Purchase this article with an account.
Johanna M. Seddon, Robyn Reynolds, Kimberly Chin, Yi Yu, Bernard Rosner; Development and Validation of Prediction Models for Advanced Macular Degeneration: 2006-2011. Invest. Ophthalmol. Vis. Sci. 2012;53(14):3321.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Age-related macular degeneration (AMD) is a complex disease. Smoking, higher body mass index (BMI), and unhealthy diet and lifestyle increase risk. Genes also influence susceptibility. We developed several prediction models incorporating these variables: 5 loci in 3 gene regions,1 genetic and environmental factors,2 genes and environment in prospective models,3 plasma complement markers,4 6 genetic variants,5 and baseline macular drusen characteristics, genes and environment in time varying analyses to predict progression.6 To further evaluate these derived models, we performed validation analyses in another large cohort of patients.
Models were derived from 2,914 individuals in AREDS as previously described.2,3,5,6 An independent longitudinal sample of 1,242 patients with no AMD, early or intermediate AMD, or unilateral advanced AMD at baseline was selected from ongoing genetic-epidemiologic studies of AMD for the validation sample. The average follow-up time in the validation sample was 6 years with 731 individuals followed for 5 years and 475 followed for 10 years. Variables in the models were age, sex, education, smoking, BMI, 6 genetic variants related to AMD, macular drusen phenotypes, and AMD grades for both eyes at baseline.
Discrimination of the models was excellent in the validation subset with C statistics of 0.765 and 0.812 for 5 year and 10 year progression, respectively. The model is useful for ranking and identifying high risk subjects within an external population. Knowledge of genetic susceptibility factors can distinguish low, medium and high risk groups for specific macular phenotypes.
The derived models were validated and performed well in an independent sample of patients with the same range of macular pathology. Our algorithms could be useful for identifying high risk individuals for clinical trials.
This PDF is available to Subscribers Only