New AMD Severity Scores Separate Dry and Wet Disease

Key Takeaways

• Separate AMD-D and AMD-W scores were derived from AREDS and AREDS2 features, rescaled to 0 to 10, and were moderately correlated in Eye ACT.
• AMD-D increased with age among untreated eyes, while AMD-W generally increased before initial anti-VEGF treatment in treated eyes.
• The approach remained usable during routine encounters with frequent missing features and may support retrospective analysis and longitudinal monitoring.

A retrospective cohort study with external validation introduced separate AMD-D and AMD-W severity measures on continuous 0 to 10 scales. Developed from AREDS and AREDS2 data using confirmatory factor analysis, the models were later applied to routine-care EMR data. In Eye ACT, the two scores showed a moderate positive association, with a Pearson correlation of 0.702 and a 95% confidence interval of 0.699 to 0.704. The scoring system provided an EMR-usable way to follow dry and wet AMD severity as distinct processes.

AMD-D drew on drusen characteristics, geographic atrophy, and pigmentary changes, whereas AMD-W used subretinal fluid, subretinal hemorrhage, scars, pigmentary changes, and non-drusenoid pigment epithelial detachment. The models were developed with confirmatory factor analysis and item response theory, and a held-out AREDS/AREDS2 test set showed general alignment with standard AREDS dry scoring, with a Pearson correlation of 0.877. After rescaling, AMD-D produced 16 possible score values and AMD-W produced 32, reflecting finer gradations than traditional categorical systems. Model fit was strong for both constructs, with CFI and RMSEA values of 0.99 and 0.021 for AMD-D and 0.97 and 0.019 for AMD-W. These parallel models were designed to represent different dimensions of AMD severity rather than a shared continuum.

In the Eye ACT cohort, ophthalmic data were extracted from routine-care EMR records with natural language processing-based text mining, covering 8,074 eyes from 4,037 participants and 151,096 eye-visit records. Among 4,412 untreated eyes from 2,248 participants, AMD-D increased slightly before age 80 and then rose more steeply through age 90. Among 220 eyes from 171 participants treated with anti-VEGF, AMD-W generally increased during the weeks or months before first treatment. In the time-to-treatment analysis, 52% had anti-VEGF on the same day as the first ICD-code diagnosis of exudative AMD, more than 70% within 7 days, and more than 80% within 35 days. The EMR cohort showed distinct longitudinal patterns for non-exudative and exudative disease states.

Routine-care missingness shaped the feature set because some findings were inconsistently documented, including hypo versus hyperpigmentation, pigment epithelial detachment subtype, and the distinction between soft and hard drusen as well as soft drusen size. Soft drusen were reduced to a present-or-absent variable, and score generation depended on a validated text-mining algorithm applied to records extending from 2005 through October 31, 2022. The work was described as a proof-of-concept analysis in a large clinical EMR database, where standard AREDS scoring is often impractical because required features are absent. Median follow-up in Eye ACT was 10.1 years, giving the cohort substantial longitudinal depth for retrospective observation. The authors concluded that the scoring system may support retrospective analysis and longitudinal monitoring in routine clinical data.