Abstract
Purpose :
Misclassifying lesion activity in neovascular AMD (nAMD) leads to under or over treatment with anti-VEGF drugs. Under treatment fails to optimise vision. Over treatment increases cost and the risk of complications. In a study of the performance of non-experts classifying lesion activity, we developed a reference classification for a sample of nAMD eyes based on experts’ assessments. Here, we identify lesion features associated with discordant classifications.
Methods :
Three clinician experts assessed colour fundus images and radial pattern spectral domain optical coherence tomograms for nAMD eyes at two time points: ‘baseline’ images showed an inactive lesion and ‘index’ images showed the same lesion at another time when the lesion was still inactive or had reactivated. Experts assessed blood, exudate, sub-retinal fluid (SRF), localised intra-retinal cysts (IRC), diffuse retinal thickening (DRT) and pigment epithelial detachment (PED)), coding these features as present or absent in index images and, if present, increased or not compared to baseline. They then classified the overall lesion as inactive, reactivated or suspicious according to pre-specified rules, so disagreements could only arise from differing judgements about features rather than how features should be interpreted. When classifications disagreed, experts assigned a consensus classification without reference to their original assessments.
Results :
Experts’ overall classifications agreed for 219/288 (76%) eyes; 33/69 eyes (11%) involved ‘major’ disagreements between active versus inactive classifications and 36 (13%) involved ‘minor’ disagreements between suspicious versus inactive (or active) classifications. Agreement for specific features was best for blood (92%), exudates (90%) and SRF (90%) and good for IRC (84%). Expert 2 judged SRF to be present more often, and IRC less often, than experts 1 and 3. Agreement was worse for DRT (66%) and PED (47%) and 13/33 major disagreements (45%) involved discordant judgements about the presence of DRT. After reaching consensus, 142 eyes (49.3%) were classified as reactivated, five (1.7%) as suspicious and 141 (49.0%) as inactive.
Conclusions :
Experts’ classifications of lesion activity agreed for 75% of eyes. Disagreements often arose when activity was indicated by DRT. Clearer definitions of PED and DRT are required before these features can be useful indicators of lesion activity.
This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.