An approach to estimation in relative survival regression

Area of Science:

• Biostatistics
• Epidemiology
• Survival Analysis

Background:

• Relative survival methodology compares cohort survival to the general population.
• Additive excess hazard models are commonly used, assuming proportional hazards for covariates.
• Existing methods may be biased due to misspecification of the baseline excess hazard.

Purpose of the Study:

• Introduce a novel expectation-maximization (EM) algorithm for relative survival analysis.
• Develop a flexible method that avoids assumptions about the baseline excess hazard.
• Generalize the Cox model for broader application in relative survival studies.

Main Methods:

• Utilize the expectation-maximization algorithm, treating cause of death as missing data.
• Employ a flexible estimation of the baseline excess hazard without additional degrees of freedom.
• Accommodate partial knowledge of cause of death for subjects.

Main Results:

• The new method reduces bias risk from misspecified baseline excess hazards.
• It provides estimates of the ratio between excess and population hazards for each subject.
• Demonstrates flexibility in estimating the baseline excess hazard, outperforming existing methods in a myocardial infarction survival dataset.

Conclusions:

• The proposed EM algorithm offers a robust and flexible approach to relative survival analysis.
• This generalization of the Cox model allows leveraging existing software capabilities.
• The method can reveal survival function forms potentially missed by traditional approaches.