Martin Tammemagi

Brock University / Henry Ford Health System

PLCO (Learn more about this study)

2008-0031

Sep 26, 2008

Lung Cancer Risk Prediction – A Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial Study

Lung cancer is the leading cause of cancer death in the developed world. Being able to identifying individuals a high risk for lung cancer is important for several reasons. Individuals identified to be at higher risk might be motivated to quit smoking if they are current smokers or might be targeted for more intensive smoking cessation programs. High risk individuals might benefit from lung cancer chemoprevention or screening programs. Lung cancer chemoprevention or screening trials efficiency may increase if they were able to enroll high risk individuals. To date several models have been developed to predict lung cancer risk: (1-6). However, current models have several limitations: (1) They incorporate only a limited number of potentially useful predictors. (2) Their overall predictive performance is generally low. (3) Some of these models were based on suboptimal study designs and the models and estimates may be biased. (4) Some studies have used suboptimal analytic approaches. The aim or the proposed study is to develop and validate a comprehensive and improved lung cancer prediction model. The study plans to use data from the screening arm of the PLCO trial to develop and validate a lung cancer risk prediction model using logistic and Cox proportional hazard regressions. Cumulative incidence of lung cancer will be estimated in the presence of competing causes of mortality. A random sample of half of the PLCO screenees will be used to develop the model and the remaining half will serve as a validation set. Calibration and discriminatory power of models will be evaluated by computing the ratio of expected number to observed number of lung cancer cases along with confidence intervals, and by the receiver operator characteristic area under the curve (ROC AUC) for logistic regression models and the equivalent concordance index (syn., c-statistic) for Cox models, respectively.