A 9-gene signature in circulating tumor DNA based on molecular residual disease detection to predict recurrence of resectable non-small-cell lung cancer

e20579 Background: Molecular residual disease (MRD) status is associated with the recurrence of non-small cell lung cancer (NSCLC) after complete resection. In clinical practice, MRD-positive patients are still uncertain of the specific recurrence time, and may even not relapse. Besides, it remains unclear whether the mutational signatures of circulating tumor DNA (ctDNA) based on MRD detection (ctDNA-MRD) may help predict recurrence. Therefore, we performed this analysis to explore whether the ctDNA-MRD mutational signatures could help predict recurrence, especially in MRD-positive population. Methods: Two investigators independently searched public databases for articles related with MRD in lung cancer published up to July 31, 2022. After screening, raw data from two published studies on resectable NSCLC were analyzed. Gene clusters were established based on the clearance of ctDNA after surgery. Kaplan-Meier and multivariate analyses were performed to find genes (Monitor genes) and clinical variables associated with recurrence-free survival (RFS). Lasso method was used to find recurrence-related genes based on perioperative ctDNA (preoperative and longitudinal ctDNA). Monitor genes and recurrence-related genes were defined as 9-gene signature. The mutational status of Monitor genes and 9-gene signature in perioperative ctDNA, as well as clinical variables, were used to establish prognostic models. The power of the models in distinguishing recurrence was shown as the area under the curve (AUC). Results: Data from 168 patients with at least one somatic mutation detected in tumors were included in the analysis. DNMT3A, EGFR, TP53, KEAP1, NAV3 (Monitor genes), and clinical variables (stage, adjuvant therapy) were factors associated with RFS. The AUC of the model based on Monitor gene mutations in perioperative ctDNA was 0.774, and the AUC reached 0.807 when clinical variables were included. The AUC of the models based on clinical variables, 9-gene signature (Monitor genes, KRAS, NTRK3, RPL5, and SOX2) in perioperative ctDNA for predicting the recurrence in the overall population and MRD-positive population were 0.840 and 0.832, respectively. Conclusions: This analysis reveals that perioperative ctDNA mutational signatures are helpful to predict postoperative recurrence, so it is necessary to pay attention to the specific gene mutations in ctDNA-MRD. [Table: see text]