The Birthplace in England national prospective cohort study: further analyses to enhance policy and service delivery decision-making for planned place of birth

Birthplace mapping study

This mapping study aimed to describe the configuration of maternity care in England and used data collected as part of the Maternity Services Review carried out by the Healthcare Commission in 200711 and a follow-up survey of units and NHS trusts carried out by the Birthplace team in late 2010. 6 The evidence from this study showed that in 2010 there were 53 alongside midwifery units (AMUs) and 59 freestanding midwifery units (FMUs) in operation in England, representing around 39% of all available maternity units. 6 Of the 152 trusts providing maternity care, around half did not offer MU care of any kind and the geographical distribution of MUs in England was uneven. MUs varied considerably in size, and within unit type the numbers of beds or bed spaces, throughput (births per bed/bed space), staffing levels and skill mix varied considerably. While, overall, births planned in MUs and at home made up fewer than 10% of all births, there was notable geographical variation in this figure. 11

Birthplace national cohort study

The Birthplace national cohort study was designed to provide robust estimates of the safety and potential benefits of births planned at home, in FMUs and in AMUs compared with births planned in obstetric units (OUs), with a particular focus on women known to be at ‘low risk’ of complications prior to the onset of labour. 7^,8 For ‘low risk’ women, the incidence of adverse perinatal outcomes was low in all settings. For all settings, adverse perinatal outcome, adverse maternal outcomes and intervention during labour were more common in nulliparous (see Glossary) women compared with multiparous (see Glossary) women. After adjusting for differences in the characteristics of women planning birth in each setting, there were no differences between birth settings in adverse perinatal outcome for multiparous women. For nulliparous women, there was no difference between outcomes in MUs and those in OUs, but adverse perinatal outcomes were more likely for nulliparous women who planned birth at home. The benefits of planned birth at home or in a MU included fewer interventions, a substantially reduced incidence of intrapartum caesarean section and a higher likelihood of ‘normal birth’ (see Glossary). 12

Birthplace cost-effectiveness study

The Birthplace cost-effectiveness study provided estimates of the cost to the NHS and the cost-effectiveness of planned births in MUs and at home compared with birth in an OU. 9^,14 For ‘low risk’ women, the cost to the NHS of intrapartum and related postnatal care, including costs associated with transfer and any clinical complications, was lower for births planned at home, in a FMU and in an AMU than for planned birth in an OU. Planned birth at home, in a FMU or in an AMU generated cost savings per additional ‘normal birth’ and per adverse maternal morbidity avoided in comparison with planned birth in an OU. Data on longer-term outcomes and costs were not available at the time of the study, and this continues to be the case. 15

Birthplace qualitative organisational case studies

The Birthplace organisational case studies aimed to describe and explore the features of maternity care systems which may affect the provision of high-quality, safe care in different birth settings and were carried out in four ‘best’ or ‘better performing’ NHS trusts, as identified in the 2007 Healthcare Commission Maternity Services Review. 10^,11 Key findings included considerable variation in the provision of maternity services between and within the case study sites, partly due to geography, and variation in the organisation of community midwifery services. Women’s choices about planned place of birth were influenced, and often limited, by geographical, organisational, service culture and provider factors. The presence of an AMU could highlight contrasts in birth philosophies between units and create issues around competition for birth rooms and staffing, and appeared to intensify the perceived workload in the associated OU, where midwives reported struggling to support normal birth. All sites were committed to multidisciplinary training, including attention to emergency skills and escalation of care, but more attention was given in this regard to the needs of midwives working in FMUs and less to the needs of midwives working in AMUs or those in the community supporting home birth.

Aims

The aim of this ‘follow-on’ project was to support the development and delivery of safe, equitable and effective maternity services by strengthening the evidence base relating to planned place of birth. In particular it aimed to:

• describe and explore the impact of service configuration and other variations in the organisation and delivery of services on birth outcomes, with a particular focus on maternal outcomes which impact on future pregnancies, such as caesarean section or complicated vaginal delivery

• further describe intrapartum transfer rates and explore the possible impact of factors relating to the organisation and delivery of services on transfers

• explore the clinical characteristics, management and outcomes of ‘higher risk’ women who opt for a non-OU birth.

Objectives

The objectives of the research were posed as a series of questions grouped as follows:

Birthplace national prospective cohort study

The Birthplace cohort study comprised 79,774 women meeting the study eligibility criteria (see Eligibility). Data were collected from the vast majority of NHS trusts providing intrapartum care in England, including 142 (97%) of the 147 NHS trusts supporting home birth, 53 (95%) of FMUs, 43 (84%) of AMUs, and a stratified random sample of 36 OUs. The random sample of OUs was stratified by unit size (< 2600 births, 2600–4850 births and > 4850 births per year) and geographic location (northern or southern England) and was sampled so that each OU in England had approximately the same probability of selection. The cohort included 18,269 births planned at home, 11,666 planned in FMUs, 17,582 planned in AMUs, and 32,257 planned in OUs. Each unit and NHS trust (for home births) collected data for a different length of time within the overall study period of 1 April 2008 to 30 April 2010.

Midwives attending the births collected data using a specially designed site-specific data collection form (see Appendix 1). Data collected include maternal characteristics, risk factors known prior to the onset of labour, labour care, intrapartum transfer details, and maternal and neonatal outcomes by planned place of birth at the start of care in labour. 5,7 In order to validate outcome events and collect more detailed resource use and other information relating to adverse outcome, more detailed follow-up data were collected on a neonatal morbidity form (see Appendix 2).

The cohort study methods are described in full elsewhere7,8 and more detailed information on study design, eligibility, classification of ‘risk status’ and the characteristics of the samples used in the analyses carried out for this ‘follow-on’ project are presented in Birthplace cohort study and data.

Birthplace staffing logs

Staffing and organisational data were collected alongside the Birthplace cohort study in the form of daily staffing and workload logs completed at around 09.00 and 21.00 by midwives in OUs and MUs. The data collected using these logs (see Appendix 3) included the number of women in the unit, the number of midwives and maternity support workers on duty and obstetric cover during the previous 12 hours. Useable staffing data were collected in 30 OUs, 32 AMUs and 29 FMUs participating in the Birthplace cohort study during each unit’s data collection period for the cohort study.

Initial inspection of the data revealed that the level of missing data relating to medical staffing in OUs was too high for the data to be useable. Data on the number of women recorded as being in the unit and the number of midwives recorded as being on duty at 09.00 hours and 21.00 hours were cleaned. Checks were applied to identify any incorrect unit codes, internal inconsistencies or unexpected or implausible values and any missing data that could be derived from other information on the forms. Records containing errors where the correct codes or values could not be inferred were excluded from the analyses.

In total, 19,686 staffing and workload logs were completed. Of these, 17,359 (88%) remained after cleaning (3431 from OUs, 9019 from FMUs and 4909 from AMUs).

Birthplace mapping study

Data from the Birthplace mapping study were collected as part of the Healthcare Commission Maternity Services Review in 200711 and in a follow-up Birthplace mapping survey carried out in 2010. 6 The Healthcare Commission survey was mandatory and included all NHS trusts providing maternity care in England at the time of the survey, covering all aspects of service provision. The 2010 Birthplace mapping survey was not mandatory and addressed a small subset of topics covered in the 2007 survey, aiming to document any changes in configuration and the organisation of care since 2007 (see Appendix 4). Responses to the 2010 survey were received from 63% of the NHS trusts providing maternity care in England at the time of the survey.

Data from the Birthplace mapping study used for the purposes of this follow-on project included NHS trust-level data on the total number of births and the number of home births in the trust, the number of delivery beds or ‘bed spaces’ in units and whether or not FMUs were staffed 24 hours per day.

Office for National Statistics data

We used data from the Office for National Statistics (ONS) on registered maternities in the financial year 2009–10 by establishment and unit type to derive the numbers of women giving birth in each OU (including births in any associated AMU) and each FMU. These were used in conjunction with Birthplace data to derive the number of births per year in each OU and to estimate the proportion of births that were planned outside an OU and ‘out of hospital’ in each NHS trust. These ONS data were licensed under the Open Government Licence v1.0 and derived by BirthChoiceUK. 17

Geographical data

We used postcode data and Google Maps (maps.google.co.uk; Mountain View, CA, USA) to calculate the distance and estimated travel time by road from each FMU to the nearest OU.

Study design

The Birthplace cohort study was a prospective cohort study with planned place of birth at the start of care in labour as the exposure and a composite measure of intrapartum and early neonatal mortality and specific neonatal morbidities as the primary outcome. Four groups of women were defined based on their planned place of birth at the start of care in labour (see Glossary for definitions):

• planned home birth

• planned birth in a FMU, that is, a MU not on the same site as an OU

• planned birth in an AMU, that is, a MU on the same site as an OU

• planned birth in an OU.

