|Home | About | Journals | Submit | Contact Us | Français|
Advances in genetics that emanated from the Human Genome Project and 100 years of prior human genetics have recently led to numerous studies associating genetic variation with common diseases such as coronary artery disease, type II diabetes, atrial fibrillation, autism and schizophrenia, amongst many others. Before 2005, the field was poorly represented by a small number of underpowered and unreplicated genetic association studies examining single genes, with only a few well-validated genetic risk factors for common disease. The fundamental advance in the field that enabled identification of important genetic associations with disease since 2005 was identification of most of the common variations in the human genome, and the technology to examine more than 500,000 of those variants in a single individual for a cost of less than $600.
In contrast to common acquired diseases that make up the majority of the health care burden in the Western world, there are numerous genetic diseases that are rare in the overall population, but are profoundly debilitating or lethal in a much smaller number of individuals. These are often obvious earlier in life, are strongly inherited, and are little modified by other non-genetic influences. Because they are rare and inherited, they are often studied in small family kindreds. These are often called Mendelian disease because of the nature of their inheritance, and the first recognized example, alcaptonuia, was discovered over 100 years ago. Other examples are sickle cell disease, malignant hyperthermia and the pseudocholinesterase deficiencies.
The purpose of this review is to outline recent developments in human genetics that have current and future impact upon Anesthesiologists in their clinical practice. Specifically, I will concentrate on the fundamental principles of genetic variation, the technology we use to identify genetic variation in studies of gene-disease associations with examples of its value, notably in common disease such as coronary artery disease and atrial fibrillation, and how identification of such relationships advances the field of medicine in general and the field of anesthesiology, specifically. In my experience, few anesthesiologists have a working knowledge of the fundamentals of genetics and genetic study methodology, especially those of us who graduated medical school before 1990. So I will offer a somewhat colloquial approach, emphasizing examples from my own experience that will hopefully engage the reader throughout the entire review!
The mechanisms of effect of genetic variation upon normal homeostasis and disease are multifactorial and are confounded by the tremendous interplay and redundancy in the human machine. An entire branch of biology, called systems biology, has been developed to try and make sense of these relationships, often with the goal of identifying key proteins and mechanisms in one or more pathways. For example, the currently-known interrelationships for the complement pathway are portrayed in Figure 1. At the genetic level, the methodologies are somewhat simpler but still look pretty ugly. In principle and somewhat simplistically, genetic variation can be categorized into variations in chromatin folding, DNA coding, RNA expression, and RNA translation into functional protein. Of these four mechanisms, we understand the most about variation in DNA code; however, the other three are likely to be just as important in human disease.
DNA is the fundamental template for protein coding, mediated through messenger RNA (mRNA) transcription. In order to create mRNA, the transcriptional machinery requires access to coding DNA; however, nuclear DNA exists in complex coils around protein units called histones that simplistically hide or expose a coding DNA sequence. Since the transcriptional machinery requires access to the coding DNA, the control of mRNA creation is under control of the way the histones and other proteins fold the DNA into the conglomerate known as chromatin. Overall, our knowledge of this important mechanism is rudimentary and will not be discussed further in this review.
DNA transcription to mRNA is regulated by regulatory proteins, called transcription factors, which bind to specific DNA sequences close to the gene. Transcription factors bind to short (5-20bp) promoter regions of DNA that are usually within a few hundred base pairs of the site where transcription starts (Figure 2). Transcription factors alter the rate of transcription of DNA sequences into mRNA by binding to the DNA and increasing the ability of the transcriptional machinery responsible for formation of mRNA. This control mechanism for mRNA production may be the most important cause of variability in complex genetic disease. If a single base pair change alters the binding of a promoter to the DNA sequence, then more or less mRNA and subsequent protein may be created. In essence, the protein sequence is the same but the amount of protein being made is different.
Other mechanisms for increasing and decreasing mRNA production also exist. Enhancers are short regions of DNA similar to promoter sequences that bind to proteins rather like the transcription factors to enhance transcription (Figure 3). The fundamental difference between enhancers and promoters are that enhancers do not need to be close to the genes they act upon. Although the enhancer sequencer may be far from the gene (as far as a million base pairs away), it is physically close to the gene when the DNA is folded around histones. Enhancers do not act on the promoter region itself, but are bound by activator proteins that enhance or repress transcription.
