Methods and reproducibility of grading optimized digital color fundus photographs in the Age-Related Eye Disease Study 2 (AREDS2 Report Number 2)

Abstract

Purpose: To establish continuity with the grading procedures and outcomes from the historical data of the Age-Related Eye Disease Study (AREDS), color photographic imaging and evaluation procedures for the assessment of age-related macular degeneration (AMD) were modified for digital imaging in the AREDS2. The reproducibility of the grading of index AMD lesion components and for the AREDS severity scale was tested at the AREDS2 reading center.

Methods: Digital color stereoscopic fundus photographs from 4203 AREDS2 subjects collected at baseline and annual follow-up visits were optimized for tonal balance and graded according to a standard protocol slightly modified from AREDS. The reproducibility of digital grading of AREDS2 images was assessed by reproducibility exercises, temporal drift (regrading a subset of baseline annually, n = 88), and contemporaneous masked regrading (ongoing, monthly regrade on 5% of submissions, n = 1335 eyes).

Results: In AREDS2, 91% and 96% of images received replicate grades within two steps of the baseline value on the AREDS severity scale for temporal drift and contemporaneous assessment, respectively (weighted Kappa of 0.73 and 0.76). Historical data for temporal drift in replicate gradings on the AREDS film-based images were 88% within two steps (weighted Kappa = 0.88). There was no difference in AREDS2-AREDS concordance for temporal drift (exact P = 0.57).

Conclusions: Digital color grading has nearly the same reproducibility as historical film grading. There is substantial agreement for testing the predictive utility of the AREDS severity scale in AREDS2 as a clinical trial outcome. (ClinicalTrials.gov number, NCT00345176.)

Keywords: AMD; color photographic imaging; reproducibility.

Figure 1

Screen capture with digital AREDS macular grid and drusen circles on color fundus photograph in IMAGEnet. See Table 1 for area definitions of grading circles.

Figure 2

Bland–Altman plot for reproducibility of area of geographic atrophy (n = 288 eyes). Area measured by two graders using planimetry had a mean difference of 0.02 mm² and Bland–Altman limits of +1.8 and −1.76 mm².