• We are sorry, but NCBI web applications do not support your browser and may not function properly. More information
Logo of ajhgLink to Publisher's site
Am J Hum Genet. Mar 3, 2008; 82(3): 593–599.
Published online Feb 29, 2008. doi:  10.1016/j.ajhg.2007.12.020
PMCID: PMC2427295

A Critical Appraisal of the Scientific Basis of Commercial Genomic Profiles Used to Assess Health Risks and Personalize Health Interventions

Abstract

Predictive genomic profiling used to produce personalized nutrition and other lifestyle health recommendations is currently offered directly to consumers. By examining previous meta-analyses and HuGE reviews, we assessed the scientific evidence supporting the purported gene-disease associations for genes included in genomic profiles offered online. We identified seven companies that offer predictive genomic profiling. We searched PubMed for meta-analyses and HuGE reviews of studies of gene-disease associations published from 2000 through June 2007 in which the genotypes of people with a disease were compared with those of a healthy or general-population control group. The seven companies tested at least 69 different polymorphisms in 56 genes. Of the 56 genes tested, 24 (43%) were not reviewed in meta-analyses. For the remaining 32 genes, we found 260 meta-analyses that examined 160 unique polymorphism-disease associations, of which only 60 (38%) were found to be statistically significant. Even the 60 significant associations, which involved 29 different polymorphisms and 28 different diseases, were generally modest, with synthetic odds ratios ranging from 0.54 to 0.88 for protective variants and from 1.04 to 3.2 for risk variants. Furthermore, genes in cardiogenomic profiles were more frequently associated with noncardiovascular diseases than with cardiovascular diseases, and though two of the five genes of the osteogenomic profiles did show significant associations with disease, the associations were not with bone diseases. There is insufficient scientific evidence to conclude that genomic profiles are useful in measuring genetic risk for common diseases or in developing personalized diet and lifestyle recommendations for disease prevention.

Introduction

Advances in genomics are expected to increase our understanding of the etiology and pathogenesis of common diseases such as type 2 diabetes, cardiovascular disease, and cancer. They are also expected to offer new opportunities for the prevention, early detection, and treatment of these diseases, in part by allowing health care providers to use individualized preventive and therapeutic strategies based on patients' genomic profiles.1 So far, the complex interactions between genetic and environmental causes of most common diseases are poorly understood and the potential usefulness of genome-based interventions is unclear. Nevertheless, several companies already offer personalized lifestyle health recommendations and nutritional supplements based on clients' genomic profiles, and many others are developing similar strategies (see Web Resources).2

Because most common diseases are caused by complex interactions among multiple genetic and nongenetic factors, each of which confer only minor increases in risk, the predictive value of genomic profiling may be insufficient as a useful basis for personalized nutritional and lifestyle recommendations. When differences in disease risk between high-risk and low-risk groups are small, we would expect both groups to benefit fairly equally from general interventions unless the interventions are proven more effective for individuals with certain genotypes; however, evidence of such gene-environment interactions is still lacking. For these reasons, it has been argued that the use of genomic profiles to devise personalized lifestyle recommendations is premature and misleading.2–6

A pervasive problem of research on genetic associations is that positive results are often difficult to replicate.7,8 Results of individual gene-disease association studies have a high probability of being false-positive, and therefore adequate replications of the same gene-disease association in independent-study populations are essential.7,9 Systematic reviews and meta-analyses of epidemiologic studies on genotype-disease associations are valuable approaches for assessing the credibility of associations reported in single studies.6,7,10 In this study, we assess the scientific evidence for the usefulness of commercially available genomic profiles by reviewing meta-analyses of gene-disease associations for the genetic variants included in the profiles.

