On a general structure for hazard-based regression models: An application to population-based cancer research

Area of Science:

• Biostatistics
• Survival Analysis
• Epidemiology

Background:

• The proportional hazards model is standard for time-to-event data.
• Existing models often fail to capture complex hazard structures.
• There is a need for flexible models in survival analysis.

Purpose of the Study:

• To propose a flexible parametric approach for time-to-event data analysis.
• To generalize beyond proportional hazards, accelerated hazards, and accelerated failure time models.
• To evaluate the model's performance in excess hazard modeling for cancer epidemiology.

Main Methods:

• Utilized a flexible parametric distribution (exponentiated Weibull) for the baseline hazard.
• Investigated a general hazard structure encompassing proportional, accelerated hazards, and accelerated failure time models.
• Conducted an extensive simulation study and applied the model to lung cancer data.

Main Results:

• The proposed model demonstrated good inferential properties.
• The exponentiated Weibull distribution effectively covers various practical hazard shapes (constant, bathtub, increasing, decreasing, unimodal).
• The Akaike Information Criterion performed well in selecting the appropriate hazard structure.

Conclusions:

• The flexible parametric model offers a robust alternative to standard survival models.
• The approach is valuable for descriptive cancer epidemiology, particularly in excess hazard modeling.
• The model's adaptability and performance are validated through simulations and real-world data application.