Flexible Cure Rate Modeling Under Latent Activation Schemes.

Area of Science:

• Biostatistics
• Medical Statistics
• Survival Analysis

Background:

• Modern medical advancements lead to more cured patients in survival datasets.
• Cure rate models, including Berkson and Gage (BG type) and Yakovlev, Chen, Ibrahim, and Sinha (YCIS type), address this phenomenon.
• Bayesian hierarchical cure models are a recent area of interest, prompting research into model relationships.

Purpose of the Study:

• To propose a unifying class of cure rate models.
• To facilitate flexible hierarchical model-building.
• To include existing BG and YCIS models as special cases within a unified framework.

Main Methods:

• Development of a unifying class of cure rate models.
• Incorporation of Bayesian hierarchical modeling.
• Discussion of regressing on the cure fraction and posterior distribution propriety.
• Application to simulation studies and real-world cancer datasets (melanoma, breast cancer).

Main Results:

• The proposed unifying class encompasses both BG and YCIS models.
• The framework allows for robust modeling by addressing uncertainty in cure mechanisms.
• The model effectively distinguishes between underlying mechanisms of relapse and cure.
• Demonstrated utility through simulation and analysis of melanoma and breast cancer data.

Conclusions:

• The proposed unifying cure rate model framework offers a flexible and robust approach to survival data analysis.
• This unified model accounts for patient cure and elucidates mechanisms of relapse and cure.
• The framework is validated through simulations and real-world cancer datasets, showing its practical applicability.