Linking primary study data with administrative and claims data in a German cohort study on work, age, health and work participation: is there a consent bias?

OBJECTIVES: We analysed the degree and impact of consent bias in the prospective study 'leben in der Arbeit (lidA)' after linking primary interview data with claims data from German statutory health insurance funds as well as with administrative data provided by the German Federal Employment Agency.

STUDY DESIGN: Prospective cohort study.

METHODS: Within two study waves (2011, 2014) primary data were collected based on computer-assisted personal interviews. During interview informed consent to data linkage was obtained. We used binary logistic regression analyses with participants' consent for record linkage as the dependent variable calculating odds ratios (ORs) and 95% confidence intervals (95% CIs) for independent variables. Several sociodemographic, socio-economic and work-related factors were modelled as potential determinants of consent.

RESULTS: A total of 4244 participants took part in both waves. After excluding invalid consent, 4178 participants were included in the analysis. About 3918 (93.8%) of these participants gave their consent to link their primary data with data from at least one source. Within regression analyses only moderate bias was found due to region of residence, apprenticeship, professional affiliations, income and number of diseases. Participants from former West Germany were less likely to have their study data linked with both data sources (OR 0.63 [95% CI 0.42-0.96]) than those from the former East Germany. Participants with no information on income were more likely to refuse consent to both data sources compared to the reference group (net income: under EUR 1000; OR 0.15 [95% CI 0.08-0.30]). Respondents with two (OR 1.37 [95% CI 1.06-1.77]) or three and more diseases (OR 1.30 [95% CI 1.02-1.66]) diagnosed by a doctor agreed more frequently to linking both data sources than participants without disease. There is just a small proportion of variance in consenting explained by the models (R(2): 0.063-0.085). Also, only small changes of factors' prevalence were observed in consenters.

CONCLUSIONS: For the first time in Germany, the lidA-study links primary survey data with health claims and administrative employment data. We conclude that there is only a minor relation between the analysed factors and consent behaviour of the participants. A linked data set may be used in further analyses without substantial biases.