Use of genomic profiling to assess risk for cardiovascular disease and identify individualized prevention strategies—A targeted evidence-based review

Abstract

Purpose: To address the key question of whether using available “cardiogenomic profiles” leads to improved health outcomes (e.g., reduction in rates of myocardial infarction and stroke) and whether these profiles help in making medical or personal decisions.

Methods: A targeted evidence-based review based on published Evaluation of Genomic Applications in Practice and Prevention methodologies.

Results: No study addressed the overarching question directly. Evidence for the analytic validity of genomic profiles was inadequate for most genes (scale: convincing, adequate, and inadequate), but based on gray data, the analytic sensitivity and specificity might be adequate. For the 29 candidate genes (58 separate associations reviewed), the credibility of evidence for clinical validity was weak (34 associations) to moderate (23 associations), based on limited evidence, potential biases, and/or variability between included studies. The association of 9p21 variants with heart disease had strong credibility with odds ratios of 0.80 (95% confidence interval: 0.77–0.82) and 1.25 (95% confidence interval: 1.21–1.30), respectively, for individuals with no, or two, at-risk alleles versus those with one at-risk allele. Using a multiplicative model, we combined information from 24 markers predicting heart disease and from 13 markers for stroke. The areas under the curves (64.7% and 55.2%, respectively), and overall screening performance (detection rates of 24% and 14% at a 10% false-positive rate, respectively) do not warrant use as stand-alone tests.

Conclusion: Even if genomic markers were independent of traditional risk factors, reports indicate that cardiovascular disease risk reclassification would be small. Improvement in health could occur with earlier initiation or higher adherence to medical or behavioral interventions, but no prospective studies documented such improvements (clinical utility).

INTRODUCTION

CVD is a major contributor to morbidity and mortality in the United States. An estimated 80 million adults have one or more types of CVD (48% occurring in individuals aged 60 years or older), and preliminary 2006 mortality data indicate that CVD accounts for one in every 2.9 deaths.¹ The 2005 overall death rate from CVD was 279 per 100,000, with death rates higher in men than women, and in blacks than whites. Consequently, the burden of CVD is high, and the cost (direct and indirect) in 2008 is estimated at 448.5 billion dollars.² Prevention and management of CVD, particularly ischemic heart disease and stroke, present a difficult challenge for health care and public health.³ Major nonmodifiable risk factors include increasing age, male gender, and heredity, whereas modifiable risk factors include smoking, hypertension, dyslipidemia, obesity, physical inactivity, and diabetes.^4–6 African Americans have more severe hypertension and increased risk of heart disease compared with whites.^7,8 In men, the average annual rate of initial cardiovascular events increases from three per 1000 at 35–44 years to 74 per 1000 at 85–94 years. Similar increases occur in women but approximately a decade later in life.²

The term “CVD” encompasses a broad range of disorders of the heart and circulatory system, and classification of CVDs is not uniformly applied in the literature. We defined two main groups of outcomes based on the International Statistical Classification of Diseases and Related Health Problems (I00-I99).⁹ The grouping of CHD includes coronary artery disease, ischemic heart disease, and MI. The second grouping of stroke includes intracerebral and subarachnoid hemorrhage, ischemic stroke, and other diseases (e.g., cerebral infarction and occlusion/stenosis of cerebral arteries). In addition to the conventional risk factors, biomarkers have also been identified for use in potentially improving the prediction of CVD risk. In one study, 10 biomarkers (e.g., C-reactive protein, fibrinogen and homocysteine) provided only modest additive contribution to risk prediction compared with the use of conventional risk factors alone.¹⁰

Over the last decade, candidate genes for CVD susceptibility have been the focus of many relatively small studies that include one or a few genes at a time.¹¹ These candidate gene-disease association studies often have lacked power, and the reproducibility in subsequent confirmatory studies has generally been poor. More recently, genome-wide association (GWA) studies have tested tens of thousands of single-nucleotide polymorphisms (SNPs) in an effort to identify associations with CVD.^11–13 Despite notable gaps in knowledge,¹⁴ genomic profiling tests for CVD (sometimes referred to as “heart health”) continue to be offered in the health care market and can be ordered online without the involvement of a physician. The goal of this evidence review is to assess what is known about the analytic validity, clinical validity, and clinical utility of these cardiogenomic profiling tests.

A recent HuGE Navigator search (Phenopedia, using the disease term “cardiovascular diseases”) identified 6493 publications, 3331 genes, 169 meta-analyses, and 60 GWA study publications. This targeted review addresses only those genes included on genomic profiles aimed at CVD or “heart health” that were available when the study was undertaken in mid 2008.

