Validation of a heteroscedastic hazards regression model

Area of Science:

• Statistics
• Survival Analysis
• Biostatistics

Background:

• Analyzing survival data with crossing hazards presents challenges for standard Cox models.
• The partial likelihood method in Cox regression cannot fully eliminate the baseline hazard function.
• Existing methods may not adequately address heteroscedasticity in survival data.

Purpose of the Study:

• To investigate a Cox-type regression model that accommodates heteroscedasticity for data exhibiting crossing hazards.
• To introduce an overidentified estimating equation (OEE) approach to handle the baseline hazard.
• To develop statistics for assessing model adequacy and testing the proportional hazards assumption.

Main Methods:

• A Cox-type regression model incorporating a power factor for baseline cumulative hazard.
• Development and application of an overidentified estimating equation (OEE) approach for parameter estimation.
• Introduction of a model checking statistic for overall model adequacy.
• Proposal of two statistics to specifically test the proportional hazards assumption under heteroscedasticity.

Main Results:

• The proposed OEE approach effectively estimates parameters in the presence of heteroscedasticity and crossing hazards.
• The developed model checking statistic provides a robust test for overall model fit.
• The proposed statistics successfully detect violations of the proportional hazards assumption.

Conclusions:

• The investigated Cox-type regression model with heteroscedasticity and OEE is suitable for analyzing survival data with crossing hazards.
• The proposed methods offer valuable tools for model diagnostics and assumption checking in survival analysis.
• The approach was successfully illustrated using a cancer clinical trial dataset.