In these ‘follow-on’ analyses the same four groups are compared. Throughout this report we refer to births planned in units or in trusts. We use ‘units’ to refer to births planned in MUs or in OUs and ‘trusts’ to describe births planned at home because home birth services are delivered within NHS trusts.

Eligibility

We used data on women who were eligible for inclusion in the Birthplace cohort study, that is ‘women who were attended by a NHS midwife during labour in their planned place of birth, for any amount of time’,7 excluding the following groups:

• women who had a caesarean section before labour

• women who presented in preterm labour (< 37 weeks’ gestation)

• women with a multiple pregnancy

• women who received no antenatal care

• women who had a stillbirth before the start of care in labour.

Samples used for analyses

The sample for the analyses reported in Chapters 3–6 was Birthplace eligible ‘low risk’ women with a term pregnancy (37 to 42 + 0 weeks’ gestation). Women were classified as ‘low risk’ if, before the start of labour, they were not known to have any of the medical, obstetric or fetal risk factors listed in national intrapartum care guidelines in England and Wales as ‘indicating increased risk suggesting planned birth in an obstetric unit’. 13 These risk factors are presented in the sample data collection form in Appendix 1. The maternal sociodemographic and clinical characteristics of the ‘low risk’ sample by parity and planned place of birth are presented in Characteristics of the ‘low risk’ sample.

Women whose pregnancy was post-term (42 + 1 to 44 weeks’ gestation) or who had at least one risk factor identified before the start of labour were classified as ‘higher risk’; analyses relating to these women are presented in Chapter 7, where their characteristics are also described.

‘Complicating conditions’ at the start of labour care

In the Birthplace cohort study attending midwives assessed women for the following risk factors present at the start of care in labour:

• prolonged rupture of membranes (> 18 hours)

• meconium-stained liquor

• proteinuria (1 + or more)

• hypertension with either:

  • diastolic blood pressure of ≥ 90 mmHg on more than one occasion 20 minutes apart or ≥ 100 mmHg on one occasion

  • systolic blood pressure ≥ 160 mmHg on at least one occasion

• abnormal vaginal bleeding

• non-cephalic presentation

• abnormal fetal heart rate

• other complications not previously identified.

New risk factors identified at the start of care in labour could not affect the woman’s planned place of birth and so these ‘complicating conditions’ did not affect the woman’s ‘risk status’ as defined in Samples used for analyses. However, differences in the prevalence of these ‘complicating conditions’ in different settings could affect the homogeneity and, therefore, comparability of the ‘low risk’ groups. The Birthplace cohort study identified a higher prevalence of women with ‘complicating conditions’ at the start of labour care in ‘low risk’ women planning OU birth than in ‘low risk’ women planning birth elsewhere. 7 For this reason we describe the ‘complicating conditions’ identified in the ‘low risk’ sample by planned place of birth in Characteristics of the ‘low risk’ sample and, for the ‘higher risk’ sample, in Chapter 7. As in the original Birthplace analysis, for some of the analyses reported here we carried out further analyses restricted to women without ‘complicating conditions’ in order to compare groups which were more homogeneous with regard to risk. In other analyses, as appropriate, we adjusted for the presence of ‘complicating conditions’.

Characteristics of the ‘low risk’ sample

The ‘low risk’ Birthplace cohort used for the analyses reported here comprised 19,739 women planning OU birth, 16,753 planning birth in an AMU, 11,210 planning birth in a FMU and 16,632 planning birth at home. The sociodemographic and clinical characteristics of these women are summarised by parity and planned place of birth in Tables 1 and 2. As almost all analyses were carried out stratified by parity, those women in the cohort for whom parity was unknown are not included in these tables.

Compared with women planning birth in an OU, women planning birth at home tended to be older; 20.2% of nulliparous women and 30.1% of multiparous women planning birth at home were aged ≥ 35 years vs. 11.3% and 22.9%, respectively, of those planning birth in OUs. Women planning home birth were also more likely to be white, have fluent understanding of English, be married or living with a partner and be living in a more socioeconomically advantaged area than were the OU group. There was little difference between the two settings in the distribution of gestational age and body mass index (BMI). The characteristics of women planning FMU or AMU birth tended to fall between the planned OU and planned home birth groups; the AMU group tended to have similar characteristics to the OU group, while the FMU group tended to be more similar to the home birth group. The planned home birth group had the lowest proportion of multiparous women who had only one previous pregnancy (≥ 24 weeks’ gestation) and the highest proportion with three or more previous pregnancies.

In all settings, the proportion of women with one or more ‘complicating condition’ identified at the start of labour care was higher in nulliparous women than in multiparous women (Tables 3 and 4). A higher proportion of women in the planned OU group had one or more ‘complicating condition’ identified than did women planning birth in the other settings (nulliparous 23.9% vs. 7.5–9.0%; multiparous 14.7% vs. 4.1–5.2%). Prolonged rupture of membranes (> 18 hours) was the most commonly recorded ‘complicating condition’ in all settings in nulliparous women and in planned home and FMU births in multiparous women. In multiparous women planning OU birth, meconium-stained liquor was also common.

Characteristics of the ‘higher risk’ sample

The sociodemographic and clinical characteristics of ‘higher risk’ women in the Birthplace cohort are described in Chapter 7.

Outcome measures

We used a number of main outcome measures in these analyses. While many of these are described as ‘maternal’ outcomes, it should be noted that several, for example, instrumental delivery and caesarean section, may also have implications for outcomes for the baby.

Maternal characteristics

Several analyses in this report examine associations after adjustment for maternal characteristics. Unless otherwise stated we adjusted for the following maternal characteristics using the categories shown in brackets:

• parity (nulliparous vs. multiparous)

• maternal age (< 20, 20–24, 25–29, 30–34 and ≥ 35 years)

• ethnicity (white vs. non-white)

• Index of Multiple Deprivation (IMD) quintile [first (least deprived), second, third, fourth and fifth (most deprived)]

• completed weeks of gestation (37, 38, 39, 40, and 41 to 42 + 0 completed weeks)

• BMI (not recorded, < 18.5, 18.5–24.9, 25–29.9 and ≥ 30 kg/m²)

• marital/partner status (married/living together vs. single/unsupported)

• understanding of English (fluent vs. some or none)

• presence of ‘complicating conditions’ identified at the start of care in labour (≥ 1 vs. 0).

For some analyses categories were combined (as described in individual chapters) for ease of interpretation or to avoid having very small numbers in some groups.

Almost all analyses were carried out separately for nulliparous and multiparous women. Where nulliparous and multiparous women were included in the same analysis parity was included as an adjustment variable. For some analyses relating to multiparous women the number of previous pregnancies of at least 24 weeks’ gestation (1, 2 or ≥ 3) was also included as an adjustment variable.

Poisson regression models

Most of the analyses estimating associations between maternal characteristics and perinatal or maternal outcomes or interventions were carried out using Poisson regression (generalised linear models with a Poisson distribution for the response and a log link). This allowed us to estimate the relative risks (RRs) of different outcomes while adjusting for the effects of other maternal factors which might differ between the groups of interest. For factors with several categories (e.g. BMI), the overall effect of each factor was determined using a Wald test, and p-values for these tests are presented in the relevant tables.

Weighting

All AMUs, FMUs and NHS trusts (for home births) in England were invited to participate in the Birthplace study; nearly all participated, but each unit and trust participated for a different length of time. OUs were selected from the population of OUs in England using stratified random sampling. As a consequence of these design features, the probability of a woman being selected to take part in the study varied. Probability weights inversely proportional to the duration of data collection in the unit/trust and the probability of selection of the unit were incorporated to adjust for this. Using probability weights, in effect, allowed us to make estimates of rates, proportions and the effects of various factors in the population of England rather than just in the sample.

Unless otherwise stated, all Poisson regression models and funnel plots used probability weighting. Where noted in the results tables, ‘weighted percentages’ were used in the same way to provide population estimates.

Robust standard errors

Standard regression models assume independence of observations. The design of the Birthplace cohort study was such that women were grouped by the unit or trust in which they planned to give birth, and it might be expected that outcomes would be more similar among women in the same unit or trust than for a random sample of women from the population as a whole. To allow for this, robust standard errors were used in all regression analyses unless otherwise stated. It should be noted that using robust standard errors does not affect the estimated RRs, only the confidence intervals (CIs), which are generally widened, and therefore gives more conservative estimates of effect by correctly taking account of the clustered (i.e. non-independent) nature of the data.

