a retrospective cohort study of the case notes of 131 women in Hong Kong in 1997 found that, of those with undiagnosed breech at labour, and excluding those in whom ECV was subsequently attempted, 11 (42.3%) had a vaginal breech delivery and 15 (57.7%) had a caesarean section (table 2, Leung et al.161). Caesarean sections are labelled as the sum of elective and emergencies, but, given that these were undiagnosed until labour, we have interpreted these as all emergency caesarean section.
Nodes B_C3a to B_C3f represent delivery modes with and without ECV, taking into account success or failure as well as spontaneous reversion (to either breech or cephalic presentation). All estimates are obtained from the POP study11 except for node B_C3b, representing delivery modes where ECV was successful but the infant subsequently reverted to the breech position, because of a lack of relevant observations in the POP study data. We assumed the same distribution as per a false-negative diagnosis of breech (57.69% probability of emergency caesarean section, node B_C2).161 Note that we assume this to be an independent probability with the same parameters as node B_C2, rather than taking the exact same value, to reflect that this is a different outcome measure from B_C2, but with the same likelihood.
Perinatal morbidity: true negative (appropriate for gestational age infants) – node D1
Node D1 represents the baseline risk of neonatal morbidity as a result of expectant management of an otherwise healthy, non-SGA infant, taken from the POP study () (see Table 11 and systematic review54). Outcomes include no, moderate and severe neonatal morbidity, and perinatal death. Moderate neonatal morbidity was defined as meeting one or more of the following criteria: a 5-minute Apgar score of < 7, delivery with metabolic acidosis (defined as cord blood pH of < 7.1 and base deficit > 10 mmol/l), or admission to the neonatal unit at term (defined as admission < 48 hours after birth at ≥ 37 weeks’ gestation and discharge ≥ 48 hours after admission). Severe neonatal morbidity was defined as hypoxic–ischaemic encephalopathy, use of inotropes, need for mechanical ventilation, or severe metabolic acidosis (defined as a cord blood pH of < 7.0 and base deficit of > 12 mmol/l).
Prevalence of no, moderate and severe neonatal morbidity in the POP study by fetal size diagnosis
The RCOG Green-top Guideline No. 20a13 states that a 0.1% risk of perinatal mortality is associated with a planned cephalic vaginal delivery. However, this figure includes all stillbirths and neonatal deaths. The relevant figure for the purpose of our model comprises intrapartum stillbirths and neonatal deaths only; deaths prior to this are assumed unrelated to orientation or size of the fetus, and thus do not affect the results of the incremental analysis. To estimate the risk of stillbirth and perinatal mortality, we used observational data from Moraitis et al.,54 because delivery before 37 weeks’ gestation was an exclusion criterion of the study. For baseline risk, we used mortality for spontaneous vaginal and assisted vaginal deliveries only. In the study, spontaneous and assisted vaginal deliveries accounted for 88.07% and 59.48% of antepartum stillbirths and delivery-related perinatal mortality, respectively. Data from the POP study showed the risk of stillbirth/perinatal mortality as a function of birthweight. Using these data, we estimated that the total number of stillbirths and perinatal mortality for spontaneous and vaginal deliveries would have been 809.66 and 455.54, respectively, if all infants had been AGA. Multiplying these numbers with the corresponding proportions of deaths resulting from spontaneous and instrumental vaginal deliveries, we estimated that the total mortality for these categories would have been 984 cases (n = 635,396). Modelling this using a beta distribution, the baseline risk (i.e. for AGA pregnancies delivered vaginally) was 0.155% (95% CI 0.145% to 0.165%).
The probabilities of no, moderate or severe morbidity and perinatal death would ideally be modelled as a Dirichlet distribution. However, as these statistics are taken from different sources, they are modelled as independent beta distributions. This may overestimate the uncertainty in morbidity risk. Furthermore, we assume that the risk of neonatal morbidity in an AGA infant is independent of delivery mode. A priori, an emergency caesarean section is expected to be associated with a higher risk of perinatal morbidity. However, the relevant population is infants who are not breech, SGA or LGA, but who are delivered by emergency caesarean section for other reasons. After factoring out these indications for emergency caesarean section, the assumption may not be so unreasonable.
Perinatal morbidity: false-negative small for gestational age infants – node S_D2
The same sources (POP study and Moraitis et al.54) for node D1 report the odds of adverse outcome in SGA infants (i.e. in the bottom decile of the distribution). The odds ratio of moderate and severe morbidity and stillbirth for SGA compared with AGA infants in the absence of intervention (i.e. induction) is 2.48, 1.88 and 4.89, respectively (node S_D2). Again, we assume that the risk of neonatal morbidity in SGA infants is solely a function of the infant’s size and not of the mode of delivery.
