Christine Lambert

M.D, Ph.D

University of Minnesota

Pulmonary Critical Care Fellow

PLCO (Learn more about this study)

PLCO-563

Jan 30, 2020

Use of Tobacco Biomarkers in Lung Cancer Risk Assessment

Lung cancer screening with low dose computed tomography has been shown to reduce mortality from lung cancer, the leading cause of cancer deaths in the United States. Screening guidelines are based on age and pack-year history. Pack-year history is difficult to estimate accurately and has limitations as a robust measure of tobacco exposure. For current smokers, a risk model that incorporates tobacco specific biomarkers may provide a more precise representation of smoking history, helping to ensure lung cancer screening is performed in those at greatest risk of lung cancer. Risk models, such as PLCOm2012 which was developed from data collected through the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial, use historical measures of tobacco exposure. We propose to test a lung cancer risk model that adds biological markers of tobacco exposure.

Current lung cancer risk prediction models include individual demographic and smoking history variables. The tobacco-specific biomarker 4-(methyl-nitrosamino)-1-(3-pyridyl)-1-butanone (NNK) is an International Agency for Research on Cancer (IARC) Group 1 carcinogen, and NNAL is a biomarker of NNK uptake. A prospective relationship between NNAL and lung cancer risk has been shown using PLCO biospecimens. Use of total NNAL, the biomarkers r-1,t-2,3,c-4-tetrahydroxy-1,2,3,4-tetrahydrophenanthrene (PheT), and cotinine can characterize tobacco carcinogen uptake and metabolism, aspects of smoking exposure that are missing from prior lung cancer risk models.

Approach
Masonic Cancer Center collaborators will provide biomarker data from 100 lung cancer cases and 100 controls from PLCO that were previously analyzed. We are requesting the additional variables from PLCO needed to calculate the PLCOm2012 risk score. Logistic regression will be used to model diagnosis of lung cancer within 6 years (the primary outcome) for all observations. Covariates included in the model will be total NNAL, PheT, cotinine and a lung cancer risk score (i.e., the predicted log odds) from the PLCOm2012 model. Due to the expected skewed distributions of the three biomarkers, log-transformations will be used. Two models will be investigated where 1) all covariates are modelled linearly, and 2) risk score is modeled linearly and all three log-transformed biomarkers are modelled nonlinearly using restricted cubic splines. The final model will be selected taking into consideration performance metrics. Cross validation will be used to internally validate the model given use of the same data for model development and validation. Performance metrics including ROC curves, AUC, and sensitivity at a set specificity of 63%, which is the specificity achieved in previous models, will be obtained. These performance metrics will be used to compare against the PLCOm2012 model. Statistical analyses will be performed using R v 3.6.0.