Development and clinical validation of a cell-free DNA methylation sequencing test for multi-cancer early detection.
Year: 2025
Type: conference-abstract
Abstract: 10537 Background: Achieving robust early-stage sensitivity in multi-cancer early detection (MCED) poses challenges, relying on large cohorts of early-stage samples and reliable prediction frameworks. Particularly for gastrointestinal cancers (GICs) with poor compliance of screening, early-stage sensitivity in MCED remains insufficient. Moreover, accurate tumor localization is crucial for choosing subsequent diagnostic procedures but remains suboptimal. We evaluate the performance of Genie-seq within the ProFuture study (NCT05874648), focusing on its capacity to detect five high-mortality cancers: lung, colorectal, liver, stomach, and esophageal cancers. Methods: The ProFuture study is a prospective multicenter case-control study that initially enrolled 3,515 participants. Following evaluations and a minimum of a half-year follow-up, 3,036 participants remained analyzable. Participants were divided into training (920 cancer; 629 non-cancer), validation (300 cancer; 215 non-cancer), and independent validation (605 cancer; 367 non-cancer) sets. Plasma cfDNA underwent a 1000X target enzymatic methyl sequencing assay (Genie-seq) targeting cancer-specific methylation patterns identified from 2,420 tumor and plasma samples. The assay normalizes abnormal fragment reads within blocks to minimize interference, using a maximization model to select sensitive and robust features. A gradient-boosted tree model was developed to integrate these features for cancer prediction, utilizing a one-vs-rest strategy to determine the tissue-of-origin (TOO). Results: Specificity remained consistently high across all phases: 99.0% (95% CI: 97.7-99.6%) in training, 99.1% (96.7-99.9%) in validation, and 99.2% (97.6-99.8%) in independent validation. Sensitivity was 68.6% (65.5-71.6%) in training, 71.0% (65.5-76.1%) in validation and 69.6% (65.8-73.2%) in independent validation. In independent validation set, stages I-III (account for 85.5% of cases) sensitivity reached 65.8% (61.5-69.9%) for all cancer types and 72.1% (66.9-76.9%) for three GICs. The TOO classifier assigned the origin in all screen-positive cases, achieving an accuracy of 87.4% (83.9-90.4%) in independent validation, including reducing misclassification from lung to esophageal cancer due to squamous similarity to 4.5%. Conclusions: This MCED test accurately identified signals from five tumor types especially in the early stages. Precise TOO localization minimizes the healthcare burden of subsequent diagnoses. The consistency of performance from training to clinical validation underscores the robustness of feature selection strategy, mitigating the risk of overfitting. Notably, this study demonstrated exceptional sensitive detection of early-stage GICs, indicating the potential efficacy of Genie-seq in MCED within the ongoing interventional Prosight study (NCT06790355). Clinical trial information: NCT05874648 .
Source: Journal of Clinical Oncology
Authors: Xiaosheng He, Shaohua Ma, Hong Xu, Zhigang Li, Yiran Cai, Yin Li, Yaxu Wang, Jianhong Lian, Leping Li, Jun Zhang, Nan Lin, Songbing He, Xiaobo Wang, Xiao-Bing Li, Qun Zhao, Yang Wang, Xiurui Zhu, Xiaohui Wu, Zhihua Liu, Ping Lan
Institutions: Sun Yat-sen University, Sixth Affiliated Hospital of Sun Yat-sen University, Peking University Cancer Hospital, First Hospital of Jilin University, Shanghai Jiao Tong University, Shanghai Chest Hospital, Chinese Academy of Medical Sciences & Peking Union Medical College, Cancer Hospital of Chinese Academy of Medical Sciences, Second Affiliated Hospital of Chongqing Medical University, Shanxi Provincial Cancer Hospital, Shandong Provincial Hospital, Shandong First Medical University, The Affiliated Yongchuan Hospital of Chongqing Medical University, Third Affiliated Hospital of Sun Yat-sen University, First Affiliated Hospital of Soochow University, Guangxi Medical University, Tumor Hospital of Guangxi Medical University, Henan University of Science and Technology, Anyang Tumor Hospital, Fourth Hospital of Hebei Medical University
Language: English
Subfield: Cancer Research
Domain: Life Sciences
Open Access status: bronze
OpenAlex ID: W4410820843