2709 - A New Method for Breast Cancer Classification: Comprehensive Application of Latent Class Analysis in Clinical and Genetic Expression Information

Q. Meng¹, K. Jiang², F. Zhang¹, X. Liu¹, and Y. Li²; ¹Department of Radiation Oncology, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, China, ²Togtoh County Hospital, Hohhot, Inner Mongolia Autonomous Region, China, Hohhot, China

Purpose/Objective(s): Breast cancer is a prevalent global malignancy that exhibits significant diversity in clinical and molecular attributes. This study leverages comprehensive clinical databases and gene expression data to conduct latent class analysis (LCA). The primary objective is to identify distinct subtypes within breast cancer and understand their associations with patient survival outcomes, aiming to refine personalized radiation therapy strategies. This approach promises to enhance treatment efficacy and patient survival by tailoring therapy to the specific characteristics of each subtype. Materials/

Methods: Latent class analysis (LCA) was performed on data extracted from the Surveillance, Epidemiology, and End Results (SEER) database. Classifying data based on four variables (race, age, pathological type, and tumor stage), we determined different breast cancer subtypes. Subsequent survival risk analysis, including non-breast cancer-specific death (nonBCSD) and breast cancer-specific death (BCSD), revealed increased risks for specific subtypes. The LCA method was then applied to The Cancer Genome Atlas (TCGA) database for result validation. Additionally, the introduction of LCATCGA(LCAt) as an optimization parameter significantly enhanced the accuracy of the PAM50 model in predicting survival rates. A secondary LCA based on the expression of 10 key genes identified unique gene expression subtypes with distinct biological characteristics and prognosis, suggesting potential new classifications of breast cancer.
Results: Analysis of the SEER database indicated significantly increased risks for nonBCSD in patients with LCAs2 subtype (HR=8.84), LCAs3 (HR=3.26), and LCAs4 (HR=4.89) subtypes. Validation in the TCGA database yielded similar classification results, affirming the stability and effectiveness of the LCA method. The introduction of LCAt parameters notably improved the accuracy of the PAM50 model. The gene expression-based LCAg classifications, named "Resistant," "Sensitive," and "Mixed," exhibited significant differences in biological characteristics and prognosis.
Conclusion: Latent class analysis effectively delineates breast cancer subtypes, enhancing survival risk prediction. The novel gene expression subtypes offer targeted avenues for personalized radiation therapy, potentially revolutionizing treatment paradigms and improving patient outcomes in radiation oncology.