Smoking and degree of occupational exposure: are internal analyses in cohort studies likely to be confounded by smoking status?