Clinical scenario

The clinical scenario focuses on individuals in the general population being offered “cardiogenomic profiling” or “heart health” tests. In general, these individuals will not have an existing diagnosis of a specific CVD event (e.g., stroke and MI) but may have intermediate findings (e.g., hyperlipidemia). The proposed clinical utility for testing is that the knowledge of a personal gene variant(s) associated with an increase in risk for CVD might improve health outcomes (e.g., morbidity and mortality) by (1) providing motivation beyond routine health messages to change health behaviors; (2) bringing potential risk to the attention of health care providers for clinical follow-up; or (3) making or revising treatment decisions. The evidence review was commissioned and will be used by the Evaluation of Genomic Applications in Practice and Prevention (EGAPP) Working Group (EWG) to inform the development of formal recommendations for clinical practice. This review addresses the overarching question: “Does the use of ‘cardiogenomic profiling’ lead to improved outcomes for the patient/consumer, or is it useful in medical or personal decision-making?”

METHODS

The process used to identify, review, analyze, evaluate, and summarize the evidence is briefly presented in this review, but an in-depth presentation of EWG methods has been published.¹⁵ Consultants from Women & Infants Hospital, Department of Pathology and Laboratory Medicine, with experience in evidence review of genetic tests were contracted by the Office of Public Health Genomics at the Centers for Disease Control and Prevention to perform the targeted review with the assistance of four EGAPP staff members and other outside consultants. Guidance was provided by a Technical Expert Panel.

An analytic framework and key questions (Table 1, Fig. 1) were developed and refined by the EWG and Technical Expert Panel members (see “Acknowledgments”), with support from EGAPP staff. The review focuses on clinical validity but addresses the limited information available on analytic validity and considers proposed measures of clinical utility in clinical practice and in direct-to-consumer settings. Standard methods used include systematic search criteria for identification of published and gray literature; application of inclusion/exclusion criteria; abstraction of data; meta-analysis; assessment of the quality of individual studies; and overall strength of evidence.¹⁵ Limiting key questions and truncating search strategies (e.g., targeted gray literature searches) are two common methods in targeted reviews.¹⁶ In reviewing the available evidence, questions from the ACCE (Analytic validity, Clinical validity, Clinical utility and Ethical, Legal and Social Implication) review framework were often used to identify and organize the specific information needed to address the key questions.¹⁷ The draft report was then revised in response to reviewers' comments, and the final manuscript was submitted to the EWG.

Table 1 Key questions relating to the analytic framework

Fig. 1

The analytic framework: Genomic profiles and cardiovascular disease. This schematic shows the analytic framework underlying the current review. The numbers indicate the four key questions contained in Table 1.

Data sources for analytic validity

PubMed®¹⁸ searches were performed for the alleles and SNPs (e.g., as “AGTR1” or “AGTR1 genotyping”) and specific terms (e.g., “analytic validity” and “clinical test”). Gray literature searches (e.g., company and genetic testing web sites) were conducted to collect information on laboratories offering testing for these markers and the methodologies used. When these sources were not sufficient, we contacted the companies offering heart health genomic panels for information, first by an e-mail questionnaire and then by phone (for those not responding).

Clinical validity literature search

A comprehensive review of all gene/CVD associations was determined by the EWG to be outside the scope of the review, which was limited to (1) addressing two major CVD outcomes groupings (i.e., CHD and stroke), (2) examining only those genes/polymorphisms included in cardiogenomic panels available in April 2008, and (3) using existing high-quality meta-analyses whenever possible. Literature searches were conducted using HuGE Navigator v1.3^19,20 after crosschecking a subset of genes by a PubMed search.¹⁸ Specific search strategies for each gene are contained in Appendix, Supplemental Digital Content 1, http://links.lww.com/GIM/A124. Reference lists of retrieved publications were examined for relevant studies. Searches were performed between June 2008 and January 2009. One investigator (L.M.N., S.M., or G.E.P.) had primary responsibility for each gene, and results were reviewed by another (usually G.E.P.). Discrepancies were resolved by discussion.

Criteria for inclusion of studies for clinical validity

To be included, a publication needed to be in English and include information about whites. The outcome needed to be a primary CVD event, heart disease, or stroke. Sufficient data needed to be present to express the effect size as an OR with confidence intervals. Existing meta-analyses were used preferentially, but we created structured summaries of original publications when no suitable meta-analyses were found.

Data analyses for clinical validity

Summary ORs and corresponding 95% CIs were derived using a random-effects model (Comprehensive Meta-analysis, Version 2, Englewood, NJ), from the original source (published meta-analysis), from a reanalysis of the reported data, or from a new literature summary. The preferred method for determining the summary ORs was to compare wild-type individuals (no at-risk alleles) with heterozygotes (one at-risk allele) and then separately compare wild type with homozygotes (two at-risk alleles). Other models were also used. For example, when the frequency of the at-risk allele was very low, heterozygotes and homozygotes were combined and compared against the wild type. Some existing meta-analyses included formal assessment of heterogeneity, usually a χ²-based Q statistic. This was converted to the I² statistic²¹ for ease of interpretation and comparison between studies. Some also included a formal examination of publication bias or stratified the summary OR by sample size. If sufficient data were available, we looked for potential publication bias.

