Transcriptomic classifiers do not reliably predict lymph node involvement in breast Cancer: Insights from three large...

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Lymph node (LN) status is a critical prognostic indicator in breast cancer.
• A molecular surrogate for LN involvement could improve risk stratification and surgical staging.
• This study explores mRNA expression profiles as potential predictors of LN status.

Purpose of the Study:

• To investigate the predictive power of mRNA expression profiles for lymph node involvement in breast cancer.
• To compare machine learning model performance for LN status prediction against histological grade prediction.
• To assess the utility of survival probability models for LN status classification.

Main Methods:

• Utilized mRNA expression data from three large cohorts (TCGA, METABRIC, SCAN).
• Applied machine learning algorithms (AdaBoost, XGBoost) to classify LN status and histological grade.
• Evaluated model performance using Area Under the Receiver Operating Characteristic Curve (AUC).

Main Results:

• Maximum AUCs for LN prediction were 0.613 (XGBoost) and 0.604 (AdaBoost), indicating limited accuracy.
• Molecular prediction of histological grade achieved significantly higher AUCs (up to 0.901).
• Survival probability models showed AUCs below 0.6 for LN prediction.

Conclusions:

• mRNA expression profiles alone are not sufficient to reliably predict lymph node status in breast cancer.
• No current mRNA-based model is a viable alternative to surgical LN assessment.
• Future research should focus on integrative approaches combining molecular, spatial, and clinical data for improved prediction.