necessarily represent the views of the funding agency.”
Human Genome Epidemiology: A Scientific Foundation for Using Genetic Information to Improve Health and Prevent Disease
Methods and Approaches 1: Assessing Disease Associations and Interactions
Epidemiologic Approach to Genetic Tests: Population-Based Data for Preventive Medicine
Marta Gwinn, Muin J. Khoury
CRC – colorectal cancer
DNA – deoxyribonucleic acid
FAP – familial adenomatous polyposis
HNPCC – hereditary non-polyposis colorectal cancer
MMR – mismatch repair
Names of genes: APC, BRCA1, MLH1, MSH2
Sequencing the human genome ahead of schedule has raised expectations for quick translation of the data into tools for medical practice. When the initial sequence was published in February 2001, Francis Collins and Victor McKusick wrote that “genetic prediction of individual risks of disease and responsiveness to drugs will reach the medical mainstream in the next decade or so.”(1) The idea that genetic tests could offer patients personal estimates of risk and interventions on the basis of their genotypes has captured the imagination of scientists and the public.
Until now, genetic tests have been used mostly to aid the diagnosis of rare hereditary disorders. In November 2000, when we reviewed the list of tests in GeneTests, a Web-accessible database that serves as the main directory of United States clinical and research laboratories offering genetic testing,(2-3) we found that fewer than 5 percent of tests available for clinical use applied to common, adult-onset diseases.(4) Most of these were tests for variants of single genes associated with disease susceptibility in high-risk families (e.g., BRCA1 for breast cancer). A recent review of entries in the online version of Mendelian Inheritance in Man(5) suggests that this situation is unlikely to change soon: although the discovery of disease-associated gene variants is accelerating rapidly, the number of identified “susceptibility genes” remains small.(6)
Genetic tests that predict future risk for disease in asymptomatic people, thereby suggesting specific strategies for prevention or early detection, are the starting point for models of individualized preventive medicine. An example that helps illustrate the expectations, limitations, and future potential of predictive genetic tests is Francis Collins’ “hypothetical case in 2010,”(7) in which a 23-year-old man named John undergoes DNA testing for genes related to several common chronic diseases. The genetic test report includes relative risks (range: 0.3-6) as well as lifetime risks (range: 7%-30%) for each of these diseases, predicted on the basis of John’s genotype for one to three genes related to each condition. John’s physician recommends that he stop smoking, undergo regular colonoscopy beginning at age 45 years, and take lipid-lowering medications.
In this example, not only the patient but the relative and lifetime risks are hypothetical. Where will we obtain the data needed to interpret and act on the results of genetic tests? Despite the media’s tendency to depict genetic tests as definitive, no test can predict with certainty the behavior of a complex biologic system (in this case, John) over a lifetime. Furthermore, risk cannot be predicted solely on the basis of individual information; it must be estimated by analysis of the characteristics, experience, and outcomes of a group of people “similar” to the individual of interest. Information about the family is used to assess risk of classic, Mendelian disorders, and the ability to predict disease based on inheritance is the foundation for the clinical specialty of genetic counseling. However, estimating the risk for complex disorders (without a clear pattern of inheritance) requires genetic information from larger population samples. Information on prevalence of gene variants, genotype-phenotype correlations, and gene-gene and gene-environment interactions must be collected systematically by epidemiologic studies conducted in populations resembling those to which inferences will be drawn.(8) These populations are likely to be much more diverse than the genetically homogeneous groups in which susceptibility genes are usually first identified.
Epidemiologic studies that collect genetic information depend on access to valid, reproducible, economical tests for the genetic variants of interest. New technology has made such tests available for use in large-scale, population-based studies.(9) Choosing among analytic methods involves practical considerations, such as availability of collaborators, as well as characteristics of the variant itself. Before a test can be used in epidemiologic research, its analytic validity must be established. Analytic validity refers to the sensitivity, specificity, and predictive value of the test in relation to genotype; these characteristics are measured by comparing the test result with a gold standard in a set of well-described samples, only some of which contain the genetic variant.
