Abstract
Purpose :
Analysis of longitudinal ophthalmologic data using the eye as the unit of analysis has to account for both inter-eye and longitudinal correlation. The purpose of this research is to compare several methods for analyzing this type of data with a continuous outcome measure.
Methods :
We describe fixed effects, mixed effects and GEE models to assess associations between continuous eye-specific longitudinal outcome measures vs. person and eye-specific covariates. These methods are applied to data from the Complications of Age-Related Macular Degenerative Prevention Trial (CAPT), which enrolled 1052 subjects with age ≥ 50, visual acuity (VA) 20/40 or better, and at least 10 large drusen in each eye. One eye was randomized to laser treatment and the fellow eye to no treatment. The outcome was ETDRS VA (letters) which was assessed at baseline and years 1, 2, 3, 4 and 5. The aims of this analysis are to compare VA loss between treated and control eyes over a 5-year period and between current and non-current cigarette smokers.
Results :
The inter-eye and longitudinal correlations of VA measures are given in Table 1. The inter-eye correlations range from 0.33 to 0.53 and were highest at baseline. The longitudinal correlations (0.31 to 0.88) diminish with a longer time between repeated measures and were higher at later time points.
Results from fitting the 3 longitudinal models are given in Table 2. There was a non-monotone effect of treatment with a small but significant effect of treatment on change in VA at 24 months for all 3 models (fixed effects: B = 0.94 +- 0.38, p = 0.014; mixed effects: B = 0.95 +- 0.38, p = 0.013; GEE: B = 1.02 +- 0.39, p = 0.008), but not at 60 months. Unexpectedly, current cigarette smokers had better VA at baseline (p < 0.05), but declined significantly more than non-current smokers at 60 months (fixed effects: B = -4.09 +- 1.91, p = 0.033; mixed effects: B = -4.13 +- 1.96, p =0.036; GEE: B = -4.29 +- 2.00, p = 0.032). Model fit was significantly better with the mixed effects vs. the fixed effects model: change(-2 ln L) = 209.6, chi-square 33 df, p < 0.001, although coefficient estimates and SE’s were similar.
Conclusions :
Longitudinal models using the eye as the unit of analysis can be implemented using available statistical software and offer the advantage of efficiently analyzing longitudinal ophthalmologic data with both inter-eye and longitudinal correlation.
This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.