Objectives To estimate a threshold value tau for the respirable quartz dust concentration and silicosis incidence (1/1, ILO 1980/2000) in the German porcelain worker cohort (Birk et al 2009, 2010, Mundt et al 2011).

Methods 17,144 porcelain workers (128,688 person-yrs) were followed for silicosis incidence (40 cases). Respirable quartz dust exposure was determined by combining detailed individual employment histories with a job exposure matrix based on 8,000 historical industrial hygiene measurements. Cox regression with age as the time variable was used to evaluate silicosis morbidity by log (cumulative quartz dust exposure + 0.01 mg/m³-yrs) time-dependently, controlling for sex and smoking status. Curvature of the relationship was explored by restricted cubic splines (RCS, ≤ 7 knots) and fractional polynomials (FP, degrees ≤ 5). Estimation of tau was performed by subtracting a series of candidate values from the annual concentration data (setting the result to zero if negative) and recalculation of the time-dependent cumulative exposures. The partial likelihood profile was used to derive point and 95%-confidence interval (CI) estimates. Non-nested models were compared by information criteria (AIC). Lagging exposures by 10 years, using different offsets in the log-function (0, 0.1, 1 mg/m³-yrs) and varying sets of covariates were explored.

Results We estimated tau = 0.25 mg/m³ (0.95-CI: 0.15 mg/m³, 0.30 mg/m³). Applying this estimated concentration threshold led to lower degree optimal FPs and returned pronounced better fits (∆ AIC > 5) in log-linear Cox models, 5-knots RCS Cox models and 2-degree FP Cox models. The overall exposure-response could be appropriately described by a Cox model on log (unlagged cumulative exposure + 0.01 mg/m³-yrs) after applying tau = 0.25 mg/m³.

Conclusions A threshold Cox model fitted the data significantly better than a non-threshold model and summarised the cohort information without a loss in extracted information and much more simply than the curvilinear procedures (RCS, FP).