BAYESIAN BIOMARKER MODELS FOR CANCER GENETICS STUDIES

描述(取自调查员？文摘)

DESCRIPTION (Taken from the Investigator?s Abstract) This proposal aims to develop statistical methods which incorporate biomarkers and genetic information to improve estimates of cancer risk. This methodology will be developed using data from an ongoing study of lung cancer, smoking history, asbestos exposure, DNA adducts (a biomarker of smoking exposure), and genes associated with adduct formation and DNA repair (including CYP1A1 MSP1, GSTM1 and XRCC1). Currently the study has 2800 subjects, approximately half of whom are cases. Genotyping for multiple genes is ongoing. Phenol-related adducts have been measured in nontumorous lung tissue for 69 cases and in mononuclear blood cells for 38 cases and 42 controls. Polycyclic aromatic hydrocarbon adducts have been measured in normal lung tissue for 143 cases. The Bayesian paradigm, which can incorporate information from prior studies, will be used throughout. This paradigm can easily handle missing data such as lung adduct counts which cannot be measured in controls. All models for lung cancer risk will allow for synergistic effects between smoking and asbestos exposure. The specific aims of the proposal are: (1) to develop a model for DNA adduct counts given genes and smoking, and to then use the predicted adduct counts in a logistic regression model for lung cancer risk which corrects for the measurement error bias induced by using blood adducts instead of lung adducts in controls; (2) to develop a pharmacokinetic model for the number of blood adducts over time as a function of genes and changing smoking behavior, and to use the predicted cumulative adduct burden in a model for cancer risk; (3) by combining aims 1 and 2, to develop a model of cancer risk which both takes into account the changing adduct load over time and corrects for measurement error bias; (4) to extend the models in aims 1-3 to allow for nonlinear relationships between covariates and lung cancer risk using generalized additive models; (5) using new data, to compare the performance of the models developed in aims 1-4 to model which do not use biomarkers, and to assess the sensitivity of the outputs to both model and prior specification; and, (6) to develop guidelines on optimal sample sizes, sample selection and timing schemes for biomarker measurement, for future biomarker and exposure-mediated cancer studies.