Material and Methods

Searching

Starting from the Genewatch 2006 report on individually tailored nutrition recommendations based on genomic profiling,2 we searched the Internet and identified seven companies that offer predictive genetic testing using multiple markers (Genelex, Genovations, Genosolutions, Integrative Genomics, Salugen, Sciona and Suracell). We obtained information about the genes and polymorphisms included in these profiles from the companies' Web sites (see Web resources), from online sample reports, and from a published article.11 The genes and polymorphisms included in the profiles are listed in Table S1 (available online). Three companies did not specify the polymorphisms that they tested. We assumed that these companies tested the same polymorphisms in genes also tested by other companies; however, when their profiles included a gene not tested by any other company, we reviewed available literature on any polymorphism of the gene. We searched PubMed for meta-analyses and Human Genome Epidemiology (HuGE) reviews published from 2000 through June 2007 regarding susceptibility to any disease associated with these polymorphisms.

Selection of Meta-Analyses

In conducting the PubMed literature search, we used both the abbreviation and the full name of each of the 56 genes tested by the seven companies, in combination with the query term “meta-analysis and (gene or polymorphism),” without specifying any polymorphisms or diseases. Next, we hand-searched all retrieved references for the exact polymorphism(s) studied, the study population, and the disease examined. We considered alternative annotations of polymorphisms and verified them in the Online Mendelian Inheritance in Man (OMIM) database. We included only meta-analyses published in English that compared the frequency of a polymorphism among people with the disease in question with the frequency among healthy or general-population controls. We did not exclude studies because of statistically significant heterogeneity in the effect sizes of the individual studies. For meta-analyses in which there was significant heterogeneity, we report the meta-analysis with one or two outlying studies removed or a subgroup analysis of three or more studies involving populations of Western European descent (the most commonly studied subgroup), if these had less or no significant heterogeneity, depending on the availability of information. If more than one meta-analysis examined the same polymorphism-disease association, we used the most comprehensive one.

Data Extraction

We retrieved data on the gene, the polymorphism, and the disease examined in each study, as well as the number of case and of control subjects involved; the odds ratio and 95% confidence interval (CI) for the polymorphism-disease association; and the heterogeneity in effect sizes of the individual studies. Many of the meta-analyses we examined presented multiple odds ratios for the same polymorphism-disease association by considering different genetic models (e.g., dominant effect of risk alleles, recessive effect, or per allele effect). We used the per-allele odds ratio if all models yielded the same conclusion about the presence of association. If the per-allele odds ratio was not statistically significant, we selected the model (dominant or recessive effect of risk allele) with the highest odds ratio, assuming that the companies were referring to that specific association. If random and fixed effects models were presented, we obtained the odds ratios from the random-effects models.

Results

The seven companies tested at least 69 different polymorphisms in 56 genes (Table S1). Only one gene (MTHFR [MIM 07093]) was tested by all seven companies, seven genes (CETP [MIM 118470], COL1A1 [MIM 120150], GSTM1 [MIM 138350], GSTP1 [MIM 134660], IL-6 [MIM 147620], TNF-α [MIM 191160], and VDR [MIM 601769]) were tested by five or six companies, and 19 genes (34%) were tested by only one company.

We identified 260 meta-analyses that met our criteria; they addressed 46 of the 69 polymorphisms and 32 (57%) of the 56 genes tested by the seven companies (Figure 1 and Table S1), as well as 160 unique associations between a genetic polymorphism and a disease. Sixteen meta-analyses (10%) addressed the two polymorphisms in the MTHFR gene (C677T and A1298C), 13 addressed polymorphisms in TNF-α, 12 addressed polymorphisms in GSTM1, 11 addressed polymorphisms in GSTP1 and GSTT1, and ten addressed polymorphisms in VDR.

