Arcsine test for publication bias in meta-analyses with binary outcomes

Area of Science:

• Biostatistics
• Medical Research Methodology
• Evidence Synthesis

Background:

• Small study effects, potentially due to publication bias, can inflate treatment effects in meta-analyses.
• Existing statistical tests for publication bias in binary outcomes (log-risk ratio, log-odds ratio) are prone to excessive false positives due to effect measure and standard error interdependence.

Purpose of the Study:

• To propose and evaluate novel statistical tests for detecting publication bias in meta-analyses of binary data.
• To address the limitations of current tests by utilizing a variance-stabilizing arcsine transformation.

Main Methods:

• Development of new tests for small study effects based on the arcsine transformation of binomial random variables.
• Simulation study under the Copas model (log OR scale) to compare the performance of proposed tests against existing methods.
• Evaluation of test size, power, and ability to handle zero events and heterogeneity.

Main Results:

• One proposed arcsine test demonstrates comparable statistical size to the best existing tests.
• The arcsine tests exhibit slightly greater power, particularly for small effect sizes and in the presence of heterogeneity.
• Arcsine tests can accommodate trials with zero events in both arms and are compatible with existing regression software.

Conclusions:

• The proposed arcsine transformation tests offer a robust and more powerful alternative for detecting publication bias in meta-analyses with binary outcomes.
• These new methods improve upon existing techniques by stabilizing variance and handling zero-event trials, enhancing the reliability of evidence synthesis.