Perinatal morbidity: false-negative large for gestational age infants – nodes L_D2a and L_D2c
Baselines
Neonatal morbidity among undiagnosed LGA infants (false negatives) was modelled to take account of specific risks for these infants, and therefore was modelled as none (no complications), respiratory morbidity, shoulder dystocia, ‘other acidosis’ or perinatal death. Shoulder dystocia can lead to no long-term complications; BPI (which can be transient or permanent); or acidosis, leading to no long-term complications, severe anoxic brain damage or perinatal mortality. ‘Other acidosis’ (secondary to other than shoulder dystocia) has the same long-term outcomes as that secondary to dystocia, namely no long-term complications, severe anoxic brain damage or perinatal mortality. The risks of neonatal morbidity (and hence mortality) are related to delivery mode. These are modelled by estimating a baseline risk for each morbidity for the general population and multiplying this by a relevant relative risk. The baseline risks are not used in the model per se, as morbidity for otherwise healthy infants is captured via ‘no/mild/moderate morbidity/perinatal death’ (node D1).
The baseline probability of respiratory morbidity was extracted from a study of the influence of timing of elective caesarean section on respiratory morbidity, conducted in Cambridge, UK.204 All deliveries between 1985 and 1993 at the centre (n = 33,289) were included in the analysis and all cases of respiratory distress syndrome or transient tachypnoea necessitating admission to neonatal intensive care were recorded. Of the entire sample, 6955 deliveries occurred at term (39+0 to 39+6 weeks’ gestation) and were delivered vaginally. Among these babies, 22 had respiratory morbidity, reported as 0.32% (95% CI 0.18% to 0.45%). Assigning a beta distribution to these figures yields a similar (but slightly different) 95% CI of 0.20% to 0.46%. This was used as the baseline risk (i.e. the risk for AGA infants).
The baseline probability of shoulder dystocia was based on figures quoted in the RCOG guidelines for the management of shoulder dystocia.205 This reported incidences in the literature of between 0.58% and 0.70%. The best-quality study informing the estimate was a retrospective analysis by Ouzounian et al.164 This reported 1686 cases of shoulder dystocia among 267,228 vaginal births, yielding an incidence of 0.63% (95% CI 0.60% to 0.66%).164
The baseline probability of other acidosis (i.e. not secondary to shoulder dystocia) was based on a Cochrane systematic review comparing induction with expectant management.16 Analysis 1.4 of the review reported incidence of birth asphyxia, with 5 out of 731 pregnancies in the expectant management arm, yielding a base probability of 0.68%.16
The baseline risk of perinatal morbidity was assumed to be the same as described above (node D1), that is an estimated risk of 0.155% (95% CI 0.145% to 165%), based on our own estimations using data from Moraitis et al.54 As this baseline risk was not specific to fetal size, we used the same baseline risk for SGA and LGA fetuses and distinguished their risk using their odds ratios instead.
To estimate the baseline risk of perinatal death, we used observational data from Moraitis et al.,54 because delivery before 37 weeks’ gestation was an exclusion criterion of the study. For baseline risk, we used mortality for spontaneous vaginal and assisted vaginal deliveries only. In the study, spontaneous and assisted vaginal deliveries accounted for 88.07% and 59.48% of antepartum stillbirths and delivery-related perinatal mortality, respectively. Data from the POP study showed the risk of stillbirth and perinatal mortality as a function of birthweight. Using these data, we estimated that the total number of perinatal deaths for spontaneous and vaginal deliveries would have been 809.66 and 455.54, respectively, if all infants had been AGA. Multiplying these numbers by the proportion of deaths resulting from spontaneous and instrumental vaginal deliveries, we estimated that the total mortality for these categories would have been 984 cases (n = 635,396). Modelling this using a beta distribution, the baseline risk (i.e. for AGA pregnancies delivered vaginally) was 0.155% (95% CI 0.145% to 165%).
Ideally, these mutually exclusive probabilities would be modelled with a Dirichlet distribution. However, as they are from different sources, they are modelled with their respective distributions. This risks generating a set of probabilities that sum to > 1. However, given the low absolute percentages, this is highly unlikely. Sampled values were verified in the model code to ensure that all were contained within [0 to 1].
Undetected large for gestational age infant (false negative), vaginal delivery (L_D2a)
No data were available on the relative risk or odds ratio of respiratory morbidity for undetected LGA with a vaginal delivery (node L_D2a). Expert opinion estimated that these infants were at either the same or a lower risk of respiratory morbidity than AGA infants. We therefore used a point estimate relative risk of 0.75, and assigned a uniform distribution between 0.5 and 1. Note that as relative risks are more intuitive than odds ratios from an elicitation point of view, we report this as a relative risk rather than an odds ratio.
The odds ratio of shoulder dystocia in a LGA infant delivered vaginally (vs. an AGA infant) is assumed to be 7.18 (95% CI 2.06 to 25.00). This is based on a systematic review reporting the incidence of shoulder dystocia in all infants with a birthweight ≥ 4000 g (table 2 of Rossi et al.165). Two source studies were meta-analysed with a random-effects model. Importantly, these data are not disaggregated by delivery method. However, it is reasonable to assume that caesarean section eliminates the risk of shoulder dystocia and, therefore, this represents the odds ratio of LGA infants delivered vaginally.
