Data-driven exploration of occupations and lung cancer risks using regularized regression models

Occupational epidemiological analyses typically focus on a priori hypotheses relating disease to specific exposures or occupations. We used regularized regression models to agnostically identify occupations associated with lung cancer while accounting for known lung carcinogens.

We pooled data from 14 European and Canadian lung cancer case-control studies. The main analysis included 13 587 cases and 16 410 controls for men and 3199 cases and 4408 controls for women. Associations between ever-employment in various occupations and lung cancer risks were evaluated using two selection approaches: a least absolute shrinkage and selection operator regression and a two-stage Bayesian hierarchical horseshoe model accounting for age, study, smoking, and occupational exposure to known lung carcinogens. Unconditional logistic regression models were used to estimate odds ratios (ORs) and 95% confidence intervals (CIs) for selected occupations.

For men, occupations consistently identified as positively associated with lung cancer included miners (OR = 1.28, 95% CI: 1.06-1.54) and bricklayers (OR = 1.30, 95% CI: 1.12-1.52), while secondary education teachers (OR = 0.37, 95% CI: 0.22-0.60), production managers (OR = 0.66, 95% CI: 0.54-0.81), and government executive officials (OR = 0.72, 95% CI: 0.62-0.83) were negatively associated. For women, secondary language and literature teachers (OR = 0.39, 95% CI: 0.20-0.72) and government executive officials (OR = 0.63, 95% CI: 0.48-0.81) showed negative associations.

Two agnostic, data-driven models identified several occupations associated with both increased and decreased lung cancer risks. These associations persisted after accounting for known exposures and covariates, highlighting the need to investigate unidentified occupational factors.