Predictive Performance of an Updated Polygenic Risk Score for Age-Related Macular Degeneration

Abstract

ObjectiveA recent genome-wide association study of age-related macular degeneration (AMD) has identified new AMD-associated risk variants. These variants can now be incorporated into an updated polygenic risk score (PRS). This study aimed to assess the performance of an updated polygenic risk score (PRS2023) in an independent cohort of older individuals with retinal imaging data, and compare performance with an older PRS (PRS2016). DesignCross-sectional study. ParticipantsA total of 4,175 participants of European ancestry aged ≥70 years with genotype and retinal imaging data from the ASPirin in Reducing Events in the Elderly trial. MethodsWe used logistic regression models and area under the curve (AUC) to assess the performance of the PRS2023 compared with PRS2016. AMD status and severity were graded using colour fundus photography. Main Outcome MeasuresAssociation of PRS2023 and PRS2016 with AMD risk at baseline. ResultsAt enrolment among 4,175 participants, 2,605 (62.4%) had no AMD, 853 (20.4%), 671 (16.1%) and 46 (1.1%) had early, intermediate and late-stage AMD, respectively. More than 27% of the participants with a high PRS2023 (top quartile) had intermediate or late-stage AMD, compared with <15% for those in the middle two quartiles and <13% for those in the lowest quartile. Both PRS2023 and PRS2016 were significantly associated with AMD after adjustment for age, sex, smoking status and lipid levels, with increasing odds ratios (OR) for worsening AMD grades. The PRS2023 outperformed PRS2016 (DeLong test comparing AUC, P=0.03 for all AMD, P=0.03 for late AMD). PRS2023 was associated with late-stage AMD with an adjusted OR of 5.05 [3.41-7.47] per standard deviation. Assuming a population prevalence of 10% for late-stage AMD, the PRS2023 explained 42.2% of disease liability. The AUC of a model containing conventional/non-genetic risk factors and the PRS2023 was 91% [87%-95%] for predicting late-stage AMD, which improved 12% over the model without the PRS (AUC=79%, P<0.001 for difference). ConclusionsA new PRS (PRS2023) for AMD outperforms a previous PRS, and predicts increasing risk for late-stage AMD (with stronger association for more severe imaging-confirmed AMD grades). Our findings have clinical implications for the improved prediction and risk stratification of AMD.