Xue et al. (2025) Risk score roulette: A cautionary tale of polygenic risk score reliability
Given the strong interest in using genome-wide association study results to infer individual-level risk of Alzheimer’s disease, we assessed the performance of several polygenic risk score approaches. While each approach was significantly associated with dementia risk with similar performance in an multi-ancestry sample as a whole, we found the inter-model reliability of individual-level risk assessment was poor, even at the extremes. This raises concerns for applying these polygenic risk scores, as fewer than half the individuals in the top/bottom 10% for one approach were also in the top/bottom 10% in the next.