BLOG: How the IRIS Registry, Verana Health helped craft research in geographic atrophy
A handful of studies at the Association for Research in Vision and Ophthalmology 2022 annual meeting demonstrated the power of the American Academy of Ophthalmology IRIS Registry.
The IRIS (Intelligent Research in Sight) Registry, powered by Verana Health’s VeraQ population health data engine, serves as a research tool for those seeking to craft a broader and deeper understanding of the current conditions in geographic atrophy (GA). The studies are also valuable for those who would like to launch GA screening in the future. Let’s examine those studies here.
Machine learning for detecting GA
A poster by Chu and colleagues detailed the development of a machine learning model used to automatically detect GA in fundus autofluorescence and infrared reflectance images. De-identified images were taken from the IRIS Registry from two large retina practices in the United States. Each patient was allowed only one image per modality for the assessment; the fundus autofluorescence and infrared reflectance images that were determined to have the highest image quality were included.
A pair of experienced human graders classified images as showing evidence or not showing evidence of GA. Approximately 300 images were used to train the machine learning’s deep learning algorithm, and images from a separate 74 patients were used to test it.
The research team determined that the machine learning model’s sensitivity was 0.88, its specificity was 0.94, and its accuracy was 0.91 for detecting any evidence of GA. This machine learning model (or one like it) could be used for screening eligible patients for trials, identifying GA in routine practice or screening the general population for evidence of early GA.
Natural history of GA
Rahimy and colleagues leveraged data from the IRIS Registry to identify approximately 22,000 real-world patients who had bilateral GA (cohort 1) and 15,000 real-world patients who had GA in one eye and choroidal neovascularization in the fellow eye (cohort 2) who had been followed for at least 36 months and fit the study’s inclusion criteria, which included no history of choroidal neovascularization in an eye with GA and no retinal disease history in either eye. Further, each cohort was subtyped as having subfoveal or nonsubfoveal disease.
Mean baseline visual acuity was better in study eyes with nonsubfoveal GA than in eyes with subfoveal GA, with a difference of approximately five letters and nine letters in favor of the nonsubfoveal eyes in cohorts 1 and 2, respectively. Mean visual acuity letter loss from baseline ranged from approximately eight to 10 letters in both cohorts.
Patients with good vision at baseline (ie, 20/40 or better) and nonsubfoveal disease in cohorts 1 and 2, respectively, lost a mean of mean 10 and eight letters from baseline; eyes with subfoveal GA lost approximately nine and 10 letters, respectively. Among those with poor vision at baseline (ie, vision from less than 20/40 to 20/100) and NSF disease, visual acuity loss from baseline was 12 and 11 letters in cohorts 1 and 2, and among those with subfoveal disease, loss from baseline was 14 and 13 letters, respectively.
This type of long-term natural history data in GA will be critical as researchers, drug developers and regulators consider how to evaluate the impact of the drug. If a drug is approved by the FDA for the treatment of GA, these data may be among the last preapproval natural history studies of long-term GA.
Vision-related quality of life in health records of patients with GA
Borkar and colleagues analyzed the unstructured notes sections of the electronic health records housed in the IRIS Registry of patients with GA to determine the volume and utility of vision-related quality of life data contained in real-world patient medical records. Researchers searched for terms linked to patient-reported outcomes, such as reading, driving, low vision, anxiety, depression, mobility, independence and disability.
The research team found that functional outcomes were rarely mentioned in EHR notes sections. Comments on a patient’s ability to read or drive, two activities commonly affected by GA, were found in less than 10% of patient records. The term low vision was found in less than 3% of notes; terms related to mood, such as anxiety and depression, were found in less than 2% of EHRs. Perhaps surprisingly, the terms mobility, independence and disability were not found in any records.
The research team concluded that, given the dearth of documentation of vision-related quality of life outcomes, EHR notes data were insufficient for stand-alone analysis of patient-reported outcomes.
What’s next?
Future conferences will continue to showcase studies leveraging data from the IRIS Registry. Keep an eye out for real-world studies with robust data sets; they might continue to harness the power of the IRIS Registry to study an expanded spectrum of diseases and novel clinical research questions.
References:
- Chu Z, et al. Automated machine learning for diagnosis of geographic atrophy using real-world fundus autofluorescence and infrared reflectance images. Presented at: Association for Research in Vision and Ophthalmology meeting; May 1-4, 2022; Denver.
- Leng MD, et al. Characterizing real-world functional outcomes in patients with geographic atrophy: an IRIS Registry analysis. Presented at: Association for Research in Vision and Ophthalmology meeting; May 1-4, 2022; Denver.
- Rahimy E, et al. Retrospective, real-world analysis of patients with geographic atrophy (GA) secondary to age-related macular degeneration followed up for 3 years. Presented at: Association for Research in Vision and Ophthalmology meeting; May 1-4, 2022; Denver.
Collapse
Read more about