🚨 Our work exploring polygenic risk scores for Coronary Heart Disease is now out in @JAMA_current and presented at #AHA24 by all-star @SarnoffCardio fellow @sabramowitz_: https://t.co/AJ5WKsYbBs

Very excited to share our preprint led by @MGLevin @skoyamamd @jakob_woerner & with @damrauer assessing genome-wide pleiotropy of >1,000 clinical traits across ~1.7M individuals with nearly 30K locus-trait associations! https://t.co/hJChZPSYNm @medrxivpreprint [1/3]

CVI's Highlighted #PapersoftheWeek https://t.co/ZmaYDfURPc

Common-variant and rare-variant genetic architecture of heart failure across the allele-frequency spectrum | ⁦@NatureGenet⁩

Check out this editorial just published in @JAMACardio by our very own💫@PennCardiology fellow Anthony Anguiera & @sabramowitz_ under the mentorship of the one & only @MGLevin! Unfolding the Link Between Transthyretin Stability and Survival https://t.co/KPDs4N9Svm

CVI's Highlighted #PapersoftheWeek @MGLevin https://t.co/FfKwlIplrK

New research by @damrauer & @MGLevin suggests that polygenic risk scores may provide conflicting results for detecting heart disease risk.

Now out in @JAMA_current: How well do individual risk estimates from different CHD polygenic scores agree? Grateful to work on this with a dream team of mentors and co-authors @damrauer @MGLevin @bpasaniuc @MarylynRitchie @dokyoon_kim @bvoight28 @BoulierKristin @karlkeat

14/ To learn more, - Check out this editorial: https://t.co/goumqQ1XGo - Listen to this AI-generated podcast summary:

13/ It was fun working with the team: @sabramowitz_, @damrauer, @BoulierKristin, @karlkeat, @MarylynRitchie, @bvoight28, @bpasaniuc and everyone else. More exciting work to come!

12/ Overall, we need better frameworks for evaluating PRS at both the individual and population level, and approaches for dealing with the variability/uncertainty if PRS are going to become clinically useful.

11/ These results don't address other uses of PRS where individual-level performance may be less critical (eg. population-health, enriching the average risk of a clinical trial cohort, etc.).

10/ Our findings raise important practical questions about the implementation of PRS: - When PRS from two different sources disagree, which should we believe? - How should we counsel patients about uncertainty? - Can CAD PRS be individually useful given these limitations?

9/ Explore how these CAD PRS vary for yourself: https://t.co/676YKN7yi2

8/ Concerningly, ~20% of participants had at least one score in both the top 5% and bottom 5%, and ~80% had at least one score in the top/bottom 20%.

7/ Quantifying this agreement at the individual-level using metrics like intraclass correlation and Light's kappa, we found generally poor agreement. The standard deviation (22%) and coefficient of variation (50%) were also high.

6/ Next, we sought to understand individual-level agreement/precision of these scores. Despite similar population-level performance, risk estimates from these scores were only modestly correlated at the individual-level.

5/ Using a "Region of Practical Equivalence" framework considering important measures of predictive performance (calibration and discrimination), we found that most scores are "practically equivalent" in their ability to identify CAD at the population-level.

4/ To start, we identified published CAD PRS from the @PGSCatalog, and evaluated population-level performance in 3 large/diverse biobanks: @AllofUsResearch, @PennMedicine, and @UCLA.

3/ Despite these applications, the individual-level agreement of PRS risk estimates has been poorly characterized. Because there's no gold-standard to benchmark the accuracy of individual-level scores, this problem is challenging.

2/ Historically, polygenic risk scores have been evaluated at the population level. However, many applications of PRS are targeted at individuals (eg. in the clinic, direct-to-consumer testing).