Advanced ensemble stacking model employing cfDNA fragmentation for early detection of esophageal and gastric cancer.
Year: 2025
Type: conference-abstract
Abstract: e16112 Background: Esophageal and gastric cancers are often detected at late stages due to asymptomatic early stages and limitations of current diagnostic tools, including endoscopy. This study aims to develop a non-invasive diagnostic model using three key cell-free DNA (cfDNA) fragmentomics features—copy number variation, fragment size profiles, and methylation patterns—integrated with advanced machine learning algorithms, enhance the accuracy of early detection and enable precise differentiation between the two cancers. Methods: This study recruited 275 healthy participants, 214 gastric cancer patients, and 61 esophageal cancer patients, dividing them into training and validation cohorts. Plasma samples underwent low-depth whole genome sequencing (WGS), and a stacked ensemble model was constructed using three cfDNA fragmentomics features: Copy Number Variation (CNV), Fragment Size Profile (FSP), and Fragment Methylation Analysis (FRAGMA). The model was trained using 5-fold cross-validation, with performance evaluated on the validation cohort. The model’s predictive power on sensitivity and specificity was assessed at a 95% specificity threshold in the training set. Results: The three features each demonstrated strong predictive accuracy, with AUC values exceeding 0.94 in the training cohort. The stacked ensemble model further improved performance, achieving an AUC of 0.972 in training and 0.947 in validation, even with the inclusion of 14 in situ cases. At 95% specificity, the model achieved sensitivities of 64.7% for stage I and 72.2% for stage II cancers in the validation cohort. Subgroup analyses confirmed the model’s robustness across differentiation statuses, stages, and demographic variables, with sensitivity reaching 100% for stage IV cancers. Additionally, the model demonstrated potential for distinguishing gastric from esophageal cancer, achieving an AUC of 0.834 in a Tissue-of-Origin analysis. Conclusions: The stacked ensemble model effectively detects esophageal and gastric cancers at early stages, offering a non-invasive, accurate alternative to endoscopy. This approach enhances the potential for early intervention, allowing for more precise and timely treatment, ultimately improving patient outcomes.
Source: Journal of Clinical Oncology
Authors: Shitong Cheng, Yusong Luo, Xiaolong Dong, Meng-yuan Liu, Zhaoqi Wu, Lu Xu, Jia Li, Chuan He, Ying Xiong, Linan Bao, Qingxin Xie, Ningyou Li, Haimeng Tang, Hua Bao, Ruowei Yang, Dongqin Zhu, Xiuxiu Xu, Xiaoxu Han, Mingfang Zhao, Hong Shang
Language: English
Subfield: Cancer Research
Domain: Life Sciences
Open Access status: bronze
OpenAlex ID: W4410810031