Multi-ancestry polygenic risk scores for the prediction of type 2 diabetes and complications in diverse ancestries.

Abstract

BackgroundPolygenic risk scores (PRSs) improve type 2 diabetes (T2D) prediction beyond clinical risk factors but perform poorly in non-European populations, where T2D burden is often higher, undermining their global clinical utility. MethodsWe conducted the largest global effort to date to harmonize T2D genome-wide association study (GWAS) meta- analyses across five ancestries--European (EUR), African/African American (AFR), Admixed American (AMR), South Asian (SAS), and East Asian (EAS)--including 360,000 T2D cases and 1{middle dot}8 million controls (41% non-EUR). We constructed ancestry-specific and multi-ancestry PRSs in training datasets including 11,000 T2D cases and 32,000 controls, and validated their performance in independent datasets including 39,000 T2D cases and 126,000 controls of diverse ancestries. In the All of Us Research Program, we compared these PRSs to those from the Polygenic Score Catalog and assessed their ability to predict diabetes micro- and macrovascular complications. FindingsAncestry-specific PRSs showed limited prediction power for T2D in AFR, AMR, and SAS compared to EUR and EAS. In contrast, multi-ancestry PRSs, built using GWAS data from five ancestries, substantially improved T2D prediction across all ancestries. Compared to those in the interquartile range, individuals at the 97{middle dot}5th percentile of their PRSs had a 6-fold increased T2D risk in AMR, EAS, and EUR, and [≥]3-fold in AFR and SAS. These PRSs were also associated with the development of microvascular complications and outperformed all previously reported PRSs for all ancestries. InterpretationWe developed and extensively validated the most up-to-date T2D PRSs across diverse ancestry groups. These PRSs are publicly available to support further evaluation of their clinical utility in diverse ancestries.