After transcription, the mRNA is further modified within the cell nucleus via splicing. Segments of the mRNA that do not code for a protein, called introns, are excised by enzymes called spliceosomes and the remaining exons (gene sequences that code for protein) are joined together. The final mature mRNA is then translocated to the cell cytoplasm, where the mRNA undergoes ribosomal translation into the corresponding amino acids of a protein. Another level of control exits at this point. Short sequences (~22bp) of RNA that partially correspond to a complementary sequence of the mRNA may bind to the mRNA, thus preventing encoding of the protein. These microRNAs (miRNA) have only recently been identified and the role of the ~700 currently known human miRNAs is still poorly understood. The majority of miRNAs are transcribed in only one or a few tissues, and in only one or a few embryologic stages from conceptum to adulthood. miRNA levels are often profoundly altered in some diseases, but their role as initiators or perpetuators of disease are rarely known. Importantly, DNA variation can alter the structure of a specific miRNA or the target of the miRNA. In addition, mRNA is degraded within the cell. The rate of degradation alters protein production.
As if we were not already overwhelmed with the options for altering the steps between DNA and protein, another vitally important step is possible. Although less than 2% of the human genome encodes the ~22,000 genes, these genes encode well over 100,000 different proteins. These different proteins result from different start, stop and joining sites for mRNA production and splicing (Figure 4). A single gene can encode hundreds of different proteins; there is even an example of a single fruit fly gene encoding over 38,000 different proteins.
Although it may seem that there is an overwhelming number of ways to create variation in the human machine from its underlying complexity, the principles are relatively straightforward. Overall, we can think of the mechanism of variation occurring as a change in the protein structure, or a change in the amount of protein being produced.
The most common type of human genetic variation is the single nucleotide polymorphism (SNP), in which two different bases are observed at the same position in the population and sometimes on the two homologous chromosomes in a single individual (Figure 5A). Genetic variation also takes the forms of insertion or deletion of one or more base pairs; sometimes entire chromosomes (Figure 5B). When the number of inserted/deleted base pairs is very large, an entire gene may be deleted or copied on the same chromosome, making a 50% decrease or 50% increase in the amount of mRNA being produced. These whole-gene insertions or deletions are called copy number variants, and are increasingly being recognized as causes of human disease. Also large segments on the chromosomal scale can be swapped between chromosomes or inverted on the same chromosome.
The replication of DNA that is required for meiosis/mitosis and passage of DNA through generations has extremely high fidelity, with less than one error per million base pairs because of the extensive proof-reading capabilities of DNA replication. Even if an error occurs, the likelihood of it having functional importance is very low. Redundancy exists in the amino acid code – 64 (43) possible combinations of base pair triplets exist to code for 21 possible amino acids, helping to limit the functional impact of errors. If the variant lies in a promoter or enhancer region, the majority of these variants have little impact upon binding of transcription factors; yet some do.
On the simplest scale, a SNP can be seen 0, 1 or 2 times on the two, of the 22, homologous non-sex chromosomes we each possess. The SNP may change the code for a protein if it lies within the coding sequence of the gene; many coding SNPs do not change the protein because of redundancy in the amino acid code. Even if the amino acid code is changed, most proteins have no alteration in function. The single base pair change that changes the β-hemoglobin sequence in sickle cell disease is a rare example of a functional protein change. Alternatively, the SNP may change binding of a transcription factor to a promoter or enhancer sequence, thus altering the quantity of protein being made, without changing the proteins structure. Finally the SNP may occur at the boundary of an intron and an exon and cause the splicing machinery to code a very different and often truncated protein.
There is one more concept we need to discuss. It is well known that variants close together on the same chromosome tend to be inherited together because it is rare for recombination to occur between adjacent chromosomal loci during meiosis. This is the phenomenon of linkage disequilibrium (LD). In simple terms, we can think of LD as the statistical correlation between two variants within a population. In general, the closer two variants are to each other on the chromosome, the more likely they are to be inherited together through future generations. However, there appear to be “hotspots” of recombination in the human genome that create a patchwork of “blocks” of variants that are almost always inherited together. These blocks average about 20kbp in length, but vary widely. Within each block there is a lot of LD. Across the block boundary created by the hotspot there is much less LD. However, this block structure can vary from being well portrayed in part of the genome, to almost non-existent in another region (Figure 6).
