The Value of Genetic and Genomic Technologies: Workshop Summary (2010)

Bruce Blumberg, M.D.

Northern California Kaiser Permanente

The Permanente Medical Group

Anne is a recently divorced 36-year-old MBA financial analyst. She has always considered herself to be both healthy and health conscious. She is an only child, with a mother of English ancestry and a father of mixed Eastern European descent. Her past medical history is notable only for a mildly abnormal glucose tolerance test at age 31 during her second pregnancy. This was not medically followed subsequent to the pregnancy. Anne prides herself on her careful diet, and she runs on a treadmill at a workplace gym at least 3 times a week. Despite these efforts she is 15 pounds overweight according to a table that she found in a popular magazine. She has never smoked. She has been tired lately, perhaps related to the demands of single motherhood.

Anne’s mother is 67 years old and was treated at age 59 for melanoma, but Anne knows no further details. Her mother has recently had mildly elevated blood sugars, and her doctor is considering beginning oral hypoglycemic therapy. Anne’s father is 70 and is taking a statin for hypercholesterolemia and a beta-blocker for hypertension. He had a very mild heart attack in his late 40s and takes prophylactic aspirin. Two of the father’s maternal first cousins are said to have died of colon cancer in their 40s, but, again, no details are available. His paternal aunt died of breast cancer in her late 30s. There may be other relevant conditions in the family history, but no physician has ever asked Anne about them.

Anne has read articles in the New York Times and elsewhere about the availability of genomic screening for health risk assessment. As is her usual

practice in health-related matters, Anne extensively reviewed the topic on the Internet and compared the tests offered by several different companies. She is especially concerned about her future risk of diabetes and coronary artery disease, based on her personal and family histories, so she selects the lab that places the greatest emphasis on these conditions on its website.

When Anne returns from vacation, a printed report awaits her. If she understands the report correctly, she is relieved to learn that her risk of type 2 diabetes is 10 percent below that of the general population. On the other hand, her future risk of coronary artery disease is 20 percent above the general population risk. It is unclear from the report if the risks have taken her family or personal histories into account or if the risks were calculated exclusively based on the genomic results. As she continues to read the report, Anne learns that her breast cancer risk is 30 percent above the general population risk and she is dismayed to read that her Alzheimer’s disease risk is double that of the general population. Finally, she is surprised and confused when she reads that she is a carrier for hemochromatosis and alpha-1 antitrypsinase deficiency, two conditions with which she is entirely unfamiliar. She wonders if these findings might explain her recent fatigue.

Anne immediately calls her doctor’s office, but the earliest available appointment is not for two weeks. When she arrives for the appointment she appears to be mildly agitated. She brings a copy of the report and has a two-page list of hand-written questions prepared for her doctor. Here are the questions on the first page:

The questions on Anne’s second page are more difficult to answer and are left to the imagination of the discussants.

The joint occurrence of two genetically determined characteristics in a population at a frequency that is greater than expected according to the product of their independent frequencies.

In simpler terms, any two events that occur together at a non-random frequency are associated. The relationship is statistical and does not imply (nor does it exclude) causality. The concept of association predates the elucidation of the human genome and, with respect to common diseases, is meaningful not only at the DNA level but also at the level of protein variations and at the level of observable physical characteristics. As an example of a protein-level association, it has been known for many decades that peptic ulcer disease is non-randomly associated with blood type, with individuals of blood type A and O being at higher lifetime risk for this illness than individuals of other blood types. As an example of an association at the level of observable physical characteristics, it is well established that the incidence of prostate cancer varies by ethnic/racial group, with Asians at low risk and African Americans at particularly high risk. Given the fact that protein and physical variations have a basis in variations at the DNA level, it is not surprising that the principles of association extend to the genomic level.

A DNA sequence variation occurring when a single nucleotide in the genome differs between individuals (or between paired chromosomes in a single individual).

