The Raine study
The recruitment and follow up of the Raine Study has previously been described in detail [24, 25]. In brief, between 1 January 1989 and 31 December 1991, 2900 pregnant women and their fetuses receiving antenatal care at King Edward Memorial Hospital, the sole tertiary referral obstetric hospital in Western Australia, were enrolled prior to 18 weeks’ gestation into a randomised controlled trial investigating the effects of repeated prenatal ultrasound examinations. Liveborn offspring of consenting parents were enrolled into a longitudinal cohort study aimed at assessing the early life origins of adult disease. Follow up assessments at ages 1, 2, 3, 5, 8, 10, 14, 17, 18, and 20 years were undertaken, collecting substantial data concerning, among others, perinatal factors, nutrition, behaviour, neurodevelopment, body composition, cardiovascular and metabolic parameters, musculoskeletal health, mental health, socioeconomic factors, stress responses, genetics and epigenetics, eye health, sleep, and reproductive health.
Linked data
Linked data were obtained with permission from the Data Linkage Branch of the Department of Health, Western Australia, using the Midwives’ Notification System and Hospital Morbidity database. The Midwives’ Notification System is a compulsory notification of every birth in Western Australia. This system reliably captures all births within hospitals and all births attended by the state-funded home birth programme, together accounting for all Western Australian births apart from a statistically negligible number of unattended births. It records details regarding the mother (age, height, marital status, gravidity, parity, ethnicity, certainty of menstrual dating), the pregnancy (prenatal complications, pre-existing medical comorbidities, onset of labour, labour procedures and complications, and mode of delivery), and the baby (weight, length, head circumference, best estimated gestation, special care nursery admission, and length of hospital stay). This is the only time point at which reliable data are collected regarding the health of every individual in the state. The Hospital Morbidity database captured 303,731 hospital admissions among 107,285 individuals between 1980 and 2010. Of relevance to this study, it contained data regarding neonatal morbidity including jaundice.
Socioeconomic status was described using the publicly available Index of Relative Socioeconomic Disadvantage (IRSD) of the Socioeconomic Indexes for Areas (SEIFA) by the Australian Bureau of Statistics [26]. These data were available either during pregnancy, at birth, or at age 1 year for approximately half of the Raine Study participants who participated in the 14-year follow up assessment. IRSD data were not available for the Western Australian (WA) population subset of mothers giving birth in Perth, so comparisons were made to the entire Western Australian metropolitan population as assessed at the 1991 census (n = 1,068,115).
Cohort subset comparisons
In order to assess how well the Raine Study cohort represents the general Western Australian population on perinatal characteristics, the liveborn offspring of the Raine Study who consented to follow up (the Raine Pregnancy Cohort, n = 2863) were compared to the remaining individuals born in Western Australia during the 3 years of recruitment (n = 99,141). Comparisons were made between five subsets of the total cohort, including: (i) participants at recruitment (n = 2868); (ii) participants of the 5 year follow up assessment (Raine 5-year Cohort, n = 2010); (iii) participants of the 20 year follow up assessment (Raine 20-year Cohort, n = 1213); (iv) participants with repeated fetal biometry and genome wide single nucleotide polymorphism data (the Raine fetal growth subset, n = 1377); and (v) participants who underwent ultrasound screening for non-alcoholic fatty liver disease at age 17 years (NAFLD) (the Raine NAFLD subset, n = 879). The latter two subsets are included as examples of deliberately non-representative samples of the entire cohort, having specifically excluded ethnic minorities in order to aid genetic association studies, to allow assessment of any selection bias introduced by deliberate sampling in subgroup analyses.
Exposure-outcome associations
Comparison methodology was adapted from Nohr et al. [23] who described an approach to assess bias in cohort studies due to non-participation, using known exposure-outcome associations within the study population and the source population. With this method, the strengths of the associations in the two populations are compared by the “relative odds ratio”, whereby a study population with identical strengths of association as its source population will have a relative odds ratio of one. The 95% confidence interval of the relative odds ratio will contain one if there is no statistically significant bias due to non-participation in the study sample for that outcome.
Associations between environmental exposures in pregnancy and adverse perinatal outcomes for various population subsets were assessed using data from the Midwives’ Notification System. In addition, linking records within the Midwives’ Notification System with the Hospital Morbidity dataset allowed associations between epidemiological characteristics of an individual at and after birth.
The perinatal exposures considered in comparisons between the Raine Pregnancy Cohort and the contemporaneously born general WA population were vacuum extraction, low birth weight (less than 2500 g at term), advanced maternal age (greater than 30 years), and preterm birth. The outcome associations examined were spontaneous vaginal delivery and elective caesarean section from the Midwives’ Notification System, and neonatal jaundice from the Hospital Morbidity database as either the primary or additional diagnosis.
More comprehensive data available for the Raine Study subsets allowed the investigation of two further exposure-outcome associations: maternal smoking and low birth weight; and pre-eclampsia and preterm birth.
Statistics
Statistical analyses were performed with the R statistical software package, version 3.0.1 [27]. Comparisons between groups were made with the Student’s t-test for continuous variables or the Chi-square test for categorical variables. Bonferroni correction for multiple testing suggested a p-value of 0.003 as the threshold for significance for exposure-outcome association comparisons between cohorts and a p-value of 0.0003 for perinatal characteristics. All assessed variables were effectively normally distributed for the purposes of parametric testing, with the exception of IRSD which was significantly skewed in distribution. Confidence intervals for IRSD were, therefore, calculated using a bootstrapping method of 0.2 trimmed mean and 2000 repetitions for the Raine Study subsets and 100 repetitions for the WA population given the large sample size of the latter group. Comparisons of IRSD between groups were performed by the Wilcoxon rank-sum test.
Ethics, consent, and permissions
The Health Research Ethics Committee of the Department of Health (Western Australia) granted ethics approval for this study (Project 2010/24, July 2010). The broader Raine Study has ethics approval from The University of Western Australia Human Research Ethics Committee. Informed consent was provided by all participants. Participant assent and parental consent was provided for minors.