EM algorithm for the additive risk mixture cure model with interval-censored data.

Area of Science:

• Biostatistics
• Survival Analysis
• Statistical Modeling

Background:

• Interval-censored failure time data are common across various scientific fields.
• Analyzing such data, especially with non-ignorable cure fractions under additive risk models, presents significant computational challenges due to complex likelihood maximization.

Purpose of the Study:

• To develop a robust statistical estimation method for regression analysis of interval-censored data with a cure fraction.
• To address the computational complexity associated with the additive risk model in the presence of non-ignorable cures.

Main Methods:

• A sieve maximum likelihood estimation (SMLE) approach utilizing Bernstein polynomials was developed.
• An expectation-maximization (EM) algorithm, incorporating Poisson data augmentation, was employed to reduce computational burden.
• Asymptotic properties of the proposed estimator were theoretically established under mild conditions.

Main Results:

• The proposed sieve maximum likelihood estimation method provides a computationally feasible approach for interval-censored data.
• The expectation-maximization algorithm effectively handles the complexity introduced by the cure fraction.
• Extensive simulations demonstrated the method's good finite sample performance.

Conclusions:

• The developed method offers a viable solution for regression analysis of interval-censored failure time data with non-ignorable cure fractions.
• The approach is computationally efficient and statistically sound, as evidenced by theoretical properties and simulation studies.
• The method's utility is further confirmed by its application to a real-world smoking cessation study dataset.