Abstract
Purpose:
Ocular hyperemia is an important efficacy, safety, and tolerability endpoint in ophthalmic clinical trials. Subjective scoring of ocular hyperemia can vary over time and suffers from inter-grader variability. A novel automated approach called Imaging System for Ocular Surface (ISOS) was applied in a double-blind interventional study in allergic conjunctivitis to explore its reproducibility and sensitivity.
Methods:
Twenty three subjects allergic to ragweed pollen were assessed by slit lamp photography after seven days of either 0.1% Maxidex ophthalmic suspension (n=13) or vehicle control (n=10) BID. Photographs were acquired before and after three hours in a ragweed pollen environmental exposure chamber (EEC) with study medication being administered just before entering EEC. Each image was immediately scored by an ophthalmologist and later by three fully-blinded expert graders following the validated bulbar redness scale from 0 to 4 in 0.5 increments. A consensus expert score for each photograph was calculated to minimize the influence of each grader’s intrinsic variability which was estimated via modeling and repeat scoring. Photographs were processed using an automatic vessel segmentation algorithm to quantify 35 morphological descriptors of the conjunctival vasculature which were then input into multivariate linear regression models of live and expert consensus scores. The sensitivity to Maxidex treatment was explored for all hyperemia scores and image descriptors using a linear mixed model of change from pre-EEC.
Results:
Vessel density alone was the best predictor of live hyperemia score (r=0.68) whereas the consensus expert score was best fit by a linear model of fourteen morphological descriptors (r=0.93) including vessel density. The resulting automated hyperemia score was within 0.36 units (95% CI) of its absolute score versus 0.34, 0.64, and 0.42-0.56 for expert consensus, live, and individual expert scores, respectively. Subjects receiving Maxidex experienced a stronger decrease in hyperemia than vehicle control per consensus expert score (p<0.05), automated score (p<0.05), vessel density (p<0.05), number of triple points (p<0.05), and total vessel length (p<0.05).
Conclusions:
ISOS-based hyperemia assessment demonstrated similar robustness as the consensus of three experts and offered improved sensitivity to treatment via the characterization of vessel morphology.