NEW YORK (GenomeWeb) – A commentary published in Genetics in Medicine last month has led to discussion about the clinical utility of polygenic risk scores (PRS) for assessing an individual's risk of developing diseases such as heart disease or cancer.
The article, titled "The illusion of polygenic risk prediction," by Nicholas Wald and Robert Old of the Wolfson Institute of Preventive Medicine at the Queen Mary University of London, makes the case that currently available PRS are bad discriminators between those who will and will not develop disease and are therefore not useful for identifying individuals at increased risk who could benefit from preventative measures.
However, proponents of PRS point out that the same argument has been made in the past for non-genetic disease risk factors that are widely used in medicine, such as high blood cholesterol for heart disease.
PRS, which aggregate dozens of genetic variants that have been linked to disease risk in genome-wide association studies into a single score, have become increasingly popular in recent years, and studies reporting new PRS for a variety of diseases appear regularly in the literature. In addition, several companies have launched commercial PRS tests to improve disease risk assessment for cancer and heart attack, including Myriad Genetics, Ambry Genetics, and Color.
Last year, an article in Nature Genetics by Sekar Kathiresan at the Center for Genomic Medicine at Massachusetts General Hospital and colleagues reported polygenic scores associated with five diseases — coronary artery disease, atrial fibrillation, type 2 diabetes, inflammatory bowel disease, and breast cancer. For each disease, the PRS identified between 1.5 percent and 8 percent of the population who are at greater than threefold increased disease risk (or having an odds ratio of more than three). In total, 19.8 percent of people were found to be at greater than threefold risk for at least one of the diseases.
In combination with clinical, environmental, and monogenic risk factors, the authors argued, PRS could be used to stratify individuals for prevention or early detection strategies, such as statin therapy for heart disease or additional mammography for breast cancer.
In their Genetics in Medicine paper, though, Wald and Old argued that PRS, while identifying important genetic causes of disease, are bad at predicting who will actually get the disease. "I think there is justified interest and excitement on the observation [of PRS] and the implications of causality, but it might in some quarters be taken too far to suggest this would be worthwhile using for population screening," Wald said.
As an example, their article cited the PRS for coronary artery disease from Kathiresan's paper, showing that it only detected 15 percent of those who will get CAD, but did not detect the remaining 85 percent. At the same time, it incorrectly classified 5 percent of those who would not get CAD as being at high risk. "Identifying about 15 percent of cases for a false-positive rate of 5 percent is poor discrimination and little better than identifying people at random," the authors wrote.
Only tests with very high odds ratios can be useful screening tests, they claimed, noting that even odds ratios of 100 detect fewer than half of those eventually affected by a disease, with a 5-percent false-positive rate. "To our knowledge, no genome-wide polygenic score meets this requirement, and none is likely to do so with polygenic scores that emerge in the future," they wrote.
Using a PRS in conjunction with existing risk factors or screening markers to stratify patients for prevention or intervention is also problematic, they suggested, because the addition of PRS may not improve things much. "You get a little gain but it's deceptively small," Wald said.
"[T]he incremental gain in screening performance needs to be quantified by the increase in the detection rate for a given false-positive rate, or vice versa, and assessed in relation to the extra cost," the paper stated
Kathiresan, however, did not agree with the assessment of the paper. "Reality of clinical medicine doesn't match Wald's theoretical arguments," he said. "The practice of medicine is full of successful use of screening tests with similar statistical properties as polygenic risk scores," he added, for example measuring blood cholesterol as a risk factor for coronary heart disease.
In fact, he said, Wald made an almost identical argument in a 1999 paper in the British Medical Journal, applying it at the time to blood cholesterol screening. "It’s basically the exact same paper if you just substitute polygenic risk scores for cholesterol," Kathiresan said. In that article, the authors argued that even though cholesterol levels are elevated in those with heart disease on average, high serum cholesterol would only detect 15 percent of men who later died of ischemic heart disease, with a 5-percent false positive rate.
Both papers actually point out an important feature of many diseases, Kathiresan said, where for any risk factor, such as blood cholesterol, broad overlap exists between those who go on to develop the disease and those who don't. "In any given complex disease, like heart attack, there are a number of different risk factors, and each risk factor basically has a modest effect, and not one factor is going to be deterministic," he said.
