Influence of errors in job codes on job exposure matrix-based exposure assessment in the register-based occupational cohort DOC*X

Objective Job-exposure matrices (JEM) may be efficient for exposure assessment in occupational epidemiological studies, but they rely on valid job information. We evaluated the agreement between JEM-based exposure estimates according to self-reported job titles converted to DISCO-88 codes and according to register-based DISCO-88 codes in the Danish Occupational Cohort with eXposure data (DOC*X). Furthermore, we evaluated the agreement between these two sets of DISCO-88 codes. Methods We used JEM regarding wood dust, lifting, standing/walking, arm elevation >90°, and noise from DOC*X. Participants from previous questionnaire studies were assigned JEM-based exposure estimates using (i) self-reported job titles converted to DISCO-88 codes and (ii) DISCO-88 codes registered in DOC*X, in four time periods (1976-78: N=7707; 1981-83: N=2193; 1991-94: N=2664; 2004: N=11 782). Agreement between the exposure estimates and between the DISCO-88 codes (four-digit levels, 1-4) was evaluated by kappa (κ) statistics. Sensitivities were calculated using the self-reported observation as the gold standard. Results We found substantial agreement (κ>0.60) between exposure estimates for all types of job-exposures and all time periods except for one κ. Low sensitivity (30-65%) was found for the period 1981-83, but for the other time periods the sensitivities varied between 60-91%. For individual 4-digit DISCO-88 codes, the sensitivities varied substantially and overall the sensitivities increased by lower digit level of DISCO-88. Conclusion The validity of the DISCO-88 codes in DOC*X was generally high. Substantial agreement was found for the JEM-based exposure estimates and the DISCO-88 codes per se, although the DISCO-88 code-specific agreement varied across digit levels and time periods.