Review Article
Synthesis of genetic association studies for pertinent gene–disease associations requires appropriate methodological and statistical approaches

https://doi.org/10.1016/j.jclinepi.2007.12.011Get rights and content

Abstract

Objective

The aim of the study was to consider statistical and methodological issues affecting the results of meta-analysis of genetic association studies for pertinent gene–disease associations. Although the basic statistical issues for performing meta-analysis are well described in the literature, there are remaining methodological issues.

Study Design and Setting

An analysis of our database and a literature review were performed to assess issues such as departure of Hardy–Weinberg equilibrium, genetic contrasts, sources of bias (replication validity, early extreme contradictory results, differential magnitude of effect in large versus small studies, and “racial” diversity), utility of cumulative and recursive cumulative meta-analyses. Gene–gene–environment interactions and methodological challenges of genome-wide association studies are discussed.

Results

Departures from Hardy–Weinberg equilibrium can be handled using sensitivity analysis or correction procedures. A spectrum of genetic models should be investigated in the absence of biological justification. Cumulative and recursive cumulative meta-analyses are useful to explore heterogeneity in risk effect in time. Exploration of bias leading to heterogeneity provides insight to postulated genetic effects. In the presence of bias, results should be interpreted with caution.

Conclusions

Meta-analysis provides a robust tool to investigate contradictory results in genetic association studies by estimating population-wide effects of genetic risk factors in diseases and explaining sources of bias and heterogeneity.

Introduction

Genetic association studies (GAS) investigate the relationship between gene polymorphisms and disease without requiring information on inheritance, and thus, are conducted on a sample of unrelated cases and controls. This gives the advantage of including large number of patients, providing greater power for statistical analysis compared to other type of studies such as family based studies [1]. GAS have been particularly popular for investigating complex diseases and the number of these papers has increased tremendously, and this trend is expected to accelerate because of their relatively ease compared to family based studies, the availability of mapped single-nucleotide polymorphisms (SNP) and advances in genotyping technologies [2]. However, GAS lack reproducibility and they frequently generate controversial or inconclusive results. These are due to a number of causes related to study design, sample size, and genetic or environmental heterogeneity between populations [3].

In contrast to GAS, the genome-wide association (GWA) studies involve a scan of genomic sequence variants [4], which enables one to examine hundreds of thousands of SNPs in cases and controls. GWA studies represent a comprehensive option to GAS when there is lack of evidence regarding the function or the location of the causal genes. However, there are already examples where initial findings by GWA studies have not been replicated by large-scale studies [5], [6], [7]. This lack of replication might be due to modest sample size to detect weak associations.

Meta-analysis estimates the overall genetic risk effects for pertinent gene–disease associations and it is a robust tool to investigate discrepant results [8]. An important role of a meta-analysis is the exploration of heterogeneity of reported findings in GAS [9]. In addition to conventional meta-analysis, cumulative meta-analysis and recursive cumulative meta-analysis provide a framework in which summary genetic risk effects are re-estimated each time a new study appears [10] to investigate the trend and the stability of risk effect as evidence accumulates.

This article addresses quality and design issues regarding meta-analysis of GAS, and additionally explores the utility of metaregression, and cumulative and recursive cumulative meta-analyses. The article focuses on dichotomous outcomes.

Section snippets

Quality/design issues

Quality and design issues of the original GAS that should be taken into account when conducting a meta-analysis include definition of phenotype, blinding, Hardy–Weinberg equilibrium (HWE) in controls, validity of genotyping method, and population stratification [11].

