Lung cancer is the number one cause of cancer death and Veterans are 25% to 76% more likely to develop this deadly disease. The main challenge in the field of lung cancer research is trying to prevent advanced lung cancers that kill patients and simultaneously minimize the potential harm caused by required invasive diagnostic techniques. Because lung cancer is so deadly, patients and providers must aggressively pursue a diagnosis to rule out cancer. The lung is not easily accessible and these biopsies often require an invasive and costly. Despite advanced imaging techniques and clinical judgment, up to 40% of the operations on patients with suspected lung cancer result in a benign diagnosis. The high rate of benign disease discovered by operative resection will continue until additional patient care. This career development award permits me to pursue research skills and investigator experience for 1) developing and validating evidence-based surgical algorithms for reducing unnecessary surgery, 2) improving patient safety by not missing cases of lung cancer, 3) implementing a safe and cost effective lung nodule clinical algorithm for patients with suspicious pulmonary nodules. Study One: To develop an evidence-based clinical algorithm for management of lung nodules referred for diagnostic surgical evaluation. We hypothesize that a new model predicting benign disease among patients presenting with suspicious pulmonary nodules will have a ROC area under the curve (AUC) of at least 0.85. Current models do not include all the epidemiological and imaging data used by surgeons to estimate the pre- surgical likelihood of cancer or benign disease and determine whether to operate on a suspicious nodule.
This aim will combine the VA-TVHS patient database, Vanderbilt Lung Nodule Cohort, and the University of Virginia database into a 950 patient Lung Nodule Cohort. A regression model will be developed from this cohort and will also include an exploratory analysis of new lung cancer biomarkers. Study Two: To evaluate the generalizability of the lung nodule clinical algorithm for management of lung nodules referred for diagnostic surgical evaluation. We will externally validate the prediction tool developed in Study One with existing datasets from the University of Alabama, Birmingham (UAB) and the completed American College of Surgeons (ACOSOG) Z4031 cooperative trial. These datasets will be combined to form a 1500 patient validation cohort. Biomarkers will also be assessed in the ACOSOG dataset from stored clinical samples. Study Three: To evaluate the predicted impact of the lung nodule clinical algorithm on patient outcomes in a multi-institutional prospective cohort. The prospective 686-patient cohort will be from VA-TVHS, VA- Birmingham and VUMC thoracic surgery clinics. This study will NOT implement the diagnostic algorithm in clinical practice but provide a safe harbor to accomplish two aims. First, we will prospectively evaluate the number of patients potentially benefiting from such algorithm by not missing cases of lung cancer and avoiding unnecessary operations. Second, we will use decision analysis to perform an incremental cost-effectiveness analysis of our algorithm in this cohort. We hypothesize that use of the prediction tool will reduce the benign diagnosis rate in surgically resected pulmonary nodules from 40% to at least 30%, the overall accuracy will be over 85% and it will be cost effective. Future studies will design a prospective multi-institutional VA pilot study to evaluate the algorithm for patients referred for surgical evaluation of pulmonary nodules.
Lung cancer is the leading cause of cancer death. Veterans are 25-76% more likely to develop lung cancer and have a higher than average death rate. The National Lung Screening Trial recently demonstrated a 20% reduction in lung cancer-specific mortality in patients screened with low dose CT scans. Abormal lung lesions were found in 27% patients but the majority were not cancer (benign). Because the lung is difficult to access, many patients with suspicious lung lesions require an invasive diagnostic surgical biopsy. Despite advanced imaging and expert clinical judgment, up to 40% of these operations result in a benign diagnosis. Accurate evidence-based non-invasive algorithms that distinguish benign disease from lung cancer do not exist for these patients and are needed. This proposal aims to use large existing databases and reduce unnecessary diagnostic operations by developing a highly accurate, cost-effective, and safe predictive model. The safety and cost- effectiveness of the model will then be tested in a prospective VA study prior to implementation.