They were significantly less likely to weigh less than 2500 g at birth or to be less than 37 weeks gestation.

This is the first prospective cohort study of maternal and neonatal 17-AAG supplier outcomes among women who planned to give birth in freestanding midwifery units in Australia. Selection bias was minimised by prospectively identifying women’s planned place of birth at booking and analysing the outcomes according to the place where women intended to give birth. The use of a population database ensured that there was a minimal loss to follow-up and minimal bias introduced due to a non-response rate. All women who planned to give birth at a freestanding midwifery unit were included in the study, regardless of identified risks at booking. In this way the outcomes reflect the current practice and function of freestanding midwifery units in Australia. The study ensured comparability of the cohorts of women by rigorously judging the tertiary-level maternity unit group at booking to be at low risk of developing obstetric complications, and also by controlling for risk at the onset of labour during analysis. The study is limited because it was not possible to randomly assign women to one or other maternity unit and system of care, therefore

leaving a potential for selection bias. In particular, the subtle differences that may exist between women who plan to give birth where there is no specialised medical support on site and those who choose to go to a tertiary-level maternity unit cannot be quantified. Thirty-four women from the tertiary-unit group crossed over to give birth in the freestanding midwifery unit group, although these women represented

less than 1% of the study population. These factors, along with not controlling for BMI and socioeconomic status, may have had a bearing on some of the outcome measures. Selecting a prospective comparative reference cohort from the referral hospitals and analysing the data Batimastat according to the place where women intended to give birth went some way in addressing the selection bias at the design stage. A further limitation of the study was the inability to retrieve data on severe morbidity recorded in databases other than the one available for the study. As a result this study could not provide the level of information relating to more complex measures of maternal and perinatal morbidity as employed in other studies.19 22 28 This reflects the fragmented nature of routine maternity information system databases. No inferential statistics were applied to some measures because of small numbers; however, the detailed reporting of adverse and rare events strengthened the study.