On robustness and model flexibility in survival analysis: transformed hazard models and average effects

Area of Science:

• Statistics
• Survival Analysis
• Biometrics

Background:

• Box-Cox transformed hazard models offer flexibility but reduce parameter interpretability.
• The interpretation of coefficients is linked to the transformation parameter, which can be poorly defined in some datasets.

Purpose of the Study:

• To develop more interpretable and robust inferential targets for transformed hazard models.
• To investigate the utility of average effects in survival data analysis.

Main Methods:

• The study proposes using average effects, calculated by averaging partial derivatives of the hazard or log-hazard.
• These average effects are considered for both fixed-form models (proportional or additive hazards) and transformed hazard models.

Main Results:

• Average effects can serve as robust inferential targets, even if the specific covariate effect form is misspecified.
• When applied to transformed hazard models, average effects are more interpretable and potentially more efficiently estimated than regression coefficients.

Conclusions:

• Average effects enhance the interpretability and efficiency of inferences in survival analysis.
• This approach offers a valuable alternative for analyzing complex hazard models, particularly when parameter interpretability is a concern.