Limitationen und methodisches Vorgehen bei Ableitung einer Risikoverdopplungsdosis aus publizierten kategoriellen Daten

For the recognition of occupational lung cancer due to exposure to hexavalent chromium, a lung cancer doubling of risk dose of 500 µg/m³-years was proposed as an orientation measure. The aim of this work is to investigate different methodological approaches to quantitatively assess the doubling of risk dose and the associated uncertainties of the derivation from published data.

In a cohort study from the German chromate industry, on which the derivation of the orientation measure of 500 µg/m³-years, was based as well as the international population-based case-control study SYNERGY, a linear model, two log linear models and, in the SYNERGY study, a thin-plate spline regression model were calculated. Different scenarios (varying upper limits for the upper exposure category, different lung cancer induction periods) and various sensitivity analyses were considered.

Depending on the methodological approach, highly variable doubling doses of risk were derived, with the choice of the upper category midpoint exerting the strongest influence on the derived dose. The sensitivity analyses also led to significant variability in the estimated doubling of risk doses. In contrast, the statistical model had little influence on the results. Analyzing the SYNERGY study based on the aggregated published categorial data provided comparable results to the study of the chromate industry; however, a direct analysis of the original data led to extremely high doubling doses of risk that would never be reached in practice.

Due to the high variability in the derived doubling of risk doses, workplace limit values derived from published data should not be used as a strict cut-off criterion but should take the workers’ individual occupational exposure and life circumstances into account. The analysis of the SYNERGY data demonstrated that deriving assessment benchmarks from aggregated data can lead to false conclusions. Reliable modelling of dose-effect relationships is therefore only possible by evaluating original data.