External validation of models to predict hepatocellular carcinoma in Hepatitis C Virus cured F3-F4 patients.

Background & aims: Several hepatocellular carcinoma (HCC) risk-models have been developed to individualise patient surveillance following sustained viral response (SVR) in Hepatitis C Virus patients. Validation of these models in different cohorts is an important step to incorporate a more personalised risk assessment in clinical practice. We aimed at applying these models to stratify the risk in our patients and potentially determine cost-saving associated with individualised HCC risk-stratification screening strategy.

Methods: Patients with baseline F3-4 fibrosis treated with new oral direct-acting antivirals who had reached a SVR were regularly followed as part of the HCC surveillance strategy. Six models were applied: Pons, aMAP, Ioannou, HCC risk, Alonso and Semmler. Validation of the models was performed based on sensitivity and the proportion of patients labelled as "high risk".

Results: After excluding 557 with less than 3 fibrosis, 12 without SVR, 18 with a follow up (FU) <1 year, 17 transplant recipients, 16 lost to FU and 31 with HCC at time of antiviral therapy, our cohort consisted of 349 F3-4 SVR patients. Twenty-three patients (6.6%) developed HCC after a median FU of 5.12 years. The sensitivity of the different models varied between 0.17 (Semmler7noalcohol) and 1 (Alonso A and aMAP). The lowest proportion of high-risk patients corresponded to the Semmler-noalcohol model (5%). Sixty-three and 90% of the Alonso A and aMAP patients, respectively were labelled as high risk. The most reliable HCC risk-model applied to our cohort to predict HCC development is the Alonso model (based on fibrosis stage assessed by liver stiffness measurements or Fibrosis-4 index (FIB-4) at baseline and after 1 year, and albumin levels at 1 year) with a-100% sensitivity in detecting HCC among those at high risk and 63% labelled as high risk. The application of the model would have saved the cost of 1290 ultrasound no longer being performed in the 37% low-risk group.

Conclusion: In our cohort, the Alonso A model allows the most reliable reduction in HCC screening resulting in safely stopping life-long monitoring in about a third of F3-F4 patients achieving SVR with DAAs.