Application of a classifier combining bronchial transcriptomics and chest CT features facilitates the diagnostic evaluation of lung cancer in smokers and non-smokers

Lung cancer screening by computed tomography (CT) reduces mortality but exhibited high false-positive rates. We established a diagnostic classifier combining chest CT features with bronchial transcriptomics. Patients with CT-detected suspected lung cancer were enrolled. The sample collected by bronchial brushing was used for RNA sequencing. The e1071 and pROC packages in R software was applied to build the model. Eventually, a total of 283 patients, including 183 with lung cancer and 100 with benign lesions, were included into final analysis. When incorporating transcriptomic data with radiological characteristics, the advanced model yielded 0.903 AUC with 81.1% NPV. Moreover, the classifier performed well regardless of lesion size, location, stage, histologic type, or smoking status. Pathway analysis showed enhanced epithelial differentiation, tumor metastasis, and impaired immunity were predominant in smokers with cancer, whereas tumorigenesis played a central role in non-smokers with cancer. Apoptosis and oxidative stress contributed critically in metastatic lung cancer; by contrast, immune dysfunction was pivotal in locally advanced lung cancer. Collectively, we devised a minimal-to-noninvasive, efficient diagnostic classifier for smokers and non-smokers with lung cancer, which provides evidence for different mechanisms of cancer development and metastasis associated with smoking. A negative classifier result will help the physician make conservative diagnostic decisions.