Machine learning based hepatic safety score predicts decompensation in hepatocellular carcinoma systemic therapy

Area of Science:

• Hepatology
• Oncology
• Artificial Intelligence in Medicine

Background:

• Hepatocellular carcinoma (HCC) often co-occurs with portal hypertension, increasing risks of hepatic decompensation (HD) and variceal bleeding during systemic therapy.
• Accurate prediction of these risks is crucial for optimizing treatment strategies in unresectable HCC.

Purpose of the Study:

• To develop and validate a machine learning-based hepatic safety score (MHSS) for predicting clinically significant portal hypertension (CSPH) and prognosis in HCC patients.
• To assess the MHSS's utility in guiding personalized systemic therapy selection.

Main Methods:

• A random forest model was trained on data from 2026 patients with unresectable HCC.
• The model, forming the MHSS, was validated in an independent cohort.
• Performance was evaluated for predicting CSPH, HD, variceal bleeding (VB), and mortality.

Main Results:

• The MHSS demonstrated robust performance (AUROC 0.840) in predicting CSPH and HD.
• High MHSS scores correlated with significantly elevated risks of HD (HR 3.25), VB (HR 4.90), and mortality (HR 2.21).
• MHSS-guided treatment selection simulation showed potential reductions in HD (24%), VB (40%), and mortality (26%).

Conclusions:

• The MHSS effectively predicts CSPH, decompensation, and survival in HCC patients before systemic therapy.
• The MHSS enables individualized risk stratification, guiding personalized treatment choices, particularly regarding bevacizumab-containing regimens.
• Implementing MHSS-guided therapy may improve patient outcomes by optimizing treatment selection and reducing adverse events.