A Machine Learning Approach Yields a Multiparameter Prognostic Marker in Liver Cancer

Area of Science:

• Oncology
• Bioinformatics
• Immunology

Background:

• Hepatocellular carcinoma (HCC) lacks a universally accepted staging system for outcome prediction.
• Machine learning (ML) offers advanced methods for integrating diverse risk factors.

Purpose of the Study:

• To develop and validate a novel machine learning-based prognostic model for hepatocellular carcinoma (HCC).
• To incorporate clinical, laboratory, and peripheral immune features for improved risk stratification.

Main Methods:

• Retrospective analysis of three HCC cohorts, including clinicopathologic, laboratory, and T-cell function data.
• Development of a gradient-boosting survival (GBS) classifier using a training dataset.
• Validation of the GBS model in two independent cohorts.

Main Results:

• A 20-feature GBS model incorporating clinical, laboratory, and T-cell parameters was constructed.
• The GBS model achieved high predictive accuracy with concordance indexes of 0.844, 0.827, and 0.806 in training and validation sets.
• The model successfully stratified patients into high, medium, and low-risk groups for mortality, independent of conventional staging systems.

Conclusions:

• A multiparameter ML algorithm integrating clinical, laboratory, and immune signatures provides a robust approach for HCC prognosis.
• The developed GBS classifier demonstrates significant potential for identifying high-risk HCC patients for targeted interventions.