Cure models with adaptive activation for modeling cancer survival

Area of Science:

• Biostatistics
• Survival Analysis
• Cancer Research

Background:

• Cure rate models are essential for analyzing cancer survival data, particularly for diseases with long-term survivors.
• Existing models like first- and last-activation may lack robustness under model misspecification.
• Analysis of colon cancer and triple-negative breast cancer survival data necessitates advanced modeling techniques.

Purpose of the Study:

• To propose a novel, flexible class of cure rate models indexed by an adaptive activation parameter.
• To establish stochastic ordering and identifiability results for the proposed model class.
• To demonstrate the proposed model's superior performance over existing methods in cases of model misspecification.

Main Methods:

• Development of a new cure rate model class with an adaptive activation parameter and a function.
• Mathematical establishment of stochastic ordering properties concerning the activation parameter.
• Derivation of two identifiability results for the proposed model class.
• Comparative analysis with existing first- and last-activation models.

Main Results:

• The proposed class of cure rate models is stochastically ordered in the activation parameter.
• Two key identifiability results were established for the new model class.
• The adaptive activation model demonstrated more appropriate inference compared to first- and last-activation models under misspecification.
• The model was successfully applied to colon cancer and triple-negative breast cancer data.

Conclusions:

• The proposed adaptive cure rate model offers a flexible and robust alternative for cancer survival analysis.
• This approach provides improved inference, particularly when standard model assumptions are not met.
• The model's applicability was confirmed in real-world cancer datasets, addressing treatment-sex interactions and tumor heterogeneity.