Abstract
Purpose: :
To determine if ethnicity-specific classification improves the diagnostic performance of the Moorfields Regression Analysis (MRA) in an Afro-Caribbean population, and to compare the outcomes to those obtained with Quantile Regression.
Methods: :
A group of 156 randomly selected residents of St. Kitts (mean age 57 y, range 50 - 87 y) underwent examination with the Heidelberg Retinal Tomograph (HRT2) as part of the St Kitts Eye Study. Based on a full eye examination including dilated fundus examination and threshold perimetry, 17 (11%) of these participants were diagnosed with glaucoma. The HRT data were analysed using the Caucasian (MRA-c) and African (MRA-a) normative limits of the commercial software (HRT viewer 3.1.2.9), as well as with limits derived from two independent Caucasian samples with Quantile Regression Analysis (QRA, Artes and Crabb, IOVS 2009).
Results: :
With specificity fixed at 90%, the sensitivities of the MRA-c (59%), the MRA-a (57%), and the QRA (59%) were similar (p>0.1). There was close agreement between the diagnostic classifications of MRA-c and MRA-a (kappa = 0.90) although the MRA-c was more conservative (more specific but less sensitive) than the MRA-a (Wilcoxon, p<0.001). With both MRA-c and MRA-a, the likelihood of "borderline" or "outside normal limits" classification increased with optic disc size (proportional-odds logistic regression; MRA-c OR = 1.14, 1.06 - 1.23, p < 0.001; MRA-a OR = 1.13, 1.05 - 1.21, p < 0.001, for each 0.1 mm2 change in disc area). When the diagnostic limits of QRA were applied, the specificity of the analysis was independent of optic disc size (p = 0.81).
Conclusions: :
The ethnicity-specific analysis (MRA-African) did not improve the diagnostic performance of the MRA in this sample. The significant disc-size-related bias of both MRA versions (lesser specificity with large optic discs) was removed with Quantile Regression. Because Quantile Regression can accurately model the margins of the rim area distribution throughout the entire spectrum of optic disc sizes, it is unlikely that separate ethnicity-specific limits are required for this analysis, even if disc sizes differ between populations.
Keywords: imaging/image analysis: clinical • imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound)