The Raine study had no evidence of significant perinatal selection bias after two decades of follow up: a longitudinal pregnancy cohort study

The Raine study had no evidence of significant perinatal selection bias after two decades of follow up: a longitudinal pregnancy cohort study

BMC Pregnancy and Childbirth, June 2017

28662683

Scott W. White, Peter R. Eastwood, Leon M. Straker, Leon A. Adams, John P. Newnham, Stephen J. Lye, Craig E. Pennell

Cohort studies may increase or decrease their selection bias as they progress through time. The Western Australian Pregnancy Cohort (Raine) Study has followed 2868 children for over two decades; from fetal into adult life. This paper analyses the cohort over time, assessing potential bias that may come and go with recruitment, retention and loss of participants. Linked data from all births in Western Australian over the 3 years the Raine Cohort was recruited were obtained to compare perinatal characteristics and subsequent health outcomes between the Western Australian (WA) contemporaneous birth population and the Raine Cohort at five time points. Perinatal exposure-outcome comparisons were employed to assess bias due to non-participation in Raine Study subsets. There were demographic differences between the Raine Study cohort and its source population at recruitment with further changes across the period of follow up. Despite these differences, the pregnancy and infant data of those with continuing participation were not significantly different to the WA contemporaneous birth population. None of the exposure-outcome associations were significantly different to those in the WA general population at recruitment or at any cohort reviews suggesting no substantial recruitment or attrition bias. The Raine Study is valuable for association studies, even after 20 years of cohort reviews with increasing non-participation of cohort members. Non-participation has resulted in greater attrition of socially disadvantaged participants, however, exposure-outcome association analyses suggest that there is no apparent resulting selection bias.