October 17, 2023

expert reaction to study on polygenic risk scores and their performance in screening, prediction and risk stratification

A study published in BMJ Medicine analyses the performance of polygenic risk scores. 

Prof Gilean McVean, co-founder and President of Genomics plc, and Professor of Statistical Genetics at the University of Oxford, said:

“The article largely misses the point about how PRS should be used in healthcare. PRS are not a diagnostic test, and should not be evaluated as you would a test for cancer or diabetes. This is not, for example, how existing risk tools – such as QRISK for cardiovascular disease – have been evaluated. PRS provide risk information ahead of screening and early diagnosis. They can be used to support the personalisation and effectiveness of screening programs (e.g. start breast cancer screening earlier and with increased frequency for those at highest risk) and preventative treatments (ensuring those at greatest risk start on statins to prevent heart disease or attend diabetes prevention programs) precisely because, as shown in the data in the paper (and in many other papers by researchers across the world), PRS can differentiate among individuals in terms of risk, not just qualitatively, but also quantitatively. This means that accurate and personal statements about individual risk can be made to support an individual’s engagement with healthcare, particularly when combined with other risk factors (not least age, sex, lifestyle and environmental factors).

“The use of genetics to personalise prevention, early diagnosis and screening is already embedded in healthcare, for example in how we care for carriers of familial mutations in known disease genes, for example BRCA for breast cancer or the genes conferring increased risk of familial hypercholesterolaemia. However, PRS can identify many more people with an inherited risk at the same level as those carriers, but who are currently invisible to the system. The opinion expressed in the press release – “Our results build on evidence that indicates that polygenic risk scores do not have a role in public health screening programmes” – argues that such risk should remain hidden, which I do not understand, and runs counter to the growing use of personalisation in treatment decisions and healthcare.”

Dr Mark Iles, Associate Professor of Epidemiology in the University of Leeds’ School of Medicine, said:

“Broadly, the authors do a good job of assessing the impact of using some existing polygenic risk scores (PRS) in clinical practice.

“They conclude that the impact of PRS is minimal – however, this is a matter of perspective.

“The authors estimate that, among their sample, adding genetic information to the standard risk score used to prescribe statins would prevent a further 21 cardiovascular events or strokes. This is a gain of 3%, which sounds reasonable.

“Given the authors have only looked at coronary artery disease and breast cancer screening, there may be other examples where the improvements are even greater – for instance in improving diagnosis or predicting side effects.

“A better approach would be to base this study on heritability – how much patients’ genes contributed to them developing disease. This would show us the limits to how well we can predict a disease using genetics.”

Prof James Ware, Professor of Cardiovascular and Genomic Medicine, Imperial College London, said:

“This is an interesting study.  In my opinion the press release reflects the science, and the analyses appear robust.

“I would not want readers to conclude that all genetic tests are useless in all contexts: this is specifically about polygenic risk scores used in population screening.

“The authors do a very good job of highlighting the limitations of applying polygenic risk scores in a population screening context, and laying out some of the concerns around potential clinical implementation.  PRS generally don’t have sufficient discriminatory power to be cost effective when used alone.

“The authors acknowledge that PRS can improve the accuracy of conventional risk scores used today, for example to decide which patients should receive a statin to prevent cardiovascular disease, but they highlight that the improvement is modest: you would need to genotype thousands of individuals to prevent a significant cardiovascular event.  You could prevent the same number of events more simply and cheaply by lowering the threshold for prescribing a statin.

“This argument does depend on the cost of genotyping.  If genetic data is already available, then this modest improvement may become worthwhile, since it is effectively free.   An increasing number of people have genome analysis undertaken for other reasons – e.g. for medical reasons, while participating in research, or recreationally to investigate their ancestry – in which case their routine cardiovascular health check could be slightly improved by the incorporation of a PRS.  There is currently a national study in the UK to investigate using genomics for newborn screening: this would provide data that could be used to calculate PRS later on.

