Integrating relative survival in multi-state models-a non-parametric approach

Area of Science:

• Biostatistics
• Survival Analysis
• Epidemiology

Background:

• Multi-state models extend survival analysis by incorporating intermediate events like relapse and remission.
• Accurately distinguishing disease-specific mortality from general population mortality is crucial, especially when cause of death is uncertain.

Purpose of the Study:

• To extend multi-state models using relative survival to differentiate disease-related and non-disease-related mortality.
• To develop a method for datasets lacking precise cause-of-death information, incorporating intermediate events.

Main Methods:

• Integration of population mortality tables into a non-parametric estimation framework.
• Utilizing the relative survival concept where total mortality hazard is a sum of population and excess hazards.
• Development of estimators for transition hazards and probabilities, including variance estimation and confidence intervals.

Main Results:

• A novel non-parametric approach for estimating transition hazards and probabilities in multi-state models with population mortality.
• Demonstration of the method's behavior through simulation studies.
• Successful illustration of the methodology using a cohort of patients post-allogeneic hematopoietic stem cell transplantation.

Conclusions:

• The proposed methodology provides a robust way to analyze mortality components in complex survival data.
• This extension of multi-state models is valuable for medical research, particularly when cause of death is not definitively recorded.
• The method has been implemented in the R package mstate for broader application.