A cautionary note on the use of attributable fractions in cohort studies