“A polygenic risk score can be measured early – potentially long before other risk factors appear.  People at higher risk of cardiovascular disease might be motivated to form healthy lifestyle habits early on (e.g. diet and exercise), which could add value.  In isolation this is unlikely to be cost-effective – it would be better to invest resources in encouraging everyone to eat well and be physically active – but a PRS might be useful if data are already available for other reasons.  The flipside of course is that some people may be reassured by a low PRS, think that they don’t need to eat well and exercise, and go on to have a cardiovascular event.  There are studies ongoing to evaluate this.

“Finally, the authors emphasise that polygenic risk scores may be helpful in other contexts beyond population screening. For example, in families with cardiomyopathy or familial hypercholesterolaemia that are caused by a faulty copy of a single gene, not all families who have the faulty copy develop disease.  A PRS might help us to understand which family members need closest surveillance, and who can be followed up less frequently.”

Professor Clare Turnbull, Professor of Translational Cancer Genetics at The Institute of Cancer Research, London, said:

“The research is of high quality with robust results.  This research shows that for the 310 diseases studied, polygenic risk scores are not predictive enough to be useful in the clinic. We typically see that the majority of cases of disease still occur in people who are not considered ‘high-risk’, as determined by a polygenic risk score. Even when we expand the window of what is considered ‘high-risk’, which lets in huge numbers of false positives (people who will never develop the disease in question), we still see that most cases of disease fall outside of the ‘high-risk’ window.

“The ambition behind using polygenic risk scores is that we could use them to determine someone’s risk of a disease, and those at highest risk could be offered screening or preventative treatment – but their predictive power is just not strong enough, and they are unlikely to be useful for this purpose either now or in the future.  This is because most diseases have a quite limited heritable component,  much of which is comprised by genetic variants that are of too low frequency or too modest effect to ever be captured by a polygenic risk score.”

Prof Michael Inouye and Assistant Prof Sam Lambert, University of Cambridge (both lead authors of the Polygenic Score Catalog which was looked at in this study), said:

“One thing that’s clear from the PGS Catalog is that polygenic scores have potential utility in many different clinical use cases – hundreds of papers from research groups around the world have demonstrated this. The current paper generally takes a narrow view of how polygenic scores can be used, e.g. estimating limited benefit of PGS for breast cancer in a two-stage screening process where screening occurs once. Many other studies use PGS in more flexible ways, for example those that vary the interval and age of screening initiation of cancer screening do show potential public health benefits and indeed cost-effectiveness of PGS. Similar results have been seen for cardiovascular disease. Considering the extensive literature on PGS, it does now look like there is ample evidence for ramping up translational studies of polygenic scores, including pragmatic trials.”

Prof Dusko Ilic, Professor of Stem Cell Sciences, King’s College London (KCL), said:

“It comes as no surprise to me that polygenic risk scores have shown poor performance in population screening, individual risk prediction, and population risk stratification. They overlook the significant impact of epigenetic regulation and the environment on an individual’s overall health.

“Reports like this one serve as a necessary reality check to be wary around exaggerated claims surrounding the influence of polygenic risk scores on healthcare and health outcomes. This applies to both individuals and healthcare professionals.”

Prof Laurence Hurst FMedSci FRS, Professor of Evolutionary Genetics, University of Bath, said:

“The principle of Polygenic Risk Scores (PRS) is clear.  From GWAS one can obtain a catalog of variants predictive of certain diseases and then employ such a catalog to consider whether, given your DNA, you are prone to a particular disease.  This can be emptied in principle in many contexts from screening a population to stratification of patients.  The key question addressed by this paper is how good and useful are these predictions.  Not very is the answer.  In short this is good quality work that acts as a necessary caution to the current “hype” and enthusiasm for PRS methods.  

“To date I am unaware of any PRS method in clinical practice.  There is related technology for screening embryos which is illegal in the UK but the first baby screened by this method was born in the US in 2021 if I recall rightly. 