The final report of the National Institutes of Health-Department of Energy Task Force on Genetic Testing(10) distinguished analytic validity from clinical validity, which they defined as the sensitivity, specificity, and predictive value of a test in relation to a particular phenotype. In contrast to analytic validity, which must be determined before a study begins, clinical validity is defined by epidemiologic studies that measure gene-disease associations. A third parameter defined by the Task Force, clinical utility, refers to the net value of the information gained from a genetic test in changing disease outcomes. Clinical utility is best assessed in clinical trials or by synthesis of observational data (e.g., as in cost-effectiveness analysis). Table 11-1 summarizes these measures of validity and utility.
Classic epidemiologic study designs include cross-sectional, cohort, and case control studies. Each design is useful for addressing particular aspects of genetic variation or gene-environment interaction in relation to disease outcomes.(11) Cross-sectional studies can be used to estimate the prevalence of gene variants, although variants associated with poor survival may be underrepresented. Cohort studies are unique in providing direct estimates of absolute risk and relative risk in people with different genotypes. “Experimental” cohort studies (randomized, controlled trials) are ideal for evaluating the effects of gene-environment interactions or specific interventions (see chapter 15). Retrospective cohort studies are attractive because genetic information is invariant and can be measured long after the study has ended; however, they are subject to the usual biases of observational studies.
Case-control studies can generally measure gene-disease associations more quickly, efficiently, and at lower cost than cohort studies. Case-control studies yield odds ratios, which approximate the relative risk of disease as long as the disease is rare or controls are sampled randomly (independent of disease status or genotype) from the source population.(12) The defining characteristic of a population based case-control study is the set of a priori criteria—applied to selection of controls as well as cases—that specifies the study’s source population (e.g., by geographic area and ethnicity). Studies comparing genotypes of patients in a clinical case series with those of an undefined convenience sample of control subjects (or worse, “control specimens”) are numerous in the published literature but they provide little basis for risk estimation. From a public health perspective, a population-based estimate of relative risk is critical because it provides the basis for estimating attributable fraction—the proportion of cases that would not occur in the absence of a particular exposure (or genotype) in the population. To examine the role of epidemiologic studies in eliciting genetic factors in common diseases, we consider the example of colorectal cancer (CRC).
Box 11.1 Example: Colorectal Cancer
Like other cancers, colorectal cancer (CRC) is a “genetic disease” caused by mutations that disable normal regulation of cell growth and differentiation. An estimated 945,000 new cases of CRC occur annually worldwide.(13) Epidemiologic studies have identified many exposures associated with increased risk for CRC, including smoking, overweight, inactivity, and dietary factors. Since familial clustering of CRC was first reported more than 100 years ago, clinical research has identified several high-risk syndromes, including familial adenomatous polyposis (FAP) and hereditary nonpolyposis CRC (HNPCC).
Table 11-2 compares estimates of the absolute (lifetime) risk and relative risk for CRC in people who have FAP,(14) pedigrees consistent with HNPCC,(15-16) or at least one first-degree relative with CRC,(17) with risks in people in none of these groups, who are considered at “average risk.” Without intervention, people with FAP are virtually certain to develop CRC (absolute risk approaching 1), usually by their mid-30s; however, FAP is very rare (prevalence approximately 1/8000) and thus accounts for a very small share of CRC cases in the U.S. population (attributable fraction <1 percent).(14) FAP is diagnosed clinically on endoscopic and histologic criteria. Although it is an autosomal dominant disorder, about one third of cases result from new mutations; thus genetic testing is useful mainly for counseling an affected person’s family members, offering increased surveillance if positive and reassurance otherwise.(18)
HNPCC is a diagnosis based on a pedigree consistent with autosomal dominant inheritance of CRC (as well as cancers at other sites in some high-risk families).(16) Unlike FAP, HNPCC cannot be diagnosed on the basis of clinical characteristics, and thus far, genotypic information has not become part of the case definition. Since 1994, mutations in several DNA mismatch-repair (MMR) genes have been found in HNPCC families, with MSH2 and MLH1 mutations by far most often implicated.(16) Identifying a mutation within an HNPCC family can be useful for testing and counseling family members. Although absolute risk for CRC is approximately 80 percent in HNPCC family members with inherited susceptibility,(16) data are insufficient to estimate the absolute and relative risks for CRC based on MMR genotype. The population prevalence of genetic variants associated with HNPCC is unknown, although one study analyzing data from Scotland, Finland, and the United States arrived at an estimate of 1/3000 for MSH2 and MLH1 variants combined.(15) Population-based data on frequency of HNPCC among CRC cases is also scarce, although recent studies suggest that it may be lower than previously estimated from case series, perhaps as low as 1 percent.(19)
Family history can capture shared, unmeasured genetic risk, as well as the potential influences of shared diet, behavior, and other non-genetic factors.(20) Accumulated evidence from epidemiologic studies of CRC suggests that having at least one first-degree relative with CRC increases the relative risk for colon cancer approximately twofold. However, because the average risk for CRC is only about 4 percent, the absolute risk of CRC in this group remains <10 percent;(17) thus, family history offers poor predictive value as a “genetic screening test” for CRC.
