Abstract
Purpose :
The e-ROP study demonstrated safety and efficacy of a telemedicine system for detection of referral-warranted retinopathy of prematurity (RW-ROP, defined as Zone 1 disease, Stage 3 ROP, or plus disease). As these systems may rely on non-physician image readers, identification of image characteristics associated with severe disease could enhance its functionality. This study evaluated the association of type and location of intraocular hemorrhage with ROP.
Methods :
All image sets from eyes ever noted to have hemorrhage either on examination by a study-certified ophthalmologist or on image grading were selected for review. Trained readers graded type (dot, blot, flame, preretinal, or vitreous) and location (Zone 1 and/or Zone 2) of hemorrhage. Cross-sectional associations of hemorrhage with any ROP, stage 3 ROP, and RW-ROP were evaluated by odds ratio (OR) using multivariate logistic regression models adjusted for birth weight (BW) and gestational age (GA). Sensitivity and specificity of using intraocular hemorrhage for detecting RW-ROP were also calculated.
Results :
Hemorrhages were present in 11% (867/7905) of e-ROP image sets, with blot (5.1%) and preretinal (5.0%) the most common. Presence of any intraocular hemorrhage was associated with a 2.46x likelihood of having ROP, increasing to 3.80x for non-dot and 12.6x for preretinal hemorrhage. For any intraocular, non-dot, or pre-retinal hemorrhage, ORs for Stage 3 disease were 2.88, 3.28, and 4.40, and for RW-ROP 3.19, 3.70, and 4.85 respectively. Any hemorrhage in both Zone 1 and Zone 2, as well as two or more types of hemorrhages were associated with higher likelihood of RW-ROP (ORs 6.85 and 5.24 respectively). Adding intraocular hemorrhage to RW-ROP features noted in an image changed sensitivity for detecting RW-ROP from 94% to 95%.
Conclusions :
Although intraocular hemorrhage is strongly associated with the presence and severity of ROP, addition of intraocular hemorrhage to the interpretation algorithm did not enhance the overall performance of the system in detecting RW-ROP on image grading.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.