Purpose:
Most prediction models for age-related macular degeneration (AMD) are based on case-control studies. These studies, however, have a tendency to over-estimate risks. The aim of this study is to develop a prediction model for Late AMD in a population-based study.
Methods:
In the population-based Rotterdam Study, participants aged 55+ yrs underwent 5 identical examinations during 1990-2010 including phenotyping for AMD on fundus photographs and genotyping of all known AMD risk variants using Illumina Infinium II HumanHap550chip v3.0® array, Hapmap imputation, and Taqman SNP genotyping assay. Genetic, phenotypic, demographic and environmental variables were tested as predictors for Late AMD in a univariate and multivariate analysis using Cox proportional hazard ratio (HR). Predictive values were estimated by reciever operating characteristic (ROC) curves.
Results:
3991 participants free of Late AMD at baseline with follow-up data (mean 9.96 yrs; sd 5.9 yrs) were successfully genotyped. Of these, 132 developed Late AMD; 2851 participants remained free of any AMD. Significant predictors for Late AMD were: age, early stage of AMD at baseline, presence of large drusen (> 250 µm), large drusen area (> 650 µm), and genetic risk variants in the CFH gene (rs1061170, rs12144939, rs800292), CETP (rs3764261), CB (rs641153) and HTRA1 (rs3793917). Predictive values for Late AMD varied from 0.793 for risk genotypes only, 0.829 for early AMD phenotype only, and 0.870 for all genetic and phenotypic variables combined.
Conclusions:
Progression to Late AMD is highly predictable, also in persons from the general population. Predictive values are highest when genetic and phenotypic risk variables are combined.
Keywords: age-related macular degeneration • genetics • clinical (human) or epidemiologic studies: biostatistics/epidemiology methodology