Genetic information obtained thus far from population-based, epidemiologic studies is relevant to only about 10 percent of CRC cases in the population. Although epidemiologic studies consistently identify CRC in a first-degree relative as one of the strongest risk factors for CRC, other factors (e.g., dietary habits, physical activity) that are less strongly associated but more prevalent in the population have higher attributable fractions.(21)
Estimating Individual Risk from Population-Based Data
Epidemiologic studies of gene-disease associations—particularly those reporting results in terms of predicted risk—have lately come under fire as an unwarranted extension of the individual risk paradigm,(22) in which relations among disease risk factors are teased out at the individual level. The exclusive focus on individuals can be criticized on both practical and philosophic grounds, for failing to prompt effective public health interventions while “blaming the victim.”(22-23) The “privatization of risk”(24) also appears to defy the concept of risk as an aggregate measure and to ignore the reality that individual risk factors generally make poor screening tools.(25) A recent commentary on the potential impact of genetics on preventive medicine echoes these concerns,(26) arguing that because most genetic tests have low predictive value, low clinical sensitivity, and little potential for stimulating tailored intervention, they will have limited value in preventing common complex diseases.
Viewing the potential contribution of a single test in isolation reflects a time-honored perspective in public health screening, as well as the traditional use of clinical genetic tests. Mass screening programs are generally delivered to whole populations without regard to prior information (such as family history or race/ethnicity), both to maximize sensitivity and to achieve social goals, such as fairness and program efficiency. On the other hand, clinical geneticists have used tests mostly for diagnosis of hereditary disorders resulting from single gene variants with very high penetrance. In this setting, a single genetic test may be definitive, although DNA sequencing is revealing increasingly diverse genotype-phenotype relations, even in classic “single gene disorders” like cystic fibrosis.(27)
Most common chronic diseases arise from interactions among multiple genes, environmental exposures, and behaviors; thus, genetic tests are most likely to be useful when combined with results of other information to uncover interactions associated with markedly elevated risks. This concept can be demonstrated using basic principles from either epidemiology(28-29) or genetics.(30) Real examples are still scarce, however, partly because they are likely to be complicated, involving multiple genes and multiple environmental exposures.
This state of affairs is familiar to most medical practitioners, who are used to considering the results of multiple clinical tests in the context of other, often incomplete, information about an individual patient, including family history, lifestyle, and physical examination. The basis for integrating and interpreting this information is experience—whether clinical experience of an individual physician, expert consensus, or data gathered systematically from scientific studies, such as clinical trials. During the last decade, the methods developed by clinical epidemiologists for critical analysis, synthesis, and application of accumulated experience have become the foundation for “evidence-based medicine.”(31) In this medical paradigm, diagnosis is a Bayesian process: as each test result is added to the body of evidence, some possible diagnoses become more likely, while others are less likely or altogether ruled out.(31, pp.121-40)
Results of genetic tests can be integrated into the same framework, as long as the association of genotype with disease outcome (genotype-phenotype correlation) has been well described.(32) When the goal is to predict future disease, rather than to make a diagnosis, a genotype becomes part of the evidence that can be used to make a probabilistic estimate of risk.(Yang Q, personal communication)(32) The underlying relationship between a susceptibility genotype (defined by variant alleles at one or more genetic loci) and disease outcome, and its reflection in measures of test performance and risk, can be summarized in the familiar framework of a 2 x 2 table (Appendix).