Evaluating credibility of the cumulative evidence for a gene-disease association

In 2007, a consensus group recommended evaluation guidelines to assess the cumulative evidence provided by genetic association studies.²² These “Venice” criteria focus on amount of evidence, replication of evidence, and protection from bias. Each of these three criteria was assigned a grade of “A,” “B,” or “C” (see Appendix, Supplemental Digital Content 2, http://links.lww.com/GIM/A125, for more detail). Epidemiological evidence for significant association was rated as “strong” if the meta-analysis received three A grades, “moderate” if it received any B grade but not any C grade; and “weak” if it received a C grade in any of the three criteria.²² These grades provide a quick but powerful overview of the likely reliability of the reported summary estimates.

There is no generally accepted process to revise the effect size estimate, if credibility is rated as moderate or low due to potential biases (e.g., publication bias). To provide an indication of the extent of potential change in effect size, we performed a cumulative effects analysis²³ for combinations having a minimum of six studies (three of which included 500 or more participants). The analysis adds one study at a time, from smallest to largest confidence interval, to create multiple estimates for the summary OR. If the range of these cumulative ORs is wide, potential for bias exists. We defined the range as the difference between the final cumulative OR and the first stable estimate (the third of three consecutive estimates all within 10% of each other). Usually, but not always, this occurred within the first three large studies. The ranges <0.1, 0.1–0.19, and ≥0.2 were considered small, medium, and large ranges, respectively.

Combining CVD genetic markers

As a way to set a reasonable upper limit on the effect size of several genetic markers, we chose a multiplicative model that assumes each marker is an independent predictor of CVD risk. Before multiplication, the ORs were adjusted, so that there was no overall impact on CVD risk.²⁴ For example, if an at-risk genotype was present in 10% of the population and was associated with a summary OR of 1.1, an OR of 1.09 would be used for the 10% at risk and 0.99 for the remaining 90%. A Monte Carlo simulation generated 100,000 individuals with, and 100,000 without, heart disease (or stroke) with their associated cumulative ORs. When multiple variants were available for a given gene, the one with the strongest evidence was used. If two variants has similar credibility, the one with the largest impact (effect size time × at risk proportion) was used.

Clinical utility

A systematic review of the literature on the clinical utility of the 29 markers was not undertaken. Rather, PubMed searches¹⁸ focused on identifying potential benefits and harms associated with addition of CVD-associated alleles/SNPs to existing risk assessment algorithms based on conventional risk factors and on the potential for genetic information to motivate behavioral change. Articles from the clinical validity review were also reviewed for information related to clinical utility.

RESULTS

Tests for cardiogenomic risk assessment

Table 2 lists the eight genomic test panels related to CVD/heart health available in April 2008, with the relevant genes/variants included. There is significant variability among these tests including the CVD outcomes (e.g., MI, coronary artery disease, and stroke); panels and platforms used for testing; populations for which testing is recommended; personal and family history information collected; pretest information provided on potential benefits and harms of testing; ways in which results are presented and interpreted (e.g., gene by gene, risk algorithm, and general or personalized health messages); and whether physicians, genetic counselors, or other health care providers are involved. A summary of this information for each genomic test or panel is available in Appendix, Supplemental Digital Content 2, http://links.lww.com/GIM/A125. Table 3 contains the health implications for each of the genes, as described by the companies offering the tests.

Table 2 Currently available genetic tests for cardiovascular disease (CVD) along with the genes included on each panel

Table 3 Genes and associated health implications based on individual company website literature

Literature searching

To determine whether HuGE Navigator was sufficient to use as the platform for literature searching, we chose three genes (GPX1, SOD2, and MTR) and searched for gene-disease association studies using both HuGE Navigator v1.3 and PubMed. For GPX1, the HuGE Navigator search identified three studies, one of which satisfied the inclusion criteria. A similar PubMed search identified an additional 59 articles, none of which met inclusion criteria. For SOD2 and MTR, PubMed searches identified 161 and 43 additional articles, respectively. None of these satisfied the inclusion criteria. Given the ease, speed of use, and apparent completeness, we chose to rely mainly on HuGE Navigator searches for the identification of publications for the clinical validity analyses. In a separate analysis of the association of 9p21 SNP markers and heart disease,²⁵ we discovered that HuGE Navigator search was less sensitive in identifying articles relating to 9p21 (a text search) than for searches involving genes (missed articles used terms such as “9p” and “9p21.3,” which were not indexed in HuGE Navigator).²⁵

Analytic validity

Identification of data sources

PubMed searches for each of the 29 genes (e.g., as “AGTR1” or “AGTR1 genotyping”) and specific terms (e.g., “analytic validity” and “clinical test”) identified no relevant articles, whereas searches for individual genes using less specific search terms yielded too many articles to review (e.g., >700 citations for [“9p21” and “assay”]). Testing methods used in the research studies included in clinical validity were not relevant, as they did not include method comparisons and were not representative of clinical practice. Websites from the eight companies (Table 2) were reviewed for information on analytic validity, but no data were found.

