Abstract
Purpose: :
Age-related macular degeneration (AMD) is a late onset disease. Controls for AMD studies are often sampled from an age group that may still develop disease in the future. Misclassification can seriously affect phenotype status accuracy and results of association studies. We propose to use our recently published predictive model (Ophthalmol 2011; vol 118: pages 2203-2211) to predict the risk of each non-advanced-AMD participant to eventually develop advanced AMD and to recode the high risk participants to cases. Using the corrected disease status, we evaluated if the new genetic variants would be more likely detected.
Methods: :
We applied this framework of predictive model based phenotype correction in an independent cohort. At baseline, there were 235 cases with advanced AMD and 992 controls. We first evaluated the AMD association signals of six new genes which were recently identified but not included in our previous predictive models. Then, the advanced AMD progression risk scores were calculated for each baseline control using our predictive model. In addition to the 235 baseline cases, 172 baseline controls who had a risk score >=3 were recoded as cases. We re-evaluated the six new genes using the corrected phenotype of 820 controls and 407 cases.
Results: :
In the baseline case-control association tests, only COL8A1 (rs13095226, P=0.03) was significant, while the other SNPs in CFI (rs10033900, P=0.85), LIPC (rs10468017, P=0.67), CETP (rs3764261, P=0.47), TIMP3 (rs9621532, P=0.27), ABCA1 (rs1883025, P=0.13) were not significant. After the phenotype correction, case-control association signals in CETP (P=0.03) and ABCA1 (P=0.04) became significant, and signals in CFI (P=0.52) and LIPC (P=0.28) were improved comparing to the baseline results. Among the 172 corrected cases, 102 of them developed advanced AMD during follow-up, while 59 of the other 70 had less than 7 years of follow-up. Only 11 of the corrected cases had not developed advanced AMD after 7 years of follow-up. The odds ratios of the six new genes estimated by the tests using corrected phenotype are closer to values reported by large GWAS studies than the values estimated by the baseline phenotype.
Conclusions: :
Correction of phenotype misclassification based on highly informative predictive models may be helpful towards identifying additional genetic and other risk factors, when validated risk factors are already offering strong discriminating ability.
Keywords: age-related macular degeneration • clinical (human) or epidemiologic studies: biostatistics/epidemiology methodology • clinical (human) or epidemiologic studies: risk factor assessment