Association of 8-gene signature with early recurrence in resected non-small cell lung cancer.

Abstract

8031 Background: Limited evidence exists defining a genetic signature associated with early-recurrence in patients with non-small cell lung cancers (NSCLC). This study aims to identify a genomic panel associated with early recurrence, defined as recurrence within 12 months from surgery. Methods: Patients with resected pT1-2aN0 NSCLC that underwent tumor RNA sequencing at a single institution from 2010-2021 were included. Exclusion criteria included neoadjuvant therapy, unknown pN status, and no recurrence. Differential gene expression (DGE) analysis was performed using DESeq2 after count normalization, identifying 50 genes with a log2 fold change > 1 or < -1 and an adjusted p-value<0.05. Recursive feature elimination (RFE) with random forests was applied to evaluate subsets of genes, utilizing repeated 10-fold cross-validation to ensure robust performance estimation. Accuracy and model stability were compared across subsets to identify the gene panel most strongly associated with early recurrence. Cross-validation was performed using a secondary RFE random forest analysis optimized for AUC, to refine the gene selection process. A multivariable logistic regression analysis was conducted to examine the association between the gene panel and early recurrence, while controlling for potential confounders. Results: 118 patients met study criteria, of whom 54% (64/118) were female, 82% (97/118) had adenocarcinomas, and 59% (70/118) underwent lobectomy. The median tumor size was 1.8 cm (IQR 1.5–2.6). Early recurrence was observed in 25.4% (30/118). DGE analysis involved 61.4% (27,009/43,959) of genes. After filtering for low total counts, using an FDR <0.1, 2% (530/27,009) were upregulated and 1.3% (351/27,009) downregulated in patients with early recurrence. After RFE with random forests, an 8-gene panel was selected, including ART3 , SLC51A , SNAP25 , CCNA1 , GRK1 , GPR63 , CNTNAP2 , and TNFRSF11B achieving an accuracy of 71.7%. The panel had an area under the curve (AUC) of 79.1, sensitivity of 93.9%, and specificity of 33.3%. On multivariable logistic regression, after adjusting for tumor size, histology, surgical procedure, pack years, and number of nodes sampled, and performance status, patients with differential expression of four or more genes within the panel were associated with early recurrence (OR: 4.11, 95% CI:1.27–13.31, p=0.02). Conclusions: A unique tumoral 8-gene signature is associated with early recurrence in resected early-stage NSCLC patients. Univariable and multivariable logistic regression analysis examining predictors of early recurrence. Variable Unadjusted OR(95% CI) p-value Adjusted OR(95% CI) p-value ≥ 4 of the 8 Gene Panel 4.17 (1.46–11.88) 0.01 4.11 (1.27–13.31) 0.02 Tumor Size 1.39 (0.86–2.26) 0.18 2.22 (1.10–4.46) 0.03 Histology Adenocarcinoma Ref Ref Squamous Cell Carcinoma 1.22 (0.42–3.49) 0.72 0.58 (0.14–2.41) 0.46 Surgical Resection Anatomic Ref Ref Sub-anatomic 1.82 (0.78–4.29) 0.17 1.85 (0.48–7.09) 0.37 Pack-Years 1.00 (0.99–1.02) 0.61 1.01 (0.99–1.03) 0.22 Performance Status >0 1.21 (0.52–2.77) 0.66 0.91 (0.82–1.03) 0.13 Number of nodes sampled 0.91 (0.83-0.99) 0.04 1.27 (0.45–3.54) 0.65