Thread by @cecilejanssens

"This glaucoma PRS will facilitate the development of a personalized approach for earlier treatment of high-risk individuals, with less intensive monitoring and treatment being possible for lower-risk groups."

Let's look at the data:
1/

nature.com/articles/s4158…

Adding the polygenic risk score to age, sex, and family history increased the AUC from 0.73 to 0.80. Individuals in the top decile had a 15-fold increased risk to those in the bottom decile. This is impressive.

The authors conclude:
"Our glaucoma PRS will be primarily *useful* for stratifying individuals into risk groups; for example, in the BMES data, screening the top *decile* of the PRS ... identifies 40% of cases."

But is that what the BMES data showed?

Not really. Not for selecting the top decile.
When a disease is 'rare', the top decile is approx where specificity is 0.90 or 90%. (Mind: the x-axis is labeled wrong here, should be 1-spec.) The top decile is then approx where 1-spec = 0.10

Around that threshold, PRS barely improves prediction of glaucoma beyond sex, age, and family history. If at all.

At the other end of the curve, the situation isn't much different. So also for selecting the lower-risk groups, the PRS doesn't add much. PRS only starts making a difference when more than 25% of the population is considered at lower-risk.

When AUC improves, always check whether the improvement is in the range where potential screening thresholds will be. If not, then the improvement, by adding PRS or other variables, has no utility.

More on how to interpret the AUC:

https://twitter.com/cecilejanssens/status/1104134423673479169

(paper in press)
/end