Impact of confounding by smoking on cancer risk estimates in cohort studies of radiation workers: a simulation study.

Abstract

Previous studies on cohorts of radiation workers have provided valuable insights into the effects of low-dose-rate radiation; however, some concerns regarding the potential confounding effects of smoking have been expressed. Although some studies have collected smoking data and adjusted for this variable, their limited numbers and the presence of other confounders obscure the extent of the impact of smoking on their results. To address this, we conducted a simulation study to quantitatively evaluate the bias from confounding and modeling conditions, similar to actual epidemiological studies. Our analysis, based on data from Japanese radiation workers, indicated that not adjusting for smoking can lead to an overestimation of radiation effects by approximately 110%. This overestimation was relatively insensitive to sample size and dose distribution parameters, but varied with radiation and smoking risk, baseline smoking probability, and heterogeneity in baseline risk. Considering the simplified settings of this simulation study and the uncertainty of the estimates of Japanese radiation workers, our simulation results were consistent with those of the real-world epidemiological study. We also compared the results using Cox and Poisson regression models, ensuring that their modeling approaches were as similar as possible, and found minimal differences between them.