Structure-Leveraged Methods in Breast Cancer Risk Prediction

Area of Science:

• Oncology
• Biostatistics
• Medical Informatics

Background:

• Accurate breast cancer risk prediction is crucial for precision medicine.
• Electronic Health Records (EHRs) contain valuable data for risk assessment, including mammography descriptors and genetic markers (single-nucleotide polymorphisms - SNPs).
• Existing methods often lack the ability to fully leverage the structural information within these diverse data types.

Purpose of the Study:

• To develop novel penalized statistical methods for improved breast cancer risk prediction.
• To integrate structured data from mammography reports and genetic information (SNPs) within EHRs.
• To enhance feature selection and integrated learning by incorporating feature dependence structures.

Main Methods:

• A retrospective case-control study design was employed.
• Utilized 49 mammography descriptors and 77 SNPs from a personalized medicine data repository.
• Developed a new methodology combining group penalty and p-fusion penalty (1 ≤ p ≤ 2) to account for feature dependencies.

Main Results:

• The proposed method demonstrated statistically significant improvements in breast cancer risk prediction.
• Effectively leveraged the structure information among mammography descriptors and SNPs.
• Showcased potential for significant impact on individuals' lives through better risk assessment.

Conclusions:

• The novel penalized methods offer a significant advancement in breast cancer risk prediction.
• Integrating structured EHR data, including imaging and genetic markers, is key to personalized medicine.
• The developed approach provides a powerful tool for identifying individuals at higher risk, enabling timely interventions.