Construction of a diagnostic model through the gene expression profile of the patients’ peripheral blood mononuclear cells and clinical characteristics for malignant lung nodules diagnosis.

Abstract

e20527 Background: Low dose computed tomography (LDCT) is widely accepted for detecting early-stage lung cancer. However, it remains a great challenge for it to distinguish malignant lung tumors from benign lung nodules because of the high false positive rate. Therefore, a better non-invasive diagnostic tool is needed to improve the specificity. Methods: Two GEO datasets of expression profiling by array of peripheral blood mononuclear cells (PBMC) from patients with malignant lung nodules or non-malignant lung conditions, as well as relevant clinical information were downloaded and disposed. R package edgeR was applied to screen out the differentially expressed genes with |log2-fold change (FC)| ≥ 0.25 and p≤0.05. Logistic regression model was constructed based on GSE135304 dataset in order to find which factors contribute to the distinguishing of malignant lung nodules from non-malignant lung conditions, and this model was then validated in GSE108375 dataset. Results: 23 differentially expressed genes from the PBMC between patients with lung cancer and non-malignant lung conditions, and the patients’ lung nodule size and age were included in the diagnostic model through the logistic regression analysis. And in the training dataset GSE135304, the ROC curve shows the specificity of the model is 0.866 and sensitivity is 0.848, and the AUC value is 0.857. In the meantime, the specificity and sensitivity of the validation dataset GSE108375 are 0.865 and 0.893 respectively, and the AUC value comes to 0.879. Conclusions: This diagnostic model shows high sensitivity and specificity in both training group and validation group, so it may be a good tool for the mini-invasive diagnosis of malignant lung nodules.