[Application of a plasma-based ctDNA multimodal model featuring GSTP1 and SFRP2 methylation for early-stage...

Objective: To investigate the diagnostic value of plasma circulating tumor DNA (ctDNA) methylation of glutathione S-transferase P1 (GSTP1) and secreted frizzled-related protein 2 (SFRP2) for early hepatocellular carcinoma (HCC), and to evaluate a multimodal model integrating smoking history, nodule size, alpha-fetoprotein (AFP), and protein induced by vitamin K absence-Ⅱ (PIVKA-Ⅱ). Methods: This single-center retrospective case-control study enrolled 180 patients with HCC or liver cirrhosis at the Affiliated Hospital of Shaanxi University of Chinese Medicine from May 2023 to December 2024, with patients with cirrhosis serving as disease controls. The training set included 63 HCC patients and 63 cirrhosis patients, and the validation set included 26 HCC patients and 28 cirrhosis patients. Baseline clinical data, tumor biomarkers, liver function parameters, coagulation indices, imaging characteristics, and plasma ctDNA GSTP1/SFRP2 methylation levels detected by methylation-specific quantitative PCR (MS-qPCR) were collected. Logistic regression was used to identify independent risk factors and construct a multimodal diagnostic model. Model performance was evaluated using receiver operating characteristic (ROC) curves, calibration curves, confusion matrix analysis, and decision curve analysis (DCA). Result: GSTP1 methylation was higher in the HCC group than in the cirrhosis group [0.87 (0.75, 0.97) vs. 0.76 (0.54, 0.86); U=-3.842, P<0.001], and SFRP2 methylation was also higher [0.83 (0.70, 0.94) vs. 0.71 (0.60, 0.83); U=-3.423, P<0.001]. Multivariate logistic regression identified smoking history (OR=18.84, 95%CI: 2.82-126.10), nodule diameter (OR=33.40, 95%CI: 2.50-446.58), AFP (OR=40.03, 95%CI: 7.39-216.95), PIVKA-Ⅱ (OR=6.18, 95%CI: 1.34-28.39), and GSTP1/SFRP2 methylation (OR=1.50/2.54) as independent risk factors for HCC. The multimodal model achieved AUCs of 0.962 and 0.955 in the training and validation sets, respectively, with positive predictive accuracies of 0.891 and 0.868, outperforming single-factor diagnostic methods. In conclusion, GSTP1 and SFRP2 methylation levels were significantly higher in the HCC group than in the cirrhosis group, and both were independent risk factors for HCC, indicating potential for early diagnosis. The multimodal model integrating ctDNA methylation and clinicoradiological indicators achieved AUCs of 0.962 and 0.955 in the training and validation sets, with positive predictive accuracies of 0.891 and 0.868, respectively. Conclusion: DCA showed greater net benefit than any single indicator, suggesting that this model could improve diagnostic performance for early HCC and provide a new strategy for liver cancer screening.