Dynamic classification using case-specific training cohorts outperforms static gene expression signatures in breast...

Area of Science:

• Oncology
• Bioinformatics
• Genomics

Background:

• Breast cancer's molecular diversity challenges universal prognostic markers.
• Existing multigene prognostic classifiers have limitations due to single training cohorts.

Purpose of the Study:

• To develop a novel dynamic predictor for personalized breast cancer prognosis.
• To overcome limitations of static prognostic models by creating case-specific predictors.

Main Methods:

• Analyzed gene expression data from 3,534 breast cancers with relapse-free survival data.
• Developed a case-specific predictor for each test case using molecularly similar patients.
• Assessed model performance using leave-one-out validation and independent case validation.

Main Results:

• The dynamic predictor demonstrated high prognostic discrimination (HR=3.68, p=1.67E-56) across all cases.
• Achieved higher overall accuracy (0.68) compared to Oncotype DX (0.64), Genomic Grade Index (0.61), and MammaPrint (0.47).
• Showed effectiveness in triple-negative cancers (HR=3.08, p=0.0093), where other classifiers failed.

Conclusions:

• A new method for personalized prognostic prediction using dynamic, case-specific training cohorts was developed.
• Dynamic predictors outperform static models and offer improved accuracy, particularly for challenging subtypes like triple-negative breast cancer.
• The dynamic classifier is accessible online for clinical application.