The use of regression analyses in a cohort mortality study of welders.
Steenland-K; Beaumont-J; Hornung-R
J Chronic Dis 1986 Apr; 39(4):287-294
Data from a cohort mortality study of 3247 welders and 5432 nonwelders was reviewed using an internal nonexposed comparison group, and two types of regression analysis. The members of the cohort worked in Western Washington from 1950 to 1976. Nonwelders, but members of the same union, served as the referent group. Relative risk was evaluated using both the Cox regression and logistic regression methods. Practically no difference was obtained in results using either age or length of followup as the time variable in Cox regression. However, when age was used, a more sophisticated program was needed to exclude individuals from the risk set who had not yet entered the cohort at the age being considered. While the results achieved in this study using the Cox regression and logistic regression were similar, the authors caution that this is unusual and probably has occurred because in this data set mortality from lung cancer was rare, the effect of welding on lung cancer risk was slight, and the effect of welding on lung cancer changed only a little over the time of the study. Usually, logistic regression would not be an appropriate method to analyze data from occupational cohort studies. Cox regression has advantages over the more traditional standardized mortality ratio in that the exposure variable and potential confounders can be treated as continuous variables and the functional form of the dose response curve can be studied using regression methods.
NIOSH-Author; Mortality-surveys; Mortality-rates; Lung-disorders; Pulmonary-system-disorders; Cancer-rates; Risk-analysis; Risk-factors; Welders-lung; Welding-industry
Journal of Chronic Diseases