Abstract
Purpose :
1) Explore the feasibility of the Eye911 protocol for the ER and urgent care setting. 2) Assess diagnostic accuracy of protocol and impact of slit lamp (SL) video for diagnosis of ocular pathologies.
Methods :
Thirteen patients presenting to the Memphis Veterans Affairs (VA) ER with ocular complaints were recruited. Following informed consent, the Eye911 protocol was carried out. Eye911 protocol includes initial eye workup (history, vision, IOP) and 2D external photos of the eyes. Then, SL video was captured for both eyes (JedMed). A single undilated fundus photo (Canon) was obtained for each eye. The patient was then seen by the on-call ophthalmologist in the traditional face to face (F2F) eye exam. The study data, images, and videos were uploaded into REDCap and sent to four reading ophthalmologists for interpretation. They were blinded to patient identity, F2F exam, and each other’s interpretations. The readers first reviewed clinical information and external and fundus photos to make an initial diagnosis (pre-SL). They then accessed the SL video and made a subsequent diagnosis (post-SL). Using percent agreement and Cohen’s kappa (κ) statistics, concordance pre and post-SL video between each of the readers and the F2F “gold standard” diagnosis was calculated.
Results :
For reader 1, 2, 3, and 4 respectively, the percent agreement pre-SL between reader and F2F diagnosis was 46.2%, 33.3%, 30.8%, and 46.2%. There was no change in percent agreement post-SL (Table 1 and Table 2). In both the pre-SL and post-SL video period, there was no agreement between reader 1 vs 2 (κ = -0.36) or reader 2 vs 3 (κ = -0.20). There was slight agreement between reader 2 vs 4 (κ = 0.00). There was fair agreement between reader 1 vs 3 (κ = 0.37), reader 1 vs 4 (κ = 0.38), and reader 3 vs 4 (κ = 0.37) (Table 1 and Table 2).
Conclusions :
Overall, percent agreement between readers and F2F exam ranged from ~30-46%, suggesting that the Eye911 protocol is feasible with moderate diagnostic accuracy. Surprisingly, the SL video did not increase percent agreement. There was also a wide range of agreement between the readers during the pre and post-SL video period ranging from none (κ < 0) to fair (κ = 0.21-0.40). These findings suggest that the protocol requires further study. Results may improve as more participants are recruited.
This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.