Figure 1
Flow Chart of Selection of Meta-Analyses

Of the 160 unique polymorphism-disease associations that had been examined by meta-analysis, 60 were statistically significant (Figure 1 and Table S2), including significant associations for 29 of 69 polymorphisms in 25 (45%) of the 56 genes. However, statistically significant associations were generally modest, with significant odds ratios (ORs) ranging from 0.54 to 0.88 for protective alleles or genotypes and from 1.04 to 1.50 for risk alleles or genotypes, except for the risk for systemic lupus erythematosus associated with TNF-α (OR 2.1, 95% CI 1.6-2.7) and the risk for Alzheimer's disease associated with APOE-4 (OR 3.2, 95% CI 2.7-3.8; [MIM 107740]). Polymorphisms were associated with significantly increased or decreased risk for 28 different diseases. The MTHFR C677T polymorphism was significantly associated with risk for seven different diseases, GSTT1 [MIM 600436] with risk for six, TNF-α with risk for five, and GSTM1 and MTHFR A1298C with risk for four. Twelve meta-analyses showed significant associations between polymorphisms and risk for cardiovascular disease (including myocardial infarction, coronary stenosis, coronary artery disease, and coronary heart disease), five between polymorphisms and stroke, and four between polymorphisms and Alzheimer's disease or acute leukemia.

On average, meta-analyses showed a significant association with disease risk for 58% of the genes included in each profile (range 38%–83%) (Table 1). These significant associations, however, were often with risk for disease outcomes other than those associated with the “profile” (Table 1). For example, statistically significant associations with disease risk were found for only two of the five genes included in osteogenomic profiles, and these associations were with risk for Alzheimer's disease, asthma, non-Hodgkin's lymphoma, obesity, and systemic lupus erythematosus, rather than with risk for bone disorders.

Table 1
Overview of Meta-Analyses of Gene-Disease Association Studies by Genomic Profile

Discussion

Although companies offering genomic profiles did not specify how they selected polymorphisms for inclusion in the profiles, they probably did so on the basis of statistically significant results from association studies. Because positive results from single gene-disease association studies are often not replicated in subsequent studies,12 one study showing a statistically significant association is considered insufficient evidence of genetic association.10 Our review of meta-analyses found significant associations with disease risk for fewer than half of the 56 genes that are tested in commercially available genomic profiles. Various polymorphisms of these genes were associated with risk for 28 different disorders. Many of these disorders were unrelated to the ostensible target condition, and the associations were generally modest.

Before interpreting our results, we need to clarify four issues regarding the review strategy we used. First, our paper addressed predictive genomic profiles that are sold online and that aim to personalize nutrition and other lifestyle health recommendations. The review did not assess the scientific basis of gene-expression profiles and pharmacogenomic applications. Although there may be applications that have stronger scientific support than others, there are clearly promising developments in this area.13,14 Second, the information on genes and polymorphisms in this study was obtained from company websites and online sample reports. As of November 2007, all seven companies were still selling the profiles, but two no longer specified on their Web sites which genes they were actually testing. Although these companies may now use other polymorphisms to profile the disease risk of their clients, the scientific evidence for the disease risk associated with these other polymorphisms is likely to be similar to that for the polymorphisms we reviewed. Third, we limited our search for meta-analyses to those on the association between polymorphisms and disease susceptibility, and we excluded those on associations with intermediate, quantitative phenotypes or risk factors such as blood pressure or bone mass density because the need for preventive intervention varies with the level of these traits. We did include meta-analyses of associations between polymorphisms and risk for conditions defined by clinically relevant thresholds, such as hypertension or osteoporotic fractures. Because the genetic profiles of the companies are offered to the general public, we restricted our search to meta-analyses of studies that included healthy or general-population controls. Because the predictive value of genetic testing depends on disease risk, genotype frequencies, and odds ratios for the association between disease risk and polymorphisms in a particular genetic profile, all of which may differ between populations, the profiles should be evaluated in the target population.15 This explains why genetic testing for APOE, Factor II [MIM 176930], and Factor V [MIM 227400] can have lower predictive value in a general population context but be very informative to persons with a family history. Fourth, we did not exclude meta-analyses on the basis of quality criteria, even though there were obvious differences in quality among meta-analyses. The authors of larger meta-analyses often selected studies according to a set of strict criteria, whereas the authors of smaller ones often combined all available studies. In addition, more than a quarter of the meta-analyses in Table S2 reported statistically significant heterogeneity in effect sizes among studies. Several of the meta-analyses that found a significant association involving heterogeneous study populations did not find a significant association when the analyses were restricted to a subgroup of more homogenous studies.16,17 Application of strict quality criteria would have reduced the number of meta-analyses in the present review substantially.

