The future of polygenic risk scores looks bright. But does it? Responsible use starts with getting real about the predictive ability and utility. And getting the science right first. nature.com/articles/s4159…
@cecilejanssens Noted by the authors as the first research gap to be filled.
@GENES_PK Yes, but how long can you keep saying that? It's already clear for years where the promising applications will be. Pretending that CAD PRS is promising is laughable ... disappointing, and misleading.
@michelnivard @cecilejanssens @GENES_PK Need to remind ourselves PRS will soon come for free, and that they're nothing special - we just treat them as another variable like age or LDL-C. But only need to measure them once. If we get that straight the bar on how much variance they need to explain is lowered ?
@michelnivard @cecilejanssens @GENES_PK its just part of the multivariable model ? clinical report e.g. produces a single value for 10 year risk of heart attack that is a little more predictive by adding in PRS. GPs dont need to know anything more than the score "includes your genetics" ?
@timfrayling @michelnivard @GENES_PK with the difference that, for many diseases, the genes also affect the clinical risk factors, and thus don't add much. I find it misleading to communicate to patients that a risk is also based on DNA when the latter may not matter much.
@cecilejanssens @timfrayling @michelnivard I do not think this is empirically true for most complex diseases. The per-PRS-sd change in log odds is roughly the same in models with or without risk factors, even though some of the PRS effect is mediated thru risk factors (which are non-exhaustive and measured with error).
@GENES_PK @cecilejanssens @michelnivard Yep. Another good example is that height PRS adds to mid-parental height in predicting your adult height. Presented at #ASHG21 by GIANT consortium.