Elsevier

Contemporary Clinical Trials

Volume 45, Part A, November 2015, Pages 139-145
Contemporary Clinical Trials

Meta-analysis in clinical trials revisited

https://doi.org/10.1016/j.cct.2015.09.002Get rights and content

Abstract

In this paper, we revisit a 1986 article we published in this Journal, Meta-Analysis in Clinical Trials, where we introduced a random-effects model to summarize the evidence about treatment efficacy from a number of related clinical trials. Because of its simplicity and ease of implementation, our approach has been widely used (with more than 12,000 citations to date) and the “DerSimonian and Laird method” is now often referred to as the ‘standard approach’ or a ‘popular’ method for meta-analysis in medical and clinical research. The method is especially useful for providing an overall effect estimate and for characterizing the heterogeneity of effects across a series of studies. Here, we review the background that led to the original 1986 article, briefly describe the random-effects approach for meta-analysis, explore its use in various settings and trends over time and recommend a refinement to the method using a robust variance estimator for testing overall effect. We conclude with a discussion of repurposing the method for Big Data meta-analysis and Genome Wide Association Studies for studying the importance of genetic variants in complex diseases.

Introduction

Three decades ago in this Journal (formerly titled Controlled Clinical Trials), we proposed a simple non-iterative method to integrate the findings from a number of related clinical trials to evaluate the efficacy of a certain treatment for a specified medical condition [1]. Our approach, the random-effects model for meta-analysis, now commonly referred to as the “DerSimonian and Laird method”, has become extremely popular in medical research and other applications. According to the Web of Science Core Collection, there are more than twelve thousand citations attributed to the article with a substantial proportion of them occurring in the more recent years.

Following the introduction of the term meta-analysis in 1976 [2] and before the publication of our article, Meta-analysis in Clinical Trials, in 1986 [1], Web of Science lists 222 articles with the term meta-analysis in the title. Almost all of these were in social sciences with 50% in psychology, 32% in education research and 10% in business economics. In contrast, a large proportion of the more than 46,000 articles since 1986 with the term meta-analysis in the title are related to medical or clinical research.

In this paper, we first review the background and the setting that led to Meta-Analysis in Clinical Trials [1] and the “DerSimonian and Laird method”, briefly describe the random-effects model for meta-analysis, assess its use in various settings and trends over time and explore the reasons for its popularity in medical and clinical research. We recommend a refinement to the method using an improved variance estimator for testing an overall effect and conclude with a discussion of repurposing the method for genetic association studies and Big Data meta-analysis.

Section snippets

Background

Eugene Glass first coined the phrase meta-analysis in 1976 to mean the statistical analysis of the findings of a collection of individual studies [2]. In the following decade, meta-analysis was primarily used in the social sciences to summarize the results of a large number of studies on many behavioral, educational and psychosocial studies and experiments. For instance, Rosenthal [3] analyzed accumulating data from studies done by others to assess if a teacher's expectations can influence a

Meta-Analysis in clinical trials

For Meta-Analysis in Clinical Trials [1], we adopted this same random-effects approach to integrate the findings from a number of related clinical trials to strengthen the evidence for the efficacy of a certain treatment for a specified medical condition. The basic idea of the approach is the same, but here we assumed the treatment effect for the i-th study, Yi, was the difference in Binomial cure rates between a treated and control group. Assuming the two groups are independent, the variance, s

Acknowledgments

We thank Jelena Follweiler for her technical assistance and Jing Wang for her help with the citation graphs.

References (32)

  • T.C. Board

    Research notes: coaching and the SAT® I

  • W.G. Cochran

    The combination of estimates from different experiments

    Biometrics

    (1954)
  • E. Kontopantelis et al.

    Performance of statistical methods for meta-analysis when true study effects are non-normally distributed: a simulation study

    Stat. Methods Med. Res.

    (2012)
  • E. Kontopantelis et al.

    Performance of statistical methods for meta-analysis when true study effects are non-normally distributed: a comparison between DerSimonian-Laird and restricted maximum likelihood

    Stat. Methods Med. Res.

    (2012)
  • D. Jackson et al.

    Extending DerSimonian and Laird's methodology to perform multivariate random-effects meta-analyses

    Stat. Med.

    (2010)
  • D. Jackson et al.

    A matrix-based method of moments for fitting the multivariate random effects model for meta-analysis and meta-regression

    Biom. J.

    (2013)
  • Cited by (1763)

    View all citing articles on Scopus
    View full text