As observed by critics of individualized preventive medicine, few risk factors for common chronic diseases have sufficient predictive ability to serve as screening tools;(25) in this respect, common polymorphisms associated with disease susceptibility are unlikely to be different. Most risk estimates useful to individuals will be obtained only by considering the joint effects of many factors; however, although technical advances have made large-scale genotyping feasible in epidemiologic studies, the ability to assimilate, synthesize, and interpret the data has not yet fully caught up. The sheer number of variables potentially available for analysis tests the limits of conventional methods.
Challenges for “Genomic” Epidemiology
Despite their independent origins, genetic and epidemiologic methods for investigating causes and predicting risks for human diseases share many concerns common to observational sciences. During the last 50 years, the synthesis of genetic and epidemiologic methods has been accelerated by growth in statistical and computing techniques and by development of molecular methods for measuring environmental exposures, biological processes, and genetic traits.(9,33)
Much recent development in genetic epidemiology and statistical genetics has focused on methods for identifying disease susceptibility genes in families; however, describing the distribution of genetic traits in populations and evaluating the role of genetic factors in disease occurrence requires larger studies of unrelated people. Epidemiologic studies of genotype prevalence, gene disease association, and gene-environment interaction are subject to the usual sources of bias, including confounding and misclassification. Confounding is analogous to “population stratification” in studies of gene-disease association;(34) misclassification of genotype occurs as a function of analytic validity. Also, type I errors are of concern when multiple gene-disease associations are tested, and type II errors are likely when results are analyzed for small subgroups defined by genotype or gene-environment interactions.
The nature of genomic data poses additional challenges for epidemiologic analysis. For example, genetic variants at different loci cannot be assumed to occur independently, even when they are found on different chromosomes.(35) Furthermore, although the genetic sequence of an individual remains the same, gene expression naturally varies tremendously among tissues, in response to environmental stimuli, and with age, reflecting cumulative effects over the course of a lifetime. Thus, the interactions of gene products with each other and with other factors in their milieu reflect an underlying complex “genetic architecture,”(36) in which health and disease are “defined by the same continuum of biological traits.”(36, p. 217) New models are needed for analyzing these relations and using them for prediction and intervention.
Most common chronic diseases result from multiple gene-environment interactions over a long period of time, involving invariant features (e.g., genotype), “context-dependent features” (e.g., diet), and chance processes.(36) Prediction from “first principles” (genotype) is thus an unrealistic goal. One strategy for capturing the effects of multiple factors pursues data more proximate to the disease outcome, such as acquired (somatic) mutations, gene expression, protein markers, or intermediate states or conditions that recapitulate prior gene-environment interactions. Predictions based on observations made closer to the outcome are likely to be more accurate. To examine the potential of this approach, we revisit the example of CRC.
