Answers are grounded in this paper and cite their sources.
Received for publication February 19, 2003; accepted for publication February 27, 2003. “The sequencing of the human genome offers the greatest opportunity for epidemiology since John Snow discovered the Broad Street Pump” (1, p. 637). A half century since the publication of the structure of the DNA molecule (2), the sequence of the human genome is complete. Some expect this achievement to be translated into advances in medicine and public health relatively rapidly (3–6). Proponents of this approach tend to focus on the potential to tailor primary prevention, secondary prevention, or therapy on the basis of genetic information. It is also possible that a better understanding of genetic effects and gene-environment interactions in disease processes will allow us to develop better interventions, such as avoidance of defined exposures and chemoprevention, to apply to the general population (7). Others are more skeptical (8–13). A key element in translation will be the application of epidemiologic studies to evaluate the role of genetic variants in the etiology of human disease (1). There has been a tremendous increase in the number and scope of peer-reviewed articles on human genome epidemiology, which has generated in turn the challenge of integrating these data, and this has led to many reviews of gene-disease associations and to gene-gene and gene-environment interaction in response. The purpose of this article is to provide an overview of the experience gained in integrating evidence in the Human Genome Epidemiology (HuGE) reviews and to suggest changes that may encourage more investigators to contribute reviews and to respond to changes in the character of the evidence. HuGE reviews were established as a means of integrating evidence from human genome epidemiologic studies, that is, population-based studies of the impact of human genetic variation on health and disease (14). HuGE reviews are systematic, peer-reviewed synopses of epidemiologic aspects of human genes, including prevalence of allelic variants in different populations, population-based information on disease risk, evidence for gene-environment interaction, and quantitative data on genetic tests and services. They are carried out according to specified guidelines (15). As of February 2003, 20 such reviews have been published (table 1), many of which were in the Journal. Five of these were concerned with gene variants associated with a high risk of disease (16–20). The other 15 have concerned common complex disorders of childhood or adult life. A further 30 reviews are in preparation (21). We now consider what we have learned from these reviews. In virtually all of the reviews, it was concluded that there was no clear immediate public health application of the data. However, several of the reviews highlighted gene-disease associations for which public health applications are being considered. For example, one review dealt with sickle cell disease, for which an intervention has been established following a randomized trial that showed that oral penicillin could significantly reduce the associated morbidity and mortality (22). The substantial differences in mortality due to sickle cell disease that were demonstrated may reflect differences in the timing of introduction and extent of coverage of newborn screening and differences in medical care, parental education, and penicillin prophylaxis to prevent infections (17). Another review considered HLA-DQ and type 1 diabetes (23), as well as the weight of evidence that has led to HLA-DQ screening for type 1 diabetes being conducted in high-risk families and the general population for intervention trials and natural history studies. The review also highlighted a critical need to reconsider the risks, benefits, and ethical, legal, and social issues regarding genetic or autoantibody testing for type 1 diabetes, as well as a need to clarify the effects of environmental exposures as independent or interacting with high-risk HLA genotypes. In the HuGE review of hereditary hemachromatosis, it was concluded that more information is needed about penetrance of clinical expression among persons with elevated transferrin saturation or HFE mutations, about the disease burden associated with hereditary hemachromatosis in the general population, about screening accuracy, and about the diagnostic tests available and the efficacy of early treatment (24). For medium chain acyl coenzyme A dehydrogenase deficiency, the main knowledge gap concerns the natural history of the disease and its clinical outcomes (16). In regard to mismatch repair genes and colorectal cancer, there is no consensus regarding the mostefficient approach of identifying mutation carriers (20). Some of the other reviews (25–33) dealt with gene variants that are part of a number of biomarkers included in test kits being marketed commercially (34), and these reviews highlighted important gaps in the evidence base. In future reviews, we encourage authors to emphasize data gaps and make recommendations for research to address these gaps. In all of the reviews dealing with gene variants associated with a high risk of disease, variable penetrance or manifestation was noted. This reinforces the point that even for single gene disorders there is wide variation in clinical phenotype (18), and for this reason HuGE reviews of these disorders are valuable. In all of the reviews, there was a lack of data on other factors contributing to variation in manifestation. The reviews highlight methodological issues such as selection bias, statistical power, and investigation of interaction or modifying factors, and they uncovered a need for unified guidelines that can be used to synthesize results of the increasing number of such studies. Progress is being made in defining quality standards for