Machine Learning Predicts Hepatocellular Carcinoma Risk from Routine Clinical Data: A Large Population-Based...

Area of Science:

• Oncology
• Bioinformatics
• Machine Learning

Background:

• Hepatocellular carcinoma (HCC) presents a significant global health challenge due to its high fatality rate.
• Accurate risk stratification for HCC is critical for early detection and intervention but remains a complex clinical problem.

Purpose of the Study:

• To develop and validate an interpretable machine learning framework for hepatocellular carcinoma (HCC) risk stratification.
• To assess the utility of multimodal clinical data for improving HCC risk prediction.

Main Methods:

• Utilized prospectively collected multimodal data from over 900,000 individuals across two large cohorts (UK Biobank and All of Us Research Program).
• Developed and trained random-forest-based machine learning models incorporating demographics, lifestyle, health records, blood markers, genomics, and metabolomics.
• Evaluated model performance against existing state-of-the-art risk scores on internal and external test sets.

Main Results:

• The developed PRE-Screen-HCC framework demonstrated significantly superior performance compared to current state-of-the-art risk scores.
• The models showed robustness across diverse ethnic subgroups, indicating broad applicability.
• Comprehensive interpretability analysis was performed, and all code and model weights were released.

Conclusions:

• PRE-Screen-HCC offers a robust, interpretable, and accurate machine learning solution for hepatocellular carcinoma risk stratification.
• The framework facilitates early detection and personalized risk assessment using routinely available clinical data.
• The open-source nature of the code and models promotes external validation and integration into clinical workflows.