Multivariate contaminated normal linear mixed models applied to Alzheimer's disease study with censored and missing...

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Clinical Research Methodology

Background:

• Multivariate linear mixed models are standard for repeated measures but struggle with outliers and missing data.
• Existing methods may not adequately handle complex clinical datasets with censored and intermittent missing responses.
• Robust statistical modeling is crucial for accurate analysis of patient data in clinical studies.

Purpose of the Study:

• To develop a robust statistical model for jointly analyzing multiple, complex, repeated clinical measures.
• To extend the capabilities of multivariate linear mixed models by incorporating a multivariate contaminated normal distribution.
• To effectively handle outliers, censored data, and missing responses in longitudinal clinical studies.

Main Methods:

• The proposed multivariate contaminated normal linear mixed model with censored and missing responses (MCNLMM-CM) was developed.
• An expectation conditional maximization (ECM) algorithm was employed for parameter estimation with missing at random responses.
• Techniques for standard error approximation, data recovery, imputation, and outlier identification were provided.

Main Results:

• A simulation study demonstrated the superior finite-sample performance of the MCNLMM-CM compared to existing models.
• The model effectively handles minor outliers, censored measurements, and intermittent missing responses.
• Parameter estimators showed robust properties in the simulation.

Conclusions:

• The MCNLMM-CM offers a robust and effective approach for analyzing complex longitudinal clinical data.
• The methodology is well-suited for datasets with data quality issues like outliers and missingness.
• The model's application to Alzheimer's disease neuroimaging data highlights its practical utility.