Multivariate Shared Frailty Cure-Rate models: a focus on Breast Cancer family history

2025-08-22

AI Summary

Conventional breast cancer risk prediction models operate at the individual level, neglecting familial clustering and shared genetic or environmental risk factors. Method: This study proposes a novel family-level statistical framework integrating a shared frailty model—capturing unobserved, heritable risk heterogeneity—with a cure-rate model—accounting for a proportion of families inherently resistant to disease—and Cox regression for hazard estimation. Parameters are estimated using Swedish multigenerational registry data and validated via simulation studies. Contribution/Results: The proposed model significantly improves identification of high-risk families, outperforming individual-level models in explanatory power, predictive accuracy, and characterization of disease progression. Notably, complete family history emerges as a critical predictor. This work establishes a paradigm shift toward family-centered risk assessment, enabling more precise screening protocols and targeted, family-level prevention strategies.

Technology Category

• Reasoning under Uncertainty: Graphical Models
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Machine Learning: Kernel Methods

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Web Mining and Content Analysis: Models for Web evolution

Abstract

We discuss a shift in perspective from traditional approaches to breast cancer risk prediction: modelling families rather than individuals as unit of analysis. By investigating the latent familial risk underlying breast cancer diagnoses, we introduce a Multivariate Shared Frailty Cure-Rate model. This model captures the familial risk as a shared frailty among members and explicitly accounts for a fraction of women not susceptible to breast cancer. We aim at identifying the high-risk families to better target screening and prevention, ultimately improving early detection. A comparative analysis with Cox models and univariate models - where a binary risk indicator acts as best guess for the latent high-risk group - is conducted using simulation studies and data from the Swedish Multi-Generational Breast Cancer registry. We demonstrate the critical importance of using complete family history of breast cancer to accurately identify high-risk families and show that the Multivariate Shared Frailty Cure-Rate model, capturing both the fraction of non-susceptible subjects and the survival distribution among susceptibles, enhances explanatory power, improves prediction accuracy, and offers a broader representation of the disease process.

Problem

• Modeling families instead of individuals for breast cancer risk prediction
• Identifying high-risk families to improve screening and prevention targeting
• Capturing familial shared frailty and non-susceptible fraction in cancer modeling

Innovation

• Multivariate Shared Frailty Cure-Rate model for families
• Captures familial risk and non-susceptible fraction
• Uses complete family history to improve prediction accuracy