A matched cohort study.
The study was conducted in North Jutland, a relatively sparsely populated region of Denmark where the local health authorities in 2001 had decided to transform two of the region's four maternity units into FMUs, opening in 2001 and 2004. The FMUs offered midwifery-led care during pregnancy and intrapartum and postnatal periods to low-risk women.
In a 3.5-year period between 2004 and 2008, data on socio-demographic factors, previous pregnancies and births, current pregnancy and birth, infants, FMU transfers, and maternal/neonatal readmissions 0–28 days postpartum were collected from patient records and the North Jutland Patient Administration System. The data collection was carried out by project staff with comprehensive professional knowledge of the field on basis of written instructions.
Data security and ethics
The project was approved by the Danish Data Protection Agency (reference number: 2005-41-5352) and the regional health authorities of North Jutland. Data were handled in strict confidentiality and in accordance with Danish law requiring neither approval from an ethics committee nor informed consent from patients for observational studies involving no risk or inconvenience to patients.36
Characteristics of the freestanding midwifery units
In Denmark care for low-risk women is midwifery-led in all birth settings. Both FMUs were located in community hospitals with an intensive care unit but without an obstetric service. The annual numbers of births in the FMUs were approximately 170 (Hobro) and 130 (Frederikshavn). Women transferred to OUs by ambulance using multidisciplinary regional criteria and continued care with an FMU or OU midwife under the supervision of an obstetrician. FMU midwives had at least 2 years' experience and training in obstetric emergencies, including ventouse delivery. FMU midwives provided antenatal care and out-of-hours postpartum care for all women in the area booked for both OU and FMU birth. FMU midwives also assisted at the nearest OU, if FMU not busy, and had 40–70 births a year. Additional contextual information is available in online table A.
Characteristics of the obstetric maternity units
Aalborg University Hospital is a one of five highly specialised Danish hospitals with a specialist OU who saw approximately 3500 births a year. Vendsyssel Hospital is a provincial hospital with 10 clinical specialities, including an OU providing care for low-risk and most high-risk pregnancies and a generalised paediatric ward. The annual number of births was approximately 1400. Mothers and infants with severe illness were transferred to Aalborg University Hospital or one of the other four, highly specialised hospitals in Denmark, depending on the condition (additional contextual information is available in online table B).
The study population was composed of an intervention group of 839 low-risk women from two FMU in Hobro and Frederikshavn, and a control group of 839 low-risk women, matched for key factors, who received routine care from the specialist obstetric unit at Aalborg University Hospital and the obstetric unit at Vendsyssel Hospital, Hjørring.
All labouring women admitted to the FMUs by their midwives on the basis of multidisciplinary, regional admission criteria were included in the study. As informed consent of participation was not required due to Danish legislation, all eligible women were included.
Women in the control group were eligible for inclusion only if they represented an individual match to the obstetric and social characteristics of a woman in the FMU group.
Women in both study groups were thus rigorously judged to be at low-risk and fulfil criteria for FMU birth, and included at the start of care in labour.
Excluded from the study were three women admitted to an FMU for emergency treatment without satisfying the criteria for FMU care; an event occurring very rarely.
Confounding is a main concern in cohort studies. The matched design was chosen because it potentially increases the statistical precision in a cohort study and effectively eliminates the association between the exposure (place of birth) and the matching variables, given a perfect balance of data is obtained on matched variables between groups.34 37 38
Matching is especially relevant in situations with non-linearity and intercorrelation between variables or where a substantial difference in the distribution of confounders between groups is expected.39
This was the case in the present study whose participants were recruited from areas characterised by varying degrees of urbanisation and heterogeneity in socio-demographic characteristics.40 41
Women in the control group were selected from the region's patient administration system which carries detailed information on the region's pregnant women. For each participant included in the FMU group, a control participant from the nearest OU was identified among the admitted low-risk women. The selection of matched control participants was conducted in accordance with strict guidelines by project staff that were blinded to the identity and the birth outcomes of women in the FMU group. The matching result was blinded until the selected control participants had given birth.
Matching was done prospectively on criteria with an established influence on birth outcomes42–45
: low-risk status, parity, smoking, body mass index (BMI), age, ethnicity, education, occupation and cohabitation status. A 100% match was carried out on: low-risk status, parity and smoking status. BMI and age were matched with a range of ±5; meaning that BMI/age scores of 22 were matchable with scores between 17 and 27. Socio-demographic characteristics such as ethnicity, education level, occupation and cohabitation status were matched within groups as shown in .
Definition of low risk
Women were judged to be at low risk if they were healthy, presented in spontaneous labour between 37+0 and 41+6 days of gestation and had an uncomplicated pregnancy and no medical/obstetric history or conditions increasing obstetric risk as outlined in the UK NICE intrapartum care guidelines.46
However, we considered healthy multiparous women as low-risk regardless of their age and BMI if their previous pregnancies and deliveries had been uncomplicated.
Variables and data measurement
The primary outcomes were Apgar score <7/5 min and caesarean section.
