249P A retrospective cohort study comparing single-incision and multi-incision video-assisted thoracoscopic surgery in early-stage non-small cell lung cancer: Long term follow-up with machine learning–based analysis using XGBoost with shapley additiv

inadequate tissue from small biopsy in non-surgical cases (79 cases).A pan-cancer CGP was performed using the TruSightOncology500HT assay on the NovaSeq6000 platform (Illumina).Statistical associations were computed in R using Fisher's exact test to determine global significance across all classes and for individual subtype comparisons against the remaining cohort.All statistical tests with p-values <0.05 were considered statistically significant.Results: Excluding from TP53 (commonly mutated across IASLC-subtypes), differences in gene mutational pattern distribution were observed: KRAS-mut in 53% of mucinous tumors, EGFR-mut in 30% of lepidic, BRAF in 30% of micropapillary pattern, STK11 in 30% of mucinous tumors and RBM10 in lepidic tumors.This different distribution was statistically significant for the EGFR (p=0.0178) and for BRAF (p=0.0275).For the early-stages EGFR was confirmed to be associated with the acinar pattern and STK11 with papillary pattern.No significant enrichments were identified for advanced stages.Conclusions: Distinct gene mutation patterns are associated with specific IASLC adenocarcinoma subtypes, even in not-advanced stage.These significant associations support the evidence of different prognosis and response to treatment across subtypes and the utility of an integrated morpho-molecular characterization.