These methodological choices partly explain why we found no meta-analyses for 24 of the 56 genes. There were meta-analyses available for many of these genes, but these meta-analyses could not evidence the utility of genomic profiling in the general population. For example, we found several meta-analyses of pharmacogenomic studies (e.g., for CYP2C9 [MIM 601130] and CYP2C19 [MIM 124020]18,19), several meta-analyses on diseases that do not affect the average individual in the general population (such as IL-10 [MIM 124092] and recurrent pregnancy loss20), and meta-analyses on health traits (e.g., smoking behavior and CYP2A6 [122720]21). Furthermore, for many genes we found meta-analyses on other polymorphisms (e.g., IL-10 G[−1082]A22 and LPL Asn291Ser23) or on related genes (e.g., Leptin Receptor gene (LEPR; [MIM 601007]), but not for Leptin [MIM 164160]).24

This review shows that the excess disease risk associated with many genetic variants included in genomic profiles has not been investigated in meta-analyses or has been found to be minimal or not significant. These results raise concern about the validity of combining tests for many different genetic variants into profiles, especially when the companies offering them do not describe how they create a composite profile from the results of tests for single genetic markers. One company reports that they use complex mathematical algorithms to produce personalized diet and lifestyle recommendations. Another recommends basic nutritional or lifestyle-change support for homozygous negatives, added support for heterozygous positives, and maximum support for homozygous positives, which suggests that they are using single genetic markers as the basis for their recommendations. This reliance on single genetic markers is particularly worrisome given the limited predictive value of results from testing single susceptibility genes with small effects.25–27 To be meaningful, a genetic risk profile should combine information about the disease risk associated with multiple genes, and creating such a profile would require extensive knowledge of gene-gene interactions, which are even less well understood than the disease risk associated with individual polymorphisms.

How the companies we examined use their clients' genetic profiles to tailor individualized nutrition-supplement and lifestyle recommendations is another intriguing puzzle. Evidence on gene-diet interactions is still preliminary because trials designed to test these interactions have thus far yielded mainly inconclusive results.28 Furthermore, several genes, such as ACE [MIM 106180], APOE, and MTHFR, increase people's risk for some diseases and decrease their risk for others (Table S2). For example, MTHFR 677TT was associated with an increased risk for depression, stroke, coronary artery disease, gastric cancer, schizophrenia, and venous thrombosis, but it was associated with a decreased risk for colorectal cancer. Hence, the putative health effects of preventive interventions tailored to a person's MTHFR genotype may not be entirely beneficial. Finally, when profiles are composed of low-risk susceptibility genes, people with purportedly “high-risk” profiles may be at only slightly higher risk of disease than are people with “low-risk” profiles. One possible danger of marketing lifestyle recommendations to people with “high-risk” profiles is that those with “low-risk” profiles could be led to mistakenly believe that they have little need to make healthy lifestyle changes. The predictive value of genomic profiling may simply be insufficient for targeting interventions when low-risk groups will receive no intervention at all.29 It also needs to be investigated whether genomic profiling can usefully identify the better from the worse responders in the choice between two treatments.

Although genomic profiling may have potential to enhance the effectiveness and efficiency of preventive interventions, to date the scientific evidence for most associations between genetic variants and disease risk is insufficient to support useful applications. Despite advances in nutrigenomics and pharmacogenomics research,30 it could take years, if not decades, before lifestyle and medical interventions can be responsibly and effectively tailored to individual genomic profiles.

Web Resources

The URLs for data presented herein are as follows:

Supplemental Data

Two tables can be found with this paper online at http://www.ajhg.org/.

