Analyzing an individual’s genome to generate risk profiles for serious disease sounds like the ultimate weapon in the fight for preventative health care, but does DNA analysis deliver on its promise?
In the U.K., a heated debate is raging in the genetics community about a not-so-new technology and its role in public health. Cheap genetic tests to discover our ancestry have become familiar consumer products, and our genes can tell us a lot about our ancestry, so it is an appealing idea that they can tell us about our susceptibility to serious diseases. Polygenic risk scores (PRS)—generated by sequencing multiple parts of a person’s DNA—are said by some to hold the key to helping people avoid everything from type 1 diabetes to cardiovascular disease and cancer. This could herald a new era of preventive medicine, and the U.K. is investing heavily, but ultimately, whether or not this a good investment is still being determined.
Genome-wide association studies have underpinned PRS for many diseases with a heritable component, and the old concept that genetics can predict disease has been given a new lease on life. It is tempting to think that a cheap genetic test can predict our future health, but they do not consider the effects of environmental or other nongenetic factors. The causes of disease are multifactorial and are spread through the whole genetic code, so can PRS really facilitate more meaningful interventions, change behaviors that contribute to risk, or target screening more effectively?
“The move came from understanding that small snips of genetic code do not provide useful prediction on their own, so counting a number of risk alleles could generate a risk score that is more informative,” says Prof. Aroon Hingorani, University College London’s Division of Biosciences, whose work focuses on the use of genetic studies to identify and validate drug targets. “That idea has been around for 15 years or more. Interest has been fuelled by the possibility of generating PRS, and the emergence of large national biobanks, like UK Biobank.”
UK Biobank is a large-scale biomedical database and research resource, containing in-depth genetic and health information from half a million U.K. participants. It is part of the U.K.’s push to use genetics to revolutionize public health.
That policy was pushed forward by former Secretary of State for Health and Social Care Matt Hancock, among others. Hancock’s speech in 2019 to the Royal Academy, in which he talked of receiving genetic analysis from Genomics plc that had calculated his predicted risk of 16 common diseases, added fuel to the PRS debate. He stated that “genomics might have saved my life” because he was made aware of his elevated risk for prostate cancer—15% by the age of 75. The groans this elicited from many in the academic community can still be heard today.
Paved with good intentions
The truth is that most men have a relatively high risk for prostate cancer, and Hancock’s statistics put him somewhere in the middle of the pack, rather than at the high end. Combining information from thousands of individual genetic variants to create a PRS can be a powerful approach to identify individuals with markedly higher or lower risk of particular diseases. This much is certainly true, the question is whether these risk profiles are useful.
According to Genomics plc, PRS is not a stand-alone diagnosis tool, but a complement to other risk factors that can help to improve screening programs by targeting at-risk individuals and better interpreting screening tests. The company also espouses the belief that improved prevention is possible for many diseases, as high-risk people can either make lifestyle changes or seek medical intervention.
With these claims in mind, the reasons for pushing PRS in the U.K. seem laudable. The National Health Service (NHS) has aggressive targets for the early detection of serious diseases, particularly cancer, and is hoping genomic medicine is the solution. Hence the 2018 launch of the NHS Genomic Medicine Service to deliver “equitable genomic testing for improved prediction, prevention, diagnosis and precision medicine.” The NHS Long Term Plan aims to significantly improve the diagnosis and treatment of cancer by 2028, with an extra 55,000 people each year surviving for five years or more following their cancer diagnosis. It further hopes that 75% of people with cancer will be diagnosed at Stage 1 or 2.
“There is an urgent need for something to change in how health care is delivered,” says Dr. Amit Sud, academic clinical lecturer in genetics and epidemiology at the Institute of Cancer Research, London. “The NHS has ambitious targets for cancer diagnosis and there is an excess of cardiovascular deaths in this country and elsewhere, and technologically we have reached a point in genetics where the translation of knowledge gained over decades can be used. But for cardiovascular disease, PRS adds little to the known risk factors.”
“There is the idea that you could offer some changes to non-genetic risk factors, but those changes should not be limited to people with a higher risk, they are good for everyone,” he adds. “It is perplexing that it is viewed in that way. People are looking for solutions to cancer and other health care issues, but PRS won’t help the NHS achieve its goal with cancer diagnosis.”