The same table in the review165 also reported the odds ratio of asphyxia in a LGA infant (vs. an AGA infant) of 2.88 (95% CI 1.34 to 6.22). We assume that this meets our definition of ‘other acidosis’ and apply the figures accordingly, but with the caveat that this is not disaggregated by delivery mode and so may overestimate the risk (e.g. asphyxia may be the reason for an emergency caesarean section).
The same table in the review165 also reported the odds ratio of perinatal death in a LGA infant (vs. an AGA infant) of 1.77 (95% CI 0.30 to 10.34). We apply this to our definition of perinatal mortality, again noting that this is not disaggregated by delivery mode. The rarity of the outcome is also reflected in the wide CI, implying a high degree of uncertainty.
Undetected large for gestational age infant (false negative), emergency caesarean section (node L_D2c)
The relative risk of respiratory morbidity for a macrosomic infant delivered via emergency caesarean section compared with an AGA infant () delivered vaginally was taken from the Cambridge cohort204 described in Baselines (table 2 of Morrison et al.163). As stated above, this study was not specific to LGA infants, but the risk of respiratory morbidity is most plausibly associated with intervention to speed delivery rather than the presence of LGA. The source table reports the odds ratio of respiratory morbidity with ‘caesarean section labour’ (assumed to meet the definition of emergency caesarean section) at 39+0 to 39+6 weeks’ gestation as 3.2 (95% CI 1.4 to 7.4) relative to the baseline of vaginal delivery at 40+0 to 40+6 weeks’ gestation. Rebasing relative to vaginal delivery at 39+0 to 39+6 weeks’ gestation yields an odds ratio of 1.674 (95% CI 1.253 to 2.001).
Risk of respiratory morbidity from emergency caesarean section
The relative risk of shoulder dystocia for emergency caesarean section was assumed to be zero.
The relative risk of other acidosis for a LGA infant delivered via emergency caesarean section compared with an AGA infant () was taken from Chongsuvivatwong et al.166 (as for elective caesarean section described above, and thus the same caveats are attached).
Risk of acidosis from emergency caesarean section
Finally, the relative risk of perinatal mortality for a LGA infant delivered via emergency caesarean section compared with that for an AGA infant was taken from the same source166 ().
Risk of perinatal mortality from emergency caesarean section
Perinatal morbidity, true-positive small for gestational age infants: induction of labour – node S_D3
If a SGA infant is induced, we assume that the relative risk is 0.7 for moderate and severe morbidity and 0.33 for perinatal death (node S_D3). These data are based on a systematic review of IOL compared with expectant management in low-risk women at or beyond term (approximately 10,000 observations; odds ratios not reported).16 Critically, this is not the treatment effect for SGA infants, for which we were unable to identify any data, and the relative risk for moderate and severe morbidity was based on data reporting a 5-minute Apgar score of < 7. However, the central estimates of relative risks (0.7 and 0.33, respectively) were considered plausible by clinical experts (GCSS and AAM), and the CIs represented plausible summaries of their epistemic uncertainty.
Perinatal morbidity, true-positive large for gestational age infants: expectant management and induction of labour – nodes L_D3a and L_D3c
An expectant management policy for true-positive diagnoses of LGA (at node MGT_LGA_TP) is identical to expectant management for a false negative, and the risk of perinatal morbidity is logically the same as for ‘undetected macrosomia (false negative), spontaneous vaginal’ and ‘undetected LGA (false negative), emergency caesarean section’ described above. Nodes L_D2a and L_D2c are therefore replicated at this point in the tree (following MGT_LGA_TP >> L_C2).
Under an IOL policy for positive diagnoses of LGA (MGT_LGA_TP >> L_C3a), delivery modes can again be spontaneous vaginal or emergency caesarean section. Where data allow, risks of perinatal morbidity are assumed to be related to IOL and the presence of LGA, as well as to delivery mode (vaginal or emergency caesarean section).
Respiratory complications
A retrospective cross-sectional study of maternal and neonatal outcomes in induced low-risk term pregnancies (n = 131,243) reported neonatal complications by week of delivery comparing IOL with expectant management.167 The adjusted odds ratio of respiratory complications at week 39 is reported as 0.540 (95% CI 0.373 to 0.783; see table 4167). This was used as odds relative to an AGA infant, whether delivered vaginally or by emergency caesarean section (L_D3a and L_D3c respectively). Of note is that these data are not LGA specific.
Shoulder dystocia
A Cochrane systematic review101 of IOL compared with expectant management for suspected fetal macrosomia estimated a relative risk of shoulder dystocia of 0.6 (95% CI 0.37 to 0.98) (analysis 1.3 of Boulvain et al.101). We therefore applied this relative risk, noting that the baseline comparator is MGT_LGA_TP >> L_C2 or MGT_LGA_TA >> L_C3. That is:
and:
Data are for ‘suspected’ LGA, and are not disaggregated by true and false positives. We therefore apply due caution and score the relevance of the data as ‘moderate’.
Acidosis
The Boulvain et al.101 Cochrane review did not report the incidence of acidosis or asphyxia. Therefore, we sourced data from the Middleton et al.16 Cochrane review, which compared IOL with expectant management in all pregnancies at term. Analysis 1.416 reported a relative risk of birth asphyxia of 1.66 (95% CI 0.61 to 4.55). We used this to represent the relative risk of ‘other acidosis’.
