Development of a Machine Learning-Based Prognostic Model for Hormone Receptor-Positive Breast Cancer Using Nine-Gene...

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Hormone receptor-positive breast cancer (HR+ BC) accounts for 80% of all breast cancer cases.
• Accurate prognosis determination is critical for improving survival outcomes in HR+ BC patients.
• Current prognostic tools are limited to clinical factors and lack universality.

Purpose of the Study:

• To develop a novel artificial intelligence (AI) framework for predicting prognosis in HR+ BC patients.
• To create a recurrence prediction model (RPM) integrating genomic and clinical data.
• To stratify HR+ BC patients into high- and low-risk categories for improved mortality prediction.

Main Methods:

• Analysis of 2,338 HR+ human epidermal growth factor receptor 2 negative (HER2-) BC cases from METABRIC, TCGA, and GEO cohorts.
• Development of an RPM by selecting nine prognosis-related genes from over 18,000 genes using logistic regression.
• Stratification of patients into high- and low-risk groups based on the RPM.

Main Results:

• The RPM significantly stratified risk in both discovery and validation cohorts.
• High-risk tumors showed enrichment in cell cycle, MYC, and PI3K-AKT-mTOR signaling pathways.
• High-risk tumors were associated with increased immune cell infiltration, cytolytic activity, and poorer response to chemotherapy and endocrine therapy.

Conclusions:

• The developed AI model effectively stratifies prognosis across multiple cohorts.
• The model's accuracy is attributed to its reflection of key BC therapeutic targets and the tumor immune microenvironment.
• Findings support the model's utility in predicting therapeutic effects and guiding treatment decisions.