Approximations for trimmed Fisher procedures in research synthesis

Area of Science:

• Biostatistics
• Medical Research Methodology

Background:

• Meta-analyses often include studies with aberrant quality or effect sizes.
• Aberrant studies, typically at the extremes, may require trimming to improve analysis robustness.
• P-values can indicate aberrant studies, but trimming alters underlying statistical distributions.

Purpose of the Study:

• To discuss the exact distribution of a trimmed Fisher procedure for meta-analysis.
• To present approximations for the complex exact distribution of trimmed statistics.
• To apply these methods to a real-world meta-analysis concerning drug dosage and mortality risk.

Main Methods:

• Developed and analyzed the exact distribution of the trimmed Fisher procedure.
• Derived and evaluated several approximation methods for the trimmed distribution.
• Applied the proposed methodology to a meta-analysis dataset.

Main Results:

• The exact distribution of the trimmed Fisher procedure was mathematically derived.
• Approximation methods were presented to handle the complexity of the exact distribution.
• The methods were successfully applied to a meta-analysis on drug dose and mortality.

Conclusions:

• The trimmed Fisher procedure provides a method for handling aberrant studies in meta-analysis.
• Approximation techniques are essential due to the complexity of the exact trimmed distribution.
• This approach enhances the reliability of meta-analyses, as demonstrated in the mortality risk example.