Perinatal mortality
The Cochrane systematic review101 of IOL compared with expectant management for suspected fetal macrosomia observed zero events in the included studies. We therefore used the Middleton et al.16 Cochrane review, Analysis 1.1,16 reporting a relative risk of 0.33 (95% CI 0.14 to 0.78) compared with AGA infants that received expectant management.
The odds ratios and relative risks for node L_D3c are identical to those for L_D3a. However, the implied probabilities at the nodes will differ because of the different baseline comparators. For respiratory morbidity, acidosis and perinatal death, the ratios are relative to expectant management for AGA infants. For shoulder dystocia, macrosomia-specific data were available, comparing induction with expectant management in cases of suspected macrosomia, so the ratio is relative to vaginal delivery or emergency caesarean section for an expectant management policy.
Perinatal morbidity, false-positive small for gestational age or large for gestational age infants: induction of labour – node D4
Following an incorrect diagnosis of SGA or following an incorrect diagnosis of LGA under the IOL policy, an AGA infant will be induced. Evidence suggests that this reduces the risk of stillbirth, but with the consequence of increasing perinatal complications; a retrospective database analysis of induction compared with expectant management at 37 weeks’ gestation found an odds ratio of 0.15 (95% CI 0.03 to 0.68) for perinatal death and 1.92 (95% CI 1.71 to 2.15) for admission to a neonatal unit or special care baby unit.160 We assumed admission to these specialist units was a proxy for moderate and severe complications, so we applied these odds ratios to the baseline risks.
Perinatal morbidity: false-positive large for gestational age infants – expectant management
Following an incorrect diagnosis of LGA, and with an expectant management policy, perinatal outcomes are logically the same as vaginal and emergency caesarean section perinatal outcomes for AGA infants. Therefore, these nodes are labelled as D1.
Perinatal morbidity: breech – false negative and true positive (B_D2a – B_D2c)
Perinatal outcomes are assumed to be dependent on whether or not the infant is presenting breech at delivery. A breech infant who reverts to cephalic positioning either spontaneously or following ECV is assumed to be at the same risk of perinatal outcomes as an AGA infant.
Vaginal breech delivery (B_D2a): perinatal death
The RCOG Green-top Guideline No. 20a13 states that vaginal delivery in the breech position is associated with a risk of perinatal mortality of 2 in 1000, but 0.5 in 1000 with elective caesarean section, compared with a 1.0 in 1000 risk for a cephalic vaginal delivery. This is based largely on a Cochrane systematic review of planned caesarean section for term breech delivery,14 the largest contributor to which was the Term Breech Trial (TBT).206
As described in Perinatal morbidity: true negative (appropriate for gestational age infants) – node D1, the risk of perinatal mortality of 1.0 in 1000 includes all deaths around the time of delivery. However, our figure of interest is solely intrapartum stillbirth and neonatal death (the implicit assumption is that prepartum deaths are due to causes other than breech, LGA or SGA). A retrospective cohort study of all term singleton births in delivery units in Scotland between 1992 and 2008 (n = 784,576) found a mortality rate of 0.04% (234/537,745) associated with cephalic vaginal deliveries.54 The same study reported a mortality rate of 0.29% (5/1719) associated with breech vaginal deliveries, yielding an odds ratio of 6.68 (95% CI 2.75 to 16.22).
Vaginal breech delivery (B_D2a): moderate and severe morbidity
We estimate the relative risk of moderate and severe morbidity associated with breech vaginal delivery compared with cephalic vaginal delivery to be 6.7 (95% CI 5.9 to 7.6). This is based on a large retrospective cohort analysis of the Swedish Medical Birth Registry from 1988 to 1997 reporting the odds ratio of a 5-minute Apgar score of < 7.170 We assume that the odds ratios are identical for moderate and severe morbidity. This may be a reasonable assumption: the odds ratio for perinatal death calculated above is 6.68, extremely close to the 6.7 reported here.
Elective caesarean section delivery (B_D2b): perinatal death
A Cochrane systematic review of elective caesarean section compared with vaginal delivery for term breech delivery (Hofmeyr et al.,14 analysis 1.3) found an overall global relative risk of perinatal death of 0.29 (95% CI 0.10 to 0.86).
Elective caesarean section delivery (B_D2b): moderate and severe morbidity
The same review14 reported a relative risk of a 5-minute Apgar score of < 7 of 0.43 (95% CI 0.12 to 1.47), and of a 5-minute Apgar score of < 4 of 0.11 (95% CI 0.01 to 0.87) (analyses 1.4 and 1.5,14 respectively). We therefore use this as the relative risk of moderate and severe perinatal morbidity, respectively, associated with elective caesarean section compared with planned vaginal breech delivery.
