A Liver Stiffness-Based Etiology-Independent Machine Learning Algorithm to Predict Hepatocellular Carcinoma

Area of Science:

• Hepatology
• Machine Learning in Medicine
• Oncology

Background:

• Existing hepatocellular carcinoma (HCC) risk scores lack accuracy and are often specific to hepatitis B.
• There is a need for improved risk stratification tools for HCC across various chronic liver diseases (CLDs).

Purpose of the Study:

• To develop and validate a liver stiffness-based machine learning (ML) algorithm for HCC prediction and risk stratification.
• To assess the performance of the developed ML algorithm against existing HCC risk scores.

Main Methods:

• Trained ML models on data from 5155 adult patients with CLDs in Korea.
• Validated the models in prospective cohorts from Hong Kong (N=2732) and Europe (N=2384).
• Assessed model performance using Harrell's C-index and time-dependent ROC curves.

Main Results:

• Developed the SMART-HCC score, identifying liver stiffness as the most crucial predictor among nine clinical features.
• Achieved high performance in validation cohorts: Harrell's C-index of 0.89 (Hong Kong) and 0.91 (Europe).
• The SMART-HCC score demonstrated superior performance compared to existing HCC risk scores and accurately stratified patients into low- and high-risk groups.

Conclusions:

• The SMART-HCC score is a validated, machine learning-based tool for HCC risk stratification.
• This algorithm is valuable for clinicians managing patients with diverse CLDs.
• Liver stiffness measurement is a key component for accurate HCC risk prediction.