When well delineated, such as seen in the F5/SELP example uppermost in Figure 6, the limited diversity of polymorphisms within a block creates groups of alleles that are almost always inherited together, known as haplotypes. There are usually only a few haplotypes within a block that describe all of the genetic variation seen with tens to hundreds of variants. The block structure of the genome allows finer mapping of disease genes and alleles to the level of the individual haplotype block. Yet use of haplotype blocks for mapping genetic polymorphisms contributing to disease complicates determination of a “responsible” variant below the level of the haplotype block, as there is strong LD between polymorphisms within the haplotype block. Often the investigator can say only that a group of highly related alleles are associated with a disease, without being able to pinpoint the specifically responsible variant. To find the actual causal variant requires molecular biology rather than genetics. The importance of this haplotype block structure will be exemplified later in this review.
The last 20 years have seen a Moore's law exponential increase in the quality and quantity of genotyping performed. Five years ago, I paid $0.50 for a single SNP measurement in a single patient. On each plate of genotyping there were 368 patients in separate wells and in each well I could concurrently genotype 5 or 6 SNPs at a time. That scale and cost was considered miraculous at the time. Last year, we paid about $0.0006 per SNP to genotype about 900,000 SNPs in a single patient – total cost per patient of about $550. In one month in 2008, we doubled the total amount of genotyping we had performed in the previous five years.
How was this possible and how was it done? There are numerous genotyping technologies available, most of which fundamentally use the polymerase chain reaction (PCR) to amplify limited amounts of a specific sequence into larger quantities for analysis. Measurement of individual SNPs usually involves identifying the two different alleles using the molecular weight of the alleles, the length of different sequences depending upon the allele, or making complementary sequences that bind to one or other of the alleles that then can be identified by florescence or some other physico-chemical property. These days, the methods are so accurate (>99.9%) and have such high success (>98%), that the choice of technique is usually dependent on price, the number of patients and SNPs being measured, and local expertise. However, one technique deserves closer review.
Whole-genome scanning (WGS) measures about 1 million SNPs and copy number variants, to cover the genome with very high precision. For the 3 billion base pairs in the human genome, genotyping 1 million variants will mean that a genotype is measuring about every 3,000 base pairs on average. Detailed analysis using real data shows that about 80% of all variation in a human is measured using about 1 million variants. This fidelity means that one or more SNPs are usually genotyped within each haplotype block. If an association is found, further genotyping of the SNPs in that region will likely identify one or more SNPs or haplotypes with the greatest association. Commercial WGS uses two technologies with similar accuracy and completeness. The chips are made by Affymetrix and Illumina, and the decision to use one or the other usually depends on local preference rather than on hard science. The physical characteristics of these chips are shown in Figure 7.
In general, the amount of mRNA produced within a cell determines the amount of protein made by the cell. There are several caveats to that statement but the general concept is correct. If a non-coding variant is found to be related to a disease, the “proof” of the relationship can be found in demonstrating increased or decreased mRNA production. Unfortunately, mRNA is a relatively unstable molecule made in small amounts, and complex techniques are required to accurately measure the amount of mRNA. Currently, the most common methods of measuring mRNA involve either: 1) measuring the amount of mRNA made through many cycles of PCR to quantify the original amount of mRNA present (slow and expensive, unsuitable for large scale work), and 2) comparing the amount of mRNA present in two paired samples, often comparing before and after an event on chips, rather like the chips used for genotyping. This second technique is fast, can operate on industrial scales, and can compare signals for most proteins produced by the genome. However, interpretation of the signals is difficult as the comparison of different amounts of red and green light produced from fluorescent markers is inaccurate. In addition, the chips require detailed sequence knowledge of the mRNA being produced and cannot reliably measure splicing variants in many situations. The latest and greatest technology for mRNA measurement is a quantal improvement because of its ability to measure even a single mRNA molecule in a cell, measure mRNA sequences that we don't even know about, and give precise numeric estimates of the amount of every single mRNA in the cell. It does this by sequencing every mRNA present in a sample. However, its cost is high and its speed is slow. Both will improve with time.