Supplemental Data

Document S1. Two Tables:

Acknowledgments

The study was financially supported by the Center for Medical Systems Biology (CMSB). A.C.J.W.J. was supported by a fellowship from the Netherlands Genomics Initiative. None of the authors had any conflict of interest concerning this manuscript. The opinions expressed by the authors do not necessarily reflect the opinions of the Centers for Disease Control and Prevention or other institutions with which authors are affiliated.

References

1. Collins F.S., McKusick V.A. Implications of the Human Genome Project for medical science. JAMA. 2001;285:540–544. [PubMed]
2. Wallace H. GeneWatch UK; Buxton, UK: 2006. Your Diet Tailored to Your Genes: Preventing Diseases or Misleading Marketing?
3. Hudson K., Javitt G., Burke W., Byers P., ASHG Social Issues Committee ASHG Statement on Direct-to-Consumer Genetic Testing in the United States. Am. J. Hum. Genet. 2007;81:635–637.
4. Vineis P., Christiani D.C. Genetic testing for sale. Epidemiology. 2004;15:3–5. [PubMed]
5. Government Accountability Office. (2006). Nutrigenetic testing: Tests purchased from four Web sites mislead consumers. (http://www.gao.gov/new.items/d06977t.pdf)
6. Haga S.B., Khoury M.J., Burke W. Genomic profiling to promote a healthy lifestyle: Not ready for prime time. Nat. Genet. 2003;34:347–350. [PubMed]
7. Khoury M.J., Little J., Gwinn M., Ioannidis J.P. On the synthesis and interpretation of consistent but weak gene-disease associations in the era of genome-wide association studies. Int. J. Epidemiol. 2006;36:439–445. [PubMed]
8. Ioannidis J.P. Genetic associations: False or true? Trends Mol. Med. 2003;9:135–138. [PubMed]
9. Ioannidis J.P., Boffetta P., Little J., O'Brien T.R., Uitterlinden A.G., Vineis P., Balding D.J., Chokkalingam A., Dolan S.M., Flanders W.D. Assessment of cumulative evidence on genetic associations: Interim guidelines. Int. J. Epidemiol. Advance. 2007 in press. Published online September 26, 2007. [PubMed]
10. Ioannidis J.P., Trikalinos T.A. An exploratory test for an excess of significant findings. Clin. Trials. 2007;4:245–253. [PubMed]
11. Blum K., Chen T.J., Meshkin B., Downs B.W., Gordon C.A., Blum S., Mangucci J.F., Braverman E.R., Arcuri V., Deutch R. Genotrim™, a DNA-customized nutrigenomic product, targets genetic factors of obesity: Hypothesizing a dopamine-glucose correlation demonstrating reward deficiency syndrome (RDS) Med. Hypotheses. 2007;68:844–852. [PubMed]
12. Ioannidis J.P., Ntzani E.E., Trikalinos T.A., Contopoulos-Ioannidis D.G. Replication validity of genetic association studies. Nat. Genet. 2001;29:306–309. [PubMed]
13. Glas A.M., Floore A., Delahaye L.J., Witteveen A.T., Pover R.C., Bakx N., Lahti-Domenici J.S., Bruinsma T.J., Warmoes M.O., Bernards R. Converting a breast cancer microarray signature into a high-throughput diagnostic test. BMC Genomics. 2006;7:278. [PMC free article] [PubMed]
14. Yin T., Miyata T. Warfarin dose and the pharmacogenomics of CYP2C9 and VKO. Thromb. Res. 2007;120:1–10. [PubMed]
15. Haddow J.E., Palomaki G.E. ACCE: A Model Process for Evaluating Data on Emerging Genetic Tests. In: Khoury M.J., Little J., Burke W., editors. Human Genome Epidemiology: A Scientific Foundation for Using Genetic Information to Improve Health and Prevent Disease. Oxford University Press; New York: 2003. pp. 217–233.
