Study design and population

This multicentre case-control study was conducted in seven of twenty New Zealand health regions (District Health Boards, DHB). Criteria for choosing these regions, was that they had the largest numbers of late stillbirths in New Zealand between 2007 and 2010 [14]. Two thirds of all New Zealand births occur in these seven health regions (Waitemata, Auckland, Counties Manukau, Waikato, MidCentral, Capital & Coast, and Canterbury). Recruitment occurred between February 2012 and December 2015. Ethical approval was obtained from the Northern “X” Regional Ethics Committee: NTX/06/05/054.

Women with multiple pregnancies and babies identified with major congenital abnormalities at any stage of the study were excluded. Cases were consenting women with a stillbirth at ≥28 weeks of gestation. We obtained the distribution of stillbirths by health region and gestation using New Zealand stillbirth data over a three-year period (2007–9) [15]. Controls were women with ongoing pregnancies randomly selected with frequency matching to a projected distribution of stillbirths, ensuring that controls would be at a similar gestation to cases in each participating health region. We planned to select two controls for each estimated eligible case in each health region.

Procedures and variables

Eligible subjects were given a brief description of the study by their midwife or doctor and asked whether the research midwife could contact them to explain the study and invite them to participate. If the woman agreed, a time and place for interview was arranged and written consent was obtained. Eligible participants were informed that the broad aim of the study was to identify modifiable risk factors for late stillbirth and specific hypotheses were not discussed. If the woman did not consent, age, ethnicity and parity were collected without identifiable information. Interpreters were organised for women who had difficulty reading or speaking in English.

Data were obtained by structured face-to-face interview (S1 Appendix) conducted by research midwives employed part-time in each recruitment centre and from medical records. For cases the interview occurred as soon after stillbirth as possible and for controls, as close to the allocated gestation as possible. Data collected in the interview included: demographic, lifestyle, general health and antenatal care details and information about a range of sleep practices [14]. Data about sleep practices were collected at the end of the interview. Height was measured by the research midwife and body mass index calculated using the earliest weight recorded in pregnancy. A single prioritized maternal ethnicity was determined as recommended by the New Zealand Ministry of Health [16]. Social deprivation was derived from the address where the participant lived during pregnancy, with category one least deprived and category five most deprived [17].

The primary outcome was the odds of late stillbirth associated with maternal self-reported going-to-sleep position, on the last night. The last night was the night before late stillbirth was thought to have occurred or the night before interview for controls. Information about usual going-to-sleep position in the last week was also collected. Going-to-sleep position on the last night was classified as: left side, right side, restless, supine (lying on the back), on the front, or propped. Restless going-to-sleep referred to women who reported that they had changed position frequently when settling to sleep and could not recall the position they finally fell asleep in. In women who did not remember any details of their going-to-sleep position, the position was recorded as unknown. Usual going-to-sleep position in the last week was classified as: left side, right side, variable side, supine (lying on the back), on the front or propped. To assist the participant to recall her going-to-sleep position the woman was asked to visualise herself lying in bed.

Participants were also asked if they got up at night to the toilet and how many hours they thought they had slept at night on the last night. Sleep duration was categorised as: <6 hours; 6 to 8 hours; and >8 hours, as per our previous study [8]. The frequency of sleeping during the daytime in the last week was also recorded.

Customised birthweight centiles, that adjust for maternal characteristics (height, early pregnancy weight, ethnicity and parity) as well as gestation at delivery and infant sex, were calculated for all infants [18]. The gestation used to calculate the centile for cases was the day the mother thought the baby had died and for controls the gestation at birth. Small for gestational age (SGA) was defined as birthweight <10^th customised centile [18].

All stillbirths were classified using the Perinatal Society of Australia and New Zealand Classification system [19]. Deaths classified as due to fetal growth restriction require either antenatal diagnosis of fetal growth restriction with abnormal Doppler studies and or specified placental pathology. If growth restriction was not identified before birth, late stillbirths with a customised birthweight centile <10^th are typically classified as unexplained [19].

Statistical analysis

We estimated that over a three year recruitment period there would be 415 eligible cases, 291 of whom would participate [15], based on 70% recruitment in The Auckland Stillbirth Study [8], and 830 eligible controls (estimated 582 participants). Given this sample size, a significance level of 0.05, power of 0.8 and prevalence of a risk factor in controls of between 30% and 60% (prevalence of non-left sleep position in The Auckland Stillbirth Study was 57%) we could detect an odds ratio (OR) of 1.50. With a 5% prevalence of a risk factor (supine sleep position in The Auckland Stillbirth Study controls) we could detect an OR of 2.2.

Statistical analyses were performed using SAS version 9.6 (SAS Institute Inc., Cary NC USA). Generalized additive models were fitted to explore the relationship between continuous variables (maternal age, BMI, birthweight centile) and stillbirth, and determine whether variables should be analysed continuously or categorically.

Differences between categorical variables were tested by chi-square tests. Continuous variables were compared using Wilcoxon Rank Sum tests. Univariable analysis was performed to evaluate the association between sleep practices and late stillbirth risk. A multivariable model was developed incorporating ethnicity and deprivation index, variables associated with increased risk of stillbirth based on previous literature (age, BMI, parity, smoking, SGA status), other sleep variables significant in univariable analysis, and variables used to select cases and controls (gestation and DHB). Unconditional logistic regression was used to adjust for potential confounders. No imputation was performed for missing data. Women who could not recall their going-to-sleep position on the last night (unknown going-to- sleep position) were excluded from the multivariable model. We also carried out a stratified analysis by term (≥37 weeks’) and preterm (≥28 to 36 weeks’) gestation. The c statistic has been calculated to assess model performance.

