Optical Coherence Tomography Predictors of 5-Year Progression to Non-Neovascular Late Age-Related Macular Degeneration

Abstract

Purpose : In the prospective AREDS2 Ancillary (A2A) SDOCT study, we found several Spectral Domain Optical Coherence Tomography (SDOCT) features preceded the appearance of geographic atrophy (GA) on color fundus photography (CFP). From this study, we aimed to build SDOCT-based models to predict 5-year new onset of GA and central GA (CGA) on CFP.

Methods : For 317 intermediate AMD eyes of 317 participants, qualitative and quantitative SDOCT variables were derived, respectively, from standardized grading and semi-automated segmentation at baseline. Up to 7 years later, CFP was assessed annually for the two outcomes: new onset of GA and CGA. For each outcome, we used repeated measures logistic regression to explore univariate models, stepwise fit a multivariate model, and build a risk calculator.

Results : Over a follow-up average of 4.1±1.3 years, while 25 of the 317 eyes had incomplete outcome data, CGA developed in 46 (15.75%) of 292 eyes with intermediate AMD, and GA in 70 (26.42%) of the subset of 265 eyes without any GA at baseline. Each final multivariate model was adjusted for age and gender [see table]. In the GA model, the strongest OCT predictors were retinal pigment epithelium (RPE) loss and hyperreflective foci (HF), followed by photoreceptor layer (PRL) thinning, OCT-reflective drusen substructures (ODS), intraretinal fluid (IRF) and the quantitative RPE-drusen complex abnormal thinning (RAT) volume. For CGA, the strongest predictors were HF, IRF and RPE loss, followed by the volumes of RAT and neurosensory retina (NSR). Logistic regression formulas from the models yielded a calculator that computes the probabilities of new onset of GA and of CGA on CFP after 1 through 5 years.

Conclusions : Based on SDOCT segmentation, drusen characteristics, and retinal pathology in intermediate AMD, we built a novel model to calculate the risk of new onset GA and CGA up to 5 years. This calculator has a promising role as a clinical prognostic tool in the future, as a possible alternative to existing scales.

This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.

 

Table: Final multivariate models are summarized: Odds Ratios (OR), 95% Confidence Intervals (CI), and p-values. Baseline distributions are reported as percentage (proportion) for binary variables, median ± IQR for Age and RAT, and mean ± SD for NSR. (Table does not include SDOCT variables non-significant in fitting of either model.)
N/A: variable not significant in the final model.