Emergency caesarean section delivery (B_D2c): perinatal death
A study of 32,776 breech presentations in Scotland between 1985 and 2004171 found 9018 emergency caesarean section deliveries (4108 pre labour and 4910 post labour), of which 14 led to perinatal or neonatal death (0.16%). As stated above, the Moraitis review54 reported a mortality rate of 0.29% (5/1719) associated with breech vaginal deliveries. This yields an odds ratio of 0.533 (95% CI 0.192 to 1.482). As this odds ratio is based on combining data from different sources, we explore this parameter in greater detail in a one-way sensitivity analysis.
Emergency caesarean section delivery (B_D2c): moderate and severe morbidity
In the absence of evidence on the effect of emergency caesarean section compared with vaginal breech delivery for the risk of moderate and severe neonatal morbidity, we assumed that the odds ratio would be the same as the odds ratio of perinatal death, that is 0.533 (95% CI 0.192 to 1.482).
Long term outcomes following no, moderate and severe perinatal morbidity (appropriate for gestational age infants, small for gestational age infants and breech presentation): nodes E1–E3
Long-term outcomes were no complications, SEN, SNM, and neonatal/infant death. The risks of each were assumed to be dependent solely on level of perinatal morbidity (where perinatal morbidity is a function of abnormality and delivery management).
A large retrospective cohort study of school children reported the risk of SEN by 5-minute Apgar score, inter alia.172 In total, 4.7% [ = 18,736/(18,736 + 376,891)] of children with a 5-minute Apgar score at birth of 8–10 had SEN. We used this as the risk of SEN for children with no neonatal complications (node E1). The same study also reported odds ratio for 5-minute Apgar scores of 4–7 and 0–3, which were used as the increase in risk for moderate and severe neonatal morbidities (nodes E2 and E3).
We used CP as a proxy for SNM. A large retrospective cohort study of births in Sweden analysed the risk of CP by 5-minute Apgar score.173 We calculated the baseline risk of CP as the sum of the number of children with CP with a 5-minute Apgar score of ≥ 7 divided by the total number of children with a 5-minute Apgar score of ≥ 7 [ = (69 + 163 + 674)/(27,664 + 129,096 + 1,037,793) = 0.08%, node E1]. The study also reported adjusted hazard ratios by individual Apgar score, rather than grouped categorisations (< 4, 4 to < 7 and ≥ 7). A weighted geometric mean hazard ratio (and 95% CI) was calculated for each group as per , and divided by the weighted 7–10 results. We interpreted the hazard ratio as the relative risk. These are different, but related concepts; the former takes account of time, whereas the latter assumes that all events happen simultaneously. Given the simple structure of our model, and the relative rarity of CP, we felt that this was a sufficient approximation.
Baseline risk of CP by 5-minute Apgar score
Infant mortality data were extracted from routine Scottish data from 1992 to 2010.174 A total of 1,013,363 neonates had a normal 5-minute Apgar score at birth (defined as ≥ 7) (see ). There were 628 neonatal (birth to 28 days) and 1446 infant deaths (29 days to 1 year), a total of 0.2%. This was assumed to form the baseline risk of neonatal/infant mortality (node E1). Adjusted relative risks of neonatal and infant mortality were reported in the appendix of the paper.174 To generate an overall relative risk over 12 months, a weighted geometric mean (and 95% CIs) of the risks reported by Iliodromiti et al.174 for neonatal and infant mortality was calculated, with weights of 1 and 12 for neonatal and infant mortality, respectively (representing the relative length of the time periods; ). Relative risks for Apgar scores of 4–6 and 0–3 were used for moderate and severe neonatal morbidity, respectively (nodes E2 and E3).
Relative risk of CP by 5-minute Apgar score
Long-term outcomes following large for gestational age infants at birth: nodes L_E1, L_F1, L_G
In our model, LGA infants are at risk of no perinatal complications, respiratory morbidity, shoulder dystocia, other acidosis or perinatal mortality. LGA infants developing shoulder dystocia are at risk of no long-term complications, BPI or acidosis. BPI can be transient or permanent. Acidosis can lead to no long-term complications, SEN, SNM or perinatal mortality. The RCOG Green-top Guideline No. 42205 states that ‘fewer than 10% resulting in permanent [injuries]’, based on findings from Gherman et al.207 These figures in turn rely on the study by Sandmire et al.169 In total, in 8 out of 145 cases BPI injuries were permanent. We modelled this using a beta distribution, yielding a risk of permanent BPI of 5.5% (95% CI 2.4% to 9.8%).
Following no perinatal complications, LGA infants are at the background risk of long-term complications, SEN, SNM and neonatal mortality (node E1).
Following respiratory morbidity, we assume that infants are at increased risk of long-term complications (SEN, SNM and neonatal/infant mortality) equivalent in severity to severe neonatal morbidity (i.e. node E3).
Shoulder dystocia can lead to no injury to the infant (in which case the background risk of SEN, SNM and neonatal/infant mortality applies), BPI (which can be transient or permanent) or acidosis.
Transient BPI leads to a background risk of long-term complications, SEN, SNM and neonatal mortality (node E1).
Permanent BPI leads to baseline risk of long-term complications, SEN, SNM and neonatal mortality, but with a decreased quality of life associated with the injury (node L_G).
Following acidosis, the risk of long-term complications, SEN, SNM and neonatal mortality is assumed to be severe neonatal morbidity (node E3).
