Abstract
Polygenic indices (PGIs) are increasingly used to identify individuals at risk of developing disease and are advocated as screening tools for personalized medicine and education. Here we empirically assess rank concordance between PGIs created with different construction methods and discovery samples, focusing on cardiovascular disease and educational attainment. We find Spearman rank correlations between 0.17 and 0.93 for cardiovascular disease, and 0.40 and 0.83 for educational attainment, indicating highly unstable rankings across different PGIs for the same trait. Potential consequences for personalized medicine and gene-environment (G × E) interplay are illustrated using data from the UK Biobank. Simulations show how rank discordance mainly derives from a limited discovery sample size and reveal a tight link between the explained variance of a PGI and its ranking precision. We conclude that PGI-based ranking is highly dependent on PGI choice, such that current PGIs do not have the desired precision to be used routinely for personalized intervention.