Calculation of adjusted rates for individual units or trusts

In the analyses reported in Chapters 3 and 5 we aimed to compare unit or trust ‘rates’ of intervention, maternal outcome and transfer. We use the word ‘rate’ as a synonym for proportion or ratio, that is, to express the proportion of women planning birth in the unit who experienced the specified intervention or outcome. While we appreciate that the use of the word ‘rate’ in this context is incorrect in strict epidemiological terms, it is commonly used and understood in these circumstances. 18

We calculated unadjusted (observed) event rates for the various outcome measures in each unit or trust separately for nulliparous and multiparous women by dividing the number of women with a given intervention or outcome by the total number of women in the Birthplace dataset in that unit or trust. We then calculated adjusted unit- and trust-level event rates using an indirect standardisation procedure to allow units or trusts to be compared as if each had the same ‘typical’ case mix of women in the sample. 18,19 Briefly, for each outcome measure, a multiple logistic regression model was used to estimate the probability of a woman having had the intervention or outcome on the basis of her demographic and clinical characteristics (see Maternal characteristics). These probabilities were then summed to give the unit’s predicted rate for each outcome. Adjusted rates for births planned in each unit or trust were calculated by dividing the unit or trust’s (observed) unadjusted rate by its predicted (or ‘expected’) rate, and multiplying this by the average rate across all units or trusts of that type in the sample.

Funnel plots

Funnel plots were used for some of the analyses reported in Chapters 3 and 5 to assess whether our data showed variation in adjusted intervention, outcome or transfer rates between individual units or trusts that was greater than we might expect to observe by chance. 19 These funnel plots show adjusted intervention or outcome rates for different units plotted against the sample size (number of women) (Figure 1).

FIGURE 1.

Example of a funnel plot.

The range of values that are likely to occur by chance for samples of that size if in reality all units have the same underlying rate are shown as ‘control limits’ on these plots. As the range of values attributable to chance becomes smaller with increasing sample size the upper and lower control limits become closer together as sample size increases along the horizontal axis, giving rise to the funnel plot’s name. We used conventional control limits at two and three standard deviations from the mean which identify the range of values in which approximately 95% and 99.8% of observed rates would be likely to fall if there was no variation in the underlying rates. The most useful interpretation of these funnel plots is, therefore, that if appreciably more than 5% (or 0.2%) of units fall outside the control limits on the graph this suggests there was more variation in the rates than could be explained by chance and that some cause(s) other than chance may, therefore, have accounted for at least some of the variation.

Linear regression

In Chapters 3 and 5 we used simple linear regression to examine associations between unit or trust characteristics and adjusted intervention, outcome and transfer rates. The adjusted intervention, outcome or transfer rate was calculated for each unit or trust, and these were regressed on selected unit/trust characteristics in turn to examine whether or not some of the variation could be explained by the unit characteristic. Individual units’ data were weighted according to the number of women in the sample for that unit and robust standard errors were used to allow for non-uniformity in the variance of unit rates with changes in some of the unit characteristics. A statistically significant result indicated that at least part of the variation in the outcome rates could be explained by that unit characteristic, with the R² statistic giving the proportion of variance explained and the sign of the coefficient giving the direction of the relationship (positive or negative). Significant relationships were also plotted graphically and, where the graph suggested the relationships might have been heavily dependent on a small number of outliers, sensitivity analyses were carried out by repeating the regressions while excluding the outliers identified on the graphs.

Study population

The main study population for these analyses was ‘low risk’ women with a term pregnancy (37 to 42 + 0 weeks’ gestation) planning a vaginal birth in any of the 43 AMUs, 53 FMUs, or 142 NHS trusts providing home birth services, and a stratified random sample of 36 OUs in which data collection for the Birthplace cohort study took place.

Unit or NHS trust characteristics

Using the methods described in Derivation of unit or NHS trust characteristics we derived the variables summarised in Table 5 to describe units or NHS trusts and configuration of care and considered these as factors which might be associated with the study outcomes.

Derivation of unit or NHS trust characteristics

Outcome measures

We used the following four main ‘outcome measures’, two capturing interventions during labour and birth, one indicating birth without complications that might affect future births and one indicating birth with low intervention:

• intrapartum caesarean section

• instrumental delivery (forceps or ventouse)

• ‘straightforward birth’, defined as birth without forceps or ventouse, intrapartum caesarean section, third- or fourth-degree perineal trauma or blood transfusion

• ‘normal birth’, defined as birth without induction of labour, epidural or spinal analgesia, general anaesthetic, forceps or ventouse, caesarean section or episiotomy. 12

For each we calculated adjusted unit or trust (for home birth) rates. For analyses of OU intervention rates we also used the following additional ‘outcome measures’, not included in the original planned analyses, to help to explain observations from the analyses of our main outcome measures:

• epidural or spinal analgesia

• augmentation with oxytocin.

Unit/trust rates of each intervention or outcome were adjusted for the maternal characteristics listed in Chapter 2 (see Maternal characteristics) and for the presence of one or more ‘complicating conditions’ identified at the start of care in labour, as described in Chapter 2 (see ‘Complicating conditions’ at the start of labour care).

Statistical methods

Characteristics of obstetric units in the study sample

For 1 of the 36 OUs included in the study we had insufficient data on the associated AMU to enable us to estimate the number of births in the OU. The remaining OUs varied in size, as measured by number of births per year, and in the number of delivery beds (Table 6). Data on midwifery staffing were available for only 30 OUs, but a relatively high proportion (median 30%) of shifts in those OUs were ‘understaffed’, with less than one midwife for each woman in labour. For six OUs it was not possible to calculate the proportion of non-OU and ‘out-of-hospital’ births because insufficient data were available to estimate the annual number of planned FMU births (four trusts) or home births (two trusts). In line with national figures, a relatively small proportion of births were planned to take place outside OUs and ‘out of hospital’, but with some variation between trusts. Nine OUs had an associated AMU in the same hospital.

Variation in intervention rates between obstetric units

We examined intervention rates, both unadjusted and adjusted for differences in maternal characteristics between OUs, to study the variation between OUs. Adjustment generally had little effect on the rates (see example in Figure 2) and so subsequent figures and tables all use adjusted rates.

FIGURE 2.

Adjusted vs. unadjusted intrapartum caesarean section rates in OUs.

For all study outcomes, funnel plots showed that there was more variation in adjusted intervention rates in nulliparous and multiparous women planning birth in OUs than would be expected by chance alone (Figures 3 and 4). There was more variation in intervention rates in nulliparous women than in multiparous women for all outcomes except ‘normal birth’, where the variation in rates in nulliparous and multiparous women was similar (Table 7). There was more variation in rates of ‘normal birth’ than for our other maternal outcome, ‘straightforward birth’. For most outcomes the number of OUs with intervention rates that were higher or lower than would be expected by chance was broadly similar (see Figures 3 and 4).

FIGURE 3.

Funnel plots showing adjusted instrumental delivery and intrapartum caesarean section rates in OUs. (a) Percentage of instrumental delivery in nulliparous women; (b) percentage of instrumental delivery in multiparous women; (c) percentage of caesarean section in nulliparous women; and (d) percentage of caesarean section in multiparous women.

FIGURE 4.

Funnel plots showing adjusted ‘straightforward birth’ and ‘normal birth’ rates in OUs. (a) Percentage of ‘straightforward birth’ in nulliparous women; (b) percentage of ‘straightforward birth’ in multiparous women; (c) percentage of ‘normal birth’ in nulliparous women; and (d) percentage of ‘normal birth’ in multiparous women.

Configuration, unit characteristics and intervention rates in obstetric units

There was no significant association between the number of OU delivery beds or the percentage of births in the trust that were planned ‘out of hospital’ and any of the main outcome measures studied (Table 8).

There was a significant association between the OU size (number of births) and the intrapartum caesarean section rate in planned OU births; larger OUs had lower intrapartum caesarean section rates in both nulliparous and multiparous women. Larger OUs also had a significantly higher ‘straightforward birth’ rate in multiparous women.

For nulliparous women there was a significant association between our measure of midwifery staffing and the intrapartum caesarean rate in planned OU births, with lower caesarean section rates observed in planned ‘low risk’ births in OUs with lower staffing levels. For multiparous women, rates of ‘straightforward birth’ were significantly higher in units with lower levels of staffing.

There were significant associations between the proportion of births that were planned outside the OU and three of the four main outcome measures. A higher proportion of non-OU births in the trust was associated with significantly higher intrapartum caesarean section rates for nulliparous and multiparous women planning OU birth, and significantly lower rates of ‘straightforward birth’ and ‘normal birth’ in multiparous women. For nulliparous women, the proportion of non-OU births was not significantly associated with rates of ‘straightforward birth’ and ‘normal birth’ (p = 0.06 and p = 0.08, respectively), but the direction of effect was the same as that seen for multiparous women.

For ‘low risk’ women planning OU birth, having an AMU in the hospital was also associated with a significantly higher intrapartum caesarean section rate in nulliparous women and significantly lower rates of ‘normal birth’ and ‘straightforward birth’ in multiparous women. These analyses were based on planned ‘low risk’ births in the OU and so women who gave birth in the unit following an intrapartum transfer from an attached AMU or elsewhere did not contribute to the rates.

None of the configuration or unit characteristics studied showed any association with rates of instrumental delivery in planned OU births.

