Random-effects meta-analysis of inconsistent effects: a time for change.

Area of Science:

• Biostatistics
• Medical Research Methodology

Background:

• Meta-analysis aims to improve treatment effect estimation by pooling study results.
• The DerSimonian-Laird (DL) estimator is widely used for pooling heterogeneous studies.
• Concerns exist regarding the DL estimator's potential for biased estimates and inaccurate precision.

Purpose of the Study:

• To explain the limitations of the DerSimonian-Laird (DL) estimator in meta-analysis.
• To present alternative methods for summarizing heterogeneous evidence.
• To advocate for a shift towards more critical synthesis in meta-analytic practices.

Main Methods:

• The article reviews the statistical properties of the DL estimator.
• It presents a classic example demonstrating potential biases and erroneous conclusions from the DL estimator.
• Alternative meta-analysis methods for heterogeneous data are discussed.

Main Results:

• The DL estimator can yield biased treatment effect estimates with falsely high precision.
• Use of the DL estimator may lead to incorrect conclusions in meta-analyses.
• Alternative methods and critical synthesis provide more accurate confidence limits.

Conclusions:

• Universal reliance on the DL estimator should be replaced.
• Analyses should critically synthesize evidence, acknowledging uncertainty.
• Random-effects estimates are recommended for more accurate confidence intervals in heterogeneous meta-analyses.