Multivariate meta-analysis: a robust approach based on the theory of U-statistic

Area of Science:

• Biostatistics
• Statistical Methodology

Background:

• Multivariate meta-analysis synthesizes multiple correlated outcomes from similar studies.
• Restricted maximum likelihood (REML) is common but assumes normality and is computationally intensive.
• Multivariate method of moments (MMM) performs similarly to REML but its performance with non-normal data is unclear.

Purpose of the Study:

• To propose a new nonparametric, non-iterative U-statistic method for multivariate meta-analysis.
• To compare the performance of U-statistic, REML, and MMM under normal and skewed data distributions.
• To evaluate the robustness of these methods to deviations from normality.

Main Methods:

• Developed a novel U-statistic-based approach for multivariate meta-analysis.
• Conducted simulation studies comparing U-statistic, REML, and MMM with normal and skewed data.
• Applied the methods to real-world meta-analysis data from hip fracture and periodontal disease studies.

Main Results:

• REML estimates showed only marginal effects from non-normal data distributions.
• MMM and U-statistic-based approaches yielded very similar estimates.
• The U-statistic method demonstrated comparable performance to REML and MMM.

Conclusions:

• The U-statistic estimation procedure is a robust and viable alternative for multivariate meta-analysis.
• This method offers advantages in terms of non-parametric nature and non-iterative estimation.
• Future research can extend U-statistic for heterogeneity testing and meta-regression.