Geographic atrophy diagnosis in the IRIS® Registry: a comparison between images and ICD-10 codes

Abstract

Purpose : To assess the accuracy of clinically documented geographic atrophy (GA) diagnosis and subfoveal involvement status using real-world imaging and electronic health record data in the American Academy of Ophthalmology IRIS® Registry (Intelligent Research in Sight).

Methods : The IRIS Registry is the nation’s first comprehensive ophthalmic clinical database. A de-identified ophthalmic imaging dataset contributed by 3 practices was curated and linked to the IRIS Registry. A subset of patients with dry age-related macular degeneration (AMD) identified through ICD-10 codes from 2016 to 2022 was sampled and labeled for the following categories: GA without subfoveal involvement, GA with subfoveal involvement, and no GA. Fundus autofluorescence (FAF), infrared reflectance (IR) and optical coherence tomography (OCT) images were used for labeling by two fellowship-trained retina specialists. A three round consensus workflow was implemented to ensure both graders agreed on all labels, with the center point of the fovea being involved as the definition of subfoveal involvement. With the image-derived ground truth, three comparisons were made: GA vs non-GA, GA without subfoveal involvement vs non-GA and GA with subfoveal involvement, and GA with subfoveal involvement vs non-GA and GA without subfoveal involvement using the ICD-10 code associated with the clinical encounter.

Results : 207 sets of FAF, IR and OCT images from 207 patients were labeled (Figure 1). Using images as the ground truth, clinical codes for GA vs non-GA resulted in a sensitivity of 82%, a specificity of 88% and an accuracy of 85%. After excluding non-GA images based on image labels, 107 patients were included in further analyses. Clinical codes for GA with subfoveal involvement resulted in an accuracy of 62% and clinical codes for GA without subfoveal involvement resulted in an accuracy of 63%. Detailed metrics and confusion matrices for all analyses are presented in Figure 2.

Conclusions : This study shows that clinically documented GA diagnosis codes are relatively accurate compared to image-derived ground truth while the clinically documented specific subfoveal involvement status is less accurate. Real-world studies should consider employing additional parameters such as requiring two of the same code within six months, removing patients with conflicting codes, as well as including additional imaging data when available.

This abstract was presented at the 2023 ARVO Imaging in the Eye Conference, held in New Orleans, LA, April 21-22, 2023.