A Predictive Model for Lymph Node Involvement with Malignancy on PET/CT in Non–Small-Cell Lung Cancer

Abstract

Accurate assessment of lymph node (LN) involvement with malignancy is critical to staging and management of non-small-cell lung cancer. The goal of this retrospective study was to determine the tumor and imaging characteristics independently associated with malignant involvement of LNs visualized on positron emission tomography/computed tomography (PET/CT). From 2002 to 2011, 172 patients with newly diagnosed non-small-cell lung cancer underwent PET/CT within 31 days before LN biopsy. Among these patients, 504 anatomically defined, pathology-confirmed LNs were visualized on PET/CT. Logistic regression analysis was used to determine the associations between nodal involvement with malignancy and several clinical and imaging variables, including tumor histology, tumor grade, LN risk category in relation to the primary tumor location, pathologic findings from additional biopsied LNs, interval between PET/CT and biopsy, primary tumor largest dimension, primary tumor standardized uptake value (SUVmax), LN short-axis dimension, and LN SUVmax. On univariate analysis, adenocarcinoma histology (p = 0.010), high LN risk category (p < 0.001), larger LN short-axis dimension (p < 0.001), and higher LN SUVmax (p < 0.001) all correlated with nodal involvement. On multivariate analysis, adenocarcinoma histology (p = 0.003), high LN risk category (p = 0.005), and higher LN SUVmax (p < 0.001) correlated with nodal involvement, whereas LN short-axis dimension was no longer statistically significant (p = 0.180). A nomogram developed for clinical application based on this analysis had excellent concordance between predicted and observed results (concordance index, 0.95). Adenocarcinoma histology, higher LN SUVmax, and higher LN risk category independently correlate with nodal involvement with malignancy and may be used in a model to accurately predict the risk of a node's involvement with malignancy.