Human gene-disease studies fall into two general classes; linkage studies and association studies, based upon the nature of the inheritance pattern. Linkage studies are used in Mendelian disease, often in families with a high prevalence of a single disease that is observed early in life, in whom multigenerational trees of inheritance of disease can be traced. The basis of linkage studies is an observed principle in genetics that homologous chromosomes (i.e. both copies of say chromosome 2) in a dividing germ cell exchange large common portions of the chromosome between each other, via a molecular process called recombination (Figure 8). Those two chromosomes will have come from the individual's parents, and by comparing a large number (>1000) of genetic markers in multi-generational families the point of chromosomal swapping in the genome that is most significantly related to the disease can be established. This point is close to the gene responsible for the disease, but the fidelity of this technique is very limited, often giving results that cover large portions (many millions of base pairs) of a chromosome. Narrowing the region down to one or more genes involves follow-up genotyping of progressively smaller regions with greater fidelity. A good example has been using linkage analysis in families with a high incidence of breast cancer at a young age to identify the BRCA1 gene on chromosome 17. In general, linkage studies are only valuable when the disease is present at a young age and not significantly modified by environmental influences.
The contrast between Mendelian disease and complex disease is many-fold and can be simplistically thought of as a comparison of a predetermined genetic fate compared to a smaller risk that is often modified by non-genetic factors (Table 1). Despite the importance of individual Mendelian disease to those affected by it, the larger genetic burden upon human health comes from complex diseases that often manifest only later in life. Simple examples are coronary artery disease, obesity and hypertension. It is obvious that occurrence of these, and other complex diseases, is highly influenced by human behaviors and circumstances. For example, obesity and coronary artery disease are more rare outside the Western world. Other complex diseases have few environmental components, such as autism and schizophrenia. Their complexity appears to come from a likelihood that many variants, rather than one single variant, cause the disease either occurring together in a single individual or occurring rarely in many individuals.
Because diseases of old age can rarely be examined in large mutigenerational families (the parents and grandparents have often died by the time the children get the disease in their 60's or 70's), association studies are used test whether a polymorphism occurs more or less frequently in many unrelated cases compared to many unrelated controls. Association studies have only become a common discovery tool now that we know of much of the variation in the human genome. We know of about 13 million SNPs in the human genome, about half of which are common (>1% frequency). This number far exceeds the density of markers used in linkage studies, so our ability to narrowly identify the polymorphism or region associated with disease is greatly improved. Another advantage is that association studies use conventional case:control methodology and do not require studying families. Consequently, association studies can allow us to examine genetic causes of diseases associated with surgery, drugs or other interventions, where few individuals in the family have undergone the same surgery or taken the same drug. However, design, analysis and interpretation of association studies require statistical rigor and expertise (Table 1).1
An important concept in our understanding of complex human disease has been the role of many small-effect genes upon common diseases. The so-called common disease / common variant hypothesis suggests that there are many common variants, each having a combined and incremental effect upon the risk of a disease. Restated, several commonly-occurring SNPs in several genes, each individually contribute a small increased or decreased risk to the overall risk of a disease in a single individual. We currently believe that this mechanism of complex disease is the most prevalent and accounts for many diseases of old age, such as diabetes and coronary artery disease. A simple and understandable example is height. There are several genes that have some role in determining height, but neonatal and childhood circumstances also play a strong role. Alternatively, the multiple rare variant hypothesis states that in any single individual, a few relatively rare variants, each with a relatively large effect, contribute to the disease - but because they are rare in the overall population, the effect appears to be complex. This mechanism is likely true for many genes of drug metabolism and obvious examples from the anesthesia realm are malignant hyperthermia and succinylcholine metabolism.
In 2005, the first genome-wide association studies (GWAS) emerged from the combination of the HapMap project (http://www.hapmap.org) with new technologies for testing hundreds of thousands of SNPs on a single chip. The studies are undertaken by measuring say one million known SNPs in say 10,000 individuals (5,000 with the disease of interest and 5,000 without). The SNPs are roughly spaced about 1 SNP every 3,000 base pairs of the 3×109 base pair human genome, thus allowing mostly complete coverage of all the variation in the genome. The coverage isn't perfect; there are gaps, but we generally believe that we are able to observe about 80% of all common variation. The words common variation are important; it is likely that our ability to find associations of disease to variation that conforms to the multiple rare variant hypothesis is probably less than associations to variation that conforms to the common disease / common variant hypothesis.