Costs
Costs of ultrasound scan for fetal size
We obtained the cost of an ultrasound scan for fetal size (and presentation) from the National Schedule of Reference Costs, 2016–17.175 We used data for ‘Ante-Natal Standard Ultrasound scan (NZ21Z)’, as reported for outpatient procedures. The reference costs contained the mean as well as lower and upper interquartile range for costs, listed by every type of service provider. We calculated a weighted average for the mean/interquartile ranges based on the reported numbers of activities over the year for each provider. We then fitted a gamma distribution to the weighted mean/interquartile range, obtaining the parameters alpha = 4.6904 and beta = 22.8062, and yielding a total cost of £107.06 per scan (95% CI £70.89 to £134.92).
Cost of ultrasound scan for fetal presentation only
Estimating a cost for an ultrasound scan for fetal presentation alone is challenging, as this type of ultrasound screening is not part of current NHS routine practice. We theorised that such a scan could be performed by a midwife in conjunction with a standard antenatal visit in primary care, using relatively basic and inexpensive equipment. However, it is uncertain whether or not implementing this is feasible. For this reason, we estimated the cost of two different scenarios of how an ultrasound scan for fetal presentation alone could be performed.
Midwife-led screening in primary care setting
We theorised that an ultrasound scan for fetal presentation alone could be provided by a midwife in conjunction with a standard antenatal visit in primary care. Although NHS reference costs are provided for ‘Ante-Natal Standard Ultrasound scan (NZ21Z)’,175 these scans frequently involve an assessment of fetal anatomy and/or biometry and, because these require much more time and training to assess than fetal presentation alone, we deemed that it was inappropriate to use this cost as an estimate for the cost of an ultrasound scan for fetal presentation alone.
Following the methodology of Wastlund et al.,11 we estimated the cost of ultrasound scanning for fetal presentation as a function of the midwife’s time, the equipment cost and the cost of the room/facilities where the scan would take place.
We obtained the cost of the midwife’s time from the Unit Costs of Health and Social Care 2017.184 We used the total hourly cost for band 5 nurses, £36; this was consistent with the costs reported for midwives in NHS Staff Earnings Estimates to September 2017 – Provisional Statistics.208 In addition to the scan itself, time would be needed to make the woman feel comfortable with the process, and to document the results of the scan; therefore, we estimated that the average scan would require 5–10 minutes in total. In the absence of data on how much it would cost to provide ultrasound equipment and sufficient training, we estimated that this could be provided for a total cost between £1000 and £20,000. We assumed that the average machine would be operated 400–3000 times annually over the 5-year time horizon. We assumed that room costs would be between £4500 and £6000 annually209 and that rooms would be in operation 1573 hours per year.184
We simulated the total cost per scan using uniform distributions and 100,000 simulations. We then fitted a gamma distribution to the resulting distribution, based on the mean and interquartile range. The resulting parameter estimation was a gamma distribution with α = 43.8259 and β = 0.2159. This resulted in a total cost of ultrasound scan for fetal presentation of £9.46 (95% CI £6.87 to £12.46) per scan.
Sonographer-led ultrasound in designated setting
If implementing ultrasound assessment in primary care (as part of a standard antenatal visit) would not be possible, the most feasible alternative would be to perform the scan at a designated ultrasonography unit. A scan for fetal presentation alone is much swifter and technically less complicated than the type of scan typically performed as part of a standard antenatal visit. For this reason, we did not consider ‘Ante-Natal Standard Ultrasounds Scan (NZ21Z)’ in the NHS reference costs175 to be a suitable cost estimate. Instead, we used the data for ‘Ultrasound Scan with duration of less than 20 minutes, without Contrast (RD40Z)’ from the reference costs175 for diagnostic imaging. The national schedule of reference costs report costs as mean (£52) and interquartile range (£37–60) only. To capture the uncertainty of this cost appropriately, we fitted a gamma distribution to the mean and interquartile range. The resulting parameter estimation was a gamma distribution with α = 9.2207 and β = 5.6395. This resulted in a total cost of ultrasound scan for fetal presentation of £52.00 (95% CI £24.05 to £90.55) per scan.
Cost for base-case scenario
Because there is genuine uncertainty about the feasibility of providing midwife-led ultrasound screening for fetal presentation only, quantifying the reasonable cost for this parameter was problematic. For the base-case scenario, we used a uniform distribution of costs, ranging between the lower end of the 95% CI if midwife-led screening was possible (£6.87) and the upper end of the 95% CI for sonographer-led screening (£90.55). This way, all plausible costs of ultrasound screening for fetal presentation alone were incorporated into the sensitivity and VOI analysis.
