RESUMO
Objective: This study aims to address disparities in risk prediction by evaluating the performance of polygenic risk score (PRS) models using the 90 risk variants across 78 independent loci previously linked to Parkinson's disease (PD) risk across seven diverse ancestry populations. Methods: We conducted a multi-stage study, testing PRS models in predicting PD status across seven different ancestries applying three approaches: 1) PRS adjusted by gender and age; 2) PRS adjusted by gender, age and principal components (PCs); and 3) PRS adjusted by gender, age and percentage of population admixture. These models were built using the largest four population-specific summary statistics of PD risk to date (base data) and individual level data obtained from the Global Parkinson's Genetics Program (target data). We performed power calculations to estimate the minimum sample size required to conduct these analyses. A total of 91 PRS models were developed to investigate cumulative known genetic variation associated with PD risk and age of onset in a global context. Results: We observed marked heterogeneity in risk estimates across non-European ancestries, including East Asians, Central Asians, Latino/Admixed Americans, Africans, African admixed, and Ashkenazi Jewish populations. Risk allele patterns for the 90 risk variants yielded significant differences in directionality, frequency, and magnitude of effect. PRS did not improve in performance when predicting disease status using similar base and target data across multiple ancestries, demonstrating that cumulative PRS models based on current known risk are inherently biased towards European populations. We found that PRS models adjusted by percentage of admixture outperformed models that adjusted for conventional PCs in highly admixed populations. Overall, the clinical utility of our models in individually predicting PD status is limited in concordance with the estimates observed in European populations. Interpretation: This study represents the first comprehensive assessment of how PRS models predict PD risk and age at onset in a multi-ancestry fashion. Given the heterogeneity and distinct genetic architecture of PD across different populations, our assessment emphasizes the need for larger and diverse study cohorts of individual-level target data and well-powered ancestry-specific summary statistics. Our current understanding of PD status unraveled through GWAS in European populations is not generally applicable to other ancestries. Future studies should integrate clinical and *omics level data to enhance the accuracy and predictive power of PRS across diverse populations.