Visual examination of scatterplots for the predictor and outcome variables that were significantly associated (see Appendix 5) did not suggest that the associations were driven by regression outliers.

Further exploratory analyses of obstetric unit intervention rates

In order to explore potential explanations for the association between the proportion of non-OU births in a trust and higher intervention rates in OUs, we carried out a number of further analyses. First, we explored whether or not epidural and augmentation rates followed the same pattern of associations as the main outcomes. Second, we explored whether or not a higher proportion of planned non-OU births in a trust was associated with a higher proportion of ‘higher risk’ births in the OU. Third, we explored whether or not the proportion of planned births that were ‘higher risk’ in an OU was associated with our main outcomes in ‘low risk’ women. Fourth, we explored whether or not intervention rates in ‘low risk’ and ‘higher risk’ women within the same OU were correlated. Finally, we examined whether or not OUs situated in trusts with a higher percentage of non-OU births tended to be those with an associated AMU.

None of the configuration or unit characteristic variables was significantly associated with the adjusted epidural rate (Table 9). For ‘low risk’ nulliparous women planning birth in an OU, a higher proportion of ‘out-of-hospital’ births in the trust was associated with significantly lower augmentation rates. In hospitals with an AMU, augmentation rates appeared to be higher, although this association was not statistically significant (p = 0.05). For multiparous women, there was a significant association between midwifery staffing and the augmentation rate; lower staffing levels were associated with higher augmentation rates.

We found no significant association between the proportion of planned non-OU births in each trust and the proportion of planned OU births that were classified as ‘higher risk’ according to National Institute for Health and Care Excellence (NICE) criteria (data not shown). Furthermore, there was no consistent relationship between the proportion of women planning OU birth who were ‘higher risk’ and our four main outcome measures in ‘low risk’ women. Exploration of intervention rates in ‘low risk’ and ‘higher risk’ women planning birth in the same OU revealed significant positive correlations for most interventions and strong positive correlations for epidural analgesia, instrumental delivery (nulliparous women) and ‘straightforward birth’ (nulliparous women); intrapartum caesarean section rates in ‘low risk’ and ‘higher risk’ women were less strongly correlated, particularly for nulliparous women [Pearson’s correlation (r) = 0.327, p = 0.05 in nulliparous women; r = 0.515, p = 0.001 in multiparous women].

There were 29 OUs for which we had data on both the percentage of non-OU births in the trust and the presence of an AMU, and, of these, five had an associated AMU. Three OUs were situated in trusts where > 20% of births were planned in a non-OU setting and all three of these had an attached AMU. The numbers of OUs with and without an AMU were too small to enable us to carry out a meaningful sensitivity analysis stratified by the presence of an AMU.

Characteristics of alongside midwifery units in the study sample

The 38 AMUs for which we could derive data on the number of planned births varied in size, but were, on average, much smaller than the OUs, with around half as many delivery beds (Table 10). On average there were two midwives on duty per shift (1.5 midwives per woman in labour) with a median of 5.4% of shifts with less than one midwife for each woman in labour.

Variation in intervention rates between alongside midwifery units

Funnel plots showed more variation between AMUs in adjusted rates of all outcomes in ‘low risk’ planned AMU births than would be expected by chance, particularly for nulliparous women. There was less variation in intervention rates for multiparous women, particularly for instrumental delivery and intrapartum caesarean section (Figures 5 and 6 and Table 11). For ‘straightforward birth’ and ‘normal birth’, most of the ‘outlier’ units had rates that were higher, rather than lower, than would be expected by chance (see Figure 6).

FIGURE 5.

Funnel plots showing adjusted instrumental delivery and intrapartum caesarean section rates in AMUs. (a) Percentage of instrumental delivery in nulliparous women; (b) percentage of instrumental delivery in multiparous women; (c) percentage of caesarean section in nulliparous women; and (d) percentage of caesarean section in multiparous women.

FIGURE 6.

Funnel plots showing adjusted ‘straightforward birth’ and ‘normal birth’ rates in AMUs. (a) Percentage of ‘straightforward birth’ in nulliparous women; (b) percentage of ‘straightforward birth’ in multiparous women; (c) percentage of ‘normal birth’ in nulliparous women; and (d) percentage of ‘normal birth’ in multiparous women.

Unit characteristics and intervention rates in alongside midwifery units

There was no significant association between the number of women per year per bed, our measure of midwifery ‘understaffing’ or the ‘staffing index’ and any of the outcome measures studied in AMUs (Table 12). For all outcome measures apart from intrapartum caesarean section in nulliparous women and instrumental delivery in multiparous women, there was a significant association between the size of the AMU (as measured by our ‘size index’) and intervention rates; in larger AMUs (with more births, more beds and more midwives) adjusted rates of ‘straightforward birth’ and ‘normal birth’ were significantly lower for nulliparous and multiparous women and adjusted rates of instrumental delivery and caesarean section appeared to be higher, but not all associations were statistically significant.

When we repeated the analysis after excluding one large AMU that appeared to be an outlier, none of the associations between AMU size and our outcome measures was significant (data not shown).

Characteristics of freestanding midwifery units in the study sample

The 48 FMUs for which we had data on the number of planned births were relatively small, with a median of two delivery beds in each unit and a smaller number of women per year per bed compared with AMUs (Table 13). Average numbers of midwives on duty per shift and per woman in labour were similar to those in AMUs, but the average proportion of shifts with less than one midwife per woman was lower than in AMUs. Most FMUs were located within 30 km of the nearest OU, with an average travel or transfer time of around 30 minutes. Forty FMUs had 24-hour staffing; this information was not available for three of the FMUs.

Variation in intervention rates between freestanding midwifery units

Funnel plots showed more variation between FMUs in adjusted rates of instrumental delivery, ‘straightforward birth’ and ‘normal birth’ in ‘low risk’ planned FMU births than would be expected by chance (Figures 7 and 8 and Table 14). There was little variation in adjusted rates of intrapartum caesarean section in nulliparous or multiparous women planning FMU birth. For instrumental delivery more ‘outlier’ units had rates that were higher than would be expected by chance (see Figure 7). For ‘straightforward birth’ and ‘normal birth’, particularly for multiparous women, more ‘outlier’ units had rates that were higher than would be expected by chance (see Figure 8).

FIGURE 7.

Funnel plots showing adjusted instrumental delivery and intrapartum caesarean section rates in FMUs. (a) Percentage of instrumental delivery in nulliparous women; (b) percentage of instrumental delivery in multiparous women; (c) percentage of caesarean section in nulliparous women; and (d) percentage of caesarean section in multiparous women.

FIGURE 8.

Funnel plots showing adjusted ‘straightforward birth’ and ‘normal birth’ rates in FMUs. (a) Percentage of ‘straightforward birth’ in nulliparous women; (b) percentage of ‘straightforward birth’ in multiparous women; (c) percentage of ‘normal birth’ in nulliparous women; and (d) percentage of ‘normal birth’ in multiparous women.

Unit characteristics and intervention rates in freestanding midwifery units

In multiparous women, in whom for most outcome measures there was less variation in intervention rates between FMUs, there was no significant association between any of the unit characteristics and the outcome measures studied (Table 15). In nulliparous women there was a significant association between the size of the FMU (as measured by our ‘size index’) and adjusted rates of instrumental delivery, ‘straightforward birth’ and ‘normal birth’; in larger FMUs instrumental delivery rates were lower and rates of ‘straightforward birth’ and ‘normal birth’ were higher. Additionally, in nulliparous women there was a significant association between our ‘distance index’ and intervention rates; FMUs that were further away, with longer travel time to the nearest OU, had higher instrumental delivery rates and lower ‘straightforward birth’ and ‘normal birth’ rates.

When we repeated the analysis after removing one very distant FMU that appeared to be an outlier, the associations between distance and intervention rates were unchanged (data not shown). Scatterplots showed that size and distance were correlated, with FMUs located further away from the nearest OU tending to be smaller (data not shown).

Characteristics of NHS trusts providing home birth services in the study sample

In most NHS trusts the number and percentage of births taking place at home was relatively small (Table 16). For three primary care trusts providing home birth services it was not possible to calculate an appropriate denominator for the percentage of births in the trust that took place at home.

Variation in intervention rates between NHS trusts providing home birth services

Funnel plots showed more variation between NHS trusts providing home birth services in adjusted rates of ‘straightforward birth’ and ‘normal birth’ in ‘low risk’ planned home births than would be expected by chance (Figures 9 and 10 and Table 17). There was little variation in adjusted rates of instrumental delivery or intrapartum caesarean section in nulliparous or multiparous women planning home birth. Almost all ‘outlier’ trusts had higher, rather than lower, rates of intervention than would be expected by chance (see Figures 9 and 10). In most trusts there were very small numbers of nulliparous women planning home birth in the study sample.

FIGURE 9.

