Lung cancer risk among bakers, pastry cooks and confectionary makers: the SYNERGY study.

INTRODUCTION: Some studies have suggested increased lung cancer risks among bakers, however the results overall were inconsistent. The authors studied lung cancer risks among bakers and baking-related occupations in the SYNERGY pooled case-control database from 16 countries. METHODS: Occupation in a baking-related job was identified from the subjects' job histories. ORs adjusted for log(age), study centre, smoking behaviour and ever employment in a job with known exposure to occupational lung carcinogens were calculated by unconditional logistic regression. Findings were stratified by sex, histological subtype of lung cancer and smoking status. RESULTS: 19 366 cases (15 606 men) and 23 670 control subjects (18 528 men) were included. 473 cases (415 men, 58 women) and 501 controls (437 men, 64 women) had ever worked in baking or a related job. We did not observe an increased risk for men in baking (OR 1.01; 95% CI 0.86 to 1.18). No linear trends were observed for duration of employment. Some results suggested increased lung cancer risks for women, for example, for working as a baker for >30 years and in never-smokers, but after exclusion of one study these increased risks disappeared. DISCUSSION: The findings from this study do not suggest increased lung cancer risks in baking-related professions.

Integration of Multiple Genomic Data Sources in a Bayesian Cox Model for Variable Selection and Prediction.

Bayesian variable selection becomes more and more important in statistical analyses, in particular when performing variable selection in high dimensions. For survival time models and in the presence of genomic data, the state of the art is still quite unexploited. One of the more recent approaches suggests a Bayesian semiparametric proportional hazards model for right censored time-to-event data. We extend this model to directly include variable selection, based on a stochastic search procedure within a Markov chain Monte Carlo sampler for inference. This equips us with an intuitive and flexible approach and provides a way for integrating additional data sources and further extensions. We make use of the possibility of implementing parallel tempering to help improve the mixing of the Markov chains. In our examples, we use this Bayesian approach to integrate copy number variation data into a gene-expression-based survival prediction model. This is achieved by formulating an informed prior based on copy number variation. We perform a simulation study to investigate the model's behavior and prediction performance in different situations before applying it to a dataset of glioblastoma patients and evaluating the biological relevance of the findings.