Abstract
Purpose: :
With the Moorfields Regression Analysis (MRA) large optic discs are commonly classified as borderline or outside normal limits more often than small optic discs. To derive accurate normal limits for the relationship between optic disc size and rim area using quantile regression and to assess the improvement in diagnostic performance obtained with this method.
Methods: :
MRA uses ordinary least squares regression (OLS) to derive lower limits of optic disc rim area from optic disc size. Quantile regression offers advantages over OLS by more accurately modelling these lower limits. Two independent datasets of healthy controls (76 + 88 = 164 eyes) were used to derive the relationship between optic disc size and lower limits of rim area (10th, 5th, 2nd, 1st percentiles). Predictions, performed separately for each dataset as well as in combination, were compared to the current "borderline" and "outside normal" limits of the MRA. Applying the limits derived from quantile regression to 106 + 146 = 252 eyes from patients with glaucoma, we investigated the effect of changed normative limits on diagnostic sensitivity across the spectrum of optic disc sizes in these samples.
Results: :
Compared to the prediction limit from standard OLS, the limits derived from quantile regression were shallower (less specific with smaller optic discs, more specific with larger discs). These findings were similar in both datasets. When limits from one dataset were applied to healthy control data of the other, the false-positive rate did not increase with optic disc size. Applying the quantile regression limits to glaucomatous optic discs increased sample sensitivity to glaucomatous damage in small discs.
Conclusions: :
The distribution of rim area in healthy optic discs broadens with increasing disc size (heteroscedasticity). Normative limits for rim area derived from quantile regression help remove the size-dependent criterion shift observed with MRA. Quantile regression may be an attractive method for establishing better normative limits in a wide range of other measurements in vision, including perimetry.
Keywords: optic disc • imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound) • clinical (human) or epidemiologic studies: biostatistics/epidemiology methodology