A closed-form estimator for meta-analysis and surrogate markers evaluation.

Area of Science:

• Statistics
• Biostatistics
• Computational Statistics

Background:

• Iterative full maximum likelihood estimators for linear mixed models face convergence issues with small, unbalanced datasets and become computationally prohibitive for large datasets.
• Existing methods present challenges in efficiently estimating complex linear mixed models, hindering their application in various statistical contexts.

Purpose of the Study:

• To introduce an unbiased two-stage closed-form estimator for multivariate linear mixed models.
• To address the computational limitations and convergence problems associated with traditional iterative estimation methods.
• To provide a computationally efficient and statistically sound alternative for estimating linear mixed models.

Main Methods:

• Developed a novel two-stage closed-form estimator based on pseudo-likelihood-based split-sample methodology.
• Assessed the statistical and computational performance of the proposed estimator through simulation studies.
• Applied the new method to a real-world dataset from a study in schizophrenia.

Main Results:

• The proposed closed-form estimator demonstrates effective performance in statistical and computational aspects.
• The method successfully overcomes common convergence problems encountered with iterative estimators.
• The estimator provides a viable alternative for analyzing normally distributed endpoints in meta-analytic contexts and beyond.

Conclusions:

• The novel two-stage closed-form estimator provides a computationally efficient and unbiased solution for multivariate linear mixed models.
• This method offers a practical alternative to iterative estimation, particularly for challenging datasets.
• The estimator has broad applicability, including meta-analysis and the analysis of schizophrenia data.