Box 11.2 Example Revisited: Colorectal Cancer
The ability to examine DNA sequence information in clinical and epidemiologic studies of CRC reveals that traditional categories for classifying CRC cases and their causes are not as distinct as they once seemed. On the other hand, insights at the molecular level may suggest more useful models for sorting out pathogenetic mechanisms of CRC, along with more specific targets for prevention, diagnosis, and treatment. For example, we now know that inherited variation in the APC gene gives rise to several different cancer syndromes with involvement at various extracolonic sites (e.g., Gardner syndrome, Turcot syndrome); a form of attenuated FAP in which far fewer polyps are found; and a susceptibility polymorphism in Ashkenazi Jews associated with only modestly increased risk for CRC.(37)
HNPCC for now remains a diagnosis based on pedigree because genotype-phenotype correlation is not well enough understood to establish sensitive and specific diagnostic criteria on the basis of genotype. Predictive tests for HNPCC (other than family history) would thus currently seem to be out of reach. Even DNA-based diagnosis remains problematic. Most tumors in affected persons exhibit microsatellite instability (MSI), which has been proposed as an initial screening test before sequencing MSH2 and MLH1; however, MSI testing itself is a costly and complex procedure that is not entirely sensitive or specific for cancer in HNPCC families.(16)
The molecular pathways that give rise to FAP and HNPCC are also important in sporadic CRC; thus events leading to loss of functional APC and MMR gene products can begin either with inherited or acquired mutations. In 1999, John Potter reviewed the evidence implicating these and other pathways in the pathogenesis of CRC, along with recognized or postulated gene-environment and gene-gene interactions.(37) He pointed out that epidemiologic studies that examine agents affecting only one or some of the pathways could be expected to find weak or inconsistent associations. However, he also predicted that as these pathways became better understood, population subgroups of similar susceptibility could be recognized for more specific preventive interventions, and that early molecular changes could serve as screening markers.
A recent study reported the feasibility of examining fecal DNA for APC mutations that occur early in the pathogenesis of CRC, suggesting future potential for new, noninvasive approaches to screening.(38) Although highly specific, the fecal DNA analysis was only 57 percent sensitive, positive in 26 of 46 patients with neoplasia (9/18 adenomas, 11/28 carcinomas). A commentary accompanying this article(39) suggested that while this study had “drawn back a curtain to reveal a tantalizing possibility, …there are other curtains and other possibilities. The basis of the next molecular screening test for CRC may not be a mutant gene but an abnormal protein that the new science of proteomics may find.”(39, p.304)
New Opportunities and Challenges for Public Health
Pathogenesis at the molecular level is far better understood for CRC than for other chronic diseases, including most other cancers. Ready access to a premalignant lesion—the adenomatous polyp—has afforded researchers a rare window for dissecting the process of tumorigenesis and public health a ready opportunity for prevention. However, new technology coupled with new analytic methods may be opening other windows onto diseases for which preventive medicine and public health have had little to offer until now. For example, collaborating researchers from several federal agencies, academic medical centers, and industry recently reported a method for using proteomic patterns in serum to identify ovarian cancer.(40) They developed and tested a new algorithm for discriminating serum protein profiles in patients with ovarian cancer from those in controls. The algorithm used data-driven methods (cluster analysis and “genetic algorithms,” in which principles of natural selection were used to select key measurements for analysis) to distinguish patterns generated by mass spectrometry. The algorithm successfully classified all 50 cancer and 50 non-cancer serum samples in a “training set,” and all 50 cancer samples in a “masked set;” however, 3 of 66 non-cancer samples were incorrectly classified (specificity 95 percent). The authors concluded, “These findings justify a prospective population-based assessment of proteomic pattern technology as a screening tool for all stages of ovarian cancer in high-risk and general populations.”(40, p.572)
New technologies are likely to continue to improve the prospects for early intervention in disease processes and prevention of morbidity, but epidemiologic studies are the key to their potential, beginning at the population level for translation into tools for individualized preventive medicine. However, evaluating the contribution of genomic data to the evidence base for prevention only begins to address the issue of clinical utility, which presents a larger array of complex issues. These include the probable timing and severity of disease outcomes, availability and effectiveness of interventions, and costs of alternative tests, interventions, and treatments.(41) Furthermore, the needs and preferences of tested individuals and their family members must be taken into account.
Even if we recognize the potential of genomics for preventive medicine, what does it mean for public health? Developments in the science and technology of genomics have prompted widespread use of the term “paradigm shift”(42) to describe the future of biological research and clinical medicine. Epidemiology, the basic science of public health, also faces challenges. Future studies of the distribution and causes of disease in human populations will be incomplete if they do not consider the potential contribution of genetic variation. Amid this sea of changes, the mission of public health remains the same: to prevent morbidity and mortality using science-based approaches that serve the interests of the total population, with special responsibility for underserved communities. However, new understanding of gene-environment interactions in disease etiology and progression may suggest interventions that require rethinking the “one size fits all” paradigm for public health interventions.
