Personalized Medicine in the Era of Genomics

Enthusiastic predictions about personalized medicine have surrounded the sequencing of the human genome. As commonly used, the term predicts a leap forward in disease prevention and drug treatment, based on knowledge of individual genetic susceptibilities.^{1 - 2} According to Guttmacher and colleagues, “genomics-based knowledge and tools promise the ability to approach each patient as the biological individual he or she is, thereby radically changing our paradigms and improving efficacy.”³ Some emerging tests support this promise: mutations in the BRCA1 and BRCA2 genes identify women who have a high lifetime risk for breast and ovarian cancer⁴ and who are candidates for breast magnetic resonance imaging screening⁵ or prophylactic surgery.⁶ Similarly, many pharmacogenetic tests are under development as a means to improve the safety and efficacy of drug treatment.⁷ In each case, the useful application of a genetic test is linked to an intervention that improves health outcomes.

Nevertheless, claims of a new medical paradigm based on genomics merit careful scrutiny. The exhortation to prepare for a “genomics revolution”⁸ often assumes that genetic risk is different in kind from other health risks. Experienced clinicians might reasonably question this assumption and also might debate how much genetic testing is likely to contribute to the medical care of individual patients. Although genomic research will certainly yield translational benefits, a realistic appraisal suggests that genomics will make only modest contributions to personalized medicine as it is traditionally practiced.

As the BRCA example demonstrates, genetic tests can sometimes predict substantial risks for common diseases. However, BRCA mutations are rare and account for only a small minority of breast cancer cases; most genetic contributors to breast cancer risk have only small effects.^{9 - 11} Indeed, the call to assemble large cohorts for genomic research^{12 - 13} derives from the understanding that most gene variants associated with common diseases will have modest effects, often generating relative risks of 2 or less. Large study populations are therefore needed to achieve sufficient power to evaluate and replicate them.¹² As with other health risk factors, population risks of this magnitude will generally perform poorly in predicting outcomes for individual patients.^{14 - 15} By extension, many valid gene-disease associations are likely to have limited utility as predictive genetic tests.

The challenge can be further demonstrated by promising research on the genetics of age-related macular degeneration (AMD). Variants in several genes have now been identified as risk factors for this condition,^{16 - 19} with odds ratios for risk alleles ranging from 1.25 to 3 in the heterozygous state and 2 to 7 in the homozygous state. Some protective variants have also been identified. The minor allele frequencies for most of the genes are high, ranging from 10% to 40% in most studies.^{16 - 19} These findings are of great scientific interest because they confirm the importance of 2 mechanisms in the etiology of AMD, involving the complement pathway and angiogenesis. Genomic research provides tools for identifying important new biological pathways involved. But do such findings represent a step forward for personalized medicine?

Genetic risk information can prevent disease only if it improves the use of behavioral or medical interventions. The strongest case can be made when a unique intervention is needed for individuals with a particular genotype. For example, a low phenylalanine diet prevents mental retardation in patients with phenylketonuria, but it is harmful in those with a normal genotype.²⁰ Testing for BRCA represents a similar if less dramatic example, providing information that justifies the use of aggressive interventions for a small number of high-risk women.⁶

On the basis of these examples, a more general tailoring of health care to individual genetic risk is intuitively appealing. However, many effective prevention strategies, such as Papanicolaou testing and childhood immunizations, are appropriately offered independent of risk status. One of the most important prevention strategies available to patients falls into this category: the constellation of behaviors that constitute a healthful lifestyle—regular exercise, a diet low in fats and high in fruits and vegetables, and avoidance of smoking and obesity.

Genetic testing has nevertheless been suggested as a guide to lifestyle changes. Both smoking and obesity are risk factors for AMD, for example, and researchers have suggested that identifying patients at increased risk for AMD might provide them with a rationale for smoking cessation or dieting.¹⁶ Aside from doubts about whether genetic risk knowledge motivates behavioral change,²¹ this strategy makes the questionable assumption that risk-based prevention is preferable to population-wide measures,¹⁵ ignoring in this case the broad population benefits for a variety of health outcomes—including the treatment or prevention of lung cancer, heart disease, hypertension, dyslipidemia, obesity, diabetes, depression—that would flow from even small improvements in diet, exercise, and smoking rates.

The information provided by genetic testing can be increased by testing for multiple disease-associated gene variants. In the case of AMD, variants in at least 4 different genes have been implicated in risk.^{16 - 19} Multigene testing may represent a means to identify unusual individuals who are at high risk due to the inheritance of multiple risk variants. For example, one study estimated that individuals who are homozygous for 2 different gene variants associated with AMD have a 50-fold higher risk for the disease.¹⁶

Identification of individuals with this level of risk would have clinical utility if specific measures could be offered to prevent or delay the onset of AMD. However, such individuals are rare—about 1 in 400—and tests to find them also identify many more individuals with lower-risk genotypes. For every individual estimated to have a 50-fold excess risk of AMD, for example, about 22 individuals with moderate risk (odds ratio, 6-10) and almost 100 with lower risk (odds ratio, 3-4) would be identified.¹⁶ Similar results were obtained from a model incorporating variants from 3 genes; this model could identify individuals across more than a 100-fold spectrum of risk for AMD, but the risk for most individuals was close to the population average.¹⁹ Age-related macular degeneration also occurs in people without known risk variants. At each level of risk, individual outcomes will vary because of other risk factors unaccounted for by the genetic test.^{14 - 15} If novel treatments are associated with any serious adverse events, the risk-benefit trade-offs are likely to vary considerably across this spectrum of risks.

