Abstract
PURPOSE: To develop a data-driven severity scoring system for age-related macular degeneration (AMD) that can generate distinct scores for exudative AMD (AMD-W) and nonexudative AMD (AMD-D) and that can be applied to routine clinical features captured in electronic medical record (EMR) data, in which complete data on features included in traditional scoring systems are rare. DESIGN: Retrospective cohort study with external validation. PARTICIPANTS: Data from participants in the Age-Related Eye Disease Study (AREDS), AREDS2, and the Eye Adult Changes in Thought (Eye ACT) study. METHODS: Severity score models were developed for non-exudative ("dry") AMD (AMD-D) and exudative ("wet") AMD (AMD_W) based on confirmatory factor analysis (CFA) of data from AREDS and AREDS2. Models were applied to an independent cohort of the Eye ACT study whose longitudinal ophthalmic clinical data were extracted from an EMR capturing routine care, using natural language processing-based text mining algorithms. MAIN OUTCOME MEASURES: Trajectories of AMD-D and AMD-W scores in the Eye ACT cohort and relationship with age and the onset of the first anti-VEGF treatment. RESULTS: In the Eye ACT cohort, AMD-D and AMD-W scores showed a moderately positive correlation (Pearson 0.702, 95% confidence interval, 0.699-0.704). In 4412 eyes from 2248 participants in Eye ACT, which never received anti-VEGF, AMD-D scores increased slightly before the age of 80 years, followed by a steeper increase through the age of 90 years. In 220 eyes of 171 Eye ACT participants, which received anti-VEGF, most showed a pattern of gradually increasing AMD-W scores in the weeks or months before the anti-VEGF treatment. CONCLUSIONS: The CFA-based scoring system enabled detailed assessments of both non-exudative and exudative severity using features collected in a routine clinical setting with ubiquitous missing data, in which standard AREDS scoring is not possible. FINANCIAL DISCLOSURE(S): Proprietary or commercial disclosure may be found after the references.