Cox proportional hazards and stratified Poisson regression are commonly used models for time-dependent data in epidemiologic studies. However, whether these methods consistently produce comparable results for the estimate of risk for both rare and prevalent outcomes is unclear.
Data from a previous study that utilized stratified Poisson regression to investigate relationships between selected causes of death and annual cumulative exposures to titanium dioxide (TiO2) were reanalysed using Cox proportional hazards modelling. The study cohort included 3,607 workers employed in three US manufacturing facilities, followed 1935–2006. Analyses were completed for cumulative doses in mg/m3-year with no lag and lagged 10 years, with all models specified similarly for covariates.
Overall, the Cox and Poisson models resulted in similar estimates in most dose categories for the selected causes of death, with no statistically significant indication of a positive association between TiO2 exposure and death from all cancers, lung cancers, non-malignant respiratory disease, or all heart disease. The Cox model routinely produced narrower 95% confidence intervals (CI), although overlapping with those from Poisson. Borderline disagreement results were associated with risk estimates lagged 10 years for heart disease at dose >80: 1.51 (CI: 1.00, 2.25) from Poisson and 1.356 (CI: 0.922, 1.995) from Cox; and for all cancers at dose 15-35: 1.35 (CI: 0.89, 2.04) from Poisson and 1.485 (CI: 1.005, 2.193) from Cox.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.