For example, if the sequences for the same DNA fragments from different individuals are AAGCCTA and AAGCTTA, we can see that the fragments are different in a single nucleotide. For such a variation to be considered a SNP, it must occur in at least 1 percent of the population. SNPs are extremely common, making up about 90 percent of all human genetic variation and occurring every 100 to 300 bases along the 3 billion base human genome. Most SNPs are “silent” and have absolutely no effect on protein structure or observable physical characteristics. The description of the human genome has provided a greatly magnified view of human genetic variation and offered a widely expanded opportunity to look for associations of these genomic variants with common diseases. Another common way of describing such associations is to speak of a SNP as a “marker” for predisposition to an associated disease.

An examination of genetic variation across the entire genome, designed to identify associations of DNA variants with observable traits.

The primary goal of most GWAS studies is the identification of disease associations that will provide biological insights into disease pathogenesis. The application of GWAS findings to personalized risk assessment may be viewed as a clinical byproduct of epidemiologically motivated research. Because of the large number of relatively weak associations that may be identified in a GWAS study, it is inevitable that some statistically significant associations will be spurious. As a general rule, associations are therefore deemed relevant only after replication in multiple GWAS studies.

To illustrate, a GWAS study of bipolar disease would require two populations, one composed of patients with bipolar disease and the other a control population consisting of people without known bipolar disease. The genomes of both groups would be analyzed, paying particular attention to SNPs (and other types of known genetic variation). Using the SNP described above, it might be found that the AAGCCTA variant was found in 4 percent of bipolar patients, with 96 percent of this group manifesting the more common AAGCTTA. If the control population revealed that only 3 percent possessed AAGCCTA, and if the sample size were large enough, the difference between 3 percent and 4 percent might be statistically significant, identifying an association between the “C” allele and bipolar disease.

It is important to point out that a “risk” allele such as “C” in this

example may confer only a small increase in the risk of the disease under study. In our example, most patients with “C” do not develop bipolar disease. It is, after all, seen in 3 percent of the healthy control population. Most patients with bipolar disease (96 percent) do not even have the “C” allele. Nevertheless, the data demonstrate that an individual with the “C” allele has a statistically higher chance of having bipolar disease than does an individual with the “T” allele. Since most genes are found in a single pair, the data might be further analyzed to show that individuals with “CC” genotypes are at higher risk of bipolar disease than are “CT” individuals, who are at higher risk than “TT” individuals.

The population prevalence of bipolar disease is approximately 1 percent. Inventing some results to continue our example, we might find that “TT” individuals have 0.9 percent risk, “CT” individuals have a 1.2 percent risk, and “CC” individuals have a 2 percent risk of bipolar disease. Care must be exercised when, in this example, the claim is accurately made that the “C” allele confers a 20 percent increased risk of bipolar disease because this effect, when expressed in this fashion, could exaggerate the practical importance of an increase from the population risk of 1 percent to a modified risk of 1.2 percent. While this example was entirely fabricated, the magnitude of risk adjustment allowed by GWAS-based associations is most commonly in a range similar to the example.

In real life, the issue is even more complicated, because a number of SNPs at different sites may be found to be associated with an increased or decreased risk of bipolar disease. The final calculation of risk requires a complex computational model that incorporates data from each of the SNPs selected for analysis. There is no universal consensus on SNP selection, so it is entirely possible for one model, using one set of SNPs, to predict an increased risk of bipolar disease, while another model, employing a different (perhaps overlapping) set of SNPs, predicts a different risk for the same individual. In the worst case scenario, one model could predict an increased risk whereas a second model could predict a decreased risk for the same individual.

The identification, among apparently healthy individuals, of those who are sufficiently at risk of a specific disorder to justify a subsequent diagnostic test or procedure or to direct preventive action.

