Thread by @cecilejanssens

New perspective in JAMA *insights* about polygenic risk scores and their importance. Here's my 'letter to the editor':
1/n

The paper has a great infographic explaining the steps in polygenic risk prediction. Great for teaching. The legend doesn't really fit the contents, but that can be cropped.

Here's where the authors started to loose me: I would not refer to the scores as risks. True, the higher the score, the higher the risk, but a score on its own doesn't say anything about risk. The score needs to be transformed into risk by considering disease incidence.

Here I got confused: are these the main variations that characterize polygenic scores? Scores can also vary in the weights, how the weights are obtained, the variants that are combined (variations and mutations), the disease incidence, to name a few.

Here's where the authors take a higher bar than I do. Ideally, the statistical model should reflect the underlying process, but, then, the simple adding of SNP effects unlikely comes close. Adding may work for prediction, but it is a huge simplification of the underlying biology.

(don't tell me you didn't know)

Here I start to disagree: to be *clinically useful*, knowledge of the polygenic risk score should enable or improve medical decisions. There should be a benefit of testing.

Then this: polygenic risk scores are not deterministic. True ...

... but that is *why* they are not equivalent to diagnostic tests. If they were deterministic, like in Huntington disease, then the scores *were* presymptomatic diagnostic tests.

Here I agree: clinical utility is the big unknown. We don't know whether polygenic risks have clinical utility and whether the benefits of testing outweigh the costs. And remember: the DNA testing might one day be free, the follow-up testing/procedures are not!

With utility still being the big unknown, it is unlikely that polygenic risk scores will become part of clinical care in the *near* future.

More on this can be found course manual on prediction research, including many references to key methodological papers (a polygenic risk version is coming soon).
cecilejanssens.org/wp-content/upl…