Proposal of a Simpler Eye-Level Risk Model Incorporating Reticular Pseudodrusen for the Clinical Prediction of Late Age-Related Macular Degeneration.

Abstract

Background: The updated simplified AREDS risk model predicts progression to late age-related macular degeneration (AMD) by person, describing up to nine observations across both eyes and 10 annual risk scores (0-4, with/without reticular pseudodrusen [RPD]). This study proposes an abridged model to enable inter-eye comparisons and potentially enhance clinical efficiency.

Methods: This retrospective cohort study included 269 participants with early/intermediate AMD over 7 years. The full, person-level updated simplified AREDS risk model was compared to eye-level candidate risk models, derived by removing the least predictive biomarkers. The main outcomes were prognostic performance (AUC) and risk score separability (χ²).

Results: At 1-3 years, the full model showed prognostic performance (AUC ± SE) up to 84.52% ± 5.93%, with overlap between most risk scores (χ² ≤ 2.08). Removing large drusen and pigmentary abnormalities in the fellow eye, intermediate drusen in both eyes, and redefining RPD presence as eye-specific maintained prognostic performance (up to 84.71% ± 4.72%). Assigning one point per retained biomarker, based on similar adjusted risks, improved risk score separability (χ² ≥ 3.85, p < 0.05) while reducing the number of annual scores from 10 to five.

Conclusions: The updated simplified AREDS risk model can be essentially halved without compromising prognostic performance by deriving eye-specific biomarkers and assigning one point per biomarker (large drusen, pigmentary abnormalities, and RPD in the primary eye, and late AMD in the fellow eye). This eye-level risk stratification may improve clinical efficiency and inter-eye study designs when one eye is of particular interest. An example of 3-year risks (scores 0-4) was ≈4%, 8%, 16%, 32%, and 64%.