Enhancing Breast Cancer Recurrence Prediction Across Treatment Scenarios with Weighted Cox Mixtures

Area of Science:

• Oncology
• Biostatistics
• Medical Informatics

Background:

• Breast cancer treatment involves multiple modalities, and recurrence risk is influenced by treatment execution.
• Accurate prediction of recurrence risk is crucial for personalized treatment strategies and patient management.

Purpose of the Study:

• To develop and validate a weighted Cox mixtures model for estimating breast cancer recurrence risk.
• To compare the performance of different weighting strategies in improving risk prediction accuracy.

Main Methods:

• A weighted Cox mixtures model was proposed, integrating treatment plans and clinical data.
• Three weighting strategies were employed: Inverse Probability of Treatment Weighting, Adaptive Weights with focal loss, and Prioritizing Subgroups.
• The model was validated using data from the Mayo Clinic (US) and the National Institute of Oncology (Morocco).

Main Results:

• Adaptive Weights with focal loss demonstrated improved predictive accuracy (C-index: 0.67-0.88) in the Mayo Clinic cohort compared to the standard Cox model.
• Similar improvements in C-index values (0.60-0.71) were observed in the Moroccan cohort, despite larger confidence intervals.
• Weighting strategies proved effective in refining recurrence risk prediction, especially in imbalanced datasets.

Conclusions:

• The proposed weighted Cox mixtures model, particularly with Adaptive Weights, enhances breast cancer recurrence risk prediction.
• The findings highlight the importance of weighting strategies for improving model reliability in diverse and imbalanced clinical cohorts.
• Expanding datasets, especially from underrepresented populations, is essential for robust model validation and clinical applicability.