Cancer mortality and incidence in an aerospace worker cohort: The extended Rocketdyne Study

We examined the occurrence of cancer deaths between 1950 and 2001 in a cohort of aerospace workers exposed to hydrazine rocket fuels and in addition, we investigated cancer incidence for the period of 1988--2000. Effect estimates were derived from Cox models using time-dependent exposure measures for hydrazine, trichloroethylene (TCE), polycyclic aromatic hydrocarbons (PAHs), mineral oils, and benzene. We found that high exposures to hydrazine rocket fuel increased colon and rectum cancer incidence, lung cancer mortality and possibly pancreatic cancer mortality; high TCE exposure increased bladder cancer incidence and possibly brain cancer mortality and kidney cancer incidence; high exposure to mineral oils increased esophagus and stomach cancer mortality, lung cancer mortality, non-Hodgkin's lymphoma (NHL) and leukemia incidence, and incidence of skin melanoma. Exposure lags of 10 and 20 years changed effect estimates only minimally and sensitivity analyses conducted for cancer incidence did not suggest major biases due to missing cases.; We examined job category effects on cancer mortality via several statistical models including the conventional Cox model and two hierarchical modeling approaches: a hierarchical multinomial logit model and a hierarchical discrete time competing risk model. The latter two models resulted in more stable estimates than the conventional one-stage Cox model as expected, but the results were quite sensitive to the specification of the second-stage variance. The hierarchical discrete time competing risk model that incorporates both the time to event and time-dependent exposure information seems to result in more reasonable parameter estimates and thus its use is recommended for the analysis of cohort data with multiple outcomes and multiple exposures.