Development and Validation of a Model Including Distinct Vascular Patterns to Estimate Survival in Hepatocellular Carcinoma

Importance

Because of tumor heterogeneity, traditional clinical variables remain insufficient to predict recurrence, which impairs long-term survival among patients undergoing radical hepatectomy for hepatocellular carcinoma (HCC). Vessels encapsulating tumor clusters (VETC) constitute a novel vascular pattern distinct from microvascular invasion (MVI), representing biological aggressiveness of HCC.

Objective

To establish a model to estimate individualized recurrence-free survival (RFS) in HCC by integrating VETC and MVI.

Design, Setting, and Participants

This prognostic study included 498 patients undergoing radical hepatectomy for HCC from 5 academic centers in China from January 1, 2013, to December 31, 2016, and consisted of 3 cohorts: training (243 [48.8%]), internal validation (122 [24.5%]), and external validation (133 [26.7%]). Follow-up was completed on March 30, 2020, and the data were analyzed from December 1 to 31, 2020.

Exposures

VETC, MVI, tumor number, and maximum tumor size.

Main Outcomes and Measures

The primary end point was RFS. The risk score for relative recurrence and nomogram for absolute RFS probability were derived from the final model, which contained variables recommended by multivariate least absolute shrinkage and selection operator Cox proportional hazards regression analysis. Their performance was quantified using the Harrell concordance index (C index), the time-dependent area under the receiver operating characteristic curve, and calibration curves and was compared with 6 prognostic systems. Recurrence-free survival was estimated by the Kaplan-Meier method, and RFS curves were compared using a log-rank test.

Results

Among the 498 patients, 432 (86.7%) were men; the mean (SD) age at diagnosis was 51.4 (11.3) years. Independent predictors for RFS identified included VETC, MVI, tumor number, and maximum tumor size, which were incorporated into the multivariate model (VMNS model). The C index (0.702; 95% CI, 0.653-0.752) for the VMNS score of the training cohort was significantly higher than those of 6 conventional systems (0.587 [95% CI, 0.535-0.638] to 0.657 [95% CI, 0.606-0.708]). Different recurrence risk groups defined by the VMNS score showed significantly different 2-year RFS (low-risk group, 81.4% [SE, 0.036]; medium-risk group, 62.1% [SE, 0.054]; high-risk group, 30.1% [SE, 0.079];P < .001). Calibration curves of the VMNS nomogram showed good agreement between the nomogram-predicted RFS probability and actual RFS proportion. The internal and external validation cohorts confirmed the results.

Conclusions and Relevance

The VMNS model enabled individualized prognostication of RFS in patients with HCC undergoing curative resection.