Background
Achieving sustained virological response (SVR) reduces the risk of liver decompensation and hepatocellular carcinoma (HCC) in patients with chronic hepatitis C virus (HCV) infection and advanced liver disease. Life-long HCC surveillance is recommended after HCV cure, but its cost-effectiveness is uncertain, especially in the context of HIV co-infection, where data remain scarce.
Methods
Multicenter prospective study conducted in the GEHEP-011 cohort (17 hospitals) in Spain. Inclusion criteria: patients with HCV chronic infection with and without HIV co-infection who: 1) Liver stiffness (LS) pre-DAA≥9.5 kPa; 2) SVR with DAA-based therapy; 3) LS measurement at SVR. Individuals with HBsAg positive, active alcohol consumption (>50g/day) and those with prior hepatic complications were excluded. The main endpoint was the occurrence of de novo HCC. The population was split as follows: 67% for model discovery, 33% for validation. A novel ad-hoc AI algorithm was developed to identify patients at low risk of HCC. Such an algorithm builds a binary tree model, where the best feature and its respective value for the cut-off at each tree node was automatically chosen by maximizing a trade-off between the sensitivity (Se) and negative predictive value (NPV), regarding only the classification made by that node.
Results
1.033 patients were included, 597 (58%) were living with HIV. The median (Q1-Q3) follow-up was 91 (69-102) months. Variables included in the model were: i) At SVR time-point: age, LS, FIB-4, GGT, albumin; ii) At 12 months post-SVR: fasting plasma glucose, GGT and LS. In the discovery population (n=691) the diagnostic performance of the model was: Se=100% (85%-100%), VPN=100% (99%-100%). No HCC were diagnosed in patients classified in the low-risk group. In this subset, HCC surveillance could have been stopped in 370 (53%) individuals. The Harrell’s C index (HCI) was 0.834. In the validation set (n=341) these figures were: Se=91% (59%-100%) and VPN=99% (97%-100%). 183 (54%) individuals were classified in the HCC low-risk subgroup. One of them developed HCC [incidence=0.08 (95% CI, 0.01-0.58) per 100 person-year]. The HCI was 0.857.
Conclusions
A simple model, based on accessible parameters measured at SVR time-point and 12 months afterwards, can identify patients, living or not with HIV, with very low risk of HCC emergence after HCV cure. The incidence of HCC in subjects considered at low risk falls below the threshold of 1.0%, for which HCC surveillance is considered not cost-effective.
Table or Figure
