The binomial distribution of meta-analysis was preferred to model within-study variability

Area of Science:

• Biostatistics
• Medical Research Methodology

Background:

• Meta-analysis of study proportions (e.g., sensitivity, specificity) commonly uses the DerSimonian and Laird random effects model.
• This approximate method can introduce bias due to its normal distribution assumption for within-study variability.

Purpose of the Study:

• To compare the performance of the standard approximate method with an exact likelihood approach for meta-analyzing proportions.
• To evaluate bias, mean-squared error, and coverage probabilities of both methods.

Main Methods:

• A simulation study was conducted, varying key parameters like overall proportion, between-studies variance, within-study sample sizes, and number of studies.
• The methods were illustrated using a published meta-analysis dataset.

Main Results:

• The exact likelihood approach consistently outperformed the approximate method, yielding unbiased estimates.
• The exact method demonstrated acceptable coverage probabilities, particularly the profile likelihood.
• The approximate approach resulted in significant bias and poor coverage in numerous scenarios.

Conclusions:

• The exact likelihood approach is the preferred method for meta-analyzing proportions due to its superior performance.
• Researchers should utilize the exact likelihood method whenever feasible for more reliable meta-analytic results.