Development of retinal biomarkers for Alzheimer's disease

Abstract

AbstractBackgroundThe retina has received increasing intention as possible non‐invasive diagnostic biomarker for (preclinical) Alzheimer’s disease (AD). Several studies suggest changes in retinal layer thickness or vasculature in patients with AD. Here, we aim to discriminate (preclinical) Alzheimer’s disease from controls using three retinal imaging modalities in a multimodal prediction model.MethodWe used images of 103 controls, 26 clinical AD cases, and 13 preclinical AD cases. Control subjects were cognitively healthy with a negative amyloid‐PET. Preclinical AD participants were defined as cognitively healthy subjects with a positive amyloid‐PET. The 26 clinical AD participants were recruited from the Amsterdam Dementia Cohort and fulfilled the NIA‐AA criteria with evidence of amyloid pathology based on amyloid‐PET or cerebrospinal fluid. Participants underwent three imaging modalities (Optical Coherence Tomography, Optical Coherence Tomography Angiography, fundus photography) and ophthalmological examinations to exclude ophthalmological comorbidities that could influence retinal parameters. First, we performed a multivariable logistic regression analyses per imaging modality, adjusted for age and sex, using all parameters provided by each device. Second, we created an explorative multimodal prediction model including the best performing variable per imaging modality, using the following parameters: retinal nerve fiber layer inner ring, macular vessel density outer ring and the central retinal vein equivalent (CRVE). We used Receiver Operating Characteristic (ROC) curves to test the overall predictive accuracy.ResultThe final explorative multimodal prediction model showed an acceptable predictive accuracy in discriminating AD from controls with an area under the receiving operating characteristic curve (AUC) of 0.987, and less so in the discrimination between preclinical AD and controls with an AUC of 0.775. However, with differences in the way the coefficients of identical parameters contributed to make the differentiation between AD and controls and preclinical AD and controls.ConclusionDespite of a relatively small sample size, our multimodal retinal imaging prediction model seems promising in discriminating between AD versus controls, and preclinical AD versus controls. We found that two parameters, the macular vessel density outer ring and CRVE behaved differently in AD cases compared to preclinical AD cases. More prospective studies with larger sample sizes are needed to confirm these findings.