A novel robust meta-analysis model using the t distribution for outlier accommodation and detection

Area of Science:

• Biostatistics
• Statistical Modeling
• Medical Research Synthesis

Background:

• Standard random effects meta-analysis models assume normal distributions, making them vulnerable to outlying studies.
• Existing robust methods using the t distribution for random effects are computationally complex.

Purpose of the Study:

• To propose a novel robust meta-analysis model, tMeta, that utilizes the t distribution for improved outlier accommodation and detection.
• To develop a computationally efficient estimation algorithm for the proposed model.

Main Methods:

• Introduced tMeta, a robust meta-analysis model where the marginal distribution of effect sizes follows a t distribution.
• Developed a simple and fast EM-type algorithm for maximum likelihood estimation.
• Leveraged the mathematical tractability of the t distribution to avoid numerical integration and enable efficient optimization.

Main Results:

• tMeta demonstrated favorable performance compared to existing methods when dealing with mild outliers in real data.
• The tMeta model maintained consistent and robust performance even in the presence of gross outliers, where other methods failed.

Conclusions:

• The proposed tMeta model offers a simple, adaptive, and robust approach to meta-analysis.
• tMeta effectively accommodates and detects outlying studies, improving the reliability of integrated results.
• The computational efficiency of tMeta makes it a practical tool for biostatistical applications.