GeneTests²⁶ queries found no US laboratories offering testing for 20 of the 29 genes (Table 2). Testing for three of the nine remaining genes, F5 (Factor V Leiden), F2 (prothrombin G20210A), and MTHFR, is widely available to assess thrombophilia risk. An American College of Medical Genetics/College of American Pathology proficiency testing program is also available.²⁷ Two systematic reviews address F5/F2 analytic validity and found adequate evidence (scale: convincing, adequate, and inadequate) that the analytic sensitivity, analytic specificity, and reproducibility of commonly used methods were high, although generalizability was limited due to lack of stratification by method/platform.²⁸ Publications providing information on the analytic validity of testing for MTHFR and the remaining six genes (i.e., ACE, AGT, AGTR1, APOB, APOE, and CBS) were not identified.

Relevant methodologies

The literature on allele/SNP genotyping described laboratory developed tests (LDTs) and a large number of commercially available reagents and platforms.^29,30 The methods used for the nine genes listed on GeneTests were reported as “sequencing and/or mutation analysis.” In general, genotyping methods have involved discrimination of alleles by primer extension, hybridization, ligation or enzymatic cleavage, and detection using fluorescence, mass, gel electrophoresis, or chemiluminescence.²⁹ Mistaken alleles, allelic dropout (i.e., amplification of only one of two alleles in a heterozygous individual), and other genotyping errors can result from a number of causes. These have included interaction with flanking DNA sequences, low quality/quantity of the DNA in samples, laboratory problems related to reagents/protocols/equipment, and human error. Questions remain about causes of errors in many newer technologies (e.g., multiplex assays, chips, and SNP arrays) used in routine clinical practice and their potential impact on patient results.³¹

A survey of companies offering genomic panels/tests

Standardized e-mails were sent to the Chief Medical Officer (or equivalent) of each of the eight companies. One company responded (deCODE, Inc., Reykjavik, Iceland). Client Services (or equivalent) from the seven remaining companies were contacted by telephone; both information about this review and a request to be contacted were left. One additional company responded (Interleukin Genetics, Inc., Waltham, MA).

Information on analytic validation of the deCODE MI™ LDT was provided. The test is based on Centaurus™ Assay (Nanogen, Inc., San Diego, CA), a methodology using DNA hybridization technology, allele-specific fluorescence detection of polymerase chain reaction products by spectrometry, and allele calling by automatic algorithm. A published reference for the real-time polymerase chain reaction methodology this test was based on was also provided.³² The information included results from comparisons with a gold standard in 100 unselected patients and 12 positive controls (100% concordance). No failures were reported. Patients/controls were run in quadruplicate, and all results agreed. Long-term quality control measures were also in place. The laboratory, located in Iceland, reported holding CLIA and College of American Pathologists accreditations.

Information on the Inherent Health™ Heart Health LDT (also provided under the name Gensona™) was provided by Interleukin Genetics (Waltham, MA). A single-base extension technology was used in a tagged fluorescent assay, and the GenomeLab™ SNPStream® Genotyping System (Beckman Coulter, Fullerton, CA) was used to identify the relevant SNPs. Genotype results using this method for 15 control DNA samples (Coriell Cell Repositories, Camden, NJ) agreed with the known genotypes (100% concordance). Results for 20 samples were compared with sequencing results from an outside laboratory (100% concordance). Replicate samples (2–9 buccal swabs) from 19 of the 20 volunteers were successfully tested for three SNPs in 98.4% of samples, and concordance was 100%. In another validation study of DNA from 104 buccal swabs and 136 control samples, 98.2% were successfully genotyped. The laboratory reported having accreditation from CLIA and four states (CA, MD, NY, and RI).

Limitations

Method descriptions and in-house analytic validation data were not provided by the other six companies. This data are considered Level 4 (the lowest quality).¹⁵ The overall quality of evidence for analytic validity of these testing panels was considered inadequate. Specific limitations included no information about the platform used for the majority of genomic tests, limited information on the impact of DNA degradation from sample collection/processing and shipment using real samples, and for some genes, limited information about the actual variant tested. For some of the genes, the prevalence of the least common genotype is low, so the positive predictive value is highly dependent on analytic sensitivity. The in-house information from two companies was encouraging, and platforms exist that could allow high sensitivity and specificity. However, estimating analytic sensitivity and specificity for the majority of genomic panels for CVD/heart health was not possible.

