Sample selection and validity of exposure-disease association estimates in cohort studies