Prediction of lymph node metastasis by analysis of gene expression profiles in primary lung adenocarcinomas

Purpose: Lymph node status is a strong predictor of outcome for lung cancer patients. Recently, several reports have hinted that gene expression profiles of primary tumor may be able to predict node status. The goals of this study were to determine if microarray data could be used to accurately classify patients with regard to pathologic lymph node status, and to determine if this analysis could identify patients at risk for occult disease and worse survival. Experimental Design: Two previously published lung adenocarcinoma microarray data sets were reanalyzed. Patients were separated into two groups based on pathologic lymph node positive (pN+) or negative (pNO) status, and prediction analysis of microarray (PAM) was used for training and validation to classify nodal status. Overall survival analysis was performed based on PAM classifications. Results: In the training phase, a 318-gene set gave classification accuracy of 88.4% when compared with pathology. Survival was significantly worse in PAM-positive compared with PAM-negative patients overall (P < 0.0001) and also when confined to pNO patients only (P = 0.0037). In the validation set, classification accuracy was again 94.1% in the pN+ patients but only 21.2% in the pNO patients. However, among the pNO patients, recurrence rates and overall survival were significantly worse in the PAM-positive compared with PAM-negative patients (P = 0.0258 and 0.0507). Conclusions: Analysis of gene expression profiles from primary tumor may predict lymph node status but frequently misclassifies pNO patients as node positive. Recurrence rates and overall survival are worse in these "misclassified" patients, implying that they may in fact have occult disease spread.