Prediction of true positive lung cancer in abnormal suspicious CXR lung screens
Principal Investigator
Name
Martin Tammemagi
Degrees
DVM; MSc; PhD
Institution
Brock University
Position Title
Professor of Epidemiology
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
201108-0010
Initial CDAS Request Approval
Aug 2, 2011
Title
Prediction of true positive lung cancer in abnormal suspicious CXR lung screens
Summary
Chest x-ray (CXR) imaging is very common in the older adult population, and finding abnormalities suspicious for lung cancer are common. But only a small minority of these is found to be true positive for lung cancer. We produced a multivariable logistic regression model predicting true positive abnormal suspicious (TP AS) for lung cancer CXR screens in the PLCO trial[1] (subsequently referred to as the Tammemagi PLCO model). Important predictors of TP AS screens included age, SES (education), pack-years smoked, smoking duration, body mass index, family history of lung cancer, and characteristics of the CXR abnormality, including lung nodule, lung mass, unilateral mediastinal or hilar lymphadenopathy, lung infiltrate, and upper/middle chest AS location. The model had a receiver operator characteristic area under the curve (ROC AUC) of 86.4%. Prediction models usually do not perform as well in external validation data as they do in the initial development dataset. So it is important to externally validate the PLCO model in an external dataset. Furthermore, the PLCO did not have available all possibly important prediction data which might be useful in such a prediction model. For example, absent were data on asbestos exposure, occupational exposure, pulmonary function, and nodule/mass size and margin characteristics. Using multivariable logistic regression modeling, the current study plans to externally validate and further refine a prediction model that will estimate the probability that an AS CXR truly represents lung cancer. Model predictive performance will be evaluated by assessing its discrimination using the ROC AUC and assessing calibration. The added benefit of new NLST data will be assessed by net reclassification improvement [2] (syn. Risk Stratification Table analysis [3]). New models will be internally validated by bootstrap correction of optimism or over-fit. Clinicians and patients will be able to make more informed clinical and personal decisions regarding a CXR image that is abnormal suspicious for lung cancer if they have an accurate probability estimate of lung cancer. It is expected that an accurate model will be produced which will have utility in health practice. (JE-5, ACRIN 665441)
Aims
Our first study aim is to externally validate the Tammemagi PLCO model predicting True Positive Abnormal Suspicious CXR lung screen using NLST data. Our second aim is to determine whether additional variables, in particular, asbestos exposure, occupational history, pulmonary function data, and enhanced radiological data (nodule/mass size and margin characteristics) can improve the prediction model, and to internally validate any new models that are developed, by preparing bootstrap corrected estimates of model’s predictive performance, including discrimination and calibration.