The general requirements for a valid and clinically useful screening program are well established:

1. The disease must be well-defined.

2. The prevalence of the disease must be known.

1. The disease must be medically important.

2. There must be an effective treatment or preventive measure available.

3. The test must be simple, safe, widely available, affordable, and reliable.

4. The test should have an acceptably low risk of ambiguous results.

5. The test must be accompanied by adequate pretest counseling, informed consent, and follow-up services.

The World Health Organization has set forth a similar set of criteria¹:

1. The condition sought should be an important health problem.

2. There should be an accepted treatment for patients with recognized disease.

3. Facilities for diagnosis and treatment should be available.

4. There should be a latent or early symptomatic stage.

5. There should be a suitable test or examination.

6. The test should be acceptable to the population.

7. The natural history of the condition, including development from latent to declared disease, should be adequately understood.

8. There should be an agreed policy on who to treat as patients.

9. The cost of case finding (including diagnosis and treatment of patients diagnosed) should be economically balanced in relation to possible expenditure on medical care as a whole.

10. Case finding should be a continuing process and not a “once and for all” project.

The Institute of Medicine has also offered a set of criteria²:

1. Genetic screening, when carried out under controlled conditions, is an appropriate form of medical care when the following criteria are met:

   1. There is evidence of substantial public benefit and acceptance, including acceptance by medical practitioners.

   2. Its feasibility has been investigated and it has been found that benefits outweigh costs; appropriate public education can be

3. carried out; test methods are satisfactory; laboratory facilities are available; and resources exist to deal with counseling, follow-up, and other consequences of testing.

4. An investigative pretest of the program has shown that costs are acceptable; education is effective; informed consent is feasible; aims of the program with regard to the size of the sample to be screened, the age of the screenees, and the setting in which the testing is to be done have been defined; laboratory facilities have been shown to fulfill requirements for quality control; techniques for communicating results are workable; qualified and effective counselors are available in sufficient number; and adequate provision for effective services has been made.

5. The means are available to evaluate the effectiveness and success of each step in the process.

A number of commercial laboratories directly offer consumers an analysis of a large panel of selected SNPs that are used to calculate the future risk of developing a number of diseases in the tested individual. Surveying the websites of the three most prominent firms, Navigenics, 23andMe, and deCODE, there are 10 common diseases included in the panels of all three companies:

1. age-related macular degeneration

2. atrial fibrillation

3. breast cancer

4. celiac disease

5. Crohn’s disease

6. prostate cancer

7. psoriasis

8. rheumatoid arthritis

9. type 2 diabetes mellitus

10. deep vein thrombosis

There are a number of additional conditions that appear on the list of one or two of these three companies. Examples include Parkinson’s disease, Alzheimer’s disease, lupus, osteoarthritis, multiple sclerosis, lung cancer, kidney stones, gallstones, and gout. Since testing is offered directly to consumers, a health professional typically has not been involved in the ordering of such tests or in pre-test counseling. The companies do have genetic counselors and other professionals on staff to address consumers’ questions that may arise. Since most governmental bodies require that medical tests be performed only at the request of a physician or other qualified health

practitioner, these commercial firms emphasize that the service they offer does not constitute a medical test, as defined in the law. In an open letter to the medical community, 23andMe thus describes its service as follows:

Genomic analysis for the prediction of common disease risk has been a controversial practice. Proponents argue that individual autonomy requires unrestricted access to any potentially available genetic information. Efforts to limit access have been branded as paternalistic. This argument continues with the assertion that the discovery of reduced risk could be reassuring and the discovery of increased risk could motivate healthy changes in lifestyle that might mitigate the increased risk. For example, an individual found to be at a risk for type 2 diabetes that is higher than that of the general population might be spurred to institute a weight-reduction diet.

