585P - Fragmentomics early detection assay leading to potential clinical benefits in colorectal cancer

Background

Colorectal adenocarcinoma (CRC) patients often experience delayed diagnosis, making liquid biopsy-based early detection a promising approach. Yet, its clinical usage has been restricted due to its insufficient sensitivity. We aimed to develop an integrated model using fragmentomic profiles of plasma cell-free DNA (cfDNA) for the accurately and cost-effectively detection of early-stage CRC.

Methods

360 participants were enrolled as the training cohort, consisting of 176 CRC patients and 184 healthy controls. Plasma cfDNA were extracted and prepared for subsequent whole genome sequencing. To differentiate healthy controls from CRC patients, an ensemble stacked model was built upon five machine learning models couple with five cfDNA fragmentomic features. The model was subsequently validated in a cohort of 236 individuals, comprising 117 CRC patients and 119 healthy controls.

Results

The contracted ensemble stacked model demonstrated remarkable ability in distinguish between CRC patients and healthy controls as evidence in multiple facets. In the validation cohort, the ensemble stacked model demonstrated superior performance compared to all other base models constructed using feature-algorithm pairs, achieving a high AUC of 0.986. At 97% specificity, the sensitivity for detecting CRC patients in validation cohort reached 93%. Additionally, the sensitivity of our model increases alongside the progression of cancer stage. Our model consistently maintained high levels of accuracy during both within-run and between-run tests, successfully predicting cancer status at different clinical settings. Through a real-world simulation, the model's effectiveness was confirmed showing a potential increase of 17.47% in the 5-year survival rate.

Conclusions

Our ensemble stacked model, which leverages the multiplex nature of cfDNA, exhibited exceptional performance in terms of sensitivity and stability for detecting CRC risk. This model has the potential to facilitate early diagnosis and benefit a larger number of patients.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

National Natural Science Foundation of China; Jiangsu Primary Research & Development Plan; Jiangsu Province TCM science and technology development plan monographic project; Jiangsu Provincial Natural Science Foundation; Jiangsu Provincial Medical Youth Talent, The Project of Invigorating Health Care through Science, Technology Education; China Postdoctoral Science Foundation; The “333 Talents” Program of Jiangsu Province; The Talents Program of Jiangsu Cancer Hospital.

Disclosure

X. Wu, W. Tang, H. Tang, H. Bao, X. Wu, Y. Shao: Financial Interests, Personal, Financially compensated role: Geneseeq Technology Inc. All other authors have declared no conflicts of interest.