Funnel plots showing adjusted instrumental delivery and intrapartum caesarean section rates in women planning birth at home in different NHS trusts. (a) Percentage of instrumental delivery in nulliparous women; (b) percentage of instrumental delivery in multiparous women; (c) percentage of caesarean section in nulliparous women; and (d) percentage of caesarean section in multiparous women.

FIGURE 10.

Funnel plots showing adjusted ‘straightforward birth’ and ‘normal birth’ rates in women planning birth at home in different NHS trusts. (a) Percentage of ‘straightforward birth’ in nulliparous women; (b) percentage of ‘straightforward birth’ in multiparous women; (c) percentage of ‘normal birth’ in nulliparous women; and (d) percentage of ‘normal birth’ in multiparous women.

Trust characteristics and intervention rates in NHS trusts providing home birth services

In nulliparous women neither the absolute number of home births in the NHS trust nor the percentage of births that took place at home was associated with rates of any of the outcome measures studied (Table 18). In NHS trusts where the absolute number and the percentage of births at home was higher, the ‘normal birth’ rate in multiparous women was statistically significantly higher, and in trusts where the percentage of births at home was higher the ‘straightforward birth’ rate was also higher and the instrumental delivery rate lower, but these characteristics explained very little of the variation observed.

Summary of main findings

Strengths and limitations

General strengths and limitations of the Birthplace cohort study dataset are discussed in Chapter 8. For the analyses presented in this chapter, we were constrained by the extremely limited availability of data on the characteristics of maternity units and NHS trusts providing home birth services in England. The data we used, assembled from a variety of sources, were sometimes of uncertain quality and validity and enabled us to evaluate a relatively limited range of unit characteristics, particularly for home births. To provide more robust measures of some unit characteristics for FMUs and AMUs, we combined several variables together to produce indices of size, staffing and distance to the nearest OU. Missing or incomplete data for some unit characteristics reduced the sample size for several analyses; we had only a small number of OUs in our sample, and so many of our findings need to be replicated in a larger sample of units.

We initially explored using multilevel modelling to investigate the extent to which unit characteristics explained variation in intervention rates, with the rationale that this method would enable us to model potentially complex inter-relationships between variables. However, because of missing data for some unit characteristics a relatively large proportion of units dropped out of these analyses and some models failed to converge. The linear regression models used could be fitted in all cases and had the advantage of allowing visual inspection of the fit of the model (see Appendix 5). We chose not to adjust for other unit characteristics in these models, as many unit characteristics were correlated with each other, making interpretation difficult.

Our analyses involved multiple testing, and should be considered exploratory in nature; our interpretation, therefore, focuses on factors which showed consistency of effect across a number of outcomes or where a number of variables measuring different aspects of the same fundamental characteristic, for example size, showed consistent associations with outcomes in the same direction.

Interpretation

Our findings confirm that there is significant variation in intervention rates in OUs and that this variation is not fully explained by the sociodemographic or clinical characteristics of women planning to give birth there. 18,25,26 We also found similar unexplained variation in intervention rates in planned AMU births, but less variation in intervention rates in FMUs and in NHS trusts providing home birth services, particularly for rates of intrapartum caesarean section.

In general, the institutional factors affecting intervention rates are poorly understood and previous studies have largely focused on intervention rates in births taking place in OUs. These have found some association between junior doctor staffing levels and intervention rates in OUs,27 and mixed or inconclusive results on the relationship between unit size or the level of neonatal care available and intervention rates. 28–30 Studies in France and Australia suggest that smaller units may have lower rates of some interventions, but in these studies smaller units differed from larger units in other ways, including, for example, the level of obstetric and neonatal care available and rurality. 28,35 There is little or no evidence on unit or NHS trust characteristics and their association with variation in intervention rates in women planning birth in non-OU settings.

In OUs we explored the effect of unit size and midwifery staffing on intervention rates and, while we observed some significant associations, these showed no consistent pattern, suggesting that there is no straightforward or clear relationship between the size of the unit or midwifery staffing and OU intervention rates, at least within the ranges observed within our samples. 36 We also found that having an AMU in the hospital was associated with significant variation in some of our outcome measures, but with no consistent pattern. AMUs vary considerably in their characteristics, for example size, number of midwives, whether midwifery staff work only in the AMU or work on rotation in the AMU and the OU, and the extent to which they are physically separate from the OU. 36 Some operate an ‘opt-in’ admissions policy, whereby women have to actively choose to plan birth in the AMU rather than in the OU, while others operate a policy whereby the default option for ‘low risk’ women is the AMU and these women have to actively ‘opt-out’ in order to plan birth in the OU. 36 We were not able to explore the effect of these varied AMU characteristics on OU intervention rates. We cannot, therefore, rule out the possibility that the presence of an AMU may have an impact on intervention rates in planned births in the associated OU.

We did find that some variation in OU intervention rates may be explained by another aspect of configuration of care: specifically, that intervention rates in planned ‘low risk’ OU births tend to be higher in OUs situated in trusts where a higher proportion of women planned birth in a non-OU setting (home, FMU or AMU). It is possible that this apparent association was confounded by the presence of a large AMU. It was notable that in our sample the three OUs situated in trusts with the highest proportions of non-OU births were also those with attached AMUs. We also observed an association between the presence of an AMU and higher rates of some interventions in planned OU births. Because of the small number of OUs in our sample with an attached AMU, we were not able to assess the independent effects of these two factors (percentage of non-OU births and presence of an AMU). The Birthplace organisational case studies found that the presence of an AMU appeared to intensify perceptions of the workload in the adjoining OU, where service providers reported struggling to support ‘normal birth’, suggesting one possible explanation for the findings relating to OU intervention rates. 10

There are also a number of possible alternative explanations for this finding. Unit intervention rates are readily available online17 and it is possible that, in NHS trusts where the OU is known to have a high intervention rate, a higher proportion of ‘low risk’ women may choose to plan birth in a non-OU setting in order to avoid unwanted interventions. However, we lack the evidence to confirm or refute this. Another possibility is that a higher proportion of ‘low risk’ births being planned in non-OU settings might result in a higher proportion of the women planning birth in the OU being, or being perceived to be,36 ‘higher risk’, possibly resulting in a more ‘medicalised’ approach to birth for women in the unit. However, we found no significant association between the proportion of planned non-OU births in the trust and the proportion of planned OU births that were classified as ‘higher risk’. Nor did we find a consistent relationship between the proportion of women planning OU birth who were ‘higher risk’ and intervention rates, suggesting that OUs with more ‘higher risk’ births did not necessarily exhibit an ‘interventionist culture’. Another possible explanation is selection bias, whereby the women opting for a non-OU birth might be those most keen to have a ‘normal birth’ without medical intervention, resulting in the planned OU group being less ‘intervention averse’. Other studies provide some support for this hypothesis. Women who opt for a hospital birth rather than a home birth in the Netherlands have been found to be more receptive to intervention,37 and one UK study found that willingness to accept intervention was a significant predictor of operative or instrumental birth. 38 Finally, there is evidence to suggest that midwives working in units with high intervention rates tend to have a higher perception of risk. 39

The vast majority of births take place in OUs,8,32 and in our sample of OUs only six (19%) were situated in trusts where > 10% of births were planned in a non-OU setting, with considerably lower levels of non-OU births in most trusts. The risk of interventions is substantially reduced in births planned in non-OU settings. 7,8,40 We lacked the data to examine intervention rates at a trust level, but the increase in interventions observed in ‘low risk’ planned OU births in trusts with more non-OU births was relatively small in comparison, suggesting that any increase would be more than offset by the reduction in interventions in the ‘low risk’ planned non-OU births. This needs to be verified using data on trust-level variations in intervention rates.

In AMUs, FMUs and NHS trusts providing home birth services we found some different patterns of association between unit or trust characteristics and intervention rates. In all these settings the interventions we considered as outcome measures can only take place after transfer to an OU. Consideration therefore needs to be given to the extent to which the associations seen are a function of measured or unmeasured characteristics of the MU or the NHS trust itself or of the OU(s) to which women are transferred, most of which were not included in our analyses because the Birthplace cohort study included only a stratified sample of 36 OUs (out of 180 OUs in England at the time).

For AMUs we found considerable variation in intervention rates, particularly in nulliparous women, and our main analysis suggested that intervention rates tended to increase with the AMU’s size. However, our sensitivity analyses revealed that these associations appeared to be driven by higher intervention rates in a single large AMU in our sample. Within the size range of the vast majority of AMUs in our sample (< ≈1500 planned births per year), AMU size did not, therefore, appear to be associated with variation in intervention rates. The observed wide variation in intervention rates between AMUs may be explained by other important unmeasured differences in the characteristics of AMUs,36 such as, for example, whether they operate an opt-in or an opt-out admission policy, whether or not they are physically separate from the OU and/or whether or not they share midwifery staff with the OU. The characteristics and intervention patterns of the associated OU may also be influential, but given the available data we were unable to investigate this. Since the Birthplace cohort study took place, the number of AMUs has increased by 71% from 51 to 87,32 and so further research on the characteristics of AMUs and intervention rates in women planning birth there is merited.

