Many diseases are linked to variations in an individual’s genetic information. For some diseases, variation of a single gene may be sufficient to cause the disease. However, for most conditions, the disease is associated with a combination of tens to thousands of single genetic changes across many genes. Genetic information on thousands of different locations in the genome can be combined together to generate polygenic risk scores (PRS) that can predict the likelihood of that person getting a particular disease. It is hoped that in the future, PRS can be used in the clinic to risk stratify patients with the aim of providing earlier or preventative treatment.
However, while many PRS have been generated for a range of diseases, there has been little research into how these different PRS compare with each other when predicting the disease risk of an individual. Using genotyping data on white British individuals from the UK Biobank, Dr Lei Clifton and colleagues from Oxford Population Health compared two PRS each for breast cancer, hypertension and dementia.
The researchers identified two main points of disagreement between the assessed PRS. Firstly, 20% or more of the genetic variations used in the first PRS did not overlap with those used in the second PRS for all three diseases. Secondly, there were large differences in the prediction of risk in individuals between the two PRS, presumably attributable to the statistical methods used to construct them. For example, 60% of individuals classed as being in the top 5% of risk when using the first PRS were not classed as this highest risk level by the second PRS.
The finding, published in Scientific Reports, that different PRS result in different risk predictions for individuals has implications for the use of PRS in medical settings. It is therefore important to develop guidelines on PRS generation so that the uncertainties in the predictions are conveyed during their clinical deployment.