Secondary outcomes were as follows: (infant) Apgar score <9/5 min, <7/1 min; neonatal asphyxia; admittance to neonatal intensive care unit (NICU); admittance to NICU >48 h; neonatal readmission 0–28 days postpartum; (maternal) spontaneous vaginal birth; intact perineum; epidural analgesia; use of water tub for pain relief; abnormal fetal heart rate leading to action; dystocia; shoulder dystocia; instrumental vaginal delivery; postpartum haemorrhage >500 ml; first-/second-degree tear; third-/fourth-degree tear; maternal readmission 0–28 days postpartum. These outcomes were, along with a range of additional outcomes, defined prior to the initiation of the study, and reported as well as all cases of perinatal mortality and severe perinatal and maternal morbidity. Unfortunately, data on umbilical blood gas could not be obtained.
The intended birthplace at the start of care in labour was considered the exposure. The study did not aim to examine differences in maternal or perinatal mortality, since their low occurrence in the Danish low-risk population (0.065‰ and 3‰, respectively) would require an extremely large and therefore unrealistic number of participants.
The data were recorded in accordance with the National Birth Register and the North Jutland Birth Register, standards and guidelines applying to all four units and with which all midwives and doctors in the region were familiar. A stop watch was used when measuring Apgar scores. Postpartum haemorrhage was routinely estimated rather than measured.
Power calculation, sample size and changes in study protocol
Clinically important differences were defined, and power calculations performed for all the above-mentioned clinical endpoints. The frequencies used in the calculations originate in the North Jutland Birth Register and the international literature. Estimations of sample sizes were based on power calculation for the primary outcomes: Apgar score <7/5 min and caesarean section. The limited number of FMU births, at 300–350 per year, was also taken into account. The study was originally planned to include data on 1027 FMU participants and 1027 control participants over a period of 3.5 years, starting 1 January 2005; however, in October 2006, the local authorities unexpectedly announced the closure of its two FMUs. The National Board of Health expressed concern that the local authorities had introduced a new model of care that had not been subjected to adequate evaluation. The power to detect differences between our two study groups was consequently reduced, and a thorough revision of the study protocol was required. At the time of the FMU closures, 550 FMU participants had been included, and in order to obtain the largest possible sample of FMU participants, we included all of the 289 eligible women who had been admitted to the FMUs since the opening of the second FMU (1 March 2004). These women were prospectively matched with women from the nearest OU, thus ensuring total samples of 839 women in each group.
After the FMU closures, power calculations were rerun. The results showed that with a sample of 839 women in each group, the study sustained the power to detect clinically relevant differences between groups on all primary and secondary outcomes. For the two primary outcomes, the revised sample provided power (5% significance level, 80% power) to detect an increase in Apgar score <7/5 min from expected 1.07% in the OU group to 3.1% in the FMU group and a reduction in the incidence rate of caesarean section from 8.8% in the OU group to 5.5% in the FMU group.
Statistical analysis of data
Analyses were based on the intention-to-treat principle and carried out using STATA software, V.11.
The two groups (matched 1:1) were compared by paired tests on all measures, the McNemar' test for paired binary data (medical data on the birth process) and the Wilcoxon signed-rank test for paired continuous data (eg, birth weight). As we were concerned that residual confounding might remain, a supplementary regression analysis adjusting for the matching characteristics was performed using both continuous and grouped variables.47 48
For ease of interpretation (eg, calculation of confidence bands), ordinal outcomes were dichotomised, but we controlled for conclusive agreement with test results based on the original data.
The analysis for occipital posterior position was performed after excluding caesarean deliveries. For all comparisons, relative risks with 95% CIs were calculated. All reported p values were two-sided, and the level of statistical significance 5%. To check for bias introduced by the inclusion of FMU women giving birth in 2004, supplementary subgroup analyses were performed on 2004 data and main data, respectively.
A low-risk match was prospectively identified for all 839 women admitted to an FMU, and full follow-up was obtained for all 1678 women. Of the 839 FMU women, 733 (87.4%) gave birth as planned in the FMU or at home, assisted by a FMU midwife (cf. ). Transit births were included in the few cases where the woman had consulted a midwife <24 h before giving birth and had been advised to stay at home longer or return home.
Ninety-seven FMU women (11.6%) were transferred intrapartum; among these, two gave birth in the ambulance. Eleven, who were in early labour, were transported in their own vehicle. Twenty-seven transfers (3.2%) took place <2 h after birth, another 13 (1.5%) during the postnatal stay. The total number of transfers was 137 (16.4%).
As shown in , the matching produced two fully comparable groups in terms of key medical and socio-demographic factors. The FMU women's background details reflected the life conditions of the local population in general.40 49
With Aalborg and Hjørring municipalities as exceptions, the educational and income levels in North Jutland rank as the lowest in Denmark. In the FMUs' predominantly rural catchment areas, unemployment rates are high, which is reflected in a slightly higher rate of FMU women without employment outside the home.