Comparative analysis of ctDNA-MRD and MVI in predicting postoperative recurrence of hepatocellular carcinoma.

Abstract

4128 Background: Microvascular invasion (MVI) is currently recognized as a pathological feature strongly associated with HCC recurrence. However, its predictive performance in clinical practice remains suboptimal. The detection of minimal residual disease (MRD), which has demonstrated significant prognostic value across multiple cancer types, represents a promising alternative. Therefore, the first goal of this study is to detail the genomic alterations that potentially drive MVI or MRD, the second goal aims to compare their predictive power for postoperative recurrence in early-stage HCC patients. Methods: This study profiles the genomic landscape of tumor samples from 126 BCLC 0/A/B stage HCC patients using WES. Patients’ longitudinal MRD status was determined through ctDNA detection in peripheral blood at preoperative, 1-, 4-, and 7-month postoperative time points using a tumor-agnostic fixed panel. Postoperative recurrence, confirmed by radiographic imaging, served as the endpoint events for comparing the sensitivity and specificity of two prediction models, with one model based on MVI classification and the other on longitudinal MRD positivity. More patients are currently being recruited and analyzed. Results: Among 126 enrolled HCC patients, 71 (56.3%) were classified as MVI-positive (M1/M2), while 55 (43.7%) were MVI-negative (M0). Of the 121 patients with at least one MRD test, longitudinal MRD positivity was observed in 27 patients (22.3%), with 94 patients (77.7%) remaining negative. Both MVI positivity and longitudinal MRD positivity were associated with postoperative recurrence, with longitudinal MRD group showing stronger statistical significance (p < 0.001 vs. p = 0.032). However, no distinct high-frequency mutation patterns were observed within either the MVI or longitudinal MRD groups. The most frequently altered genes in both groups included TP53, CTNNB1, JAK1, ARID1A, CDKN2A, and AXIN1. Of note, we also developed two prediction models based on MVI classification and longitudinal MRD positivity. The longitudinal MRD-based model demonstrated superior performance, with higher AUROC (0.835 vs. 0.715), PPV (0.357 vs. 0.162), accuracy (0.836 vs. 0.512), and TPR (0.833 vs. 0.786). Conclusions: Longitudinal MRD monitoring using a tumor-agnostic fixed panel demonstrates superior predictive performance over MVI classification for postoperative HCC recurrence with higher sensitivity, specificity, and accuracy. The findings highlight the potential of longitudinal MRD monitoring as a more reliable tool for guiding personalized postoperative management in early-stage HCC patients.Moreover, the combination of the two models may potentially offer superior predictive performance for recurrence, which is also a promising direction worthy of further exploration.