Abstract 950P
Background
Early detection of primary liver cancer (PLC), including hepatocellular carcinoma (HCC) and intrahepatic cholangiocarcinoma (ICC), is essential for patients’ survival. Previous studies mainly focused on HCC detection using DNA methylation or combined epigenetic and genetic profiles. This prospective study aims to develop an accurate and cost-effective method for PLC early detection and differentiating ICC from HCC using plasma cell-free DNA (cfDNA) fragmentomic profiles.
Methods
Whole-genome sequencings (WGS) (∼5X) were performed for plasma cfDNA samples of 381 PLC patients (316 HCC, 52 ICC, 13 mix; 244 stage I, 93 stage II, 44 stage III) and 335 non-cancer controls (including 104 liver cirrhosis[LC]/hepatitis B virus[HBV]-positive individuals). 192 PLC patients and 170 non-cancer controls were recruited in the training cohort. An ensembled stacked model for PLC detection was constructed using cfDNA high-resolution fragment distribution profile. The model performance was then assessed in the test cohort (189 PLC and 165 non-cancer cases).
Results
Our predictive machine learning model yielded an Area Under the Curve (AUC) of 0.995 in the test cohort while reaching an overall sensitivity for cancer detection of 96.8% at 98.8% specificity. Additionally, our model showed excellent sensitivities in detecting early-stages PLC (stage I: 95.9%, stage II: 97.8%), patients with smaller tumors (<=3cm: 98.2%), and HCC (96.0%) or ICC (100%). Furthermore, the AUC for distinguishing PLC patients from LC/HBV controls reached 0.985 (96.8% specificity at 96.1% specificity). Promisingly, our model maintained consistent performances during the downsampling process, even while using 1X coverage WGS data (AUC: 0.994, 94.4% sensitivity at 98.8% specificity). Fragmentomic profiles also showed great potential in distinguishing ICC from HCC (AUC: 0.776).
Conclusions
Our predictive model has out-performed previously reported methods with solely low coverage WGS data, and therefore exhibits excellent potential in clinical practice for cost-effective PLC early detection and more accurate diagnosis of PLC subtypes.
Clinical trial identification
Editorial acknowledgement
Legal entity responsible for the study
Liver Cancer Institute, Zhongshan Hospital, Fudan University.
Funding
Liver Cancer Institute, Zhongshan Hospital, Fudan University.
Disclosure
W. Tang: Financial Interests, Personal, Full or part-time Employment: Nanjing Geneseeq Technology Inc. R. Liu: Financial Interests, Personal, Full or part-time Employment: Nanjing Geneseeq Technology Inc. H. Bao: Financial Interests, Personal, Full or part-time Employment: Nanjing Geneseeq Technology Inc. X. Chen: Financial Interests, Personal, Full or part-time Employment: Nanjing Geneseeq Technology Inc. S. Wu: Financial Interests, Personal, Full or part-time Employment: Nanjing Geneseeq Technology Inc. X. Wu: Financial Interests, Personal, Full or part-time Employment: Nanjing Geneseeq Technology Inc. Y. Shao: Financial Interests, Personal, Member of the Board of Directors: Nanjing Geneseeq Technology Inc. All other authors have declared no conflicts of interest.