Hepatocellular carcinoma (HCC), the third leading cause of cancer deaths, urgently requires innovative early detection strategies. This prospective study aimed to evaluate the feasibility of cell-free DNA (cfDNA) methylation, mutation, and/or alpha-fetoprotein (AFP) for HCC early detection model development. Peripheral blood samples were prospectively collected from 635 participants (288 HCC, 347 non-HCC), who were randomly assigned (6:4) to the training and validation sets to develop and validate a multi-omics early detection (MOED) model. The model was externally validated on 797 subjects (160 HCC, 637 non-HCC) and further blindly tested on 452 community-recruited high-risk individuals. The methylation-based model showed superior performance over AFP and mutation-based models in the training set. In the validation set, the MOED model, integrating methylation with AFP levels ≥ 400 ng/mL, slightly increased sensitivity from 87.1% (95% CI: 79.6%-92.6%) to 88.8% (95% CI: 81.6%-93.9%) at 95.7% specificity (95% CI: 91.0%-98.4%), whereas adding mutation data did not improve the performance. In the independent validation set, the locked MOED exhibited 91.9% (95% CI: 86.5%-95.6%) overall sensitivity, 83.3% (95% CI: 51.6%-97.9%) stage 0 sensitivity, and 98.4% overall specificity (95% CI: 97.1%-99.2%). In high-risk individuals, the model demonstrated 92.9% (95% CI: 64.2%-99.6%) sensitivity, 90.6% (95% CI: 87.4%-93.1%) specificity, 24.1% (95% CI: 13.9%-37.9%) positive predictive value (PPV) and 99.7% (95% CI: 98.4%-100.0%) negative predictive value (NPV). The MOED model, integrating cfDNA methylation and AFP, is highly effective for HCC detection and promising for screening in high-risk populations.
Wang et al. (Mon,) studied this question.