AIMS: To investigate the relative validity of retrospectively calculated pack-years (py-retro) by comparing py-retro with prospectively calculated pack-years (py-pro).
DESIGN: A 23-year ongoing cohort study (1977-2000).
PARTICIPANTS: One hundred and fifty-four males and females, 13 years old in 1977 and 36 years old in 2000.
SETTING: Amsterdam, the Netherlands.
MEASUREMENTS: To calculate py-pro, current smoking and quitting efforts were investigated nine times in a period of 23 years with the help of an interview or a questionnaire. At the age of 36, subjects filled out a comprehensive questionnaire about their smoking history, to calculate py-retro. Individual differences between py-pro and py-retro were calculated. In addition, Cohen's kappa was calculated after categorising py-pro and py-retro into three groups.
FINDINGS: (1) Py-retro does not under- or overestimate life-time tobacco smoking. (2) The relative validity of py-retro was moderate due to large individual differences between py-pro and py-retro. (3) The individual differences between py-pro and py-retro became larger, the higher the number of pack-years. (4) Mean difference (and 95% limits of agreement) between py-pro and py-retro was -0.039 (-5.23, 5.32) when average pack-years was < 5.2 and -1.17 (-10.00, 14.65) when pack-years > or = 5.2. 5. Cohen's kappa between categorized py-pro and py-retro was 0.79.
CONCLUSIONS: Future researchers in the field of smoking should be aware of the moderate relative validity of py-retro. Categorizing py-retro into smoking groups results in a misclassification error that is smaller than the quantitative error in continuous py-retro, but goes together with a loss of information.