Identification and external validation of a novel miRNA signature for lymph node metastasis prediction in submucosal-invasive gastric cancer patients.

Endoscopic resection (ER) has been increasingly performed in the treatment of early gastric cancer (GC). However, lymph node metastasis (LNM) can cause treatment failure with ER, especially in T1b patients. Here, we attempted to develop a miRNA-based classifier to detect LNM in T1b patients. Based on high-throughput data from The Cancer Genome Atlas, we identified 20 miRNAs whose expression significantly changed in T1-2 GC with LNM vs T1-2 GC without LNM. We then developed a miRNA signature to predict LNM of T1b GC using the LASSO model and backward step wise elimination approach in a training cohort. Furthermore, the predictive accuracy of this classifier was validated in both an internal testing group of 63 patients and an external independent group of 114 patients. This systematic and comprehensive in silico study identified a 7-miRNA signature with an area under the receiver operating characteristic curve (AUROC) value of 0.843 in T1-2 GC and 0.911 in T1 EGC. The backward elimination was further used to develop a 4-miRNA (miR-153-3p, miR-708, miR-940 and miR-375) risk-stratification model in the training cohort with an AUROC value of 0.898 in cohort 2. When pathologic results were used as a reference, the risk model yielded AUROC values of 0.829 and 0.792 in two cohorts of endoscopic biopsy specimens. This novel miRNA-LNM classifier works better than the currently used pathologic criteria of ER in T1b EGC. This classifier could individualize the management of T1b patients and facilitate treatment decisions.