Clinical validity

Association of individual gene variants with coronary heart disease

Table 4 provides a summary of the evidence. The gene acronym (and common aliases) is shown in the first column, along with the variant(s) reviewed, the proportion of the population at risk, and the model(s) used in the literature in the next three columns. For example, the first row summarizes the information for the ACE gene insertion/deletion variant. The at-risk genotype (present in 27% of whites) is the presence of two at-risk deletion alleles (DD). Also provided is the Venice assessment of cumulative credibility. None of the grades are based on effect size. For the ACE I/D variant, the cumulative credibility is classified as “weak.” Although the amount of evidence is large (indicated by the “A”), the extent of replication is poor (indicated by the “C”). The protection from bias is not evaluated, because if any of the three grades was a C, the credibility is considered weak regardless of the grade assigned for bias. The consensus OR is 1.22, with the “s” indicating significance (P < 0.05). In the example of ACE, several meta-analyses showed a strong association between sample size and effect size (smaller studies having larger effects), but no attempt to change the effect size was undertaken. A description of the literature search, how the at-risk genotypes were defined, sources for the at-risk frequencies, included confidence intervals, tests of heterogeneity, and evaluation of publication bias are all included in Appendix, Supplemental Digital Content 1, http://links.lww.com/GIM/A124. Overall, 38 different variants/models were reported for 28 genes (no data were found for SOD3), and 17 (45%) had at least one statistically significant OR. Only one marker (2.4%) had strong credibility (A,A,A)—the 9p21 SNPs. Among the remaining combinations, 15 (37%) and 25 (61%) have moderate and weak credibility, respectively. Without considering the 25 combinations with weak credibility, the ORs range from a high of 1.25 to a low of 0.80 (both estimates are found using the 9p21 SNPs).

Table 4 Summary of observed gene/disease associations for cardiovascular disease in whites

Cumulative effect of multiple genes tested

The bolding of the OR (Table 4) indicates which of the effects was chosen for modeling for each gene. The distribution of cumulative ORs (cORs) resulting from the Monte Carlo simulation are shown in Figure 2A. Although there is considerable overlap, the cORs are higher among cases. Overall, 10% of controls and 23% of cases are assigned a cOR of 1.38 or higher.

Fig. 2

Theoretical overlapping distributions of cumulative odds ratios (cORs) in individuals with and without disease. A, The cOR using markers from 24 genes associated with heart disease from Table 4. The dashed line (with open circles) indicates the distribution for individuals with CHD/MI, and the solid line (with filled circles) indicates the distribution in control individuals. B, A similar analysis with stroke as the outcome of interest. The cOR for stroke is based on markers from 13 genes (Table 4). At a false-positive rate (FPR) of 10% and 11%, these markers have detection rates (DR) of 23% and 15% for heart disease and stroke, respectively.

The cOR distributions can also be displayed as a receiver operator characteristic curve (Fig. 3). The AUC is 64.7% and, by itself, would not be considered a useful test for heart disease risk stratification.^33,34

Fig. 3

CHD: The receiver operator characteristic (ROC) curves for the cumulative odds ratio (cOR). The genomic test panel detection rate (sensitivity) is on the vertical axis, whereas the false-positive rate (1-specificity) is on the horizontal axis. The 45[degree] dashed reference line indicates a “useless” test where detection and false-positive rates are equal. The ROC curve for the 24 CHD/MI markers is shown with open circles, whereas the corresponding curve for the 13 stroke markers is shown with filled circles. The areas under the curve for the CHD/MI and stroke markers are 64.7% and 55.2%, respectively.

Association of individual gene variants with stroke

The last two columns in Table 4 provide similar data for the 28 genes in association with stroke. Less data were available, with only 19 genes having identified publications that include 20 combinations of variants/models. Six of the 20 combinations (30%) were associated with a statistically significant OR. None received a strong credibility rating. Overall, 50% (10 of 20) had moderate credibility, with the remaining having weak credibility. Figure 2B shows that the distributions of cORs generated by a separate Monte Carlo simulation have only modest separation between cases and controls. Overall, 15% of cases and 11% of controls had cORs of 1.38 or higher. Figure 3 also shows the receiver operator characteristic curve for the stroke model, with an AUC of 55.2%.

Assessing the potential for biases in the effect size

A cumulate effects analysis was performed for 17 of the 38 combinations for heart disease. Of these, 2 (12%) had a large range, 5 (29%) had a moderately wide range, and the remaining 10 (59%) had a relatively small range. In all instances of a large or moderate range, the final consensus effect size was larger than the first consistent estimate. Only 7 of the 13 combinations for stroke had sufficient data. Of these, one (14%) had a large range, three (43%) had a moderately wide range, and three (43%) had a relatively small range. Among the four combinations with moderate or large ranges, one (MTHFR) had a first stable estimate (1.35) that was larger than the consensus (1.1). Overall, this suggests that the findings shown in Table 4, and Figures 2 and 3 are upper limits, with an “unbiased” effect size for many of the gene marker combinations likely being closer to 1.00.

Risk prediction: Combining traditional risk factors and genomic markers

Limited information is available on whether combining genomic markers with traditional risk factors (TRFs) improves prediction of heart disease. Only three studies^35–38 report the accuracy of 10-year risk prediction using traditional risk factors with, and without, measurement of the 9p21 SNPs, and a summary of these results has been published.²⁵ The net reclassification index, a measure of improvement³⁹, ranged from −0.2% (a nonsignificant decrease in prediction by adding 9p21 SNP information) to 4.9% (a small, but significant improvement in risk prediction). However, the mechanisms of 9p21 markers in modifying risk for heart disease are not yet fully understood.^35–37,40

The change in the heart disease risk is relatively small, when compared with change in risk estimates possible using conventional risk factors, and other tests (e.g., electrocardiogram) and biomarkers.^41,42 How to analyze the potential improvement over risk prediction using traditional risk factor (e.g., AUC, event-specific reclassification, integrated discrimination improvement test, and other new approaches) remains an active area of current and future research.^39,43–45

Limitations

Most analyses have been restricted to whites, usually of European descent; results in other race/ethnic groups might differ in effect size, prevalence of at-risk genotypes, variant of interest, or some combination of these. Some of the gene-disease associations under review involve a small effect size for relatively uncommon genotypes. The current GWA studies may be underpowered or limited in SNPs coverage of the genome to reliably confirm or reject real associations in the genes we reviewed. The finding of weak associations between gene variants and CVD outcomes does not exclude possible roles for these variants in the pathway of lipid metabolism or in other pathways associated with CVD. Many studies have confirmed both gene/lipid associations and that the variants identified are functional. However, the function of the marker with the highest credibility, 9p21, is not well described. Understanding its role in disease etiology might result in an improved understanding of the process, leading to improved treatments or prevention activities. Publication bias seems to be an important problem in the literature regarding genomic tests for CVD. Finally, the genes included in this study were, on average, described more than five years ago, with the newest one (9p21) first reported in 2007.^46,47 Since that time, one GWA study⁴⁸ has reported new markers that may also have strong credibility, once confirmatory studies have been reported.

Clinical utility

Health outcomes and clinical scenarios

A test has clinical utility when the results change clinical management, which, in turn, leads to measurable improvements in primary health outcomes (i.e., reduced morbidity and mortality related to heart disease). Secondary outcomes of interest include the impact on health care providers' decision making and the individuals' personal motivation. The potential clinical utility of testing is likely to differ depending on the clinical scenario, most importantly the setting in which the testing is offered (e.g., clinical and direct-to-consumer) and the population targeted to be offered testing (e.g., high risk or general population). Three likely clinical scenarios are (1) genomic testing is not performed, and actions continue to be based on risk estimated using TRFs; (2) testing is ordered by a physician or other health care provider and the results considered in the context of CRFs; or (3) testing is ordered by a consumer (who may or may not have a CRF) who may or may not seek advice and/or interpretation from a health care provider.

Potential interventions and related benefits and harms

No studies were identified on long-term CVD outcomes after personalized clinical interventions based on genomic test results. In the clinical setting, physician changes to clinical management based on increased heart disease risk might include earlier initiation or higher rates of pharmacotherapy (e.g., statins and antihypertensives), increased use of other medical tests or interventions (e.g., health plan prevention and management programs), and provision of targeted recommendations for lifestyle changes (e.g., smoking cessation, diet, and exercise). Increased use of diagnostic or prognostic tests, such as exercise stress testing, myocardial infusion imaging, and echocardiography, could improve clinical outcomes when targeted to those with the highest risk (potential benefit) or increase costs and patient anxiety without improving health outcomes (potential risk). There is preliminary evidence that physician knowledge of CHD risk scores may translate into short-term benefits, such as increased prescription of drugs and modest reductions in risk factors (e.g., patient blood pressure), without identified clinical harms.⁴⁹ However, further studies are needed to determine whether such findings are replicable, apply across different practice settings, and lead to improved long-term CHD outcomes.⁴⁹

The challenge of motivating long-term behavioral change is key in addressing modifiable risk factors (e.g., compliance with treatment, diet/exercise, weight loss, and smoking cessation),⁵⁰ whether in the clinical or direct-to-consumer setting. However, measuring behavioral change is uniquely important in demonstrating the utility of CVD genomic profiling test result reported directly to consumers. No studies were identified that assessed the clinical utility of providing genetic risk information to consumers outside the clinical setting. In fact, very little is currently known about all aspects of the direct-to-consumer scenario, including how many tests are ordered, who is ordering the tests, what their reasons and motivation are for having the tests, and how the information is being used.

