A systematic analysis of trans-ethnic portability and within-family validation of polygenic risk score prediction
Over the last ten years, polygenic risk scores have advanced from academic subject matter to clinically useful tools, capable of assisting diagnoses and setting screening standards. Yet, as medical professionals embrace the use of these tools in the US healthcare system, researchers have warned that these predictions are worse than expected for non-Europeans and cannot stratify risk within families as well as expected. Over the last several years, statistical geneticists have developed theory for and empirically analyzed the causes of these differences in prediction accuracy by ancestry.
Prediction accuracy tends to get worse as the genetic distance between training cohort and test cohort increases, and this can even occur when applying polygenic risk score models trained on a British cohort to an Italian test cohort. To help fix these problems, researchers have developed a number of tools that help construct risk prediction models that take ancestry into account.
Within the next 36 months, we aim to systematically assess these methods across a variety of traits to compare their utility risk prediction in non-European cohorts. We further intend to investigate whether these gaps differ by the trait or disease analyzed.
The proposed research fills an important gap in the current literature because there is no standardized baseline. Since differences in risk prediction by ancestry create a barrier to equitable use in the clinic, this research addresses an important need in public health.