Reanalysis of the Mayo Lung Project data: the impact of confounding and effect modification

Abstract

Objectives: To examine whether age at entry, history of cigarette smoking, exposure to non-tobacco lung carcinogens, or previous pulmonary illnesses were confounders or effect modifiers of the relation between screening and lung cancer mortality in the Mayo Lung Project.

Setting: The Mayo Lung Project was a randomised, controlled, clinical trial conducted between 1971 and 1986 in 9211 male smokers over the age of 45 in Minnesota (USA). The group screened received chest x ray examination and sputum cytology every four months for six years. The unscreened group were recommended to obtain usual care (annual chest x ray examination and sputum cytology). After follow up, lung cancer mortality was similar in both groups.

Methods: Proportional hazard models were used to analyse data. A variable was considered a confounder if its inclusion in a model changed the rate ratio for screening by more than 15%; a variable was considered an effect modifier if its stratum-specific rate ratio for screening differed by a factor of two.

Results: None of the four aforementioned variables changed the rate ratio associated with screening (1.07) by more than 2%. The effect of screening may have differed by years smoked (rate ratio for smoking fewer than 30 years 2.4; rate ratio for smoking 30 or more years 1.0), though we suspect that this result occurred by chance.

Conclusion: Adjustment for or stratification by four established lung cancer risk factors did not alter the original findings of the Mayo Lung Project.