Purpose:
To compare the performance of 3 predictive models for the development of POAG that included baseline age and baseline eye-specific variables that were calculated using the: 1) means of both eyes, 2) means of both eyes plus their asymmetries, 3) the worse eye.
Methods:
Sample: Participants with complete baseline data in both eyes in the OHTS observation group (n=717) and EGPS placebo group (n=324). Analysis: Cox proportional hazards models which included baseline age and baseline eye-specific variables calculated as follows:1. Means model: age, the means of right and left eyes for IOP, CCT, vertical C/D, and PSD2. Means plus asymmetry model: age, the means of right and left eyes for IOP, CCT, vertical C/D, PSD plus their absolute differences between eyes.3. Worse eye model: age, IOP, CCT, vertical C/D and PSD for the worse eye as defined by a weighted multivariate risk score calculated for each eye. When the score was equal in both eyes, one eye was selected randomly.We calculated the C-statistic for discrimination (50 random to 1.00 perfect agreement) and the calibration chi-square for the distribution of predictive accuracy (0.00 best fit to >20.00 worse fit).
Results:
Baseline age, CCT, IOP, vertical C/D and PSD were statistically significant predictors in all 3 models. Asymmetries in IOP and vertical C/D were statistically significant when added to the model that included the means of both eyes. The median absolute difference between eyes in IOP and vertical C/D was 1mmHg and 0.05 respectively. The correlation between risk estimates among the 3 models ranged from 0.95 to 0.98.
Conclusions:
Differences in the performance of the 3 models were small. The usefulness of the asymmetry model and worse eye model is limited by measurement variability. The prediction model using the means of right and left eyes is the most robust to measurement variability.
Clinical Trial:
www.clinicaltrials.gov NCT00000125
Keywords: clinical (human) or epidemiologic studies: risk factor assessment • clinical (human) or epidemiologic studies: biostatistics/epidemiology methodology • visual fields