Efficient Hybrid EM for Linear and Nonlinear Mixed Effects Models with Censored Response

Area of Science:

• Biostatistics
• Pharmacometrics
• Computational Biology

Background:

• Medical laboratory data frequently exhibit censoring due to technological limitations, such as quantification limits in assays.
• Accurate analysis of censored data is crucial for pharmacokinetic measurements and understanding disease progression, like HIV particle concentration.
• Linear and nonlinear mixed-effects models are standard but require adaptations for censored data.

Purpose of the Study:

• To present a novel hybrid Monte Carlo and numerical integration Expectation-Maximization (EM) algorithm.
• To compute maximum likelihood estimates for mixed-effects models with censored data efficiently.
• To address challenges in analyzing medical data with lower and upper quantification limits.

Main Methods:

• Developed a hybrid EM algorithm combining Monte Carlo simulation and numerical integration.
• Implemented efficient block-sampling, automated convergence monitoring, and QR decomposition for dimension reduction.
• Utilized numerical integration for clusters with up to two censored observations, replacing Monte Carlo simulation.

Main Results:

• Achieved a several-fold reduction in computation time compared to existing methods.
• Demonstrated the algorithm's effectiveness using data from an HIV/AIDS clinical trial.
• Simulation studies confirmed the performance and advantages of the proposed Monte Carlo EM approach.

Conclusions:

• The hybrid EM algorithm provides a computationally efficient solution for mixed-effects models with censored data.
• This method enhances the analysis of challenging medical datasets, particularly in pharmacokinetics and infectious disease research.
• The developed algorithm offers a significant improvement for statistical modeling in the presence of quantification limits.