Cost per mode of delivery
We obtained data on costs for different modes of deliveries from the national schedule of reference costs.175 For a (cephalic) vaginal delivery, we used data for a normal delivery without epidural or assistance. For all modes of deliveries, the reference costs were presented for different levels of complications (CC scores), and we calculated a weighted average cost for all levels. The reference costs reports the mean as well as the lower and upper interquartile range for costs, listed by types of clinical setting (e.g. elective inpatient, non-elective inpatient, outpatient procedures). We calculated a weighted average for the mean/interquartile ranges based on the reported numbers of activities over the year for each setting. For each of the three modes of deliveries (cephalic vaginal, planned caesarean section and emergency caesarean section), we fitted a gamma distribution to the resulting weighted mean/interquartile range. For vaginal delivery, this yielded the parameters alpha = 7.2606 and beta = 252.5824, with a total cost of £1834.47 (95% CI £1750.43 to £2236.05). The corresponding values for planned caesarean section were alpha = 11.1212 and beta = 307.0169, with a total cost of £3411.93 (95% CI £2679.80 to £4038.29). For emergency caesarean section the values were alpha = 14.7329 and beta = 318.1354, for a total cost of £4688.27 (95% CI £3816.15 to £5443.02).
As the National Schedule of Reference Costs, 2016–17175 does not list separate costs for vaginal breech deliveries, we made the simplifying assumption that these costs would have the same ratio to the costs of elective caesarean section as reported by Palencia et al.177 For that study, the costs were CA$7255 and CA$8440 for elective caesarean section and vaginal breech delivery, respectively, with a mean cost difference of CA$1185 (95% CI CA$719 to CA$1663). We fitted a normal distribution (mean 1.1633, standard deviation 0.0332) to calculate the relative cost increase from vaginal breech delivery to elective caesarean section. This yielded a relative cost increase of 1.1633 (95% CI 1.0982 to 1.2284). To obtain the cost of vaginal breech delivery for our model, we multiplied the cost of elective caesarean section (as calculated above from the National Schedule of Reference Costs, 2016–17175) by the relative cost increase from vaginal breech delivery.
Cost of external cephalic version
We obtained the cost of ECV from the cost analysis of offering ECV in the UK reported by James et al.178 The authors provided two different estimates of costs, using low (£186.70) and high (£193.30) staff costs. To convert to 2017’s price level, we used the HCHS inflation index: compared with baseline, the index was £302.30 for year 2017,184 and £196.50 for year 2001.210 The resulting cost per ECV was £287.20 and £297.40 for low and high staff costs, respectively. We interpreted this as the feasible range that costs could assume, and let the model sample from this interval using a uniform distribution.
Cost of neonatal unit admission
To capture the cost of admission to neonatal care following delivery, we used cost data from the National Schedule of Reference Costs, 2016–17.175 We divided neonatal critical care into three levels: ‘intensive care’, ‘high-dependency’ and ‘special care’. For intensive and high-dependency care we used currency codes XA01Z and XA02Z, respectively, and for special care we used a weighted average of currency codes XA03Z to XA05Z. We assumed that the proportion of admittance to each level of neonatal care and length of stay was the same as the one reported by Alfirevic et al.179 This meant that 19%, 7% and 74% of admitted neonates went to intensive, high-dependency and special care, respectively, and that the length of stay was 2, 1.5 and 2 days, respectively. To capture the uncertainty in the cost of care, we fitted a gamma distribution based on the mean and interquartile values, as reported in the reference costs.175
To estimate the number of neonates admitted to neonatal care as a function of neonatal morbidity at delivery, we reanalysed data from the POP study.8 We used 5-minute Apgar score as a proxy for neonatal morbidity at delivery: a 5-minute Apgar score of ≥ 7, 4–6, and 0–3 was equivalent to no, moderate and severe neonatal morbidity, respectively. This meant that the risk of admittance was 7.4% (95% CI 6.6% to 8.2%) with no morbidity and 47.4% (95% CI 31.9% to 63.1%) with moderate morbidity; we modelled this using the beta distribution. For severe morbidity, we instead made the simplifying assumption that all neonates with severe morbidity would be admitted to a neonatal unit because of the small sample number of infants with severe neonatal morbidity in the POP study. In the absence of evidence as to how the level of neonatal morbidity at birth affects the chance of ending up in each tier of neonatal care, we assumed that the proportions were constant, and that the level of neonatal morbidity affected the level of overall admittance only.
Cost from respiratory morbidity
Morrison et al.163 reported the incidence and length of stay at hospital for respiratory morbidity. A total of 28% of the morbidities consisted of respiratory distress syndrome and the rest of transient tachypnoea of the newborn. The average stay at the NICU was 4 days for respiratory distress syndrome and 0.6 days of transient tachypnoea of the newborn. The NHS cost of NICU admission is £1295 per day (interquartile range £1015–1541).175 Given this, the average cost for a case of respiratory distress syndrome is £5180 (interquartile range £4060–6164), and the cost for transient tachypnoea of the newborn is £777 (interquartile range £609–925). Assuming that respiratory distress syndrome and transient tachypnoea of the newborn make up 28% and 72% of respiratory morbidities, respectively, the average cost of a case of respiratory morbidity would be £2010 (interquartile range £1575–2392). Owing to the very low mortality rate from respiratory distress among infants born at term, we made the simplifying assumption that respiratory distress could lead to NICU admission, but would otherwise have no consequences.211 To capture the uncertainty of the cost of respiratory morbidity in one parameter, we fitted a gamma distribution based on the mean and interquartile range. The resulting distributions had parameters α = 10.7125 and β = 187.6316, yielding a total cost of £2011 (95% CI £993 to £3381).