In FMUs we explored associations between unit characteristics, measuring size, midwifery staffing levels and distance to the nearest OU and intervention rates, and found that larger FMUs had lower intervention rates and that FMUs located further away from the nearest OU had higher intervention rates, both in nulliparous women. However, as more distant units, many of which serve small rural communities,41 were also more likely to be smaller, it was not possible to clearly separate out the independent effects of size and distance. None of the FMU characteristics considered, including size and distance, was associated with variation in intrapartum caesarean section rates. For planned home births, we also found indications that NHS trusts with higher absolute numbers or proportions of births that took place at home might have lower rates of intervention. The observed associations for both FMUs and home births are consistent with other analyses reported in Chapter 5, which showed similar associations between the size and location of FMUs and number of home births in a trust and transfer rates. Possible explanations for these findings on transfer, which may explain the results reported in this chapter, are considered in Chapter 5.

Study population

The main study population for these analyses was ‘low risk’ women in the Birthplace cohort with a term pregnancy (37 to 42 + 0 weeks’ gestation) planning a vaginal birth in an OU, in an AMU, in a FMU or at home. Women with unknown parity were excluded.

Outcome measures

We used four main outcome measures, two capturing interventions during labour and birth, one indicating birth without complications that might affect future births and one indicating birth with low intervention:

• instrumental delivery (forceps or ventouse)

• intrapartum caesarean section

• ‘straightforward birth’, defined as birth without forceps or ventouse, intrapartum caesarean section, third- or fourth-degree perineal trauma or blood transfusion

• ‘normal birth’, defined as birth without induction of labour, epidural or spinal analgesia, general anaesthetic, forceps or ventouse, caesarean section or episiotomy. 12

Statistical methods

We used multivariable Poisson regression to study associations between the following and each outcome in turn:

• planned place of birth (OU, AMU, FMU, home)

• ethnicity (white, non-white)

• IMD (less deprived: IMD quintiles 1, 2 and 3; more deprived: IMD quintiles 4 and 5)

• maternal age in years (< 20, 20–24, 25–29, 30–34; ≥ 35).

Ethnicity and IMD were dichotomised to ensure that all groups had large enough numbers for statistical testing.

Relative risks were adjusted for these and other maternal characteristics and the presence of ‘complicating conditions’ identified at the start of labour care as described in Chapter 2 (see Poisson regression models).

Initially, the absolute and adjusted RRs for each maternal characteristic and for AMUs, FMUs and home births compared with OUs were estimated with adjustment for the variables listed in Chapter 2 (see Maternal characteristics) using Poisson models (main effects models). We then investigated whether the effect of planned place of birth (OU, AMU, FMU and home) on each outcome measure was modified by ethnicity (white vs. non-white), area deprivation score (less deprived vs. more deprived) or maternal age (< 35 years vs. ≥ 35 years) by carrying out tests for interaction using Wald tests. To interpret these we also calculated unadjusted and adjusted absolute and RRs of each outcome in each setting for each category of ethnicity, deprivation or maternal age also using Poisson models. When modelling interactions, we adjusted for other maternal characteristics using the variables and categories listed in Chapter 2 (see Maternal characteristics).

Characteristics of the sample

The study group consisted of 61,820 ‘low risk’ women with a ‘term’ pregnancy (37 to 42 + 0 weeks’ gestation), of whom 27,497 were nulliparous and 34,323 were multiparous; their characteristics are presented in Tables 1–4.

Association between planned place of birth, maternal characteristics and interventions during labour or birth

The following tables show the associations between planned place of birth, maternal characteristics and the outcomes studied: instrumental delivery (Table 19), intrapartum caesarean section (Table 20), ‘straightforward birth’ (Table 21) and ‘normal birth’ (Table 22), by parity. While the tables include results for all maternal characteristics included in the models, in order to show the effect of potential confounders, the commentary focuses on the explanatory variables of interest in this chapter; that is, ethnicity, area deprivation and maternal age.

Is the association between planned place of birth and intervention modified by ethnicity, area deprivation or maternal age?

Area deprivation

We found significant interactions between planned place of birth and area deprivation for instrumental delivery (p = 0.013), intrapartum caesarean section (p = 0.002) and ‘straightforward birth’ (p = 0.028) in nulliparous women and for ‘normal birth’ (p = 0.023) in multiparous women, indicating that the association between planned place of birth and these outcomes was different in more deprived areas compared with less deprived areas (Table 24).

TABLE 24 - Tests for interactions between planned place of birth and area deprivation

Test for interaction between planned place of birth and area deprivation using the Wald test.

Outcome	Nulliparous, p-valuea	Multiparous, p-valuea
Instrumental delivery	0.013	0.306
Intrapartum caesarean section	0.002	0.552
‘Straightforward birth’	0.028	0.051
‘Normal birth’	0.466	0.023

Tables 25–28 present absolute and RRs (unadjusted and adjusted) for the outcomes where we found significant interactions. Absolute risks for each of these outcomes by planned place of birth and area deprivation, adjusted for other maternal characteristics and ‘complicating conditions’, are also shown in Figures 11–14. In OUs the level of area deprivation did not modify the risk of any of the outcomes considered (instrumental delivery, intrapartum caesarean section, ‘straightforward birth’ and ‘normal birth’). Examination of the absolute and RRs of each outcome stratified by level of area deprivation revealed that although there were statistically significant interactions between planned place of birth and area deprivation for these outcomes, the pattern of association did not differ markedly between the more and less deprived groups. Irrespective of level of area deprivation, women who planned birth in a non-OU setting experienced a significant reduction in interventions, compared with women who planned birth in an OU.

TABLE 25 - Risk of instrumental delivery by planned place of birth and area deprivation in ‘low risk’ nulliparous women (n = 27,477)

Using same sample as adjusted analysis.

IMD quintiles 1, 2 and 3.

IMD quintiles 4 and 5.

Area deprivation and birth setting	Unadjusted		Adjusted
	Absolute risk (%)a	RRa (95% CI)	Absolute risk (%)	Adjusted RR (95% CI)	p-value
In OUs only
Less deprivedb	23.0	1	21.8	1
More deprivedc	21.9	0.95 (0.86 to 1.06)	23.5	1.08 (0.97 to 1.21)	0.179
Less deprived
OU	23.0	1	21.8	1
Home	12.7	0.55 (0.47 to 0.65)	11.2	0.51 (0.44 to 0.60)	< 0.001
FMU	12.1	0.52 (0.43 to 0.64)	11.8	0.54 (0.45 to 0.66)	< 0.001
AMU	18.3	0.79 (0.67 to 0.94)	17.7	0.81 (0.69 to 0.96)	0.015
More deprived
OU	21.9	1	23.5	1
Home	12.7	0.58 (0.47 to 0.72)	11.9	0.51 (0.41 to 0.63)	< 0.001
FMU	8.8	0.40 (0.32 to 0.51)	9.5	0.40 (0.32 to 0.51)	< 0.001
AMU	14.2	0.65 (0.52 to 0.81)	15.3	0.65 (0.52 to 0.81)	< 0.001

FIGURE 11.

Adjusted absolute risks (%) of instrumental delivery in ‘low risk’ nulliparous women by planned place of birth and area deprivation.

TABLE 26 - Risk of intrapartum caesarean section by planned place of birth and area deprivation in ‘low risk’ nulliparous women (n = 27,477)

Using same sample as the adjusted analysis.

IMD quintiles 1, 2 and 3.

IMD quintiles 4 and 5.

Area deprivation and birth setting	Unadjusted		Adjusted
	Absolute risk (%)a	RRa (95% CI)	Absolute risk (%)	Adjusted RR (95% CI)	p-value
In OUs only
Less deprivedb	16.4	1	15.6	1
More deprivedc	14.7	0.90 (0.78 to 1.03)	14.9	0.96 (0.84 to 1.09)	0.513
Less deprived
OU	16.4	1	15.6	1
Home	9.3	0.57 (0.46 to 0.69)	9.6	0.62 (0.50 to 0.76)	< 0.001
FMU	5.8	0.35 (0.29 to 0.43)	6.7	0.43 (0.35 to 0.53)	< 0.001
AMU	8.8	0.54 (0.43 to 0.67)	9.8	0.63 (0.49 to 0.80)	< 0.001
More deprived
OU	14.7	1	14.9	1
Home	7.1	0.48 (0.36 to 0.64)	7.4	0.49 (0.37 to 0.66)	< 0.001
FMU	8.0	0.54 (0.41 to 0.71)	10.1	0.67 (0.52 to 0.87)	0.002
AMU	6.8	0.47 (0.37 to 0.59)	8.1	0.54 (0.42 to 0.70)	< 0.001

FIGURE 12.

Adjusted absolute risks (%) of intrapartum caesarean section in ‘low risk’ nulliparous women by planned place of birth and area deprivation.

