Comparative analysis of ctDNA-MRD and MVI in predicting postoperative recurrence of hepatocellular carcinoma: a prospective multicenter cohort study

Background Microvascular invasion (MVI) predicts recurrence after curative hepatocellular carcinoma (HCC) resection, but its accuracy is limited. Dynamic circulating tumor DNA (ctDNA)–minimal residual disease (MRD) monitoring has strong predictive value in solid tumors, yet head-to-head comparisons with MVI in HCC are lacking. Methods This prospective, multicenter cohort study enrolled 152 patients with HCC in Barcelona Clinic Liver Cancer stages 0–B undergoing curative resection. HCC tissue was collected for whole-exome sequencing, and ctDNA sequencing with an 804-gene panel was performed on peripheral blood obtained preoperatively and at 1, 4, and 7 months postoperatively to monitor MRD. The primary endpoint was recurrence-free survival. The predictive performance of MVI, longitudinal ctDNA-MRD, and their combined model was assessed using survival analysis, time-dependent receiver operating characteristic (ROC) curves, concordance index, integrated discrimination improvement, and net reclassification improvement. Results MVI-positive patients (52.0%) had poorer recurrence-free survival (hazard ratio [HR] = 3.52, p < 0.01). Longitudinal MRD status, newly emerged or persistently positive versus clearance or persistently negative, strongly predicted recurrence (hazard ratio [HR] = 16.01) and remained independently prognostic (HR = 12.45, p < 0.01). In 87.5% of cases, longitudinal MRD positivity preceded radiological recurrence (median lead time, 86 d). Time-dependent ROC analysis showed that longitudinal MRD (area under the curve [AUC] 0.785–0.849) consistently outperformed MVI (AUC 0.655–0.752) at 4, 7, 12, and 15 months postoperatively. A risk stratification model combining MVI and longitudinal MRD (low-, medium-, and high-risk) further improved predictive performance (AUC 0.853–0.899) and clinical net benefit. Integrated discrimination improvement and net reclassification improvement analyses confirmed that longitudinal MRD significantly improved predictive performance compared with MVI. Conclusions Dynamic longitudinal ctDNA-MRD monitoring significantly outperforms static MVI assessment for recurrence prediction after HCC curative resection. This combination enables more precise postoperative risk stratification, providing insights to guide personalized adjuvant therapy decisions and follow-up strategies. Trial registration ClinicalTrials, NCT06449846. Registered 30 April 2024, https://clinicaltrials.gov/study/NCT06449846?term=NCT06449846&viewType=Card&rank=1 .