As more genetic tests are developed and marketed for use in public health and healthcare settings, it will be important to evaluate the value they add to existing interventions. Public health policies, backed by strong epidemiologic research, must provide a balance to intense commercial pressures, which have identified high-technology screening tests, including those based on genomics, as a new opportunity for direct marketing to individuals.(43-44) Ultimately, the public will benefit only when genetic tests are used appropriately, interventions are tailored to those at risk, and access is assured. Thus public health institutions clearly have a role in helping realize the potential of genetic information to prevent disease and improve health, by developing appropriate research and policies, and by helping educate healthcare providers and the public.
- Collins FS, McKusick VA. Implications of the Human Genome Project for medical science. JAMA 2001;285:540-544.
- Pagon RA, Tarczy-Hornoch P, Baskin PK, et al. GeneTests-GeneClinics: Genetic testing information for a growing audience. Hum Mutat 2002;19:501-509.
- GeneTests. Web site. Available at http://www.genetests.org/ .
- Yoon PW, Chen B, Faucett A, et al. Public health impact of genetic tests at the end of the 20th century. Genet Med 2001;3:405-410.
- Online Mendelian Inheritance in Man. Web site. Available at http://www.ncbi.nlm.nih.gov/sites/entrez?db=omim.
- Peltonen L, McCusick VA. Genomics and medicine: dissecting human disease in the postgenomic era. Science 2001;291:1224-1229.
- Collins FS. Shattuck lecture--medical and societal consequences of the Human Genome Project. N Engl J Med 1999;341:28-37.
- Khoury MJ, Little J. Human genome epidemiologic reviews: the beginning of something HuGE. Am J Epidemiol 2000;151:2-3.
- Ellsworth DL, Manolio TA. The emerging importance of genetics in epidemiologic research. I. Basic concepts in human genetics and laboratory technology. Ann Epidemiol 1999;9:1-16.
- Task Force on Genetic Testing. Ensuring the safety and effectiveness of genetic tests. In Holtzmann NA, Watson MS (eds) (1997). Promoting safe and effective genetic testing in the United States: final report of the Task Force on Genetic Testing.
- Whittemore AS, Nelson LM. Study design in genetic epidemiology: theoretical and practical considerations. Monogr Natl Cancer Inst 1999;26:61-69.
- Rothman KJ, Greenland S. Case-control studies. In Modern Epidemiology, 2nd ed, Rothman KJ (ed). Philadelphia: Lippincott Williams & Wilkins. 1998. pp.93-114.
- Ferlay J, Bray F, Pisani P, Parkin DM. Globocan 2000: Cancer Incidence, Mortality and Prevalence Worldwide. International Agency for Research on Cancer, World Health Organization. Lyon, France: IARC Press, 2001.
- Bodmer W. Familial adenomatous polyposis (FAP) and its gene, APC. Cytogenet Cell Genet 1999;86:99-104.
- Dunlop MG, Farrington SM, Nicholl I, et al. Population carrier frequency of hMSH2 and hMLH1 mutations. Brit J Cancer 2000;83:1643-1645.
- Lynch HT, Lynch JF. Hereditary nonpolyposis colorectal cancer. Semin Surg Oncol 2000;18:305-313.
- Fuchs CS, Giovannucci EL, Colditz GA, et al. A prospective study of family history and the risk of colorectal cancer. New Engl J Med 1994;331:1669-1674.
- Cole TRP, Sleightholme HV. ABC of colorectal cancer: the role of clinical genetics in management. BMJ 2000;943-946.
- Samowitz WS, Curtin K, Lin HH, et al. The colon cancer burden of genetically defined hereditary nonpolyposis colon cancer. Gastroenterology 2001;121:830-836.
