|| Checking for direct PDF access through Ovid
To develop a job-exposure matrix (JEM) addressing smoking to allow for confounder adjustment in register-based occupational health studies. We combined and harmonised questionnaire and interview data on smoking from several Danish cohort studies and surveys in the time-period 1981–2013 for 2 64 054 employees registered with a DISCO-88 code (the Danish version of ISCO-88) in the Danish nationwide JEM database, DOC*X. We modelled the probability of being a smoker, and the amount of smoking (g/d) among smokers. In mixed models, age and sex were included as fixed effects and DISCO as random effect for six different time-periods. The proportion of smokers decreased linearly from 56% in 1981–90% to 19% after 2010, whereas the amount increased from 15.9 g/d in 1981 to 16.5 g/d in 1991–95, and then declined to 13.2 g/d after 2010. In general, the quality of the JEM increased by calendar year, as 23% and 71% of the DISCO-codes were represented in the first and latest time-period, respectively, on the most detailed 4 digit DISCO-level. This was also reflected in the calculated interclass correlation coefficient (ICC), which increased by calendar year. The within job-group variation was large relative to the between jobs variation, but the range between jobs was in general high, as the probability ranged from 6% to 40% and the amount from 8.0 to 19.5 g/d after 2010. We succeeded addressing a smoking JEM with substantial variability between jobs, which may prove a useful tool for confounder adjustment in register-based occupational studies.