Background: Numerous risk-scoring systems for HCC amongst CHB patients have been developed to stratify ‘high-risk’ patients into screening programs.

Purpose: We developed and assessed a random survival forest (RSF) model to identify patients with an elevated 5-year risk of developing HCC.

Methods: We used the Canadian Hepatitis B Network, a cohort sourced from 19 centers nationally. Inclusion criteria were: patient data from January 1st, 2012 to December 31st, 2021, mono-infection, and minimum 7/10 features (race, age, sex, platelets, albumin, ALT, INR, AFP, HBV DNA, and bilirubin). RSF model was created with 4-fold cross-validation with 5-year HCC risk assessed through AUC. Complete-case analysis of two existing models (PAGE-B, aMAP) was also conducted.

Results: Of 6243 patients with CHB, 1058 patients met criteria. AFP, HBV DNA status, platelets, and age were most statistically significant determinants of HCC in the RSF cross-validation. The RSF model had the highest AUC (0.89), sensitivity (0.99), specificity (0.57), PPV (0.25), and NPV (1.00). Results in Table 1.

Conclusion: The RSF model had an improved predictive capacity for 5-year HCC risk in CHB patients compared to more traditional scoring models.