Nevertheless, PRS is becoming part of the platform for the U.K.’s future delivery of health care services. “PRS is now being rolled out in a government venture—Our Future Health—a national effort to recruit five million people,” says Anneke Lucassen, a professor of genomic medicine and director of the Centre for Personalised Medicine, St Anne’s College, Oxford University. “It is a great research tool but feedback of information to individuals about predispositions to diseases is less useful. UK BioBank has it right, because there is no such feedback, though the broader advances in research it delivers will be fed back into health care.”
True sceptics—and they are serious academics, not foil hat-wearing conspiracy theorists—note that there are strong commercial interests at play in the push for PRS to become a mainstay in public health programs. For instance, Illumina, the goliath of genome sequencing, has been embedded into NHS public health projects, including sequencing genomes for a U.K.-wide COVID-19 study. That process was started by former Prime Minister David Cameron when in office. After resigning, he joined Illumina as a consultant and seemingly played a role in securing a £123m contract after meeting with then health secretary Hancock.
What can PRS do for public health?
Whether commercially driven or not, PRS would seem to be an appealing concept for public health at first glance, but it appears to fall short of its promises.
“If something is shiny and new people think it is good, but in cancer, we know that smoking, being fat, and drinking too much have a huge impact on risk,” remarks Prof. Richard Houlston, head of the Division of Genetics and Epidemiology at the Institute of Cancer Research. “The government has grappled with that for years, so it is not inspiring, but the idea that genetics are deterministic is very attractive, and it doesn’t come with any advice to stop smoking.”
“What is lost in that excitement is that you need to interpret the information in a broader sense, not just look at bits of genetic code and think that shows you what to do,” says Lucassen. “The wider you look, the more variation you find and the harder it can be to interpret. That is where we need a more widespread understanding of what PRS can and cannot do. Interpreting anyone’s genetic code is complex, so we need to understand the interaction of genetic factors with each other and with our environment and predictions are often far less clear than people expect. PRS predict differently in different populations.”
The key flaw with PRS seems to be the narrow definition of “high risk.” Knowing a person’s lifetime risk of prostate cancer is 12% rather than the average of 11% does not result in better treatment or intervention. “It may not be that useful to know a person’s lifetime risk of a common cancer is a couple of percentage points particularly if that knowledge doesn’t result in better treatment or intervention,” says Lucassen. “For many PRS there is no follow up test—such as a biomarker or scan that improves the prediction or tells you whether the disease is actually present—so it’s debatable whether knowing really gives you more power.”
The counterargument is that these scores are just another risk factor that could be useful in context with other factors like blood pressure and cholesterol. All too often, however, genetics are seen as deterministic, so the expectations of PRS may be too great. PRS is only as good as the screening process that goes with it.
“Cancer screening programs, for example breast screening, are costly,” says Clare Turnbull, professor of translational cancer genetics in the Division of Genetics and Epidemiology at the Institute of Cancer Research (Figure 1). “Whilst there may be modest impact on breast cancer-specific mortality, it is still debated whether breast screening impacts on all-cause mortality. There is an ambition is to improve early detection and survival, but PRS is only as good as the efficacy of the cancer screening tool applied in the ‘high-risk’ group, since the cancers arising in that group are biologically and clinically equivalent to those arising in the population as a whole.”
Turnbull believes that if a screening tool trialed in the general population has demonstrated poor sensitivity for specificity, has limited impact on survival, or results in substantial overdiagnosis, these outcomes will likewise hold true for the PRS-defined high-risk group. A screening paradigm that is problematic remains problematic despite stratification into PRS-defined groups.
Some genes have been identified as high-risk for hypercholesterolaemia (LDLR gene), breast cancer (BRCA1/2 genes), colorectal cancer (Lynch genes), and other conditions, for which genetic testing can identify a small number of people at very high risk.
“However, for the most part, common cancers and diseases such as Type 2 diabetes and ischemic heart disease have only a modest genetic contribution, which is determined by many hundreds—or more likely thousands—of variants, the majority of which have been shown to be too rare or of too modest effect to be identified by our genome-wide association studies,” she adds. “Mother Nature is far more complex than we thought 20 years ago when we coined the common variant hypothesis.”
PRS takes a set of associated variants—there are currently 300 for breast cancer—and gives a score. You could take the top 20% of the group and start screening earlier—perhaps at 40 instead of 50. Residual risk over ten years is around 1.7%. For the top 20%, it rises to 3.2%—nearly double, but still not that high.