TABLE 27 - Chance of ‘straightforward birth’ by planned place of birth and area deprivation in ‘low risk’ nulliparous women (n = 27,312)

Using same sample as the adjusted analysis.

IMD quintiles 1, 2 and 3.

IMD quintiles 4 and 5.

Area deprivation and birth setting	Unadjusted		Adjusted
	Absolute risk (%)a	RRa (95% CI)	Absolute risk (%)	Adjusted RR (95% CI)	p-value
In OUs only
Less deprivedb	57.0	1	58.9	1
More deprivedc	60.2	1.06 (1.02 to 1.10)	58.6	1.00 (0.96 to 1.03)	0.808
Less deprived
OU	57.0	1	58.9	1
Home	73.6	1.29 (1.22 to 1.36)	77.5	1.32 (1.25 to 1.39)	< 0.001
FMU	78.4	1.38 (1.31 to 1.45)	76.6	1.30 (1.24 to 1.36)	< 0.001
AMU	68.0	1.19 (1.13 to 1.27)	67.5	1.15 (1.09 to 1.21)	< 0.001
More deprived
OU	60.2	1	58.6	1
Home	76.8	1.28 (1.20 to 1.35)	78.4	1.34 (1.26 to 1.42)	< 0.001
FMU	79.6	1.32 (1.24 to 1.40)	73.2	1.25 (1.19 to 1.31)	< 0.001
AMU	74.9	1.24 (1.17 to 1.32)	70.6	1.20 (1.14 to 1.27)	< 0.001

FIGURE 13.

Adjusted absolute risks (%) of ‘straightforward birth’ in ‘low risk’ nulliparous women by planned place of birth and area deprivation.

TABLE 28 - Chance of ‘normal birth’ by planned place of birth and area deprivation in ‘low risk’ multiparous women (n = 34,174)

Using same sample as the adjusted analysis.

IMD quintiles 1, 2 and 3.

IMD quintiles 4 and 5.

Area deprivation and birth setting	Unadjusted		Adjusted
	Absolute risk (%)a	RRa (95% CI)	Absolute risk (%)	Adjusted RR (95% CI)	p-value
In OUs only
Less deprivedb	74.9	1	75.9	1
More deprivedc	78.9	1.05 (1.02 to 1.09)	78.3	1.03 (1.00 to 1.07)	0.051
Least deprived
OU	74.9	1	75.9	1
Home	95.5	1.27 (1.23 to 1.32)	95.1	1.25 (1.21 to 1.30)	< 0.001
FMU	94.7	1.26 (1.22 to 1.31)	94.2	1.24 (1.19 to 1.29)	< 0.001
AMU	90.1	1.20 (1.15 to 1.25)	89.4	1.18 (1.13 to 1.23)	< 0.001
More deprived
OU	78.9	1	78.3	1
Home	96.1	1.22 (1.18 to 1.26)	94.8	1.21 (1.18 to 1.24)	< 0.001
FMU	94.2	1.19 (1.16 to 1.23)	92.3	1.18 (1.15 to 1.21)	< 0.001
AMU	91.4	1.16 (1.12 to 1.20)	89.3	1.14 (1.11 to 1.17)	< 0.001

FIGURE 14.

Adjusted absolute risks (%) of ‘normal birth’ in ‘low risk’ multiparous women by planned place of birth and area deprivation.

Maternal age

We found significant interactions between planned place of birth and maternal age for ‘straightforward birth’ in nulliparous (p < 0.001) and multiparous (p = 0.009) women and for ‘normal birth’ (p < 0.001) in nulliparous women, indicating that the association between planned place of birth and these outcomes was different in older and younger women (Table 29).

TABLE 29 - Tests for interactions between planned place of birth and maternal age

Outcome	Nulliparous, p-value	Multiparous, p-value
Instrumental delivery	0.194	0.762
Intrapartum caesarean section	0.469	0.420
‘Straightforward birth’	< 0.001	0.009
‘Normal birth’	< 0.001	0.214

Tables 30–32 present absolute and RRs (unadjusted and adjusted) for the outcomes where we found significant interactions. Absolute risks for each of these outcomes by planned place of birth and maternal age, adjusted for other maternal characteristics and ‘complicating conditions’, are also shown in Figures 15–17. In OUs the absolute and relative chances of ‘straightforward birth’ and ‘normal birth’ were significantly reduced in older nulliparous women; the chances of ‘straightforward birth’ were also significantly reduced in older multiparous women, although the difference was small (around 3%). In nulliparous women, the association between planned place of birth and these two outcomes was significantly modified by maternal age. Older nulliparous women were less likely than younger women to have a ‘straightforward birth’ or ‘normal birth’ but, irrespective of age, those who planned birth in a non-OU setting experienced a significant increase in the chances of ‘straightforward birth’ and ‘normal birth’, compared with women who planned birth in an OU. The relative increase in the chances of a ‘straightforward birth’ or ‘normal birth’ was greater in older women. In multiparous women, the association between planned place of birth and ‘straightforward birth’ differed significantly with maternal age, but the size of the differences was small.

TABLE 30 - Chance of ‘straightforward birth’ by planned place of birth and maternal age in ‘low risk’ nulliparous women (n = 27,312)

Using same sample as the adjusted analysis.

Maternal age and birth setting	Unadjusted		Adjusted
	Absolute risk (%)a	RRa (95% CI)	Absolute risk (%)	Adjusted RR (95% CI)	p-value
In OUs only
< 35 years	60.4	1	60.5	1
≥ 35 years	43.6	0.72 (0.67 to 0.77)	45.6	0.75 (0.70 to 0.81)	< 0.001
< 35 years
OU	60.4	1	60.5	1
Home	76.1	1.26 (1.21 to 1.32)	74.2	1.23 (1.17 to 1.28)	< 0.001
FMU	79.9	1.32 (1.27 to 1.38)	76.9	1.27 (1.22 to 1.32)	< 0.001
AMU	72.9	1.21 (1.15 to 1.27)	69.9	1.15 (1.10 to 1.21)	< 0.001
≥ 35 years
OU	43.6	1	45.6	1
Home	69.2	1.59 (1.43 to 1.76)	68.4	1.50 (1.35 to 1.67)	< 0.001
FMU	67.9	1.56 (1.39 to 1.75)	66.6	1.46 (1.30 to 1.64)	< 0.001
AMU	60.6	1.39 (1.24 to 1.56)	59.7	1.31 (1.17 to 1.47)	< 0.001

FIGURE 15.

Adjusted absolute risks (%) of ‘straightforward birth’ in ‘low risk’ nulliparous women by planned place of birth and maternal age.

TABLE 31 - Chance of ‘straightforward birth’ by planned place of birth and maternal age in ‘low risk’ multiparous women (n = 34,023)

Using same sample as adjusted analysis.

Maternal age and birth setting	Unadjusted		Adjusted
	Absolute risk (%)a	RRa (95% CI)	Absolute risk (%)	Adjusted RR (95% CI)	p-value
In OUs only
< 35 years	87.9	1	88.0	1
≥ 35 years	84.4	0.96 (0.94 to 0.98)	85.3	0.97 (0.95 to 0.99)	0.002
< 35 years
OU	87.9	1	88.0	1
Home	97.4	1.11 (1.09 to 1.12)	96.2	1.09 (1.08 to 1.11)	< 0.001
FMU	97.1	1.10 (1.09 to 1.12)	96.2	1.09 (1.08 to 1.11)	< 0.001
AMU	95.1	1.08 (1.06 to 1.10)	93.8	1.07 (1.05 to 1.08)	< 0.001
≥ 35 years
OU	84.4	1	85.3	1
Home	97.0	1.15 (1.12 to 1.18)	96.3	1.13 (1.10 to 1.16)	< 0.001
FMU	96.6	1.14 (1.11 to 1.17)	96.1	1.13 (1.10 to 1.16)	< 0.001
AMU	93.1	1.10 (1.07 to 1.13)	92.5	1.08 (1.06 to 1.11)	< 0.001

FIGURE 16.

Adjusted absolute risks (%) of ‘straightforward birth’ in ‘low risk’ multiparous women by planned place of birth and maternal age.

TABLE 32 - Chance of ‘normal birth’ by planned place of birth and maternal age in ‘low risk’ nulliparous women (n = 27,394)

Using same sample as the adjusted analysis.

