The importance of selection bias in prospective birth cohort studies