Purchase this article with an account.
Bernard Rosner, Gui-Shuang Ying, Robert Glynn, Maureen G Maguire; Comparative study of different methods for analyzing correlated continuous data. Invest. Ophthalmol. Vis. Sci. 2016;57(12):3694.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To introduce and demonstrate the appropriate linear regression methods for analyzing correlated continuous eye data.
We used random effects models and population average models using the Generalized Estimating Equations (GEE) approach, under various covariance structures, to account for inter-eye correlation. We demonstrated using SAS software the application of these linear regression models to analyze continuous eye data from two real clinical studies, one for comparing the baseline refractive error between study eyes with choroid neovascularization (CNV) and fellow eye without CNV among 355 subjects in the Comparison of Age-related Macular Degeneration Treatments Trials (CATT), another for determining factors associated with visual field among the elderly in a cross-sectional study of 197 patients age 65 or older who were seen at the Glaucoma Service at the Massachusetts Eye and Ear Infirmary.
For the 1st example (see Table 1), in the standard multivariable linear regression analysis (adjusted for age, gender, smoking status; presence of geographic atrophy, and glaucoma), mean refractive error difference between the study eyes with CNV and their fellow eyes without CNV is 0.15 diopters (95% confidence interval (CI): -0.03 to 0.32, p=0.10). Using random effects models or GEE, the difference was the same but with narrower 95% CI (0.01 to 0.28) and statistically significant (p=0.03).In the 2nd example (see Table 2), the standard regression model for visual field data without accounting for inter-eye correlation provided biased estimates of standard error (generally underestimated), and smaller p-values than those from random effects models and GEE (e.g., p-value for age is 0.008 from the standard regression model, and 0.02 from mixed effects models or GEE).
In vision research, ignoring the inter-eye correlation can lead to invalid inference. The random effects or GEE models should be used to appropriately account for the inter-eye correlation.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.
The comparison of baseline refractive error between study eyes with active CNV and fellow eyes without CNV
Comparison of results from various regression approaches for visual field data
This PDF is available to Subscribers Only