Likelihood-based meta-analysis with few studies: Empirical and simulation studies

TL;DR

This paper evaluates likelihood-based meta-analysis methods for few studies, highlighting the limitations of traditional approaches and demonstrating Bayesian methods as a promising alternative through empirical and simulation analyses.

Contribution

It provides a comprehensive comparison of frequentist and Bayesian meta-analysis methods for small sample sizes, proposing Bayesian approaches as more reliable in such settings.

Findings

Frequentist methods often have below-nominal coverage with few studies.

Bayesian methods achieve better coverage and narrower credible intervals.

Caution is advised when applying meta-analysis to few, unbalanced studies due to coverage issues.

Abstract

Standard random-effects meta-analysis methods perform poorly when applied to few studies only. Such settings however are commonly encountered in practice. It is unclear, whether or to what extent small-sample-size behaviour can be improved by more sophisticated modeling. We consider several likelihood-based inference methods. Confidence intervals are based on normal or Student-t approximations. We extract an empirical data set of 40 meta-analyses from recent reviews published by the German Institute for Quality and Efficiency in Health Care (IQWiG). Methods are then compared empirically as well as in a simulation study, considering odds-ratio and risk ratio effect sizes. Empirically, a majority of the identified meta-analyses include only 2 studies. In the simulation study, coverage probability is, in the presence of heterogeneity and few studies, below the nominal level for all frequentist methods based on normal approximation, in particular when sizes in meta-analyses are not balanced, but improve when confidence intervals are adjusted. Bayesian methods result in better coverage than the frequentist methods with normal approximation in all scenarios. Credible intervals are empirically and in the simulation study wider than unadjusted confidence intervals, but considerably narrower than adjusted ones. Confidence intervals based on the generalized linear mixed models are in general, slightly narrower than those from other frequentist methods. Certain methods turned out impractical due to frequent numerical problems. In the presence of between-study heterogeneity, especially with unbalanced study sizes, caution is needed in applying meta-analytical methods to few studies, as either coverage probabilities might be compromised, or intervals are inconclusively wide. Bayesian estimation with a sensibly chosen prior for between-trial heterogeneity may offer a promising compromise.