Recent findings from genome-wide association studies have demonstrated their considerable potential for identifying genetic determinants of common diseases of public health significance such as cancer, heart disease, and diabetes (1), but they have also highlighted the continued importance of targeted genotyping to replicate genome-wide association findings (2). Approaches to the integration of evidence in human genome epidemiology have evolved rapidly in the last few years. The combination of results from multiple studies, often known as meta-analysis, has a key role both in enhancing power and in characterizing relative risks (3). As evidence accumulates on genetic variants that confer identifiable effects on disease susceptibility, so does the need to summarize the evidence in digestible and accessible formats. Here, we describe how the Human Genome Epidemiology Network (HuGENet) is keeping abreast of developments in methods for collating and synthesizing the evidence.
HuGENet was established in 1998 to integrate epidemiologic evidence on the role of genetics in human health and disease, and to develop an online searchable, updated, knowledge base (4). HuGENet's main activities are compilation and evaluation of epidemiologic research, facilitating of collaborations, training and technical assistance, and information exchange through the World Wide Web. A “road map” for human genome epidemiology outlines a vision for the future of this important field (5), and activities of the network are now facilitated by four coordinating centers in Atlanta, Georgia (6); Cambridge, United Kingdom (7); Ottawa, Canada (8); and Ioannina, Greece (9).
An important part of the HuGENet initiative is conducting “Human Genome Epidemiology (HuGE) reviews” on genotype-disease associations, including joint effects of genes and of genes with environmental exposures (10, 11). Indeed, HuGENet's new logo (figure 1) highlights the central role of gene-environment (GE) interactions in predisposition to disease. HuGE reviews are typically systematic, aiming to identify, appraise, and synthesize evidence from all relevant existing studies on the topic in question (12). Regular readers will have noticed an increasing number of HuGE reviews in the American Journal of Epidemiology, and nine additional journals have agreed to be publication venues for these systematic reviews (6). HuGE reviews may also be accessed from the HuGENet website; the 62 HuGE reviews published as of June 1, 2007, have covered a wide array of topics ranging from rare, single-gene disorders such as neurofibromatosis to common conditions such as preterm birth, cancer, and heart disease (6).
The HuGE Review Handbook (13) is an evolving, online document that offers guidance to researchers undertaking HuGE reviews. It is inspired partly by the Cochrane Handbook for Systematic Reviews of Interventions (14). The Cochrane Collaboration undertakes systematic reviews of the effects of health-care interventions and has published more than 2,500 such reviews to date (15). Cochrane reviews implement rigorous methods in an attempt to minimize bias either from individual studies or during the review process, and similar rigor is being used in HuGE reviews. The Handbook will be updated over time as methodology and understanding develop.
The Handbook resulted from a methodology workshop held in Cambridge, United Kingdom, in November 2004. The workshop brought together epidemiologists, geneticists, statisticians, and other health-care researchers to develop methodological guidance for authors of systematic reviews and meta-analyses in human genome epidemiology, and to identify any potential developments that could improve their validity. Before this workshop, the original (4) and updated (11) guidelines for HuGE reviews did not specify in detail the recommended methods for searching published and unpublished literature, analyzing data, or synthesizing information. Furthermore, the initial concept of a “full” HuGE review—to cover prevalence, association, interaction, and implications for genetic testing and public health (16)—was relatively broad in scope. Thus, early HuGE reviews varied in their methodology and particularly in the application of formal meta-analytic methods. This reflected concern about the application of meta-analysis to observational studies (16, 17). However, meta-analysis has become widely applied and accepted in human genome epidemiology in recent years (18, 19), and all but four of the 17 HuGE reviews published since the beginning of 2006 include a formal meta-analysis. Furthermore, over 500 meta-analysis articles have already been published in this field outside the HuGENet effort (6). With the advent of genome-wide association studies, it has become common practice that prospective validation of identified variants through combined analysis (or meta-analysis) of data from multiple teams is accomplished as part of the very first publication of the new data (3). Meta-analysis of genome-wide association studies themselves is also increasingly applied (20, 21).
Some key recommendations in the HuGE Review Handbook for improving the methodology of HuGE reviews include the following:
Encouraging consortia of primary research investigators as the most reliable approach for performing combined analyses or meta-analyses (based on individual participant data) (22)
Adopting methods to minimize human error in the literature-based reviews, such as duplicating selection of studies and data extraction
Conducting comprehensive (yet practically realistic) searches for eligible studies, considering sources beyond MEDLINE (National Library of Medicine, Bethesda, Maryland)
Considering in more detail the potential for bias in individual studies and in the total body of available evidence (17)
Encouraging quantitative synthesis of results from multiple studies (meta-analysis) where appropriate (23)
Encouraging incorporation of intermediate phenotypes (such as molecular markers) so that “Mendelian randomization” can be exploited to examine the causal effects of such phenotypes (24)
Meta-analyses can offer both enhanced power to detect associations and increased precision of estimates of its magnitude. Consistency of findings across studies can be formally assessed and heterogeneity explored. Of course, the potential for selective availability of findings on the basis of their statistical significance must always be borne in mind. It is essential that the scientific community continues to progress toward making all findings, positive and negative, available to all. Registers of DNA collections (akin to existing registers of randomized controlled trials (25)) and online repositories for negative results would go some way toward realizing the vision of an unbiased and data-rich environment within which to evaluate gene-disease associations. Genome-wide association investigations offer a unique opportunity for full, transparent availability of detailed databases to other researchers, such as those already adopted by the Wellcome Trust Case Control Consortium (26), the National Institutes of Health's Database of Genotype and Phenotype (dbGaP) (27), and the European Genotype Archive (28). Wherever possible, we encourage the development of consortia of investigators to analyze individual participant data on at least a retrospective basis and, ideally, a prospective basis.
The fast pace of development in the field creates new challenges such as the need to continually revisit the inferences of meta-analyses and the definition of replication in the context of massive testing ability (29, 30). Inferences on the cumulative evidence on genetic associations may change over time. As part of a HuGENet initiative, interim guidelines have been developed to assess the epidemiologic strength of the cumulative evidence (31). We suggest that these guidelines be applied to the final inference for each HuGE review and other meta-analyses or field synopses (6). We expect the Handbook to be a dynamic enterprise that will be regularly updated to recognize consensus on best methods for new challenges, as these arise. We encourage others interested in human genome epidemiology to contribute to this process and to contact us via one of the websites listed below:
HuGENet United States (main site): www.cdc.gov/genomics/hugenet
HuGENet Canada: www.hugenet.ca
HuGENet Greece: www.dhe.med.uoi.gr/hugenet.htm
HuGENet United Kingdom: www.hugenet.org.uk
A decade ago, Shpilberg et al. declared that “the sequencing of the human genome offers the greatest opportunity for epidemiology since John Snow discovered the Broad Street Pump” (32, p. 637). With the completion of the Human Genome Project in 2003, the era of developing the handle for the pump formally began (16). With the emergence of new tools now available to epidemiologists, we hope the pump handle will begin to turn, albeit slowly, to uncover the secrets of gene-environment interactions in common human diseases.
Human Genome Epidemiology
Human Genome Epidemiology Network
The findings and conclusions in this report are those of the authors and do not necessarily represent the views of the Centers for Disease Control and Prevention.
Conflict of interest: none declared.