gms | German Medical Science

Background: Several classifications systems for age-related macular degeneration (AMD) have been described in the literature. Although these systems all stratify disease according to severity, it is unclear how well they predict advanced disease. We compared various well-known AMD classification systems and their capability to predict occurrence of late AMD.

Methods: Color fundus photographs of 9,990 participants from the Rotterdam Study, with up to five follow-up visits during 25 years, were graded by two experienced graders. AMD lesions were graded according to the international classification system. Baseline gradings of drusen type, size and area, hyperpigmentation and RPE-degeneration were used to score patients according to six AMD classifications systems. Late AMD was considered present when central geographic atrophy or choroidal neovascularization was observed. Incidence rates were calculated for late AMD per classification and subclass, and stratified for age groups <75 and >75 years at baseline. Overall prediction of late AMD was determined by the area under the operating characteristic curve (AUC).

Results: 183 cases of incident late AMD were observed. Incidence rates increased with a higher subclass, independent of classification system. The AREDS 9-step scale score (2005) at subclass 8 and the 3-Continent harmonization classification (2014) at subclass 4 showed the highest incidence rates for patients <75 years at 30 cases/1000py. The Rotterdam classification (1993) at subclass 3 showed the highest rate for patients >75 years at 65 cases/1000py. Overall, the AREDS 9-step scale had the highest AUC at 0.79, while the 3-Continent harmonization classification showed the best linear trend for prediction of late AMD.

Conclusion: This study compared frequently used classification systems in their prediction of late AMD. Systems differ in their advantages and disadvantages, which may drive their use. The findings will help researchers and clinicians select appropriate classification systems for applications such as clinical trials and deep learning algorithms.