Abstract
Purpose :
Age-related macular degeneration (AMD) is characterized by complex interactions between genetic and environmental factors. The aim of this study is to develop a prediction algorithm for AMD risk assessment based on individual genetic profile and environmental factors.
Methods :
DNA samples were genotyped using the AutoGenomics INFINITI platform. AMD panel included ABCA1(rs1883025); APOE(rs429358, rs7412); ARMS2/HTRA1(rs10490924, EU427539, rs11200638); C3(rs2230199); CCDC109B(rs17440077); CETP(rs3764261); CFB(rs4151669, rs522162); CFH(rs1048663, rs1061170, rs10737680, rs1329428, rs2274700, rs3766405, rs412852, rs800292); CFI(rs10033900); COL8A1(rs13095226); LIPC(rs493258, rs10468017); LPL(rs12678919); TIMP3(rs9621532) and VEGFA(rs3025000, rs943080). Class prediction model building and testing were performed using TreeNet software.
Results :
We genotyped 30 SNPs from 1254 AMD patients (382 Intermediate AMD, 269 GA and 603 CNV) and 423 normal control patients. For AMD risk assessment, we tested if TreeNet can be applied to genotyping data. The known published odds ratios for 30 SNPs from the panel were used as numeric variables in the model building for 872 AMD patients and 423 controls. Receiver operating characteristic curve (ROC) for both models were highly comparable (ROC=0.88 for numeric and ROC=0.94 for categorical) suggesting that TreeNet can be used for descriptive genotyping data. Next, we analyzed the performance of the prediction model for AMD progression risk assessment. The ROC for Normal Control vs Intermediate AMD Stage was 74% sensitivity and 82% specificity. The ROC for Intermediate AMD vs GA was 82% sensitivity and 77% specificity. The ROC for Intermediate AMD vs CNV was 91% sensitivity and 58% specificity. The variable ranking showed the most important SNPs associated with AMD at the top of the list including rs412852 (CFH), rs2230199 (C3) and rs10490924 (ARMS2/HTRA1) which is in accordance with previously published data.
Conclusions :
These results suggest that the prediction algorithm can be used for the AMD risk assessment. It can also incorporate environmental factors such as age, smoking and other variables into the model building. Having a larger group of patients with well characterized AMD phenotypes, the algorithm could also be applied to predict disease progression and therapeutic response, and aid personalized precision medicine.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.