These GWAS studies have successfully identified >30 genes associated with obesity, >30 genes associated with type II diabetes, >20 genes associated with coronary artery disease or myocardial infarction, >10 genes associated with breast or endometrial cancer, >5 genes associated with each of prostate cancer, schizophrenia, autism and psoriasis, and three genes associated with atrial fibrillation (AF). In all, >400 genes have been associated with >75 complex diseases, as I write this in mid-2009.
I'd like to demonstrate a simple example of gene:disease association studies in order to illustrate the general principles and to serve as a basis for understanding the studies you will find in the Anesthesiology literature. The methods used depend on the outcome being examined, the number and density of variants being examined and the clinical factors that may alter the relationship between gene and disease.
The outcome of interest may be a continuous (such as duration of hospitalization), ordinal (such as ASA class) or dichotomous (such as postoperative myocardial infarction, or not) variable. Analysis methods exist for all these types of outcomes. The simplest example to illustrate is a dichotomous variable. The advantage of this type of analysis in unrelated individuals is its simplicity. It looks and behaves like the rudimentary 2×2 tables of epidemiology and χ2 statistics. Let's say we have a population of 2,000 patients who underwent CABG surgery performed at a single center and that 10% of them had an MI. Our hypothesis is that a single SNP in the chromosome 9p21 region is associated with MI. However, we could know that several pre- and perioperative factors affect the frequency of MI, such as age, race, current smoking, prior MI, severity of coronary disease, duration of cardiopulmonary bypass and surgeon. We know this for our population because before we embarked on the genetic analysis, we had examined many variables that may or may not predict MI and created a robust logistic regression model of the clinical variables that are associated with MI. We made this model for two reasons: 1) reducing other causes of variability in the relationship between SNP and MI may allow a stronger identification of the relationship and, 2) there may be situations where the signal is only present when a particular clinical covariate is also present. A theoretical example is that the SNP may only have an effect when the patient is a smoker. Although construction of a logistic regression model is essential in genetic studies of clinical outcomes, I will demonstrate the overall principles using a 2×2 table.
Looking at the 2×2 table in Table 2A, we have made a few assumptions, the minor (less frequent) allele occurs one or more times in 40% of the population, MI occurred at a rate of 10%, and there is a dominant genetic model. The null hypothesis of no relationship between SNP and MI would result in the distribution of individuals in the 2,000 patient cohort illustrated in the upper panel. There is no difference in the frequency of MI between those with and without the SNP. However, let's say we actually observed the numbers in Table 2B. The chance of having an MI if you carried the SNP was observed to be 15%, but if you didn't carry the SNP the chance of having an MI was 6.67%. The relative risk was increased 2.25 fold by carrying the SNP. The confidence interval of the relative risk of 2.25 is 1.72 – 2.94 and the result is statistically significant (P<0.0001). This is an unrealistically simple example. There are several factors that may confound this relationship between SNP and MI. One possible example is race. Let's say that the study population included 1,000 Caucasians and 1,000 African Americans. The frequency of carrying one or more copies of the SNP is 50% in Caucasians and 30% in African Americans and the MI frequency is 12.5% in Caucasians and 7.5% in African Americans. In Caucasians (Table 2C), the relative risk from carrying the allele is 1.27 (0.91 – 1.77) and is not significant (P=0.15) in this study. In African Americans (Table 2D), the relative risk from carrying the allele is 4.67 (2.94 – 7.40) and is very significant (P<0.0001). The effect of the SNP is only present in African Americans; an example of the relationship being stratified by population structure. We can opine on the biology that may cause such a difference, but the actual cause cannot be derived within the experiment above. This probably-unrealistic example is emblematic of the importance of including demographic and clinical variables in defining the relationship between a SNP and an outcome or disease.
There are several potential uses of the results of genetic studies. We can think of them as being potentially valuable for a group of individuals, say white males with prostatic symptoms or white females with a family history of breast cancer. Alternatively, the benefit may only accrue to a single individual (Table 3). It's important to appreciate that the majority of examples are still theoretical. We are in the early stages of applying genomics to medicine and there are many limitations to our knowledge and its applicability.
