Abstract 5096: Exploring novel prognostic genetic markers using elastic-net regularization and association rule mining in HER2-negative metastatic gastric cancer patients

Abstract

Abstract Trastuzumab is the standard of care and has been shown to prolong survival of HER2-positive gastric cancer (GC) patients. However, HER2-positive is present only in about 20% of GC patients, there is a need to explore prognostic genetic markers for 80% of HER2-negative GC patients. The aim of this study was to identify the prognostic markers by elastic-net regularization and association rule mining and evaluate prognosis with selected genetic markers in HER2-negative metastatic GC (mGC) patients. A total of 143 HER2-negative mGC patients who have been treated with first-line chemotherapy in Yonsei Cancer Center, Korea between 2011 and 2018 were retrospectively assessed. Patients were categorized into two groups (responder or non-responder) based on the median progression-free survival (PFS: 7.8 months (95% confidence interval (CI), 6.3 - 8.9)) of the first-line chemotherapy. Targeted sequencing data of patients was obtained from CancerSCANTM and CANCER MASTER, and were analyzed with common variants of two panels. We used combined tools of elastic-net with 1,000 times of re-sampling bootstrap method and association rule mining to filter out and select critical genetic markers distinguishing responder or non-responder groups. To assess prognosis significance of selected genetic markers, we developed prognostic score based on the sum of sub-scores. If the right-hand-side of the rules signifies a responder, the patient is given a value of 0 when applying to the rule and 1 when they do not. PFS were determined by the Kaplan-Meier method and compared using the log-rank test. The 85 candidate genetic markers were selected by elastic-net with 60% of variable inclusion probability. Among the 85 candidate genetic markers, 12 genetic markers involved in 26 rules with lift greater than 1.5 were related to the responder group from association rule mining. The selected 12 genetic markers are included as follows: EGFR/KRAS normal copy number, APOBEC3B deletion, FGFR2/MLH1/PHLPP2/PARP4/SH2B3/BARD1/ALK/GNAS wild type, and NCOA3 mutation. The prognostic scores generated by selected genetic markers were 3, 4 and 5 respectively, and patients with higher scores had significantly poor prognosis than those with lower scores. Median PFS of each score group were 9.8 months (95% CI, 8.8 - 11.6), 7.5 months (95% CI, 5.7 - 8.6), and 5.2 months (95% CI, 1.2 - 6.7) for patients with consecutive scores of 3, 4 and 5 (P=0.017). The combined tool of elastic-net and association rule mining effectively found prognostic genetic markers and can be applied to develop the prognostic scoring system. The results of novel combined tool identified 12 multiple prognostic genetic markers and demonstrated significant survival differences in prognostic scores among HER2-negative mGC patients, although further studies are required. Note: This abstract was not presented at the meeting. Citation Format: Sejung Park, Woo Sun Kwon, Inhye Jeong, Hyun Jeong Kim, Jong Soo Hong, Jaeyoung Kim, Jingmin Che, Chung Mo Nam, Hyun Cheol Chung, Sun Young Rha. Exploring novel prognostic genetic markers using elastic-net regularization and association rule mining in HER2-negative metastatic gastric cancer patients [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 5096.