Semiparametric regression analysis of panel binary data with a dependent failure time.

Area of Science:

• Biostatistics
• Health Research Methodology
• Survival Analysis

Background:

• Panel binary data with recurrent events are common in health research.
• Existing methods often fail to account for dependent failure times (e.g., death) that truncate observation windows.
• Analysis of hospitalization data highlights the need for methods accommodating both recurrence and failure time.

Purpose of the Study:

• To propose a novel semiparametric joint-modeling procedure for analyzing panel binary data with dependent failure times.
• To address limitations of generalized linear models and existing literature in handling recurrent events and failure times simultaneously.
• To provide a robust statistical framework for health and clinical research involving longitudinal event data.

Main Methods:

• Developed a semiparametric joint-modeling approach.
• Implemented a computationally efficient Expectation-Maximization (EM) algorithm for model fitting.
• Provided theoretical guarantees for consistency and asymptotic normality of estimates, enabling valid statistical inferences.

Main Results:

• The proposed EM algorithm provides computationally efficient model fitting.
• Estimates derived from the method are shown to be consistent and asymptotically normal.
• Simulation studies validated the method's performance in practical health research scenarios.

Conclusions:

• The developed joint-modeling procedure effectively analyzes panel binary data with dependent failure times.
• The method offers a statistically sound approach for identifying risk factors in longitudinal health studies.
• This work advances the analysis of recurrent event data in the presence of competing risks, exemplified by hospitalization data analysis.