16. Casas J.P., Cavalleri G.L., Bautista L.E., Smeeth L., Humphries S.E., Hingorani A.D. Endothelial nitric oxide synthase gene polymorphisms and cardiovascular disease: A HuGE review. Am. J. Epidemiol. 2006;164:921–935. [PubMed]
17. Ye Z., Liu E.H., Higgins J.P., Keavney B.D., Lowe G.D., Collins R., Danesh J. Seven haemostatic gene polymorphisms in coronary disease: Meta-analysis of 66,155 cases and 91,307 controls. Lancet. 2006;367:651–658. [PubMed]
18. Sanderson S., Emery J., Higgins J. CYP2C9 gene variants, drug dose, and bleeding risk in warfarin-treated patients: A HuGEnet systematic review and meta-analysis. Genet. Med. 2005;7:97–104. [PubMed]
19. Padol S., Yuan Y., Thabane M., Padol I.T., Hunt R.H. The effect of CYP2C19 polymorphisms on H. pylori eradication rate in dual and triple first-line PPI therapies: A meta-analysis. Am. J. Gastroenterol. 2006;101:1467–1475. [PubMed]
20. Daher S., Shulzhenko N., Morgun A., Mattar R., Rampim G.F., Camano L., DeLima M.G. Associations between cytokine gene polymorphisms and recurrent pregnancy loss. J. Reprod. Immunol. 2003;58:69–77. [PubMed]
21. Munafo M., Clark T., Johnstone E., Murphy M., Walton R. The genetic basis for smoking behavior: A systematic review and meta-analysis. Nicotine Tob. Res. 2004;6:583–597. [PubMed]
22. Warle M.C., Metselaar H.J., Hop W.C., Tilanus H.W. Cytokine gene polymorphisms and acute liver graft rejection: A meta-analysis. Liver Transpl. 2005;11:19–26. [PubMed]
23. Hu Y., Liu W., Huang R., Zhang X. A systematic review and meta-analysis of the relationship between lipoprotein lipase Asn291Ser variant and diseases. J. Lipid Res. 2006;47:1908–1914. [PubMed]
24. Paracchini V., Pedotti P., Taioli E. Genetics of leptin and obesity: A HuGE review. Am. J. Epidemiol. 2005;162:101–114. [PubMed]
25. Vineis P., Schulte P., McMichael A.J. Misconceptions about the use of genetic tests in populations. Lancet. 2001;357:709–712. [PubMed]
26. Holtzman N.A., Marteau T.M. Will genetics revolutionize medicine? N. Engl. J. Med. 2000;343:141–144. [PubMed]
27. Janssens A.C.J.W., Gwinn M., Valdez R., Venkat Narayan K.M., Khoury M.J. Predictive genetic testing of type 2 diabetes. BMJ. 2006;333:509–510. [PMC free article] [PubMed]
28. Ordovas J.M., Mooser V. Nutrigenomics and nutrigenetics. Curr. Opin. Lipidol. 2004;15:101–108. [PubMed]
29. Janssens A.C.J.W., Aulchenko Y.S., Elefante S., Borsboom G.J.J.M., Steyerberg E.W., Van Duijn C.M. Predictive testing for complex diseases using multiple genes: Fact or fiction? Genet. Med. 2006;8:395–400. [PubMed]
30. Kaput J., Ordovas J.M., Ferguson L., van Ommen B., Rodriguez R.L., Allen L., Ames B.N., Dawson K., German B., Krauss R. The case for strategic international alliances to harness nutritional genomics for public and personal health. Br. J. Nutr. 2005;94:623–632. [PubMed]

Articles from American Journal of Human Genetics are provided here courtesy of American Society of Human Genetics
PubReader format: click here to try

Formats:

Related citations in PubMed

See reviews...See all...

Cited by other articles in PMC

See all...

Links

  • Cited in Books
    Cited in Books
    PubMed Central articles cited in books
  • MedGen
    MedGen
    Related information in MedGen
  • PubMed
    PubMed
    PubMed citations for these articles

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...