This is an observational study involving secondary analysis of maternity records, in which information was recorded contemporaneously by health professionals as pregnancies progressed. In the UK, even if a home birth is planned, a pregnant woman receives maternity care from health care professionals who are based at an individual hospital, so the hospital records included planned home births as well as planned hospital births. The study data were taken from the St Mary’s Maternity Information System (SMMIS), a computerised records system which was used by most of the hospitals within the former North West Thames Regional Health Authority (RHA) area during the study period. Between 1988 and 2000 (inclusive), 15 National Health Service (NHS) hospitals contributed data relating to all the pregnancies for which they provided any maternity care. The participating hospitals came from a wide range of types and locations, so there is no reason to suppose that the results are unrepresentative of the region as a whole.
A total of 585,291 pregnancies from the 15 hospitals were included in the SMMIS database. Studies have concluded that the completeness and quality of the information recorded within SMMIS is good. For example, studies comparing the information recorded on the database against case notes found a very high degree of corroboration (at least 95% agreement for most variables, but with somewhat lower levels of corroboration for maternal blood pressure and haemoglobin levels), and a high level of consistency across different hospitals [14
Figure illustrates the groups excluded from this analysis. All pregnancies which did not end in either a live birth or stillbirth were excluded because they were not relevant to the research question, ie miscarriages and terminations were not part of the analysis. Pregnancies for which the intended place of birth was not known were also deleted (0.4% of the study population), because it was not possible to determine whether the place of birth was planned or unplanned. Because unplanned home births are known to have worse outcomes than planned home births and planned hospital births [16
], this is a crucial distinction.
Numbers and types of pregnancies included in this analysis.
Unplanned home births were excluded from the analysis because they would have all been classed as having intended a hospital birth and their inclusion would have artificially increased the risks associated with planning a hospital birth. It could be argued that unplanned home births are similar to planned home births which were transferred to hospital during labour (because birth did not take place in the intended location), and that not getting to hospital in time is a risk of planning a hospital birth, and for this reason we have run the analysis both with and without unplanned home births (see ‘results’ section). However, there is a fundamental difference between the two situations: transferring from home to hospital generally involves a considered decision made by the labouring woman and her partner in consultation with the attending midwife. Unplanned home births do not involve a considered decision – they are an unavoidable response to circumstances such as very quick labour. Planning a home birth would not have avoided this risk completely, because there is still a chance that that the midwife would not have arrived in time for the delivery. By contrast, planning a hospital birth would generally avoid the risk of having to travel to hospital during established labour.
Research on the comparative safety of different birth settings tends to exclude ‘high-risk’ pregnancy; conventional wisdom states that women with ‘high-risk’ pregnancies should plan a hospital birth because they are at higher risk of negative pregnancy outcomes. In fact, there is little hard evidence to suggest that, if the pregnancy is ‘high-risk’, a negative outcome is more likely if a home birth is attempted, so there is an argument for including ‘high-risk’ pregnancies in this type of analysis, and this was attempted as part of this research project. The inclusion of ‘high-risk’ pregnancies in the analysis necessitated a significantly more complex model to control for the fact that the ‘intended hospital birth’ group contained a higher proportion of ‘high-risk’ pregnancies. Numerous conditions render a pregnancy ‘high-risk’, and the attempt to control for them all led to a model containing 27 covariates, which introduced the possibility of statistical problems such as collinearity. For this reason, the main results shown exclude ‘high-risk’ pregnancies. However, a model which included them was also fitted, with results given alongside the main findings for comparison.
Women who were classed as not having had a birth attendant were also excluded. These were mostly women who had intended a home birth and had had very rapid labours (the mean labour length for all those who intended a home birth was 6.0 hours, but among those who were unattended by an appropriate health professional it was 2.1 hours). Presumably in these cases the baby was delivered before the midwife arrived at the woman’s home. It was therefore possible to categorise them with unplanned home births in terms of the circumstances surrounding their labours and births. As with unplanned home births, an argument could be made for including unattended births in the analysis, as giving birth unattended is a risk of planning a home birth, so, as a further sensitivity analysis, the model was re-run with these observations included (see ‘results’ section).