Opposing arguments point to the lack of consensus on the genetic markers selected for study and the consequent inconsistency of risk prediction. Risk modification is most typically of small magnitude (as in the example above) and often does not exceed the risk stratification that

could be achieved by more traditional methods (e.g., an assessment of weight and family history in predicting the risk of type 2 diabetes). Some of the conditions being assessed offer no clear risk-modifying intervention (e.g., rheumatoid arthritis) and some risk-modifying interventions are strongly indicated independent of one’s genetic risk (e.g., smoking cessation is always recommended regardless of an individual’s precise risk of lung cancer). Fears have been expressed that knowledge of reduced risk might encourage unhealthy lifestyle choices (i.e., a real risk of testing). Furthermore the paucity of pre-test counseling and the large panel of assessed risks raise concerns that patients will be unprepared to deal with the results. A patient motivated to seek testing because of a strong family history of prostate cancer may or may not be prepared to learn of his increased risk for the future development of Alzheimer’s disease. Patients’ personal physicians may not be in the best position to assist with result interpretation, as they had no role in ordering the tests and are unlikely to have received the requisite education to contribute any real expertise.

The majority of observers probably occupy some middle ground between these opposing opinions. This middle group points to the lack of evidence upon which to judge the clinical utility, safety, or cost-effectiveness of genomic risk assessment. There are many examples from the pre-genomic era of the failure to translate statistically validated risk stratification into an effective screening regimen or intervention. (Prostate cancer screening in African American men is a familiar example.) Neutral observers question the adherence of genomic risk assessment to the previously described principles of population screening. The current state of direct-to-consumer testing also runs the risk of exacerbating health disparities by offering an expensive test that is unlikely to be covered by health insurance (because of the lack of evidence of clinical utility or cost effectiveness). For example, the Navigenics website offers testimonials from its customers in a section titled “Success Stories.” The selected group consists of an Internet entrepreneur, a psychotherapist, a software analyst, a venture capital executive, a journalist, an Internet executive, an attorney, a marketing consultant, and a marketing executive. Even by Silicon Valley standards, the tested group is highly unrepresentative of the general population. Such unequal access to service leads even those who recognize the potential health benefits of genomic risk assessment to wonder if it is possible to reconcile personalized medicine with public health.

A number of professional societies have published policy statements regarding direct-to-consumer genomic testing. Depending on the focus of these societies, the statements either broadly or more narrowly address the

technical, clinical, ethical, legal and/or social aspects of such testing. These policy statements include:

The American College of Clinical Pharmacology:

Ameer, B., and N. Krivoy. 2009. Direct-to-consumer/patient advertising of genetic testing: A position statement of the American College of Clinical Pharmacology. J Clin Pharmacol 49:886–888.

The American College of Medical Genetics:

www.acmg.net/AM/Template.cfm?Section=Terms_and_Conditions&termsreturnurl=Section=Policy_Statements&Template=/CM/ContentDisplay.cfm&ContentID=2975, published 2008.

The American Society of Human Genetics:

Hudson K., G. Javitt, W. Burke, and P. Byers, with the ASHG Social Issues Committee. 2007. ASHG Statement on Direct-to-Consumer Genetic Testing in the United States. Am J Hum Genet 81:635–637.

The American Society of Clinical Oncology:

http://jco.ascopubs.org/cgi/reprint/JCO.2009.27.0660v1

The National Society of Genetic Counselors:

www.nsgc.org/about/position.cfm#DTC, adopted 2007.

Caulfield, T., N. M. Ries, P. N. Ray, C. Shuman, and B. Wilson. 2010. Direct-to-consumer genetic testing: Good, bad, or benign? Clin Genet 77:101–105.

Evans, J. P., and R. C. Green. 2009. Direct-to-consumer genetic testing: Avoiding a culture war. Genet Med 11:568–569.

Khoury, M. J., A. Berg, R. Coates, J. Evans, S. M. Teutsch, and L. A. Bradley. 2008. The evidence dilemma in genomic medicine. Health Affairs 27:1600–1611.

Ng, P. C., S. S. Murray, S. Levy, and J. C. Venter. 2009. An agenda for personalized medicine. Nature 461:724–726.

Wasson, K., E. D. Cook, and K. Helzlsouer. 2006. Direct-to-consumer online genetic testing and the four principles: An analysis of the ethical issues. Ethics Med 22:83–91.