Occupations associated with lung cancer in Sao Paulo, Brazil. Application of hierarchical regression analysis

摘要

Lung cancer is an important public health problem, and a large proportion of known carcinogens occurs in occupational settings. There have been very few studies of exposure to occupational carcinogens in developing countries. This is case-control study carried out on 12 hospitals in the Metropolitan Region of Sao Paulo, Brazil, from July 1989 to June 1990, to determine the extent to which time employed in an occupation affects incidence of lung cancer. Information on sex, age, race, socio-economic variables, smoke history and job history was obtained for 430 lung cancer cases, and 880 hospital controls. Jobs with similar exposure to carcinogens were grouped into 29 occupations. I used a Bayesian multiple imputation method to assign values for missing information on job history, and spline regression to model effects of continuous covariates.;Occupational case-control studies are typically studies of multiple exposures, and the usual strategies to deal with multiple comparisons leads to an increased probability of making incorrect statements regarding null hypothesis. They often generate large and unstable estimates. An alternative is the use of a hierarchical model that can give estimates more plausible and stable than conventional estimates. Here I develop a two-stage hierarchical regression model. This model yields semi-Bayes incidence density ratio estimates for occupations by using a second-stage model to pull estimates toward each other when they have similar levels of carcinogen exposure. The first stage is a maximum-likelihood logistic regression on 29 occupations and covariates. The second-stage is a linear regression of occupational effects on 12 known carcinogenic substances or physical agents. The estimates obtained with one-stage maximum-likelihood logistic regression were very stable, having a small change when using the hierarchical regression model, on the contrary to that expected if their magnitude were just a multiple comparison artifact. I found an incidence density ratio of 1.65 (95% CI - 1.03 , 2.65) for working in agriculture after 1960, when the use of pesticides in large scale was introduced in Brazil, and an incidence density ratio of 2.18 (95% CI - 1.20 , 3.98) for working as a glass former or potter. Further studies on these occupations are recommended.