Abstract 1266: Development and performance of a multi-cancer early detection test utilizing plasma cfDNA fragmentomics: A large-scale, prospective, multicenter study

Abstract

Abstract Background: The implementation of the multi-cancer early detection (MCED) test offers a valuable adjunct to existing screening methods, enabling more efficient detection of cancer and potentially leading to improved treatment outcomes and prognoses for patients. Here, we report on the performance of an MCED test, which utilizes plasma cfDNA and leverages genome-wide fragmentomics-based characteristics to identify cancer signals and predict the signal origin across a diverse range of cancer types. Methods: Plasma cfDNA from evaluable blood samples was analyzed using an MCED blood test called MERCURY, a robust machine learning classifier leveraging the low-coverage whole-genome sequencing and a comprehensive set of genome-wide features derived from cfDNA fragmentomics. A carefully selected cohort of 3076 cancer patients representing 13 cancer types, in addition to 3477 healthy controls, were pre-specified into the training and internal validation sets to train and internally validate the models to assess cancer and tissue of origin (TOO). The classifier was trained to a target specificity of 99% and locked before analysis of the independent validation set. The independent validation was enrolled prospectively and consists of 1465 participants (cancer: n= 732; non-cancer: n= 733). Results: The performance metrics in the internal validation set demonstrated that the sensitivity and specificity for cancer detection were 0.865 (95% CI [0.840, 0.887]) and 0.989 (95% CI [0.980, 0.994]), respectively. These impressive results were further substantiated in the independent validation set, where the overall sensitivity and specificity were found to be 0.874 (95% CI [0.848, 0.897]) and 0.978 (95% CI [0.965, 0.987]) respectively. Notably, the sensitivity showed an incremental increase with the stage of cancer (Stage I: 0.769, 95% CI [0.708, 0.821]; Stage II: 0.840, 95% CI [0.784, 0.886]; Stage III: 0.923, 95% CI [0.874, 0.954]; Stage IV: 0.971, 95% CI [0.901, 0.995]. Regarding the TOO model, a total of 10 cancer types with more than 100 patients in the model construction cohort were considered. The TOO model achieved a prediction accuracy of 83.5% (95% CI [80.7%, 86.6%]) and 91.8% (95% CI [89.6%, 94.1%]) for the top predicted origin and the top two predicted origins, respectively, amongst the true positive cases in the independent validation set. Conclusions: In this pre-specified, large-scale study, the MCED test, utilizing cfDNA fragmentomics, demonstrates its remarkable ability to assess cancer signal with an elevated level of sensitivity and specificity across 13 distinct types of cancer. Moreover, it displays noteworthy accuracy in predicting the tissue of origin. The performances are to be further validated in a prospective cohort study (NCT06011694). Citation Format: Hua Bao, Xiaoxi Chen, Min Wu, Shiting Tang, Xuxiaochen Wu, Wanxiangfu Tang, Dongqin Zhu, Shanshan Yang, Shuang Chang, Peng He, Xiuxiu Xu, JinPeng Zhang, Yi Shen, Shuyu Wu, Ya Jiang, Sisi Liu, Xian Zhang, Xue Wu, Yang Shao. Development and performance of a multi-cancer early detection test utilizing plasma cfDNA fragmentomics: A large-scale, prospective, multicenter study [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 1266.