Elective Caesarean sections were also excluded from the analysis. PPH was more common if the baby was delivered by Caesarean section than if born vaginally (in SMMIS, PPH occurred in 6.7% of emergency Caesareans and 4.3% of elective Caesareans, compared with just 1.1% of vaginal births). Because elective Caesareans occurred only in the ‘intended a hospital birth’ group, their inclusion in this analysis would have artificially inflated the risk of PPH for hospital births, because elective Caesareans tend to be performed in response to fears about the safety of vaginal delivery, eg if the foetus is malpresented. For similar reasons, medical inductions (ie those using oxytocin and/or prostaglandins) were excluded. These too only occurred in the ‘intended a hospital birth’ group, so that comparisons would have become irrelevant if the cases had been included.
There are a number of definitions of PPH. According to the Royal College of Obstetricians and Gynaecologists (RCOG), although an estimated blood loss of at least 500ml counts as a PPH, in the UK a case should be considered an “emergency” only when the blood loss exceeds 1000ml [18
]. For this reason, the definition of PPH adopted for this analysis was the loss of at least 1000ml of blood. In both the ‘home’ and ‘hospital’ groups, the amount of blood lost was recorded by clinicians and later inputted into a field in the SMMIS database.
Women were classed as having intended a home birth if: (a) a home birth was intended at booking and the baby was delivered at home, (b) a hospital birth was intended at booking but the baby was delivered at home, and SMMIS recorded the change in intention as having taken place before labour commenced, or (c) a home birth was intended at booking but the baby was delivered in hospital, and SMMIS recorded the change in intention as having taken place during labour. Thus, intrapartum transfers from home to hospital were included in the ‘intended a home birth’ group. Women were classed as having intended a hospital birth if: (a) a hospital birth was intended at booking and the baby was delivered in hospital, or (b) a home birth was intended at booking but the baby was delivered in hospital, and SMMIS recorded the change in intention as having taken place before labour commenced. Maternities were classed as unplanned home births if a hospital birth was intended at booking but the baby was delivered at home, and SMMIS recorded the change in intention as having taken place during labour.
The risk status of a pregnancy was defined using a mixture of maternal International Classification of Disease (ICD) codes [19
] and individual fields in the SMMIS database, and was based on a 2007 clinical guideline from the National Institute for Health and Clinical Excellence (NICE) which contained lists of medical and obstetric conditions which indicate increased risk of negative pregnancy outcomes [20
]. Some were listed as “suggesting planned birth at an obstetric unit” and some as “indicating individual assessment when planning place of birth”. Pregnancies with conditions in the former list were classed as ‘high-risk’, and those with conditions in the latter list as ‘medium-risk’. All other pregnancies were classed as ‘low-risk’.
The analysis was carried out using a logistic binary regression model, with PPH as the outcome variable and built using manual forward selection (with p
0.05 as the cut-off). Because SMMIS contained over 200 items of information for each pregnancy, the list of potential covariates was a long one. Covariates were selected after a literature review of characteristics associated with intended place of birth [10
] and/or PPH [12
Some potential covariates were excluded from the modelling despite being associated with PPH: mode of delivery, type of health professional attending delivery, type of pain relief used in labour and augmentation of labour. This is because these factors may act as mediators and may explain the difference between home and hospital birth, and therefore holding them constant would have led to controlling for the effect of planned place of birth on PPH. Had the aim of this analysis been to identify characteristics associated with PPH, clearly these covariates would have been included (as would many of the maternities excluded from the analysis as described earlier), so it would not be appropriate to use these results to draw conclusions about the association between PPH and covariates other than intended place of birth.
A number of covariates were not included in the final model because, after adjustment for the other model covariates, there was no significant association between them and PPH. These covariates are listed in Table .
Covariates excluded from model due to having no significant association with postpartum haemorrhage
Once the final additive model was built, interaction terms were tested, involving intended place of birth and: pregnancy risk factors, year, parity, maternal age and time of birth. None made a statistically significant improvement to the model fit.
There were no missing data for the outcome variable. The approach for handling missing data for the explanatory variables depended on the extent of the problem. If fewer than 0.1% of records had data missing on a variable, these records were deleted. If between 0.1% and 12% of records had data missing on a variable, a ‘missing’ category was created and included as a separate measure within the model. If more than 12% of records had missing data on a variable, that variable was not included as a covariate.
The Riverside Research Ethics Committee (REC) approved the project (REC reference number 08/H0706/42) on 17 April 2008.