For a group of individuals, identification of a gene or region that is associated with a disease provides information on which genes or pathways are integral to disease. With extensive work, the molecular biology of the association may be established, a biomarker or drug target defined and perhaps a drug developed. These examples are currently few and far between. It is likely they will become more common; however, the greatest value of identifying a drug target or some other therapy may only be present for genetic factors with very high relative risks. Illustrating the example, there a newly-found regions that increase the risk of coronary disease by 15%. Many individuals and companies are working on identifying drug targets and drugs for this genetic risk factor. But, will I take a drug to prevent coronary disease if I am reducing my risk by only 15%? Probably not and especially if, on a population basis, everyone has to take the drug. In contrast, not starting, or stopping, smoking reduces risk by 50%. However, if I have severe coronary disease and a new drug will reduce my risk of MI by 50%, that reduction will likely be important to society and to me.
Another illustrative scenario is warfarin dosing. The CYP2C9 and VKORC1 genes are implicated in warfarin and vitamin K metabolism, and variants in these genes are strongly and consistently associated with bleeding while taking warfarin. Variation in these two genes outweighs all other clinical predictors including age, gender and body weight. My father takes 1-2mg of warfarin a day, most likely because he has one or both of these variants. There's a good chance my warfarin requirements will be similar, if I ever need warfarin. If I didn't have that prior knowledge my warfarin dosing will likely be based on a population average until I have my INR measured several times. That may incur a risk of bleeding. Someone else may need much more warfarin than the population average and be at risk of thrombosis in those first few weeks of drug treatment. Several companies have developed rapid turn-around genotyping tests with high accuracy that cost about $400; however, the cost of saving a life using these tests is about $170,000 per quality-adjusted life-year.2 Payment for the test was rejected by the Centers for Medicare and Medicaid Services earlier this year.3
Another example may be illustrative. If a woman has a history of breast cancer in her family, there may be value in testing for the several variants that are associated with breast cancer (BRCA1 and BRCA2). The lifetime risk of breast cancer for a woman is approximately 12%; but is increased to 60% for carriers of one of the BRCA1 or BRCA2 variants. At least nine other genes have been associated with hereditary breast or ovarian cancers, but the majority of hereditary breast cancers can be accounted for by inherited mutations in BRCA1 and BRCA2. However, at a societal level there is no value in testing every woman for these variants. The reason is the low efficacy of determining risk of breast cancer from these tests. Overall, BRCA1 and BRCA2 variants account for only 5 to 10 percent of breast cancers and 10 to 15 percent of ovarian cancers among Caucasian women, but one or more of these variants are carried by only 2.2% of Caucasian women. By contrast, the variant is present in 8.5% of Ashkenazi Jewish women (a population with a high rate of inherited breast cancer) and only 0.5% in Asian women (a population with a low rate of inherited breast cancer).4 The guidelines for testing for these mutations should reflect the relative frequency of the variants and family history.5 Another example is Crohn's disease, which has at least 32 genetic variants strongly associated with it. Because the average prevalence of the disease in the general population is less than 0.2%, people with several high-risk genes may have an approximately 20-fold increased risk, but still have a small probability (<4%) that they will get Crohn's disease.
We are all familiar with classification of patients into risk classes. Daily examples are Mallampati classification for intubation difficulty, ASA class, and the many indices of cardiac risk. We are also aware of how these indices are not perfect predictors. Would having more information make them better predictors? The remainder of this discussion is predicated on several statements with varying degrees of truth:
The value of any additional information to a risk classification index is measured and determined by how many patients are correctly switched from one risk class to another and that a different and presumably better therapy is provided due to that information. Let's look at an example. The 9p21 chromosomal region has been strongly associated with coronary artery disease and MI. Papers that describe the association have P values better than 10-12 and have risk ratios of ~1.25 (i.e. risk is increased by about 25% over the general population). You can go to the web and order a kit to swab buccal cells, mail it back, and have your 9p21 status returned to you in a couple of weeks. You will be about $200 poorer; will you be wiser? By contrast, knowing several non-genetic facts may provide just as much, or likely more, predictive value. Using identified predictors from the Third Report of the National Cholesterol Education Program Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (ATP III) risk score, well-conducted studies have strongly predicted subsequent risk of cardiovascular events. As a perfect example of the relative value of genetic information the study by Paynter et al.6, using 22,129 Caucasian participants in the Women's Genome Health Study showed the ATPIII index (age, systolic blood pressure, cholesterol, HDL cholesterol, smoking, anti-hypertensive use and diabetes) was strongly predictive of subsequent cardiovascular events. A measure of this prediction is the c-index which in this case was 0.803 ± 0.019, which is very strong. However, adding genetic information from the strongest genetic variant associated with CAD in the 9p21 region (rs10757274) only trivially improved the index to 0.805 ± 0.019. It's not statistically significant, and tells us little about the value of the genetic marker as a risk classifier for individual patients.