“To put this more into context, we knew before this analysis that in the case of cardiovascular disease the method can be really rather imperfect. A fairly recent study was found to miss 85% of individuals with the disease.  That makes for a pretty expensive not very good method.  This paper extends the same word of caution – the method is not strongly predictive – they consider cardiovascular disease in god detail coming to similar conclusions.   

“I don’t however think one should throw the baby out with the bathwater quite yet, especially in cases where current diagnostics are poor.  I’m thinking especially of prostrate cancer where I think a case can be made for PRS. Men in their 50s often go to their doctor complaining of problems with their lower urinary tract. They might need to get up in the night to wee, they may need to go very often or the flow of urine may be weak. Sometimes this is owing to a relatively harmless growth of the prostate gland. However, 3.5% of individuals complaining of these same symptoms will go on to develop prostate cancer within 2 years. Prostate cancer accounts for about a quarter of all new cancers in men, about 52,000 per year in the UK (and increasing by 4% each year). 10-20% of deaths could be avoided with earlier detection. 

“GWAS suggests there to be 269 risk variants for prostate cancer. From these for any man we can estimate a risk score – the more of the “risky” mutations you have the more at risk you will be. I am aware of a recent analysis that suggest that we can effectively use a genetic risk score  to decide which of the men should be fast-tracked to the clinic and which can safely not go for further investigation. Those with the lowest risk scores (bottom 20%) had less than 1% chance of developing cancer within 2 years, whilst those in the top 20% had an 8.8% chance.  As an approach, when combined with male age, this is more accurate than the current best test looking at levels of prostate specific antigen (PSA). It is an open question whether a combination diagnosis using genetics and PSA levels might work even better. 

“Even if the genetic predictions were used on their own, this could have important consequences for hospital capacity. In the UK frontline doctors make about 800,000 referrals for possible prostate cancer annually. If we added in information from PRS to this, about 320,000 men could safely avoid referral (which involves both unpleasant investigation and anxiety) while 160,000 could be fast tracked. I’d like to see that properly tested before fully rejecting the approach. But I agree with the current authors, that the current promises (“sell”) for a brighter medical future with PRS seem more like optimistic marketing.”

Reference for study described in comment above: Green HD, Merriel SWD, Oram RA, Ruth KS, Tyrrell J, Jones SE, et al. Applying a genetic risk score for prostate cancer to men with lower urinary tract symptoms in primary care to predict prostate cancer diagnosis: A cohort study in the uk biobank. Br J Cancer. 2022;127(8):1534-9. doi: 10.1038/s41416-022-01918-z.

‘Performance of polygenic risk scores in screening, prediction,  and risk stratification: secondary analysis of data in the Polygenic  Score Catalog’ by Aroon D Hingorani et al. was published in BMJ Medicine at 23:30 UK time on Tuesday 17^th October.

DOI: 10.1136/bmjmed-2023-000554

Declared interests

Prof Gilean McVean: “I am co-founder and President of Genomics plc, a UK company that has developed the tools and technologies required to measure and understand the inherited risk for disease we carry in our genomes and use this information to improve healthcare, working with healthcare partners, drug developers and organisations looking to support their employees and customers. Genomics plc provided the PRS that UK Biobank has released to their research community. We are also the providers of genomic risk information to Our Future Health, the government-funded program that is looking to realise the potential of such information within the UK.”

Prof James Ware: “I was not involved in the research.

My research does look at polygenic risk scores in a different context: using them to understand the variable penetrance of genetic variants that can cause monogenic disease.

I have acted as a consultant for Bristol-Myers Squibb, Pfizer, Foresite Labs, and Health Lumen.  None of that work relates to the use of PRS for population screening.”             

Prof Michael Inouye and Assistant Prof Sam Lambert: No COIs.

Prof Laurence Hurst: “I am an exPresident of the Genetics Society and work in evolutionary genetics as allied to medicine.”  

Prof Dusko Ilic: No conflicts of interest.

For all other experts, no reply to our request for DOIs was received.