Temporal Association Rule Mining: Race-Based Patterns of Treatment-Adverse Events in Breast Cancer Patients Using...

Area of Science:

• Oncology
• Data Mining
• Health Disparities

Background:

• African American (AA) patients face disparities in breast cancer treatment and survival.
• Adverse events (AEs) from systemic therapy also differ, but data are limited.
• Understanding these differences is crucial for equitable care.

Purpose of the Study:

• To identify differences in associations between breast cancer treatments (TRs) and adverse events (AEs) between AA and White women.
• To leverage temporal association rule (TAR) mining on the SEER-Medicare dataset to analyze these disparities.
• To compare care settings: hospital outpatient units versus private practitioners.

Main Methods:

• Utilized the SEER-Medicare dataset for women aged 65+ diagnosed with breast cancer.
• Consolidated 141 drug codes into 46 mechanistic treatment (TR) categories and ICD-9 codes into 18 AE categories.
• Applied TAR mining (FPGrowth algorithm) to identify associations between TRs and AEs, considering patient factors like race and stage.
• Defined significant differences as at least one unit of lift between groups.

Main Results:

• Specific treatment-AE associations are significantly influenced by patient race, cancer stage, and the venue of care.
• The study successfully identified race-dependent patterns in treatment-related adverse events.

Conclusions:

• The TAR mining approach effectively highlights differences in TR-AE associations across diverse patient populations.
• This method provides a valuable reference for predicting AEs in different breast cancer patient groups.
• Unsupervised learning with temporal focus enhances predictive power as patient health status evolves.