Predicting Late Age-Related Macular Degeneration Development: Clinical Performance With Multimodal Imaging and Comparison to a Prediction Model

Abstract

Purpose : The ability to accurately predict which individuals progress to late age-related macular degeneration (AMD) is a critical and challenging task. This study sought to examine how well this is performed by clinicians, the value-add of multimodal imaging (MMI), and how this compares with a prediction model using conventional AMD risk factors, as clinical performance is often used as a benchmark for evaluating new predictive models.

Methods : Four retinal specialists independently graded a baseline dataset consisting of 280 eyes from 140 participants with bilateral large drusen, without late AMD. These participants underwent MMI, including optical coherence tomography (OCT), fundus autofluorescence, near-infrared reflectance and color fundus photographs (CFPs) at baseline, and then at 6-monthly intervals for 3-years to determine late AMD development. The retinal specialists assessed the likelihood at baseline that each eye would progress to late AMD over 3 years with CFPs only, and then again with MMI. Their predictive performance (based on the average area under the receiver operating characteristic curve [AUC]) with CFPs only or with MMI were compared to each other, and to a basic prediction model using age, presence of pigmentary abnormalities, and OCT-based drusen volume.

Results : The clinical performance for predicting late AMD using CFPs alone (AUC = 0.75; 95% confidence interval [CI] = 0.68 to 0.82) improved when using MMI (AUC = 0.79; 95% CI = 0.72 to 0.85; P = 0.034). However, the clinical performance was lower when compared to a basic prediction model (AUC = 0.85; 95% CI = 0.78 to 91; P = 0.001 for CFPs alone and P = 0.002 for MMI).

Conclusions : Clinical specialist prediction of late AMD development was improved by using MMI compared to using CFPs alone. However, a basic prediction model using conventional AMD risk factors was better than the clinical predictions. Hence, a basic prediction model may be a better benchmark than clinicians when evaluating new predictive models to determine if they provide additional prognostic information compared to conventional AMD risk factors.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.