Cox regression model under dependent truncation.

Area of Science:

• Statistics
• Biostatistics
• Epidemiology

Background:

• Time-to-event studies, particularly in neurodegenerative diseases, often face statistical truncation.
• Standard Cox regression models are inadequate when survival time is subject to left, right, or double truncation.
• Existing methods often assume independence between survival and truncation times, which may not hold true.

Purpose of the Study:

• To develop a novel statistical approach to address truncation in survival analysis.
• To relax the restrictive independence assumption in Cox regression models under various truncation scenarios.
• To accurately assess the impact of risk factors on survival time.

Main Methods:

• Proposed an expectation-maximization algorithm to handle left, right, or double truncation.
• Relaxed the independence assumption to conditional independence on observed covariates.
• Evaluated estimator performance through extensive simulations.

Main Results:

• The proposed expectation-maximization algorithm provides consistent and asymptotically normal regression coefficient estimators.
• Simulations showed the new estimator has minimal bias and competitive or superior mean-squared error compared to existing methods.
• The approach was successfully applied to study occupation's effect on survival in Alzheimer's disease patients.

Conclusions:

• The developed expectation-maximization algorithm offers a robust solution for survival analysis with truncated data.
• This method overcomes limitations of existing approaches by relaxing the independence assumption.
• The findings have significant implications for analyzing risk factors in neurodegenerative disease research and other time-to-event studies.