Serum models accurately predict liver-related clinical outcomes in chronic hepatitis C.

Abstract

This study developed liver outcome scores in chronic hepatitis C (CHC) that directly predict liver-related death, hepatocellular carcinoma (HCC), and liver decompensation. Six hundred seventeen CHC patients were followed up for a mean of 6 years and randomized into a training set (n = 411) and a validation set (n = 206). Clinical outcomes were determined using a population-based data linkage system. In the training set, albumin, gamma-glutamyl transpeptidase, hyaluronic acid, age, and sex were in the final model to predict 5-year liver-related death (area under receiver operating characteristic curve [AUROC] 0.95). Two cut points (4.0 and 5.5) defined three risk groups with an incidence rate for liver-related death of 0.1%, 2%, and 13.2%, respectively (P < 0.001). Albumin, gamma-glutamyl transpeptidase, hyaluronic acid, age, and sex were used to predict 5-year liver decompensation (AUROC 0.90). A cut point of 4.5 gave a sensitivity of 94% and a specificity of 84% to predict 5-year decompensation and defined two groups with an incidence rate for decompensation of 0.2% and 5.8%, respectively (P < 0.001). Alkaline phosphatase, α2-macroglobulin, age, and sex were used to predict 5-year HCC occurrence (AUROC 0.95). A cut-point of eight had a sensitivity of 90% and specificity of 88% to predict 5-year HCC occurrence and defined two groups with an incidence rate for HCC of 0.2% and 5.6%, respectively (P < 0.001). Similar results were obtained using the validation set. All three liver outcome scores had excellent predictive accuracy and were able to stratify risk into clinical meaningful categories for CHC patients.