Mixture regression models for the gap time distributions and illness-death processes

Area of Science:

• Biostatistics
• Survival Analysis
• Epidemiology

Background:

• Illness-death models are crucial for understanding disease progression where subjects may experience illness before death, or only death.
• Dependent censoring, where the time to censoring is related to the event times, can introduce bias in standard survival analyses.
• Accurate estimation of event times and model parameters is vital for clinical decision-making and understanding disease dynamics.

Purpose of the Study:

• To analyze gap event times within the illness-death model framework, considering both illness-death and death-only pathways.
• To investigate the influence of illness duration on the subsequent death event.
• To develop a robust statistical method that accounts for dependent censoring in successive events.

Main Methods:

• Generalized semiparametric mixture models for competing risks data were extended to include subsequent events.
• A copula function was employed to model the dependency structure between successive events.
• The counting process approach and nonparametric maximum likelihood estimation were used, avoiding the need to estimate censoring time survival functions.

Main Results:

• The proposed method provides consistent estimation of model parameters, mitigating bias introduced by dependent censoring.
• Simulation studies demonstrated the effectiveness and performance of the developed analysis technique.
• The method was successfully applied to a clinical study on chronic myeloid leukemia, showcasing its practical utility.

Conclusions:

• The generalized semiparametric mixture model with copula effectively handles dependent censoring in illness-death models.
• This approach offers a reliable method for analyzing successive events and their dependencies in survival data.
• The findings have significant implications for clinical research, particularly in understanding complex disease trajectories like chronic myeloid leukemia.