Meta-analysis statistical approach

The integral parts of a typical meta-analysis are the estimation of a summary metric such as odds ratio (OR) and the exploration of heterogeneity between studies. The heterogeneity between studies in terms of degree of association is tested using the Cochran Q-statistic, which is traditionally considered statistically significant for P < 0.10 [8]. The pooled OR is estimated using fixed effects (ORFE; Mantel–Haenszel) and random effects (ORRE; DerSimonian and Laird) models [8]. Random effects

Evolution of risk effect in time: Cumulative and recursive meta-analyses

Cumulative and recursive cumulative meta-analyses are additional methods to explore heterogeneity in risk effect for a genetic model in time [10]. They provide a framework for updating a genetic effect from all studies and a measure of how much the genetic effect changes as evidence accumulates. Recursive cumulative meta-analysis also predicts major changes in risk effect that may occur in the future.

In cumulative meta-analysis, studies are ordered by a covariate (e.g., publication year), and

Replication validity

In meta-analysis of GAS, it is frequent that the results of the first GAS do not correlate with the results of the subsequent studies, and usually the first studies tend to overestimate the magnitude of an association. This can be due to bias or to genuine diversity of the populations involved in the GAS. The impact of the first study(ies) is evaluated using the z-statistic, which estimates the effect of the first study vs. the remaining studies. The z-statistic is defined as the difference of

Haplotypes

Haplotypes of gene polymorphisms play an important role in investigating gene–disease associations and the interaction of polymorphisms within haplotypes might be a major determinant of disease susceptibility than the individual polymorphism [32]. Summary ORs for haplotypes can be estimated using the expectation-maximization algorithm in combination with log-linear modeling [25]. The expectation-maximization algorithm assigns haplotypes to heterozygous individuals [25].

However, discrepant

Discussion

Although GAS are very popular, they tend to lack the power to detect statistically significant associations. Although some large candidate-gene studies and GWA studies have been undertaken recently, the sample sizes needed to show association will be far beyond what is currently available and no single institution or entity alone will be able to provide a reasonable number of patients for the most complex analyses [1]. Meta-analysis of multiple studies clearly has a role in offering an analysis

References (50)

  • D. Clayton et al.

    Epidemiological methods for studying genes and environmental factors in complex diseases

    Lancet

    (2001)
  • J.N. Hirschhorn et al.

    A comprehensive review of genetic association studies

    Genet Med

    (2002)
  • D.J. Schaid

    Likelihoods and TDT for the case-parents design

    Genet Epidemiol

    (1999)
  • E. Zintzaras et al.

    Trends in meta-analysis of genetic association studies

    J Hum Genet

    (2008)
  • W.Y. Wang et al.

    Genome-wide association studies: theoretical and practical concerns

    Nat Rev Genet

    (2005)
  • N.J. Samani et al.

    Genomewide Association Analysis of Coronary Artery Disease

    N Engl J Med

    (2007)
  • E. Zintzaras et al.

    Heterogeneity testing in meta-analysis of genome searches

    Genet Epidemiol

    (2005)
  • J. Lau et al.

    Cumulative meta-analysis of therapeutic trials for myocardial infarction

    N Engl J Med

    (1992)
  • J. Little et al.

    Reporting, appraising, and integrating data on genotype prevalence and gene-disease associations

    Am J Epidemiol

    (2002)
  • E. Zintzaras et al.

    Association between the GLUT1 gene polymorphism and the risk of diabetic nephropathy: a meta-analysis

    J Hum Genet

    (2005)
  • E. Zintzaras et al.

    Gene polymorphism as a risk factor for diabetic nephropathy: a meta-analysis

    J Hum Genet

    (2007)
  • J. Xu et al.

    Positive results in association studies are associated with departure from Hardy–Weinberg equilibrium: hint for genotyping error?

    Hum Genet

    (2002)
  • E. Zintzaras et al.

    Association of paraoxonase 1 gene polymorphisms with risk of Parkinson's disease: a meta-analysis

    J Hum Genet

    (2004)
  • F. Pompanon et al.

    Genotyping errors: causes, consequences and solutions

    Nat Rev Genet

    (2005)
  • D.G. Clayton et al.

    Population structure, differential bias and genomic control in a large-scale, case-control association study

    Nat Genet

    (2005)
  • Cited by (324)

    View all citing articles on Scopus
    View full text