A Multi-Omics-Based Prognostic Model for Elderly Breast Cancer by Machine Learning: Insights From Hypoxia and Immunity of Tumor Microenvironment

Abstract

Introduction: Older adult breast cancer (OABC) patients (≥ 65 years) frequently experience poorer prognoses compared to younger adults, attributed to complex tumor biology and age-related factors. The present study employs a multiomics approach combined with machine learning to develop a novel prognostic model for OABC, with a focus on the hypoxic and immune characteristics of the tumor microenvironment.

Methods: Genetic and molecular data from 503 OABC and 589 younger adult breast cancer (YABC) patients were analyzed using The Cancer Genome Atlas (TCGA) database. An ensemble machine-learning model was developed, integrating multiomics data-including mRNA, miRNA, lncRNA, copy number variations (CNVs), and single nucleotide variants (SNVs)-along with clinicopathological features, to predict survival outcomes. The model was trained on 300 OABC samples and validated on 203 samples.

Results: The ensemble machine-learning model achieved a predictive accuracy of 69.5% for survival outcomes in OABC patients. Distinct hypoxia-related gene expression patterns and reduced immune cell infiltration were observed in OABC compared to YABC. Hypoxia was significantly associated with poorer disease-free survival (DFS) in OABC (P = .037), but not in YABC (P = .38).

Conclusions: The multiomics-based prognostic model developed for OABC showed clinical potential, and the findings highlight the critical role of hypoxia and the immune microenvironment in the prognosis of OABC. Further research is warranted to validate this model in larger cohorts and to explore its potential application in guiding personalized treatment strategies for OABC patients.

Keywords: Elderly Breast Cancer; Hypoxia; Machine Learning; Multi-Omics Integration; Tumor Immune Microenvironment.