Individuals at risk based on one or more CRFs and a positive genetic test result seem most likely to receive the potential benefits. However, more studies are needed to determine (1) the balance between potential benefits, potential risks (e.g., false reassurance and perceived genetic determinism), and implementation issues (e.g., costs, available, and effective interventions); (2) whether tests should be routinely offered for risk screening or targeted to refining risk in patients with identified CRF; (3) what level of increased risk warrants a management change; and (4) whether such changes will improve long-term heart disease outcomes. Arguments about the inherent value of information are difficult to support if no action is taken (e.g., clinical or personal decision making) or if interventions do not improve health outcomes.

Summary

A test with demonstrated clinical validity does not necessarily result in improved health outcomes (clinical utility). A recent survey found that only approximately 37% of US physicians reported regular use of any heart disease risk score (93% of whom used the Framingham system).⁵¹ Patient adherence to lipid-lowering medications is also an important factor in attaining improved heart disease outcomes, but preliminary data suggest that only approximately half of patients reach target lipid levels, and only one in four continue drug treatment long term.⁵² Other complicating factors may be access to medical care and medications. Therefore, a positive impact of reclassification based on addition of genetic markers to CRF cannot be assumed, even if risk assessment were to be improved. In fact, it has been proposed that heart disease may be more effectively prevented by implementing an inexpensive standardized multidrug intervention (i.e., the polypill) in an easily defined population (i.e., all people aged 55 years or older), regardless of individual risk levels.⁵³

Clinical trials are needed to demonstrate that use of genetic tests is associated with changes in physician management decisions, patient motivation and long-term behavioral changes, improved outcomes, and/or reduced costs to the health care system (clinical utility). For genomic tests to add substantially to existing predictive models for heart disease, it would likely require many markers such as 9p21 to be included.^54,55

DISCUSSION

Quality of evidence

Analytic validity

The quality of evidence for analytic validity is inadequate. For most of the offerings (Table 2), it was not possible to identify the specific variant tested within a specific gene. We found no published literature on the specific testing platforms, analytic sensitivity/specificity, or predictive values for the eight genomic panels relating to heart health from the eight companies advertising these tests. Two of the companies provided in-house data and referrals to a publication. Testing for the gene variants described in the review can be done reliably with currently available technologies (e.g., F2 and F5), but there is inadequate evidence that this is so for all the companies identified in this report.

Clinical validity

The quality of evidence for clinical validity varied widely among the 29 genes/variants. For this reason, we used the Venice grading system²² to rate the credibility of gene/disease association studies. The most credible evidence is for the 9p21 SNP markers and heart disease (but not stroke). This effect is highly reproducible in multiple large studies and unlikely to be influenced by major biases. On the other extreme are several gene/variant combinations that are based on only a few heterogeneous small studies, often with an effect size that is suspect due to important potential biases. Some of the strongest reported effect sizes are associated with weak credibility (credibility grades of C, C, -). This review focused on the outcomes of heart disease and stroke. The identification of gene/variants has helped greatly in understanding the relevant pathways. Finding little or no association with a given health outcome does not mean the gene does not play an important part in, for example, lipid metabolism.

The literature is not consistent in how genotypes are grouped when reporting gene/disease association studies. When the at-risk allele frequency is low, those rare homozygotes are often combined with heterozygotes to improve the power of the study. These ad hoc decisions make combining information from studies more difficult and may mask larger effect sizes in the smaller groups. By contrast, an allele-specific OR fits the 9p21 data very well (i.e., the change in risk from no to one at-risk allele is equivalent to the change from one to two at-risk alleles).²⁵ The reliability of this finding is partly due to the relatively high frequencies of the three 9p21 genotypes and to the large study sizes.

Although we rated the credibility of gene/disease associations, it was not possible to determine the extent to which biases might account for some (or all) of the effect size when the credibility is not rated as strong. Lohmueller et al.⁵⁶ addressed this issue and concluded that up to 25% of reported gene-disease associations might represent real associations. Ntzani et al.⁵⁷ examined the relationship between AT1R and MI and found a highly significant per-allele OR of 1.13. However, there was high heterogeneity, with the larger studies showing no effect. Thus, the point estimates for the effect size for genes/variants graded as weak (or even moderate) must be viewed skeptically, until sufficient information is available to determine the extent to which potential biases might be influencing the size or even the presence of a reported effect.

