PO.CL01.18 · 临床研究
OncoSeek 2.0:一种增强早期检测并辅助癌症诊断的先进多癌种血液检测
OncoSeek 2.0: An advanced multi-cancer blood test enhancing early detection and aiding cancer diagnosis
作者与单位 Authors & Affiliations
摘要 Abstract
中文摘要
背景:早期癌症检测和及时诊断对改善患者预后和降低医疗成本至关重要。OncoSeek是一种AI驱动、低成本的多癌种早期检测(MCED)检测方法,整合了七种蛋白肿瘤标志物(PTMs)与临床数据,此前在超过15,000名受试者中显示出58.4%的敏感性、92.0%的特异性和70.6%的组织起源准确性。在此,我们介绍OncoSeek 2.0,这是一个升级版本,纳入了三种额外的PTMs以增强对前列腺癌、肺癌和鳞状细胞癌的检测,并评估其在验证队列中的性能及其在组织肿块患者中辅助癌症诊断的扩展应用。
方法:OncoSeek 2.0保留了原有的机器学习框架,并扩展了生物标志物输入。在一个包括722例癌症和355例非癌症受试者的回顾性队列中评估MCED性能。为进一步评估临床效用,分析了一个由732例组织肿块患者组成的前瞻性队列,这些患者经临床判断很可能为恶性并计划接受手术,以评估该检测在辅助癌症诊断中的有效性。
结果:与OncoSeek 1.0相比,OncoSeek 2.0在15种预先设定的癌症类型(合计占全球癌症死亡的76.5%)中实现了更高的AUC(0.934 vs. 0.888),并在90.1%特异性下将敏感性从70.4%提高到83.5%。敏感性提升在肺癌(79.9%→89.6%)、前列腺癌(58.8%→94.1%)、宫颈癌(44.4%→72.2%)和食管癌(41.6%→64.4%)中尤为显著。在前瞻性组织肿块队列中,OncoSeek 2.0正确识别了682例确诊癌症中的573例(84.0%敏感性)。在相同的15种癌症类型中,12种的敏感性超过80%,卵巢癌(76.9%)、乳腺癌(72.3%)和淋巴瘤(63.6%)的值略低。在其余50例经病理确诊为良性病变的患者中,28例(56.0%)被正确分类为非癌症,可能使其免于不必要的手术操作。
结论:OncoSeek 2.0在保持高特异性和经济性(每次检测试剂成本约30美元)的同时显著提高了敏感性。升级后的检测强化了其在多癌种早期检测中的效用,并在临床疑似癌症病例中展示了经验证的诊断价值。其在组织肿块队列中正确识别84%确诊癌症病例的能力,展示了有意义的诊断支持,有助于优先处理需要确诊性操作的患者。这些发现使OncoSeek 2.0成为一种切实可行且可扩展的解决方案,既适用于人群层面的癌症早期检测,也适用于真实世界的诊断支持,尤其适用于那些没有USPSTF推荐筛查手段或既定无创诊断路径、通常需要手术操作以明确诊断的癌症。
查看英文原文 English abstract
Background: Early cancer detection and timely diagnosis are critical for improving patient outcomes and reducing healthcare costs. OncoSeek is an AI-driven, low-cost multi-cancer early detection (MCED) assay integrating seven protein tumor markers (PTMs) with clinical data, previously demonstrating 58.4% sensitivity, 92.0% specificity, and 70.6% tissue-of-origin accuracy in over 15,000 participants. Here, we present OncoSeek 2.0, an upgraded version that incorporates three additional PTMs to enhance the detection of prostate, lung, and squamous cell carcinomas, and evaluate its performance in validation cohorts and its expanded use for assisting cancer diagnosis in patients with tissue masses.
Methods: OncoSeek 2.0 retained the original machine learning framework with expanded biomarker inputs. MCED performance was assessed in a retrospective cohort including 722 cancer and 355 non-cancer subjects. To further evaluate clinical utility, a prospective cohort of 732 patients with tissue masses, clinically deemed likely malignant and scheduled for surgery, was analyzed to assess the assay's effectiveness in aiding cancer diagnosis.
Results: Compared with OncoSeek 1.0, OncoSeek 2.0 achieved a higher AUC (0.934 vs. 0.888) and improved sensitivity from 70.4% to 83.5% at 90.1% specificity across 15 prespecified cancer types collectively accounting for 76.5% of global cancer mortality. Sensitivity gains were notable in lung (79.9% → 89.6%), prostate (58.8% → 94.1%), cervical (44.4% → 72.2%), and esophageal (41.6% → 64.4%) cancers. In the prospective tissue-mass cohort, OncoSeek 2.0 correctly identified 573 of 682 confirmed cancers (84.0% sensitivity). Across the same 15 cancer types, sensitivities exceeded 80% in 12, with slightly lower values observed in ovary (76.9%), breast (72.3%), and lymphoma (63.6%). Among the remaining 50 patients with pathologically confirmed benign lesions, 28 (56.0%) were correctly classified as non-cancer, potentially sparing them from unnecessary surgical procedures.
Conclusion: OncoSeek 2.0 significantly improves sensitivity while maintaining high specificity and affordability (~$30 reagent cost per test). The upgraded assay strengthens its utility for multi-cancer early detection and demonstrates validated diagnostic value in clinically suspected cancer cases. Its ability to correctly identify 84% of confirmed cancer cases in a tissue-mass cohort demonstrates meaningful diagnostic support to help prioritize patients needing confirmatory procedures. These findings position OncoSeek 2.0 as a practical and scalable solution for both population-level cancer early detection and real-world diagnostic support, particularly for cancers without USPSTF-recommended screening modalities or established non-invasive diagnostic pathways, where surgical procedures are commonly needed for definitive diagnosis.
利益披露 Disclosure
M. Mao,
SeekIn Employment, Stock Option.
Y. Shen, None.
S. Li,
SeekIn Employment, Stock Option.
W. Wu,
Shenzhen Employment, Stock Option.
Y. Chang,
Shenyou Bio Employment.
P. Xing,
Shenyou Bio Employment.
C. Ding,
Shenyou Bio Employment.
D. Zhu,
Shenyou Bio Employment.
Q. Xu, None..
W. Cui, None.