On the estimation of interval censored destructive negative binomial cure model

Area of Science:

• Biostatistics
• Survival Analysis
• Statistical Modeling

Background:

• Competitive risk survival models are crucial for understanding diseases with multiple outcomes.
• Handling interval-censored data and missing risk information presents significant analytical challenges.
• Existing models often struggle with complex data structures involving destructive mechanisms on initial risk counts.

Purpose of the Study:

• To develop and evaluate novel estimation algorithms for a competitive risk survival model with a destructive mechanism.
• To address challenges posed by interval-censored data and missing information on initial and remaining risks.
• To compare the performance of an Expectation Maximization (EM) algorithm and its stochastic variation (SEM) for parameter recovery.

Main Methods:

• Development of two distinct estimation algorithms: an Expectation Maximization (EM) algorithm and a Stochastic EM (SEM) algorithm.
• Utilizing conditional distributions of missing data to decompose and maximize the expected complete log-likelihood.
• Conducting a Monte Carlo simulation study to assess bias, root mean square error, and confidence interval coverage probability.

Main Results:

• The SEM algorithm demonstrated superior performance in parameter recovery compared to the standard EM algorithm.
• Simulation results indicated that the SEM algorithm offers improved accuracy and reliability for the proposed survival model.
• The study successfully applied the SEM algorithm and destructive model to real-world data from a children's mortality study.

Conclusions:

• The stochastic EM (SEM) algorithm is the preferred method for analyzing this type of complex survival data.
• The proposed destructive risk model and SEM algorithm provide a robust framework for handling interval-censored data with missing risk information.
• The findings have significant implications for epidemiological studies, particularly in understanding disease progression and risk factor dynamics, as exemplified by the children's mortality data analysis.