Predicting risk of cancer in 12 sites using the screened arm of the PLCO
colon/rectum, esophagus, head and neck, liver/bile-duct, lung, lymphoma, ovary, pancreas. plasma cell neoplasm and stomach. Potential predictors for the model include variables in the background questionnaire such as sex, age, smoking history, race, bmi, family history of cancer, use of HRT, and others.
Build a model for incidence of the 12 cancers listed above. Covariates will be entered as main effects of categorical variables. Model selection will be done using lasso with nested cross-validation to determine the regularization parameter.
Validate the model internally via cross-validation
The validated risk model will be used in a net benefit analysis of potential MCED screening trial designs.
Paul Pinsky
Jian-Lun Xu