Polygenic risk researchers can publish PRS after PRS, but their studies will not bring us an inch closer to implementation of PRS in practice when the studies lack clinical or public health relevance
1/
1/
Finding PRS associations is of little interest.
When PRSs are constructed from SNPs that are *associated* with a disease, then PRS is *associated* with the risk of disease.
The strength of association is of interest, not the p-value.
When PRSs are constructed from SNPs that are *associated* with a disease, then PRS is *associated* with the risk of disease.
The strength of association is of interest, not the p-value.
The strength of association determines how well PRS can predict. This *strength* of association and thus predictive performance varies between populations (among others). That's why prediction needs to be studied in a representative sample of ...
... the population in which the PRS will be applied if proven predictive.
An example:
1 clinical risk model, 1 PRS
5 different populations
--> 5 wildly different predictive performances.
Does PRS improve prediction here?
An example:
1 clinical risk model, 1 PRS
5 different populations
--> 5 wildly different predictive performances.
Does PRS improve prediction here?
Predictive performance in one population doesn't guarantee similar performance in others.
That's why PRS prediction needs to be studied in clinically or public health relevant populations ... if we want medicine and public health to be evidence-based.
/end
That's why PRS prediction needs to be studied in clinically or public health relevant populations ... if we want medicine and public health to be evidence-based.
/end