As gene panels expand, genetic testing will also inevitably identify genotypes with unknown clinical significance,²² including combinations of high-risk and protective-gene variants with unknown effects. Thus, the evaluation of a panel test for multiple variants must take into account both the health outcomes for high-risk individuals and the potential harms of identifying patients with modest and poorly characterized risks. These harms might include anxiety, discrimination, and exposure to unproven therapies or to therapies that may have serious adverse effects.

Each new genetic test will need to be evaluated and assessed to demonstrate that the overall health benefits exceed the harms before it is implemented in practice. The fundamental principle is that genetic risk information will be useful only if it guides more effective, or more cost-effective, use of medical interventions than can be achieved without the risk information. As described in the methodology of the US Preventive Services Task Force,²³ definitive evidence requires controlled studies that evaluate all the relevant outcomes of testing and the effects of any associated clinical interventions. Genetic tests for AMD risk might be useful, for example, if they could be shown to result in improved population vision outcomes through more effective use of ophthalmic screening or pharmacological treatment in risk groups identified by testing and did not result in unacceptable harms.

In essence, testing for genetic susceptibilities needs to be justified by evidence that the gene variant or profile modifies the effect of treatment (eg, by increasing efficacy or safety) before the test can be recommended as a method for allocating treatment. Even if a treatment effect were present, issues of cost, acceptability, or harms of the testing process might argue against widespread use of the test. Of note, when risk information is useful to identify candidates for interventions, genetic risk may not be the best approach. For example, focusing AMD screening on smokers and individuals with a history of intense light exposure²⁴ might be more useful than screening based on genetic risk.

New clinical interventions based on genomic knowledge are another, although indirect, route to clinical utility. As with AMD, many findings from genomic research are likely to provide new clues to disease biology by the identification of genes and biological pathways unexpectedly associated with the disease process.^{15 ,25} Over time—typically decades—a better understanding of disease biology may produce new approaches to treatment or prevention. Genomic research on AMD might in this way make important contributions to prevention of vision loss associated with AMD. This method is a slow process at best, and success is not guaranteed: 60 years after defining the molecular cause of sickle cell disease, definitive treatment is still lacking. More important, if interventions are developed based on genomic research, their use will not necessarily depend on genetic testing for appropriate use. New therapeutics for AMD, for example, might well prove effective for all patients with early signs of the disease.

Use of genetic risk information to guide intervention must be justified by data demonstrating improved outcomes, reduced costs, or both. Given the uncertainties of risk prediction and the small effect size of most genetic risk factors for common diseases, predictive genetic testing is likely to play only a modest role in the prevention of these diseases, with relatively little medical care improved by knowledge of a patient's genetic profile. The limitations of genetic risk prediction do not reduce the importance of personalized medicine. Rather, they serve as a reminder that genuinely personal health care, as practiced by physicians for centuries, is based on the relationship between patient and physician rather than on any particular technology. Even in the genomic era, the focus on individual patient needs and concerns will remain at the core of health care; and if genetic testing diverts physicians' attentions away from the specific concerns of the patient, it may interfere with the practice of personalized medicine.

Knowledge of a patient's life circumstances, sometimes acquired slowly over the course of longitudinal care, allows a clinician to tailor both therapeutic choices and the decision-making process. Family circumstances, such as the care of an ailing spouse, may limit a patient's options. Competing morbidities may make some medications, diets, or exercise programs more appropriate than others. A patient who requires a wheelchair needs a different cardiac prevention program than a marathon runner—and genetic profiling may not help either to make sound choices.

A sympathetic understanding of a patient's values and decision-making style can also improve care. Prevention efforts often involve weighing the pros and cons of different choices, as in decisions about the use of prostate-specific antigen testing or hormone replacement therapy. Similar pros and cons are likely to arise with genetic profiling, particularly if tests to identify rare high-risk individuals involve the cost of finding many people with moderate risk and uncertain clinical options. Tailoring care to the individual includes the clinician's effort to provide information and support about these test choices in the manner best suited to each patient.

Personalized medicine has always been a component of good medical practice. Genetic tests may provide new tools, but they do not change the fundamental goal of clinicians to adapt available medical tests and technologies to the individual circumstance of their patients. As genetic tests become widely available, personalized medicine will include assisting patients to make wise use of genetic risk assessment, taking into account the cautions discussed in this article. When genetic testing is used, the personalized nature of the care will extend well beyond the patient's base pair sequences.