Wald's requirements for what is a useful clinical test are too stringent, he argued, and should not be the same as what is needed for a diagnostic test. "Nobody thinks you’re going to have a heart attack just because you have high cholesterol. It’s not a one-to-one relationship," Kathiresan said. "That’s what he’s asking as the requirement, [but] there’s no factor for heart attack that will meet those criteria."
"We already know that cholesterol is very useful in clinical practice despite the statistical properties that he finds challenging in terms of screening," he added, and the same will be true for PRS tests. "It’s just a matter of figuring out in which clinical scenario this new test will be useful."
Another reason why PRS are clinically useful, he said, is that they can be independent of other risk factors. For example, Kathiresan's team recently found that in about 20 percent of patients with early heart attack, a high PRS was the only factor that distinguished them. "This polygenic score being high is what stands out. Their cholesterol doesn’t stand out, their blood pressure doesn’t stand out," he said.
He acknowledged that work remains to be done to figure out in what clinical scenarios PRS will be most useful. One opportunity, he said, is identifying young patients in their 30s to 50s who are at risk of heart attack, "where all these other things that we do right now are not that effective."
Another area where PRS could be applied, he said, is in breast cancer risk prediction, as the vast majority of breast cancer patients do not have mutations in one of the high-risk breast cancer genes, including BRCA1 and BRCA2.
Myriad Genetics, for example, has added a test called RiskScore to its MyRisk Hereditary Cancer test. RiskScore uses a PRS that analyzes more than 80 genetic markers, which the company has validated in a study of 17,000 patients who were negative for mutations in breast cancer risk genes, and combines it with family history using the Tyrer-Cuzick model. The resulting combined risk score is a superior predictor of breast cancer risk compared to Tyrer-Cuzick alone, according to Myriad, and its clinical implementation "may offer significant potential for the management of greater than 90 percent of high-risk women who test negative for monogenic mutations in breast cancer susceptibility genes," according to the firm's website.
Jerry Lanchbury, CSO of Myriad Genetics, explained that the Tyler-Cuzick model alone identifies about 30 to 35 percent of women who have a lifetime breast cancer risk of over 20 percent, which he said is often used as a threshold for enhanced screening, such as with MRI. The PRS used in RiskScore adds about 10 percent of women to the high-risk group, he said, and takes women out of the high-risk group who were originally included based on the Tyler-Cuzick model.
As such, he said, PRS are useful in risk stratification, combining different types of risks to identify patients for screening or intervention, whereas the Genetics in Medicine paper "is really much more about traditional yes/no diagnostics rather than determining risk."
Brigette Tippin Davis, senior vice president of research and development at Ambry Genetics, argued similarly that PRS are not screening tests to diagnose diseases but serve as risk assessment tools for future disease. "For example, PSA is measured as a screening test to identify asymptomatic patients who may have prostate cancer, whereas a prostate cancer PRS is used to estimate risk for future development of prostate cancer," she said in an email.
Ambry, which is owned by Konica-Minolta, is offering two PRS tests, AmbryScore for breast cancer and AmbryScore for prostate cancer. "We and many others have shown that PRS models for breast and prostate cancer are better able to assess risk for disease than existing clinical models or family history alone, based on comparison of predictive performance assessed in large case-control studies," she said.
The tests are only offered as an add-on for patients already undergoing genetic testing for mutations in moderate- and high-risk cancer genes, she said, and offer additional information at no additional cost to those who test negative in those genes.
Tippin Davis pointed out, though, that in order to use PRS for clinical decision-making about enhanced screening, disease-specific thresholds need to be established, adding that for breast cancer PRS, clinical trials are underway, such as the Women Informed to Screen Depending on Measures of Risk (WISDOM) trial and Ambry Genetics' Clinical Implementation of a Polygenic Risk Score for Breast Cancer study.
Wald said companies adding PRS to their risk assessment tests should be required to state by how much this improves the detection of individuals at increased risk. "I think what they should provide is estimates of the detection rate and false-positive rate of their combined test because that's what matters in screening," he said.