Statistical significance was defined at the 5% level. OR and adjusted odds ratios (aOR) with 95% confidence intervals (CI) were used to estimate risk. Population Attributable Risk (PAR) was calculated using unadjusted OR for the primary outcome, supine sleep position, and other potentially modifiable risk factors that remained significant in multivariable analysis [20].

Biological plausibility of findings

Reduced vena-caval diameter has been demonstrated by magnetic resonance imaging in supine compared with the left lateral position in late pregnancy [23]. Another study, using Doppler ultrasound, showed that blood flow in the uterine artery was less in the supine position than in the left lateral [10]. Adverse fetal effects of the supine position are suggested by reduced middle cerebral artery Doppler resistance, a fetal response to hypoxia [11], and reduced fetal oxygen saturation in labour in the supine position [24]. Furthermore we have recently reported that in healthy late pregnancy, when the mother is in the supine position, the fetus spends more time in behavioural state 1 (fetal quiescence) and less time in active fetal behavioral state 4, compared to when the mother is on her left side [25]. This provides additional evidence to support a relatively hypoxic environment for the fetus when a healthy mother is in the supine position in late pregnancy.

Supine position is also associated with sleep disturbed breathing, which increases in pregnancy, and has been associated with pregnancy complications including hypertensive disorders, gestational diabetes and fetal growth restriction [26]. Hence there are multiple mechanisms which could mediate the effects of supine position in late pregnancy on fetal wellbeing.

Comparison with other studies

Compared with controls in the Auckland Stillbirth Study [8], there has been an increase in left sided going-to-sleep position (43% versus 57.6%) and a small decrease in supine going-to-sleep position (5% versus 3.8%) in the current study, which demonstrates that going-to-sleep position is modifiable. In addition we have recently completed a survey of sleep practices in late pregnancy in a multi-ethnic community with high perinatal mortality [27]. A large majority of women surveyed (227/263, 86%) who reported they did not currently go-to-sleep on their left side responded that they could change to going-to-sleep on their left side if this was better for their baby [27].

In the current study, the risk of late stillbirth was not elevated in women who reported going-to-sleep on their right on the last night and the last week; whereas in our previous study there was a borderline increase in risk with right sided compared with left sided going-to-sleep position on the last night [8]. Further evidence is required to confirm whether right sided going-to-sleep position is associated with similar risk of late stillbirth to left sided going-to-sleep position.

Contrary to our hypothesis, we did not demonstrate an association between sleeping during the day or long night-time sleep duration and late stillbirth risk. We did demonstrate an association between short sleep duration on the last night and late stillbirth risk (aOR 1.81, 95% CI 1.14 to 2.88), similar to findings in The Auckland Stillbirth Study. One possible explanation for this association is that women with short sleep duration may have been experiencing discomfort from symptoms of impending labour on the last night.

Consistent with data from meta-analyses of stillbirth risk factors [6, 7], the highest population attributable risks for late stillbirth were associated with SGA infants (24.7%) and maternal obesity (20.4%). The large majority of SGA stillborn infants are not currently recognised as SGA before birth [28]. Programmes that hold promise for reducing mortality in SGA pregnancies, such as the UK based Growth Assessment Protocol [29], require further evaluation. Consistent with a recent study [30], we demonstrated a dose dependent association between increasing BMI and stillbirth risk. Multiple mechanisms are likely responsible for the association between obesity and stillbirth [31]. Our findings again highlight the importance of entering pregnancy with a normal BMI to optimise the likelihood of a healthy pregnancy.

Strengths and limitations of study

A case-control study is the most appropriate study design to address our hypothesis. As late stillbirth is relatively rare (3/1000 births), a prospective cohort study design is not feasible or affordable. Such a design would involve recruitment of hundreds of thousands of women with regular completion of questionnaires about sleep practices and other lifestyle factors.

In any case-control study there is potential for recall bias. This was reduced by using a structured interview and ensuring that participants and research midwives were not informed of specific study hypotheses. Information about going-to-sleep position was by necessity collected by self-report. The reliability of self-reported sleep position has been questioned [32], however, our group has recently demonstrated that there is good correlation between maternal short term recall of going-to-sleep position and going-to-sleep position recorded by video technology [33].

The median time to interview in women with late stillbirth was 24 days. This is unlikely to have resulted in recall bias which is a systematic difference between cases and controls in the accuracy of recalled information. The time lapse could influence the accuracy of recalled information but this would not be biased towards the exposure.

Maternal report was also used to estimate the timing of fetal death, with potential that in some cases the ‘last night’ was not the night before fetal death, or the night during which the baby died. However, a similar risk of late stillbirth was seen for women who reported they usually went to sleep supine in the last week (aOR 3.46, 95% CI 1.49 to 8.03) as for the last night. Furthermore 70.7% of women who reported going-to-sleep supine on the last night also reported usually going-to-sleep supine in the last week. This suggests that self-report of ‘last night’ is likely to be reliable and adopting a supine going-to-sleep position on the last night is not a response to the baby’s death.

During recruitment there was a reduction in late stillbirths after 37 weeks of gestation in New Zealand resulting in fewer cases recruited compared with projected numbers. With this decrease in late stillbirth we were underpowered to investigate interactions between going-to-sleep position and measures of fetal vulnerability such as small for gestational age infants and maternal obesity and smoke exposure [34]. We used very similar study design and questionnaires as The Auckland Stillbirth Study [8] and the Midlands and North of England Stillbirth Study [35] and are therefore planning an individual participant data meta-analysis to investigate these interactions and also to determine whether right sided going-to-sleep position is associated with increased risk (PROSPERO registration: CRD42017047703).

Controls in our study were selected from the pregnant population at similar gestation to the stillbirths, allowing comparison of maternal lifestyle at similar gestational age. The participants were also broadly representative of the eligible population and our results are likely to be generalisable to other similar populations.