Combining multiple genomic makers together is likely to increase the predictive power over that possible with any one marker. However, little or no evidence exists as to how these combinations should be made. The most common method, and the one we used in our modeling, is to assume independence of the markers and multiple the effect sizes. Using only the combination of 24 gene/variants to predict heart disease, it was possible to identify 24% of individuals with disease along with 10% of those without disease, at a cut-off OR of 1.38. Poorer performance was found for stroke. This modeling might overestimate the effect if some of the markers are related (e.g., in the same biological pathway), but insufficient data are available to create more sophisticated models. It is also possible, but unlikely, that some markers interact with each other to provide effect sizes that are larger than the product of the two. Thus, the results of modeling the effect size for genomic panels should not be considered definitive.

Clinical utility

This review did not rate quality of evidence for clinical utility. Rather, it focused on an overview of issues related to clinical utility as part of the framework for evaluating genomic testing in clinical practice. No study was identified that actually demonstrated an improvement in health outcomes based on use of genomic markers. Assessment of risks and benefits associated with genomic testing for CVD risk must consider whether CRFs are included and whether the testing scenario is through a health care provider or direct to consumer. Preliminary work on the interest, barriers, intent and behavioral changes with respect to genomic testing is underway. An important collaborative effort between the National Institutes of Health and two health plans (Multiplex Initiative) is underway to investigate “… the interest level of healthy young adults in receiving genetic testing for eight common conditions …,” as well as who decides to take the testing, how they interpret the results, and how the results impact their future health care decisions.⁵⁸

Gaps in knowledge

• Little or no available information on the analytic validity of genomic panels, either in the published literature, or on the company websites. Often, it was not possible to even determine the testing platform or methodology being used.

• The specific genes and variant(s) included on the genomic panel.

• How genotypes should be grouped/modeled for risk (e.g., allele-specific OR, recessive model, and dominant model) and what evidence is needed to decide.

• Which of the gene/variant associations identified might benefit from further validation and/or analysis to improve their credibility.

• How information gained from GWAS might be helpful in determining the effect size and credibility of existing gene/disease associations.

• Which, if any, of the gene/disease associations identified with moderate or weak credibility might be overestimated due to potential biases (e.g., publication bias).

• How multiple genomic markers for CVD should be combined, and the types of data needed to inform these models.

• What methodology should be used to determine the extent to which genomic (or nongenomic) markers add useful information to an existing risk model.

• How information about genomic tests for CVD can be kept current, given the rapid increase in knowledge and technology.

• Alternative strategies for prevention of heart disease and how genomic markers might impact these strategies.

• How genomic markers that modify CVD risk derived from traditional risk factors will change the pattern of clinical practice.

• Are there behavioral changes related to providing the results of genomic testing, and would these changes plausibly lead to improved health, and what factors might influence these changes (e.g., setting, method of delivery, and change in risk).

Research agenda

Laboratories performing analytic validation studies for genomic panels should consider publishing their detailed results in peer-reviewed journals, as a way to build the evidence base for reliable testing. A consensus method of handling data with poor credibility and/or the existence of possible bias that could have a nontrivial impact on the effect size should be developed. This would allow more consistent and reliable modeling to occur. Further work on standardizing genotype models, summarizing/evaluating the literature, combinations of genomic markers, and combinations of genomic and nongenomic markers should be continued.^22,44 Such work should capitalize on the existing efforts occurring in various complex disease-specific areas such as breast cancer and diabetes. A recent scientific statement from the American Heart Association⁴⁴ proposed six phases in the evaluation of novel risk makers for CVD. The 9p21 markers are the most developed of the novel genomic markers and would be considered to be in Phase 3 (show incremental value over standard risk markers) or Phase 4 (change risk sufficiently to modify therapy). It has already passed through proof of concept (Phase 1) and prospective validation (Phase 2). Assuming that clinical utility can be demonstrated by a change in recommended therapy, Phase 5 would require a randomized trial to demonstrate that a health outcome is improved, whereas Phase 6 would determine whether the marker's use is cost-effective.

Update on the availability and content of genomic panels for heart disease risk

This targeted review addresses only those genes included on genomic profiles aimed at CVD or “heart health” that were available when the study was undertaken in mid 2008. Because it is possible that additional or updated panels may now exist, we repeated our search strategy in June 2010. Of the genomic panels included in Table 2, one had added new markers (deCODE MI), two panels no longer seem to be available (Genovations and Genosolutions), and the remaining five were unchanged. No new genomic panels for heart health were identified. The updated offering from deCODE was released in May 2009 after this review had been conducted. That “heart attack” panel currently includes markers in nine genes not included in the present review (BRAP, CELSR2/PSRC1, CXCLI2, MIA3, MRAS, PHACTR1, SH2B3, SLC5A3/MRPS6/KCNE2, and WDR12).