Polygenic risk scores (PRSs) summarise genome-wide genotype data into a single variable that measures genetic liability to a disorder or a trait. Technically, the PRS is calculated from genome-wide association study (GWAS) summary statistics, summing the number of risk alleles carried by an individual, weighted by the effect size from the discovery GWAS. The PRS is seductive in its simplicity, summarising several million genotyped and imputed common genetic variants, and it is easily calculated using standard software [1]. The PRS is widely used in research studies but does it have potential as a clinical tool for risk prediction, prognosis or stratification?
Currently, the PRS is most often used to follow up GWAS, testing the prediction of case–control status or a continuous trait in an independent study. The disease or trait tested may be the same as that in the discovery GWAS or different; for example, testing the hypothesis that the type 2 diabetes PRS predicts depression case–control status. Such studies give a measure of predictive ability, such as the proportion of variation in trait status that is explained.
The PRS is often standardised for easy interpretation, rescaling so that scores have a mean of 0 and a standard deviation of 1. This allows the conversion of an individual’s PRS to quantiles; for example, identifying the 10% of the population with the highest PRS. We expect that the average PRS in cases will be higher than that in controls (indicating a higher genetic risk for the disorder), but the difference may be small. Many individuals will have a PRS value close to the population mean, implying that the PRS adds little information, and the individual’s predicted risk will be close to the population life-time disease risk.
For clinical application, the perspective moves from comparing PRS values in cases and controls to assessing where an individual’s PRS lies on the population distribution. For example, individuals with the highest 1 or 5% of PRS values, depending on the estimated risk for the disease and its severity, might be offered regular screening, encouraged to participate in lifestyle modifications or prescribed therapeutic interventions. The potential value of using the PRS in defining screening algorithms has already been observed in breast cancer, where the PRS was used to stratify breast cancer risk and to explore the implications for screening [2]. In the UK, mammogram screening is initiated at the age of 47, based on a 10-year risk of breast cancer in the average woman. Mavaddat et al. [2] showed that women in the top 5% of PRS risk reach this level of risk at the age of 37, while those with the lowest 20% of PRS will never reach it. This study suggests that, even with our incomplete knowledge of breast cancer genetics, a PRS-based population cancer screening programme could be defined. However, there are substantial barriers to implementation. These tests will require extensive training of medical professionals, access to large-scale genotyping and interpretation; most importantly, the tests are likely to be controversial, and would need to overcome negative public attitudes towards genetic testing [3].