OP18 Using linked administrative data to reduce bias due to missing outcome data in exposure-outcome estimates: a study of the association between breastfeeding and iq using simulations and data from a birth cohort