A pairwise likelihood augmented Cox estimator for left-truncated data.

Area of Science:

• Biostatistics
• Survival Analysis
• Epidemiology

Background:

• Survival data from prevalent cohorts often exhibit left truncation, complicating analysis.
• Existing methods like conditional approaches can be inefficient, while length-biased methods may introduce bias.

Purpose of the Study:

• To develop a robust semiparametric method for analyzing left-truncated survival data under the Cox model.
• To address the limitations of existing methods by efficiently utilizing information from truncation times.

Main Methods:

• A novel semiparametric approach using conditional and pairwise likelihoods to eliminate truncation distribution dependence.
• An iterative algorithm for simultaneous estimation of regression coefficients and baseline hazard function.
• Theoretical validation using empirical process and U-process theories for consistency and asymptotic normality.

Main Results:

• The proposed method demonstrates substantial efficiency gains over the conditional approach for both regression coefficients and cumulative baseline hazard.
• When truncation times are uniform, the estimator shows reduced bias and efficiency comparable to full maximum likelihood.
• The method was successfully applied to a chronic kidney disease cohort study.

Conclusions:

• The proposed semiparametric method offers an efficient and accurate way to analyze left-truncated survival data without assuming truncation time distributions.
• This approach provides a valuable tool for epidemiological studies, particularly those involving prevalent cohorts like chronic kidney disease research.