Retrospective assessment of shift-work in epidemiological studies - Lessons learned.

The reliable assessment of characteristics of shift-work exposure remains a critical methodological issue in epidemiological studies. A working group of the International Agency for Research on Cancer developed recommendations for the assessment of shift-work. These were translated into a detailed interview for the 10-year follow-up of the Heinz Nixdorf Recall (HNR) cohort. This study investigated the agreement of shift-work characteristics between three different assessments that were administered in interviews at two time points. At the study baseline (2000-2003), 4,814 participants were enrolled, and brief shift-work information was collected for 2,121 working participants (1,244 men and 877 women aged 45-75 years). Of 2,613 cohort members in the prospective 10-year follow-up between 2011 and 2013, 2,444 (also non-working) individuals participated in detailed shift-work interviews that consisted of (a) key summary questions and (b) period-based shift-work histories. participants' shift-work exposures up to the study baseline were compared in 1,217 subjects who were interviewed during both the baseline and the follow-up. Within the follow-up, participant responses to key summary questions were compared with calculated parameters from period-based histories. Agreement was measured by simple agreement (%), Gwet's agreement coefficient 1 (Gwet's AC1), and intraclass correlation coefficients (ICC) with 95% confidence intervals (CI). Beta-regression models were applied to investigate potential associations between age and sex with the reliability of shift-work characteristics. A high level of agreement was found between ever having worked shift-work (ever shift-work) and duration that each participant worked shift-work (duration of shift-work) reported during baseline and at follow-up (ever shift-work until study baseline in men: Gwet's AC1 = 0.77 (CI 0.72-0.82)). When comparing key summary questions and detailed shift-work histories, the duration of shift-work showed a high level of reliability that marginally decreased with age (in women ICC = 0.96 (CI 0.95-0.97), linear effect of age groups on µ: p = 0.08). Participants had problems remembering more detailed shift-work information. Ever shift-work and duration of shift-work can be assessed with key summary questions but reliability slightly decreases with age.