New models for describing outliers in meta-analysis

Area of Science:

• Statistics
• Biostatistics
• Meta-analysis

Background:

• Between-study variation in meta-analysis is typically modeled using a normally distributed random effect.
• Outliers and unusual estimates can challenge conventional random effect models.
• Alternative distributions have been proposed for heterogeneous datasets.

Purpose of the Study:

• To propose two new marginal distributions for modeling heterogeneous meta-analysis datasets.
• To offer a more computationally robust alternative to hierarchical models.
• To avoid the need for numerical integration in likelihood evaluation.

Main Methods:

• Development of two novel marginal distributions for random effects in meta-analysis.
• Focus on marginal distributions to bypass direct modeling of study-specific effects.
• Evaluation of computational robustness by avoiding numerical integration.

Main Results:

• The proposed marginal distributions provide a viable alternative for heterogeneous data.
• The methodology demonstrates improved computational robustness compared to traditional approaches.
• Application to four diverse datasets showcases the utility of the new distributions.

Conclusions:

• The new marginal distributions offer a robust and computationally efficient method for meta-analysis.
• These distributions are particularly beneficial when dealing with heterogeneous datasets and potential outliers.
• The proposed approach enhances the reliability of meta-analysis when standard assumptions are violated.