Furthermore, for early detection to significantly improve outcomes, the focus would need to be on young people if the goal is to buy more years of life. “Because these diseases—common cancers and other conditions such as Type 2 diabetes and ischemic heart disease are mainly due to environmental exposures accrued during aging, and because we can only identify a modest proportion of the genetic component, polygenic risk scores are only weakly predictive of disease,” says Turnbull. “This means that even in the so-called ‘high-risk’ groups, disease incidence is relatively low. Depending on where your cut off between high and low risk is, much or most of disease will arise in the latter low-risk group.”
A screening paradigm that is a problem remains a problem with PRS. With screening for many cancers, overdiagnosis is already a problem and increased screening may have a limited impact on survivability. Hingorani agrees that population-wide polygenic scoring is inherently limited because disease frequently occurs in people who do not have high-risk scores, and some with high risk do not develop disease.
“If you are talking about using PRS in screening for disease or stratifying the population into groups that are higher or lower risk, then you need metrics such as sensitivity—the proportion of cases you detect—and false positive rate,” he says. “You want the first number to be high and the second low.”
Mammography for the detection of breast cancer has a sensitivity of 75% against a 7% false positive rate. For the faecal immunochemical test for colorectal cancer, it is approximately 79% against 7%. When Hingorani turned PRS scores into the sensitivity and false positive rates, their typical detection rate was 12% for a false positive rate of 5%. “You miss 78% of cases,” he explains. “Yet coronary artery disease remains an area where people are pushing PRS while detection rate is only 12%.”
“There is very little value in PRS, as it provides little differentiation,” says Houlston. “It doesn’t change our understanding of the risk factors. For prostate cancer, what modification can you do? The only risk factor is being old. For lung cancer, there is only a small difference between high-risk and low-risk groups, while smoking carries a huge risk factor.”
Therein lies a key problem: PRS does not account for—or affect—powerful environmental and social factors. “People hear low, medium, or high risk PRS as a total risk, measuring a minority of the total risk—a ballpark figure of approximately 20% for many common diseases—so prediction is still very difficult if 80% of the risk comes through other factors,” Lucassen explains. “We also have very little evidence that telling people about their risk changes their behavior.”
A meta-analysis published in the British Medical Journal in 2016, titled “The impact of communicating genetic risks of disease on risk-reducing health behaviour: systematic review with meta-analysis,” found that people tend not to change their lifestyle on the basis of genetic results. In fact, a study of nearly 1,000 blood donors found that the provision of PRS information for coronary heart disease did not affect objectively measured levels of physical activity or other health-related behaviors.
A call for calm
The antipathy between the different camps in the PRS debate has become unnecessarily exaggerated. Critics admit that there are potential benefits. Houlston, for example, sees PRS as useful for designing trials for new interventions that could prevent serious diseases, such as osteoporosis in women, because the treatment population could be enriched with high-risk candidates. But for early detection of disease, many academics give a firm “no” to PRS.
“If PRS were cheap and simple without any downsides, and our screening data systems afforded super-slick individualized programs of when and how often an individual was screened, implementation of PRS might afford a modest boost in cancers detected per screen performed,” says Turnbull. “However, PRS won’t solve whether screen-detection of that cancer results in improved outcome or just causes overdiagnosis. Moreover, it would take an almost unfeasible investment to set up data systems enabling such individualized screening protocols.”
“Furthermore, there are some potential harms,” she adds. “Requirement for a genomic testing ahead of screening may trigger disengagement. There may be behavioral change in those assigned as low risk with them disregarding lifestyle advice or ignoring symptoms. Also, PRS perform more poorly in non-European ancestries, which presents a significant issue. The U.K. has a long history of excellence in clinical genetics and more recent prominence in our early adoption of clinical whole-genome sequencing. But for most putative scenarios, introduction of PRS-stratification will add negligible clinical benefit. We need to be mindful of genomics for genomics’ sake. It is not ‘world-leading’ to be the first in the world to introduce types of genomic tests that have no clinical utility.”
PRS could be a useful research approach that adds to the understanding of the risk mechanisms of common diseases but may be less useful in the context of a public health organization like the NHS that is already struggling.
“Prevention is better than cure, and that is certainly an appealing rhetoric, but do we have realistic expectations of what PRS can deliver?” wonders Lucassen. “I’m not sure. For example, your childhood postcode might be just as good an indicator of your risk of some common diseases, and our public patient involvement and engagement (PPIE) group were very surprised to hear that.”
For now, PRS is a beguiling idea but expensive and of limited use in disease prevention. The belief that genes can be tested once to create a roadmap for better preventative medicine remains, for now, science fiction. Genetic medicine is apparently not as deterministic as the politicians would like to believe.