genetic-epidemiologic research, but ongoing evaluation is needed to make sure that such guidelines are refined and implemented. In 2001, an expert panel sponsored by the Centers for Disease Control and Prevention and the National Institutes of Health developed guidelines and recommendations for the reporting, evaluation, and integration of data from human genome epidemiology with emphasis on studies of 1) prevalence of gene variants and gene-disease associations, 2) gene-environment and gene-gene interactions, and 3) evaluation of genetic tests. Conclusions and recommendations from this workshop have been published (35, 36). In addition, other groups have proposed guidelines for gene-disease association studies (37–42). Many of the recommendations are similar, and the use of these guidelines in reporting studies should facilitate the integration of evidence in the future. Similarly, there is increasing interest in standardized approaches to the evaluation of genetic tests (43–45). The use of meta-analysis or pooled analysis as a tool to synthesize evidence has been left to the discretion of the authors of HuGE reviews, in part because of concern about the lack of comparability of study methods and in part because of concern about the validity of meta-analysis of observational studies (46, 47). Meta-analysis was used as a tool for synthesizing evidence in two of the reviews (25, 32). In the future, with the application of guidelines for reporting human genome epidemiology studies, more comparability among published data will make meta-analysis a more feasible option. For the present, as the potential value of using meta-analysis is likely to vary between different gene-disease associations, we prefer to leave this decision to the authors of reviews. In one of the reviews, pooled analysis (which requires data on individual subjects) was used in addition to meta-analysis (32). Interestingly, the results of the pooled and meta-analyses were very similar. Pooled analyses require much greater resources than meta-analyses (48, 49) and would be preferred to meta-analysis only when a high degree of precision of the measures of effect is required. For example, as data on the penetrance of HFE mutations accumulate, a pooled analysis might be of considerable value. More generally, there has been considerable concern about nonreplication of gene-disease association studies (37, 38, 42, 50–52). Nonreplication has also been an issue in other areas of epidemiologic research, so much so that epidemiology has been occasionally viewed as having reached its limits (53); for example, the results of recent cohort studies are challenging the inverse association between cancer and consumption of vegetables and fruit (54–58). The investigation of gene-disease associations differs from the investigation of exposure-disease associations in two important respects. First, the assessment of genotypes by DNA assays (polymerase chain reaction methods) is generally more accurate than for exposure assessment, and it is less heavily dependent on study design. Second, because of “Mendelian randomization” (59), an association between a disease and a genotype is unlikely to be due to confounding, provided that the study is designed according to the principles of population-based studies (60). Although there has been concern about population stratification (36, 61, 62), empirical studies in non-Hispanic White Americans and modeling suggest that bias from this source may not be substantial when epidemiologic principles of study design, conduct, and analysis are rigorously applied (63, 64). In this context, it is interesting that, in an analysis of 301 published studies covering 25 associations in which the first positive report was excluded, grouping studies by ethnicity generally did not remove heterogeneity (65). In the same meta-analysis, there was an excess of studies replicating the initial report that seemed unlikely to be due to publication bias (65). For eight of the associations, the combined estimate of relative risk was statistically significant; this proportion is similar to the findings of another set of meta-analyses (51). Thus, it is possible that, as an area of investigation matures with a move from small innovative studies which might best be viewed as pilot studies to large well-resourced studies in which potential biases are minimized, more consistent associations will be observed than predicted by the rather bleak commentaries based on early studies. This raises the challenge of keeping overviews of evidence up-to-date. In the early stages of an area of investigation, publication bias may be of critical importance (51, 52), as suggested for example by the pattern of accrual of evidence regarding the association between the angiotensin-converting-enzyme insertion/deletion polymorphism and myocardial infarction (66). Differences in timing may account for some differences between the results of meta-analyses as evidence accrues. Later, publication bias may be less of an issue as large high-quality studies are likely to be published irrespective of their findings. The best solution to the problem of publication bias appears to be the establishment of a research register for studies of gene-disease associations and of gene-gene and gene-environment interactions, analogous to those for other areas of medicine (67–69). This would help to address the problem of integrating all available evidence (70), taking into account its quality. There is also the challenge of the ever-increasing number of human genome epidemiology studies. For example, in the literature database maintained in the Centers for Disease Control and Prevention Genomics and Disease Prevention Information System (71), 2,436 primary studies of this type were published in 2001, and 2,922 studies were published in 2002. Moreover, as a result of the increasing