Leveraging cfDNA fragmentomic features in a stacked ensemble model for early detection of esophageal squamous cell carcinoma.

• DOI: 10.1200/jco.2024.42.16_suppl.4054
• Publication Date: 2024-06-17T14:18:51Z
• Authors: Zichen Jiao; Xiaoqiang Zhang; Yulong Xuan; Xiaoming Shi; Zirui Zhang; Ao Yu; Ningyou Li; Shanshan Yang; Xiaofeng He; Gefei Zhao; Ruowei Yang; Jianqun Chen; Xuxiaochen Wu; Xiaoxi Chen; Hua Bao; Fufeng Wang; Wei Ren; Hongwei Liang; Qihan Chen; Tao Wang
• Keywords: 03 medical and health sciences; 0302 clinical medicine

4054 Background: In this study, we developed a stacked ensemble model that leverages cell-free DNA (cfDNA) fragmentation for the early detection of esophageal squamous cell carcinoma (ESCC). The combined four fragmentomics features obtained from whole genome sequencing (WGS) and employed machine learning algorithms. We evaluated model’s generalizability in an independent validation cohort external collected at different center. Additionally, robustness repeatability were assessed across low coverage repeated measured samples. results underscore promising potential our as effective strategy diagnosis management ESCC clinical settings. Methods: 256 healthy individuals 243 patients diagnosed with cancer enrolled including 47 participants 44 cohort. Plasma samples all profiled by whole-genome (WGS). Fragmentomic encompassed copy number variation (CNV), size (FSC), distribution (FSD), nucleosome positioning (NP) incorporated alongside models to develop optimized classification model. generated score ranging 0 1 each noncancer sample, closer indicating higher probability cancer. performance was using cohort, reproducibility examined. Results: A integrating cfDNA This integrated exhibiting remarkable sensitivity 91.8% (89/97) specificity 98.1% (102/104), AUC 0.986 cohorts cut-off 0.69 selected level 98% (105) training demonstrated 86.4% (38/44) 95.7% (45/47) same cut-off. model's remained consistent even depths 0.5× (AUC 0.978), providing valuable insights into resilience stability methodology ascertained its practical applicability scenarios limited resources or constraints. Furthermore, identifying pathological features, 94.1% (16/17) stage I 91.4% (53/58) tumors smaller than 3 cm. Conclusions: implemented utilizing exceptional carcinoma. anticipated impact is enhancement strategies setting.