Polygenic risk and hazard scores for Alzheimer's disease prediction

• Published in: Annals of clinical and translational neurology 2019-03, Vol.6 (3), p.456-465
• Main Authors: Leonenko, Ganna, Sims, Rebecca, Shoai, Maryam, Frizzati, Aura, Bossù, Paola, Spalletta, Gianfranco, Fox, Nick C., Williams, Julie, Hardy, John, Escott‐Price, Valentina
• Format: Article
• Language: English
• Subjects: Age; Alzheimer Disease - epidemiology; Alzheimer Disease - genetics; Alzheimer's disease; Apolipoproteins E - genetics; Case-Control Studies; Female; Genetic Predisposition to Disease - genetics; Genome-Wide Association Study; Genomes; Health risk assessment; Human health and pathology; Humans; Life Sciences; Male; Multifactorial Inheritance; Polymorphism, Single Nucleotide; Proportional Hazards Models; Risk Factors
• ISSN: 2328-9503; 2328-9503
• DOI: 10.1002/acn3.716

Objective Genome‐wide association studies (GWAS) have identified over 30 susceptibility loci associated with Alzheimer's disease (AD). Using AD GWAS data from the International Genomics of Alzheimer's Project (IGAP), Polygenic Risk Score (PRS) was successfully applied to predict life time risk of AD development. A recently introduced Polygenic Hazard Score (PHS) is able to quantify individuals with age‐specific genetic risk for AD. The aim of this study was to quantify the age‐specific genetic risk for AD with PRS and compare the results generated by PRS with those from PHS. Methods Quantification of individual differences in age‐specific genetic risk for AD identified by the PRS, was performed with Cox Regression on 9903 (2626 cases and 7277 controls) individuals from the Genetic and Environmental Risk in Alzheimer's Disease consortium (GERAD). Polygenic Hazard Scores were generated for the same individuals. The age‐specific genetic risk for AD identified by the PRS was compared with that generated by the PHS. This was repeated using varying SNPs P‐value thresholds for disease association. Results Polygenic Risk Score significantly predicted the risk associated with age at AD onset when SNPs were preselected for association to AD at P ≤ 0.001. The strongest effect (B = 0.28, SE = 0.04, P = 2.5 × 10−12) was observed for PRS based upon genome‐wide significant SNPs (P ≤ 5 × 10−8). The strength of association was weaker with less stringent SNP selection thresholds. Interpretation Both PRS and PHS can be used to predict an age‐specific risk for developing AD. The PHS approach uses SNP effect sizes derived with the Cox Proportional Hazard Regression model. When SNPs were selected based upon AD GWAS case/control P ≤ 10−3, we found no advantage of using SNP effects sizes calculated with the Cox Proportional Hazard Regression model in our study. When SNPs are selected for association with AD risk at P > 10−3, the age‐specific risk prediction results are not significant for either PRS or PHS. However PHS could be more advantageous than PRS of age specific AD risk predictions when SNPs are prioritized for association with AD age at onset (i.e., powerful Cox Regression GWAS study).