Nonparametric estimation in an illness-death model with component-wise censoring.

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Survival Analysis

Background:

• Composite endpoints are frequently used in clinical trials for diseases with risks of both death and serious nonfatal events.
• Standard survival analysis methods often fail to adequately address component-wise censoring, a common issue when nonfatal events are detected only at clinic visits.

Purpose of the Study:

• To develop a nonparametric estimator for event-free survival that correctly accounts for component-wise censoring.
• To propose and analyze estimators for state probabilities and restricted mean time in state within illness-death models under component-wise censoring.

Main Methods:

• A kernel smoothing method, originally for marker processes, is adapted for intermittently observed time-dependent binary variables.
• The proposed method treats nonfatal event status as an intermittently observed binary variable, differing from traditional interval-censored approaches.
• Large-sample properties of the proposed estimators are derived.

Main Results:

• The novel kernel smoothing method provides a more accurate nonparametric estimation of event-free survival in the presence of component-wise censoring.
• Simulation studies demonstrate the method's performance compared to existing multistate survival techniques.
• The methods were successfully applied to data from a large randomized trial in men at high risk for coronary heart disease.

Conclusions:

• The adapted kernel smoothing method offers a robust solution for analyzing composite endpoints with component-wise censoring in clinical trials.
• This approach enhances the accuracy of survival estimations, particularly in complex disease settings.
• The study provides valuable tools for biostatisticians and researchers conducting clinical trials with similar data structures.