Cost of acidosis without long-term consequences
In the absence of data on the costs associated with short-term acidosis (i.e. acidosis that requires neonatal treatment but resolves without any other health consequences), we made the simplifying assumption that treatment would be required at the NICU for 1–4 days, with equal probabilities. To obtain per-day costs, we fitted a gamma distribution for the unit cost of NICU care using cost data from the National Schedule of Reference Costs, 2016–17,175 based on mean and interquartile range. Combining the time and per-day costs, we obtained a total cost distribution. To be able to capture total cost uncertainty in a single parameter, we fitted a gamma distribution to the total cost. The resulting parameter (α = 3.6143 and β = 895.6169) had a total cost of £3240 (95% CI £806 to 7328).
Cost of transient and permanent brachial plexus injury
To estimate the costs associated with BPI, we assumed the same resource use as that reported by Culligan et al.180 Transient BPI costs included a hospital consultation by a specialist, weekly physical therapy for 4 months and one needle electromyography test. Permanent BPI costs included the costs from transient BPI but with weekly physical therapy for 3 years instead, plus one outpatient visit to a specialist, and magnetic resonance imaging of the shoulder.180 We obtained costs for the specialist consultations and weekly physiotherapy treatments from the Unit Costs of Health and Social Care 2016;212 these were £199 and £87, respectively. The costs for electromyography and magnetic resonance imaging were taken from the National Schedule of Reference Costs, 2016–17 (AA33D and RD01C);175 these were £269.20 and £106.59, respectively. All costs were updated to the price year 2016–17 using the HCHS index.184 We assumed that all costs except the cost of physiotherapy were incurred in the first year of life and discounted accordingly; the discount rate was 3.5% as recommended by NICE.188 The total discounted costs from transient and permanent BPI were £2066 and £14,133, respectively.
To account for uncertainty, Culligan et al.180 expanded their cost estimate into a plausible range of costs, which ranged between 50% and 200% of the point estimate. However, directly incorporating this plausible range into our own estimation (after adjusting for cost differences) by using uniform distribution would have been inappropriate, as this would have overestimated costs. Instead, we interpreted the plausible range as a 95% CI for total costs, and then fitted a log-normal distribution to the appropriate mean and 95% CI range. This way, the lower and upper 95% CI were still 50% and 200% of the point estimate, respectively, but in this case following a log-normal distribution. For transient BPI, the resulting distribution had a logged standard error of 0.3536, and the total costs were £2066 (95% CI £1033 to £4132). The corresponding figures for permanent BPI were a logged standard error of 0.3536, and a total cost of £14,133 (95% CI £7067 to £28,264).
Cost of perinatal death
We used the cost of stillbirth as a proxy for the cost of perinatal death. The direct costs of stillbirth were obtained from Mistry et al.181 The authors estimated that the costs would be between £1242 (core investigation and counselling only) and £1804 depending on the clinical scenario surrounding the stillbirth and what tests were needed. The authors chose not to present a most plausible estimate within this, but instead just reported these costs as the full range of costs for stillbirth. For this reason, we interpreted these costs as the upper and lower boundaries that the cost of perinatal death could reasonably assume. We updated these costs to the price year of 2016–17 (the original source used price year 2010) using the HCHS index,184 and used a uniform distribution.
Cost of special educational needs
We obtained the cost of SEN from Barrett et al.,182 using the difference in costs between SEN and typically developing groups. The cost difference was £6315 (95% CI £3798 to £8832). These costs were estimated for the cost year of 2007–8; hence, we inflated these to the value of price year 2016–17 using the HCHS index,184 resulting in a cost difference of £7428 (95% CI £4467 to £10,389). This cost was applied annually for years 6–17 of life (the typical school years) and discounted using a discount rate of 3.5%, as recommended by NICE.188
The cost of severe neurological morbidity
We used CP as a proxy for SNM. In the absence of English cost data that are detailed enough to provide an annual cost for the relevant payer perspective, we obtained the annual cost of CP from Cerebral Palsy Australia.183 We used total per-capita cost for the health system, as well as indirect costs (e.g. programme services, aids and home modifications), but we omitted productivity losses, dead-weight losses from financial transactions and costs for informal carers. The annual average cost per case of CP in 2005 was AU$5362. We converted this to Great British pounds (£) using the exchange rate at 31 December 2005, and updated to the price level of 2016/17 using the HCHS index.184 This gave a total annual cost of £2929.60. Because the data were derived from the nationwide population of people with CP, this average annual cost is applicable to any year of life.
Capturing the uncertainty in these costs was problematic as costs are not easily transferable between different health-care systems. Furthermore, Cerebral Palsy Australia did not provide any estimates of cost uncertainty. For this reason, we chose to assume that English costs could reasonably fluctuate between half and double those quoted in Australia. We interpreted this as a 95% CI stretching between £1465 and £5859, and fitted a log-normal distribution to this interval.
