8
Freeman, Cataract Surgery and Age-Related Macular Degeneration
because cataract surgery and AMD status may differ by eye. The standard errors were adjusted
using Generalized Estimating Equations to account for the correlation between eyes.
18
Age-adjusted analyses were done using logistic regression to determine the association of
cataract surgery and AMD for all eyes. Next, multiple logistic regression was used to determine
the association of cataract surgery and AMD adjusting for the following other covariates besides
age: race (SEE and BES only), gender, smoking (never, past, current smoking), and study
(combined analyses only). In the combined analyses, race was coded as Caucasian versus non-
Caucasian so as to avoid colinearity between Hispanic race and study population, as all the
PVER participants were Hispanic.
A fixed effects model that assumed no heterogeneity in the true odds ratios across studies
was assumed. With only three studies, it was not possible to reliably estimate the across-study
component of variance for the log odds ratios as would be done in a random-effects model
19
.
However, the combined odds ratios under different assumptions about the among-study variance
were estimated. The moment estimate of the among-study variance was 0.0 because the variance
among the three estimated log odds ratios was smaller than the average statistical variance across
the studies. Hence, the best indication available from the data was that the fixed effects model
was appropriate.
Interactions were assessed by stratification and by the addition of interaction terms into
the regression models. All analyses were done using the SAS statistical software (SAS Institute,
Cary, NC).
RESULTS
Descriptive Statistics
