TORNA ALL’ELENCO

Development and validation of an artificial intelligence PIVKA-II-based model for the prediction of hepatocellular carcinoma development in patients with HCV-related cirrhosis successfully treated with direct-acting antivirals in Digestive and Liver Disease

2022
AOU Città della Salute di Torino

Tipo pubblicazione

Conference Abstract

Autori/Collaboratori (9)Vedi tutti...

Caviglia GP
Unit of Immunology, Rheumatology, Allergy and Rare Diseases (UnIRAR), IRCCS San Raffaele Scientific Institute, Vita-Salute San Raffaele University, Milan, Italy
Fariselli P
Computational Biomedicine, Department of Medical Sciences, University of Turin, Turin, Italy
D'Ambrosio R
Division of Gastroenterology and Hepatology, Foundation IRCCS Ca' Granda Ospedale Maggiore Policlinico, Milan, Italy

et alii...

Abstract

Introduction: Patients with hepatitis C virus (HCV)-related cirrhosis witha a sustained virological response (SVR) to direct-acting antivirals (DAA) remain at risk of hepatocellular carcinoma (HCC). Recently, serum protein induced by vitamin k absence or antagonist-II (PIVKA-II) showed promising results as HCC predictive biomarker. Aim: To develop and validate an artificial intelligence PIVKA-II-based model for the stratification of the risk of HCC development in cirrhotic patients with an SVR to DAA. Materials and Methods: Two independent cohorts of consecutive patients were analyzed: the Turin training cohort (n=574; males 57%, median age 65, 28–90 years; follow-up [FU] 38.6, 21.3–52.3 months; incident HCC: 54 [10.0%]) and the Milan external test cohort (n=372; males 56%, median age 66, 24–92 years, FU 48.3, 34.6–51.9 months; incident HCC: 22 [5.9%]). Post-DAA serum PIVKA-II was measured by Lumipulse G system (Fujirebio, Japan). Model performance was assessed by Harrell's C-index and cumulative dynamic area under the curve (AUC). Results: Using a penalized Cox regression, a model including PIVKA-II values combined with age, gender, BMI, AST, ?-GT, albumin, total bilirubin, and triglycerides was derived from the training cohort (3-fold cross-validated C-index=0.80±0.03). In the external test cohort, the model showed C-index=0.78±0.01 with a 12-month AUC=0.89±0.03. When patients were grouped into 3 risk categories, HCC cumulative incidence was 1.6%, 3.5%, and 14.2% in the low-, medium-, and high-risk groups, respectively (log-rank test medium- vs. high-risk, p=0.020; Figure). Notably, no HCC occurred within 12-months FU in low- and medium-risk groups. Tests with more complex models, such as Random survival forests and gradient boosting, showed no improvement over the penalized Cox, indicating an intrinsically linear and additive risk. Conclusions: Our artificial intelligence PIVKA-II-based model accurately predicted HCC development and may be usef

DOI : 10.1016/j.dld.2022.08.002

Keywords

validation process; sustained virologic response; protein blood level; prediction; male; nonhuman; major clinical study; log rank test; liver cirrhosis; liver cell carcinoma; Japan; human tissue; human; high risk population; Hepatitis C virus; albumin; antivirus agent; bilirubin; decarboxyprothrombin; endogenous compound; gamma glutamyltransferase; triacylglycerol; aged; area under the curve; artificial intelligence; aspartate aminotransferase level; body mass; cancer model; cancer patient; cancer survival; cohort analysis; conference abstract; controlled study; cumulative incidence; drug therapy; female; follow up; forest; gender;