limitations include the possibility of unmeasured or residual confounding (for example, we did not have information on family history of ADHD), the mental health outcome measures were based on a brief questionnaire (SDQ), sample attrition involving missing data on the exposure at 32 weeks gestation and response rates for the outcome measures, the influence of multiple testing as outcomes were assessed using information from different sources, and the representativeness of responders to the parent-completed SDQ. However, there was less bias in the availability of teacher or KS2 data and there is evidence that differential attrition within ALSPAC does not affect estimations of risk of behavioural disorder [38].
