Conditional modeling of panel count data with partly interval-censored failure event.

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Survival Analysis

Background:

• Panel count data in longitudinal studies often involve recurrent events.
• Partly interval-censored failure events can provide informative data on recurrent events.
• Existing methods using latent variable models offer indirect interpretations of failure event effects.

Purpose of the Study:

• To propose a novel statistical model for panel count data with informative, partly interval-censored failure events.
• To develop an estimation procedure that provides direct interpretation of failure event effects.
• To address limitations in existing statistical methods for recurrent event data analysis.

Main Methods:

• A failure-time-dependent proportional mean model with an unspecified link function was developed.
• A two-stage estimation procedure using conditional expectation of least squares was employed.
• B-spline functions were used to approximate unknown baseline mean and link functions, treating failure time distribution as a nuisance parameter.

Main Results:

• The proposed method allows for direct interpretation of the failure event's effect on recurrent events.
• Theoretical derivations established the convergence rate and asymptotic normality of the estimators.
• Extensive simulation studies confirmed the finite-sample performance aligns with theoretical results.

Conclusions:

• The developed statistical model and estimation procedure effectively handle panel count data with informative, partly interval-censored failure events.
• The method offers a more direct and interpretable way to analyze the impact of failure events on recurrent processes.
• The approach was successfully illustrated in a longitudinal healthy longevity study, yielding insightful conclusions.