availability of mapped single nucleotide polymorphism markers (72, 73), this trend is expected to accelerate. Therefore, integration of evidence will become increasingly important as a means of dealing with potentially unmanageable amounts of information. Certainly, the Human Genome Epidemiology Network (HuGE Net) will continue to benefit from the contributions of researchers writing HuGE reviews in their own specialty areas. However, we would also like to suggest some priorities with the hope of encouraging others to invest effort in integrating evidence about the gene-disease associations (and related gene-gene and gene-environment interactions) most likely to expand our knowledge and ability to apply research results. Some proposed criteria for prioritizing HuGE reviews are presented in table 2. The type of evidence is also relevant. An analysis of abstracts of published human genome epidemiologic papers for 2001–2002 shows that, of the 5,358 published articles, 601 (11.2 percent) reported only on the population prevalence of gene variants, 4,657 (86.9 percent) reported on gene-disease associations, 978 (18.3 percent) reported on gene-gene and gene-environment interactions, and 173 (3.2 percent) dealt with evaluation of genetic tests and population screening (71). Much of the evidence on the population prevalence of genetic variants in HuGE reviews has been derived from data on controls in gene-disease association studies. We recognize that assembling these data for the purposes of a HuGE review may be very labor intensive. There is increasing emphasis on the importance of biologic data, in particular on gene function and gene expression, in the interpretation of gene-disease associations and gene-gene and gene-environment interactions (74–77). As in the use of biologic data in making causal inference in other contexts, caution is warranted (38). Molecular biologic research has enjoyed explosive development, and it is difficult to identify and organize the information that would be useful in considering the biologic plausibility of an association or putative interaction (78). In regard to the quality of such information, it has been noted, for example, that the lack of standard methodologies or nomenclature for DNA expression studies has made it difficult to compare results (79, 80); recommendations for standardization have now been made (80). We anticipate that development of methods for the synthesis of biologic data will enhance the understanding of the functional effects of gene variants, particularly of multiple genes operating in pathways and networks, and that it will be relevant to consider this evidence in future HuGE reviews. In recognition of the increasing volume of evidence and the distribution of types of evidence, we will now propose two additional categories of HuGE review: 1) reviews of gene-disease associations and related gene-environment or gene-gene interactions only and 2) reviews of genotype prevalence only. The relations between these and the existing formats for HuGE reviews to be published by the Journal are summarized in table 3; instructions for these are presented on the HuGE Net website (21). There has been a change in the emphasis of research on genetic susceptibility from single candidate genes to multiple genes operating in pathways and, indeed, networks and systems (81, 82). For example, because the substrates resulting from phase 1 activation may be more reactive and potentially more carcinogenic than the starting xenobiotic compound, coordinated expression of phase 1 and phase 2 genes is likely to be critical in the metabolism of xenobiotic compounds (83). Moreover, many xenobiotic compounds can be metabolized by more than one cytochrome P450 enzyme (84). Similarly, there are genetic polymorphisms of several key proteins involved in folate metabolism (85). Therefore, we encourage authors to submit reviews to the Journal involving more than one gene operating in a pathway. Overall, the task of characterizing the human genome is at the beginning. The concern that the potential value of this exercise to public health has been exaggerated or that the amount of time needed for information relevant to public health to be accrued has been underestimated underlines, more than ever, the need for integration of evidence from carefully conducted population-based studies. We hope that the suggested changes to HuGE reviews will stimulate more investigators to contribute to this task. Correspondence to Dr. Muin J. Khoury, Office of Genomics and Disease Prevention, National Center for Environmental Health, Centers for Disease Control and Prevention, 4770 Buford Highway, MS K-28, Atlanta, GA 30341 (e-mail: muk1@cdc.gov). HuGE* reviews, 1999–2002 * HuGE, Human Genome Epidemiology; acyl-CoA, acyl coenzyme A; NAD(P)H, nicotinamide adenine dinucleotide phosphate (reduced form). HuGE* reviews, 1999–2002 * HuGE, Human Genome Epidemiology; acyl-CoA, acyl coenzyme A; NAD(P)H, nicotinamide adenine dinucleotide phosphate (reduced form). Proposed criteria for prioritizing HuGE* review topics * HuGE, Human Genome Epidemiology. Proposed criteria for prioritizing HuGE* review topics * HuGE, Human Genome Epidemiology. Formats of HuGE* review† * HuGE, Human Genome Epidemiology; HuGE Net, Human Genome Epidemiology Network. † Detailed instructions are available at http://www.cdc.gov/genomics/hugenet/reviews/guidelines.htm. ‡ This was not requested explicitly in the previous guidelines (15). Formats of HuGE* review† * HuGE, Human Genome Epidemiology; HuGE Net, Human Genome Epidemiology Network. † Detailed instructions are available at http://www.cdc.gov/genomics/hugenet/reviews/guidelines.htm. ‡ This was not requested explicitly in the previous guidelines (15).
No takes yet. Share an insight, caveat, or question.
Julian Little (2003) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: