BACKGROUND

Large hepatocellular carcinoma (HCC) is difficult to resect and accompanied by poor outcome. The aim was to evaluate the short-term and long-term outcomes of patients who underwent liver resection for large HCC, eventually drawing prediction models for short-term and long-term outcomes.

METHODS

1710 large HCC patients were recruited and randomly divided into the training (n = 1140) and validation (n = 570) cohorts in a 2:1 ratio. Independent risk factors were identified by regression model and used to establish three nomograms for surgical risk, overall survival (OS), and recurrence-free survival (RFS) in the training cohort. Model performances were assessed by discrimination and calibration. The three models were also compared with six other staging systems.

RESULTS

Platelet (PLT), gamma-glutamyl transpeptidase (GGT), albumin-bilirubin (ALBI) grade, blood transfusion and loss, resection margin, tumor size, and tumor number were established in a nomogram to evaluate surgical risk ( https://largehcc.shinyapps.io/largehcc-morbidity/ ). The model had a good prediction capability with a C-index of 0.764 and 0.773 in the training and validation cohorts. Alpha-fetoprotein (AFP), resection margin, tumor size, tumor number, microvascular invasion, Edmondson-Steiner grade, tumor capsular, and satellite nodules were considered to construct a prognostic nomogram to predict the 1-, 3- and 5-year OS ( https://largehcc.shinyapps.io/largehcc-os/ ). The C-index of the model was 0.709 and 0.702 for the training and validation cohorts. Liver cirrhosis, albumin (ALB), total bilirubin (TBIL), AFP, tumor size, tumor number, microvascular invasion, and tumor capsular were used to draw a prognostic nomogram to predict the 1-, 3- and 5-year RFS ( https://largehcc.shinyapps.io/largehcc-rfs/ ). The C-index of the model was 0.695 and 0.675 in the training and validation cohorts. The discrimination showed that the models had significantly better predictive performances than six other staging systems.

CONCLUSIONS

Three novel nomograms were developed to predict short-term and long-term outcomes in patients with large HCC who underwent curative resection with adequate performance. These predictive models could help to design therapeutic interventions and surveillance for patients with large HCC.