Maternal age and birth setting	Unadjusted		Adjusted
	Absolute risk (%)a	RRa (95% CI)	Absolute risk (%)	Adjusted RR (95% CI)	p-value
In OUs only
< 35 years	43.6	1	43.8	1
≥ 35 years	29.8	0.68 (0.63 to 0.74)	31.4	0.72 (0.66 to 0.78)	< 0.001
< 35 years
OU	43.6	1	43.8	1
Home	68.6	1.57 (1.47 to 1.69)	65.3	1.49 (1.38 to 1.60)	< 0.001
FMU	70.9	1.63 (1.51 to 1.75)	66.7	1.52 (1.41 to 1.64)	< 0.001
AMU	62.9	1.44 (1.33 to 1.57)	58.9	1.35 (1.24 to 1.46)	< 0.001
≥ 35 years
OU	29.8	1	31.4	1
Home	61.6	2.07 (1.82 to 2.35)	59.6	1.90 (1.66 to 2.17)	< 0.001
FMU	61.9	2.08 (1.80 to 2.41)	59.1	1.88 (1.62 to 2.18)	< 0.001
AMU	50.5	1.70 (1.47 to 1.96)	48.3	1.54 (1.32 to 1.79)	< 0.001

FIGURE 17.

Adjusted absolute risks (%) of ‘normal birth’ in ‘low risk’ nulliparous women by planned place of birth and maternal age.

Summary of main findings

These results confirm that ‘low risk’ women who plan birth in a non-OU setting are more likely to labour and give birth without intervention than comparable women planning birth in an OU.

Compared with white women, women from non-white ethnic groups were more likely to experience an intrapartum caesarean section, but, overall, if non-white women planned birth in a non-OU setting they experienced the same reduced risk of intervention as white women.

Women living in more socioeconomically disadvantaged areas were no more or less likely to experience an intervention than women living in relatively advantaged areas. We found statistically significant differences between more and less disadvantaged areas in the reduction in the risk of intervention associated with planning non-OU birth. However, the magnitude of these differences was small and, irrespective of area deprivation, women who planned birth in a non-OU setting experienced a reduction in interventions, compared with women who planned birth in an OU.

In ‘low risk’ nulliparous women the risk of interventions during labour and birth increased with increasing age, while in multiparous women there was no consistent pattern across age groups in any of the interventions considered. In nulliparous women the association between planned place of birth and ‘normal birth’ and ‘straightforward birth’ was different for younger and older women. Older nulliparous women who planned a non-OU birth had a greater relative increase in ‘normal birth’ and ‘straightforward birth’ than younger women. This greater relative increase at older ages partly reflects the comparatively low proportion of older nulliparous women who plan birth in an OU and have a ‘straightforward birth ‘ or ‘normal birth’ (45.6% and 31.4%, respectively); this ‘magnifies’ the relative effect of any given absolute improvement in rates in older women. For example, an absolute increase of 10 percentage points in the ‘normal birth’ rate from 31.4% (the rate observed in planned OU births in older nulliparous women) to 41.4% corresponds to a relative increase of 40%; a 10 percentage-point increase in the rates from 43.8% (the rate observed in planned OU births in younger nulliparous women) represents a 23% relative increase.

Strengths and limitations

The large national data set provided by the Birthplace cohort has enabled us to compare the risk of intervention across birth settings and within different groups of women in a relatively homogeneous population of ‘low risk’ women, while controlling for important potential confounders such as BMI. Previous analyses have shown that a higher proportion of women planning birth in an OU had ‘complicating conditions’ at the start of care in labour than did women planning birth in other settings, and these could potentially have affected intervention rates. In these analyses we additionally controlled for the presence of ‘complicating conditions’ at the start of care in labour. There remains the potential for confounding by unmeasured maternal characteristics, but it seems unlikely that the significant reductions in risk observed for women planning non-OU birth could be explained by differences in the clinical characteristics of the women planning birth in the different settings.

Despite the large overall size of the cohort, the number of women in some groups defined by ethnicity, area deprivation and maternal age was relatively small in some non-OU settings. For this reason, and additionally for ease of interpretation, when testing interactions between planned place of birth and these maternal characteristics we therefore combined groups. Our non-white ethnic group comprised women from many different ethnic backgrounds; by combining these groups, differences in outcomes between women from different ethnic backgrounds might have been concealed. Similarly, by combining the least disadvantaged three quintiles and the most disadvantaged two quintiles in our measure of area deprivation we will also have combined groups of women living in very different areas and with very varied individual socioeconomic status. 53 In the case of maternal age, we dichotomised maternal age into two groups (< 35 years and ≥ 35 years) for the purposes of analysing age interactions. Because intervention rates are strongly affected by age, analyses based on this crude dichotomisation may be affected by differences in the age structures of women planning birth in different settings. For example, the younger age distribution of nulliparous women aged < 35 years in the OU group is likely to have led to the relative difference in intervention rates between planned OU and non-OU births being underestimated in women aged < 35 years in analyses where age was dichotomised.

We tested for interactions using a common approach for all three variables. However, analyses reported elsewhere54 indicate that the association between maternal age and many interventions is broadly linear within the age range 16–40 years, so that we could have modelled age as a continuous variable for the purposes of testing interactions. Results from these alternative analyses are discussed below.

Interpretation

Other large studies conducted in the UK have found that women from some black and minority ethnic (BME) groups are at increased risk of an intrapartum caesarean section,26,46 although it is difficult to assess which specific groups are at increased risk because of differences in study design and in classification of ethnicity. Studies in other countries similarly show increased risk of intrapartum caesareans section in some, but not all, BME groups. 55 Our findings relating specifically to ‘low risk’ women add to the evidence that women from a BME background may be at increased risk of intrapartum caesarean section, but it seems probable that risks differ for different ethnic groups. None of the other outcomes considered showed a significant difference between white and non-white women, although we cannot rule out the possibility that risks may differ for some specific ethnic groups. Our findings suggest that planned birth in a non-OU setting appears to confer equal benefits on white and BME women in terms of reduced risk of intervention.

We did not find that the risk of intervention varied significantly with the level of disadvantage of the area in which the woman lived. This appears to be consistent with other UK studies that have examined the effect of area and individual level socioeconomic status on caesarean section rates. 25,47 For example, analysis of trends in emergency caesarean section rates in women from different socioeconomic backgrounds in Scotland showed that a social gradient (with higher rates among women from more disadvantaged areas) disappeared over the period 1980 to 2000. 47 Our findings did suggest that the association between planned place of birth and risk of intervention might be different for more and less advantaged areas, but the differences were small and did not appear to consistently ‘favour’ more or less advantaged areas.

Our finding that intervention rates increased with increasing age in ‘low risk’ women, particularly in those having a first baby, even after adjustment for maternal characteristics and the presence of ‘complicating conditions’ is generally consistent with previous studies56 and is also consistent with the age related pattern of intrapartum transfer reported in Chapter 5. Our findings also confirm that planning birth in a non-OU setting has a ‘protective effect’ in terms of a reduction in the risk of intervention in both older and younger women, although our findings do suggest that the ‘effect’ may vary with age. For reasons discussed above, in the analyses reported here we considered maternal age as a binary variable, with the age of 35 years as the cut-off point. In a more detailed analysis of the Birthplace cohort relating to maternal age, planned place of birth and intrapartum outcomes, published elsewhere, we considered age as a continuous variable. 54 This showed that up to the age of 40 years the risk of maternal interventions and adverse outcomes requiring obstetric care increased in a broadly linear way, with no step-change in risk below the age of 40 years and, furthermore, that there was no significant increase with maternal age in adverse neonatal outcome within the age range 16–40 years. That analysis similarly showed that the relative reduction in risk of instrumental delivery and intrapartum caesarean section associated with planned birth in a non-OU setting was not significantly affected by age.

Study population

The study population for the analyses reported in this chapter was ‘low risk’ women in the Birthplace cohort with a ‘term’ pregnancy (37 to 42 + 0 weeks’ gestation) planning birth in an AMU, in a FMU or at home (see Chapter 2, Characteristics of the ‘low risk’ sample).

Objective 1: maternal characteristics associated with transfer

Objective 2: variation in transfer rates and the impact of unit or trust characteristics

Objective 3: variation in transfer by time of the day and day of the week

For women who were transferred to an OU, data on the date and time of the following events were available: start of care in labour (data for all women), decision to transfer, start of transfer, first assessment by a midwife or an obstetrician in the OU and birth (data for all women). If these times did not follow a logical sequence then women were excluded from the analysis of transfer timings.

Objective 4: urgency and duration of transfer from freestanding midwifery units and planned home births

Weighting and clustering

For all analyses apart from histograms we adjusted for the varying duration of participation of units and NHS trusts and for the ‘clustering’ of women within units/NHS trusts, as described in Chapter 2 (see Probability weights and Robust standard errors).

Characteristics of the sample

The study population consisted of 16,632 ‘low risk’ eligible women with a ‘term’ pregnancy (37 to 42 + 0 weeks’ gestation) planning birth at home, 11,210 planning a FMU birth and 16,573 planning an AMU birth; their characteristics are summarised in Table 33. As has previously been reported, women planning birth in an AMU tended to be more ethnically and socially diverse than women planning birth at home or in a FMU. 7,8 More women planning birth at home, and more women transferred from planned home births, were aged > 30 years and were having a second or subsequent baby, compared with women planning birth in, and transferred from, FMUs or AMUs.