Development and Validation of a Prediction Model for Identifying Small Malignant Pulmonary Nodules
Principal Investigator
Name
Gerard Silvestri
Degrees
MD
Institution
Medical University of South Carolina
Position Title
Professor of Medicine
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
200910-0040
Initial CDAS Request Approval
Oct 19, 2009
Title
Development and Validation of a Prediction Model for Identifying Small Malignant Pulmonary Nodules
Summary
Lung cancer is the most common cause of cancer death worldwide.1 Today, computed tomography (CT) of the chest is routinely undertaken and frequently identifies nodules that might be lung cancer. In some studies as many as 25% of individuals have CT-detected pulmonary nodules. Only a small fraction of these are eventually confirmed to be lung cancers (<2%). It takes costly diagnostic testing and time-consuming follow-up to establish whether a CT-detected nodule is lung cancer or not. Considerable psychological distress may be experienced by the individual and family waiting for a definitive diagnosis. Thus, a prediction model that can provide accurate probablilities of a CT-detected pulmonary nodule
being lung cancer would be of great value to patients, clinicians and researchers.
The proposed study will prepare logistic regression prediction models to estimate the probability that a CT-detected pulmonary nodule (>4mm to <15 mm) is lung cancer. Potential predcitors will include sociodemographic, exposure, medical history and imaging variables. Model building will attempt to maximize goodness-of-fit, calibration (do predicted probabilities correspond to observed probabilities) and discrimination (classification accuracy). Internal validation of the model will be carried out using bootstrap resampling methods to correct estimates of predictive performance for overfitting or optimism. The study sample used for modeling will include all available National Lung Screening Trial (NLST) participants in the CT-screening arm who had one or more pulmonary nodules on a CT screening. External validation will use Early Detection of Lung Cancer - a Pan-Canadian Study data.
being lung cancer would be of great value to patients, clinicians and researchers.
The proposed study will prepare logistic regression prediction models to estimate the probability that a CT-detected pulmonary nodule (>4mm to <15 mm) is lung cancer. Potential predcitors will include sociodemographic, exposure, medical history and imaging variables. Model building will attempt to maximize goodness-of-fit, calibration (do predicted probabilities correspond to observed probabilities) and discrimination (classification accuracy). Internal validation of the model will be carried out using bootstrap resampling methods to correct estimates of predictive performance for overfitting or optimism. The study sample used for modeling will include all available National Lung Screening Trial (NLST) participants in the CT-screening arm who had one or more pulmonary nodules on a CT screening. External validation will use Early Detection of Lung Cancer - a Pan-Canadian Study data.
Aims
(1) To develop a prediction model that will discriminate true lung cancers from false positive CT-detected nodules (>4-<15 mm) detected on chest screening, using sociodemographic, exposure, medical history and imaging data.
(2) To internally validate the predictive performance of the model, including discrimination and calibration, using bootstrap methods.
(3) To externally validate the predictive accuracy of the model using data from the Early Detection of Lung Cancer – A Pan-Canadian Study and the British Columbia Cancer Agency Screening Study (PI's Drs. Stephen Lam and Ming Tsao).