- Yoon P, Peterson-Oehlke KL, Scheuner MT, et al. Can family history be used as a tool for public health and preventive medicine? Genet Med 2002;4:304-310.
- Tomeo CA, Colditz GA, Willett WC, et al. Harvard report on cancer prevention. Volume 3: Prevention of colon cancer in the United States. Cancer Causes and Control 1999;10:167-180.
- Rockhill B, Kawachi I, Colditz GA. Individual risk prediction and population-wide disease prevention. Epidemiol Rev 2000;22:176-180.
- Pearce N. Traditional epidemiology, modern epidemiology, and public health.
Am J Public Health. 1996;86:678-683.
- Rockhill B. The privatization of risk. Am J Public Health 2001;91:365-368.
- Wald NJ, Hackshaw AK, Frost CD. When can a risk factor be used as a worthwhile screening test? BMJ 1999;319:1562-1565.
- Holtzman NA, Marteau TM. Will genetics revolutionize medicine? New Engl J Med 2000;343:141-144.
- Wang X, Moylan B, Leopold DA, et al. Mutation in the gene responsible for cystic fibrosis and predisposition to chronic rhinosinusitis in the general population. JAMA 2000;284:1814-1819.
- Khoury MJ. Will genetics revolutionize medicine? (letter) New Engl J Med 2000;343:1497.
- Khoury MJ, Wagener DK. An epidemiological evaluation of the use of genetics to improve the predictive value of disease risk factors. Am J Hum Genet 1995;56:835-844. [Erratum, Am J Hum Genet 1996;58:253.]
- Pharoah PDP, Antoniou A, Bobrow M, Zimmern RL, Easton DF, Ponder BAJ. Polygenic susceptibility to breast cancer and implications for prevention. Nat Genet 2002;31:33-36.
- The Evidence-Based Medicine Working Group. Users’ guides to the medical literature: a manual for evidence-based clinical practice. Guyatt G, Rennie D (eds.) Chicago: American Medical Association, 2002.
- a. Sotos JG, Rienhoff Y Jr. Will genetics revolutionize medicine? (letter). New Engl J Med 2000;343:1496.
b. Yang Q, Khoury MJ, Botto L, Friedman JM, Flanders W. Improving the predicition of complex diseases by testing for multiple disease-susceptibility genes. Am J Hum Genet 2003; 72:636-649
- Beaty TH, Khoury MJ. Interface of genetics and epidemiology. Epidemiol Rev 2000;22:120-125.
- Satten GA, Flanders WD, Yang Q. Accounting for unmeasured population substructure in case-control studies of genetic association using a novel latent-class model. Am J Hum Genet 2001;68:466-477.
- Zerba KE, Ferrell RE, Sing CF. Genetic structure of five susceptibility gene regions for coronary artery disease: disequilibria within and among regions. Hum Genet 1998;103:346-354.
- Sing CF, Haviland MB, Reilly SL. Genetic architecture of common multifactorial diseases. Ciba Foundation Symposium 1996;197:211-229.
- Potter JD. Colorectal cancer: molecules and populations. J Natl Cancer Inst 1999;91:916-932.
- Traverso G, Shuber A, Levin B, et al. Detection of APC mutations in fecal DNA from patients with colorectal tumors. N Engl J Med 2002;346:311-20.
- Schwartz RS. A needle in a haystack of genes. N Engl J Med 2002;346:302-304.
- Petricoin EF, Ardekani AM, Hitt BA, et al. Use of proteomic patterns in serum to identify ovarian cancer. Lancet 2002;359:572-7.
- Evans JP, Skrzynia C, Burke W. The complexities of predictive genetic testing. BMJ 2001;322:1052-1056.
- Subramanian G, Adams MD, Venter JC, Broder S. Implications of the human genome for understanding human biology and medicine. JAMA 2001;286:2296-307.
- Lee TH, Brennan TA. Direct-to-consumer marketing of high-technology screening tests. N Engl J Med 2002;346:529-531.
- Personalized medicine: opportunities and challenges for commercialization. http://www.lifesciencesinfo.com/email/2707/.