There is a better method. In the same study, they divided the ATPIII-predicted risk of cardiovascular events into four risk classes (<5%, 5% to <10%, 10% to <20%, and ≥20% risk of a cardiovascular event over the next 10 years). These classes are commonly used to tell patients about their risk. They then compared the assigned ATPIII risk categories for each participant without including any genetic information and the same participant while including additional genetic information from the rs10757274 genotype (Table 4). The proportion of participants who were reclassified into a different risk group by the ATIII index plus genetic information, versus the ATIII index only, can be estimated. Some of these reclassifications will be correct (i.e. they correctly predict an event), and some will not be correct. If we added a variable to the model that had no effect on developing a cardiovascular event (e.g.: left vs. right handedness), then the number of participants who correctly changed risk class would roughly equal the number who incorrectly changed risk class. By contrast, if the additional information was highly informative then the number correctly changed would far outweigh the number incorrectly changed. In this study, they observed 606 of 22,129 participants were reclassified (Table 4). Overall, 526 (87% of those reclassified) were reclassified correctly. This was a modest but significant (P=0.02) improvement. But does it help us? This depends on what clinical value we assigned to each of the classes. If an effective therapy is only provided to patients with ≥5% risk, then 205 additional patients will correctly receive the therapy and 181 will correctly not receive the therapy. No-one will be incorrectly denied therapy. By contrast, if the effective therapy is only given to those with ≥20% risk, 31 additional patients will correctly get the therapy and 26 will correctly not get it. Although it may seem that these examples are pretty robust, they are actually trivial in numeric terms. Overall, very few patients in this very large cohort had a significant advantage conveyed by the additional genetic information. Why is that? The additional risk of having the rs10757274 genotype is small – about 25% more – and is outweighed by risks from smoking and other factors.
By contrast, the risk of atrial fibrillation during an individual's lifetime is approximately doubled by possessing a variant of a chromosome 4q25 SNP called rs2200733.7 Similarly, we have recently demonstrated that the risk of AF after cardiac surgery is approximately doubled by carrying minor alleles of rs2200733.8 This effect is independent of other well-known risk factors such as older age, a past history of AF, and the type of operation being performed. In a population of 959 patients undergoing CABG with or without concurrent valve surgery we used a statistical model that included these variables, amongst others, to derive a predicted risk for each patient. We then classified individuals into two classes of risk of developing AF (Table 5) based on whether or not the risk was greater than or less than the population average AF rate of 30%. After adding the patients rs2200733 genotype status to the model, 53 patients in the low-risk group (<30% risk) are reclassified into the high-risk group (≥30% risk). Of these 53 patients, 21 (40%) develop postoperative AF and so can be construed as being correctly reclassified. After adding the patients rs2200733 genotype status to the model, 69 patients in the high-risk group (≥30% risk) are reclassified into the low-risk group (<30% risk). Of these 69 patients, 49 (71%) do not develop postoperative AF and so can be construed as being correctly reclassified. So, in this example, we have correctly reclassified 70 patients to a different risk group, but have incorrectly reclassified 52 patients, for a net gain of 18 patients. If we had put the patient on a drug, or not, based on the revised classification, we may have done some “good” or the effect may have been minimal. That would principally depend on the efficacy and side effects of the drug.
There is considerable promise to genetic studies for identification of pathways of disease and determining therapies for individual patients. We should make ourselves aware of the unbiased evidence that supports a genetic test, its overall value to the population and to the individual and its implications in determining clinical decision-making. However, much work needs to be done and most of what you hear over the next 10-20 years will be hype rather than reality.
We thank all study subjects who participate in the CABG Genomics Program and the Surgeons who collaborated by identifying their patients.
Sources of Funding: This work was supported by a grant from the National Heart, Lung, and Blood Institute (K23HL068774).
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.