Contact Us Menu
Healthcare Providers Life Sciences Research

Unlocking the Promise of Ophthalmic Images as a Real-World Data Source

Author: Sonya Li, Strategic Initiatives and Partnerships, Verana Health; Zhongdi Chu, PhD, MSc, Quantitative Sciences, Verana Health
September 2022
Back

Imaging is an integral component of ophthalmic clinical practice. Beyond offering signposts for diagnosis, ophthalmic images supply help to ophthalmologists and researchers to ascribe observed functional impairments to structural changes in the eye. Advances in imaging have enabled ophthalmologists to detect ocular diseases1 earlier and with more precision, to track disease progression,2 and to evaluate responses to treatments.3 Already used as a method for clinical trial endpoints,4 ophthalmic images are a valuable source of healthcare data. 

Yet, the value of ophthalmic images as a real-world data (RWD) source remains underutilized. Real-world insights require massive de-identified population databases, and while such a database exists for electronic health record (EHR) data — the American Academy of Ophthalmology (Academy) IRIS Registry (Intelligent Research in Sight) — it has lacked image data. Linking image data to EHR data has the possibility to enable new data-driven insights to drive future care and innovations in ophthalmology. 

In the years to come, harmonizing patient imaging reports with patient profiles will usher in a new era of real-world data–driven insight.

David W. Parke II, MD, Executive Chairman of Verana Health

Imaging as Untapped Real-world Data

Ophthalmologists often focus on two types of outcomes: functional outcomes and anatomical/structural outcomes. The former describes the function of the eye — its area of vision (visual field), its ability to perceive details (visual acuity), etc. These items are routinely tested as they impact quality of life, and therefore are generally documented in the EHR. The latter is the domain of ophthalmic imaging. Both outcomes contribute to the full picture of health captured by ophthalmology practices.5   

The Academy launched a comprehensive clinical disease registry, IRIS Registry, in 2014 to help ophthalmology practices improve the quality of care they deliver as well as meet federal reporting requirements. The IRIS Registry—the largest specialty society clinical data registry in all of medicine—is comprised of EHR data from more than 75 million de-identified patients across over 60 EHR systems and 16,000 contributing clinicians. Ophthalmology practices and researchers alike analyze this data to garner insights on care and therapeutics, helping to improve eye care across the U.S. While valuable in powering real-world evidence, EHR data alone often paints only part of the picture. 

The other part is ophthalmic imaging. Imaging allows providers to see the shape, size, and depth of structures within the eye. Observations of features such as tissue thickness, fluid build-up, and lesion size can help provide objective, quantifiable data to support diagnosis and treatment decisions.6 This data complements the functional outcomes commonly captured by EHR with structural correlates. 

Given their complementarity, the value of linking the two data sources is obvious: when images are connected with the clinical information housed in EHRs, we can better infer real-world behaviors such as the rationale for treatment or the pathology of disease. 

Yet linking image and EHR data is easier said than done. 

There are a half-dozen imaging modalities available to ophthalmologists, and each one tells its own story about the health of the eye. Harmonizing image data is difficult and requires the collaboration of various stakeholders such as medical device companies, EHR companies, and healthcare providers. Even within a single modality, images may originate from different vendors, devices, and scanning protocols. A key step to linking image and EHR data is standardizing ophthalmic images to the internationally accepted Digital Imaging and Communications in Medicine (DICOM) format.9 Standardization would allow ophthalmic images to be manipulated at scale.  

Siloing is also an obstacle. Many ophthalmic images exist only in local systems, lacking duplication in the cloud. Ophthalmic images often need to be aggregated into a data lake before they can be leveraged to improve eye care across ophthalmology practices. As its steward, Verana Health is actively working to link image data that Academy members contribute to the IRIS Registry, and deepen its value through its VeraQ™ population health data engine where the images are de-identified, tokenized, harmonized, curated, and linked to EHR data.  

The Power of Imaging to Drive Real-world Research

Geographic Atrophy

The relationship between imaging and patient treatment can be clearly illustrated in geographic atrophy (GA). GA, estimated to affect 5 million people worldwide,10 describes the atrophy of the macular in late-stage age-related macular degeneration (AMD). This degenerative process eventually leads to permanent vision loss and blindness. Because visual acuity impairment emerges late in GA, assessment by imaging is essential. Several imaging modalities can be used to characterize GA, including color fundus imaging, FAF imaging, and OCT. These methods provide ophthalmologists with detailed retinal images that can be used to diagnose GA and monitor its progression in clinical trials and in the real world.11 

At Verana Health, we recently developed a machine learning-aided algorithm that can be used to help confirm GA in dry AMD eyes using images collected in routine practice. The pipeline was trained and tested on a de-identified clinical imaging dataset obtained from 2 large retina practices. The dataset contains 2,019,625 unique ocular images from 228,739 unique visits of 49,372 unique patients and is linked to EHR data in the IRIS Registry. For the pipeline, we focused on FAF and IR images from this dataset, removing low-quality and duplicate images. The machine learning model achieved an accuracy of ≥0.91 for training, validation, and testing, meaning the model correctly confirmed 9 out of every 10 GA diagnoses.12  

Models like ours have the potential to be deployed in clinical trial settings to screen patients for GA and in the real world to monitor future treatment effectiveness. In the future — when more images are accessible to machine learning — similar models could be leveraged to predict future disease progression in ophthalmology patients.

Macular Edema

Macular edema is another condition in which image data could play an outsized role in improving care. Macular edema occurs when fluid buildup causes the macula to swell, resulting in blurry vision. It is observed in patients with wet AMD, diabetic macular edema, uveitic macular edema, retinal vein occlusion, retinitis pigmentosa, and sometimes it could be caused by eye surgery or certain medications. OCT imaging has become the gold standard13 in diagnosing macular edema.14 Fortunately, effective treatments are available in the form of anti-VEGF injections, but the treatment effects of different types of anti-VEGF drugs vary from patient to patient.15 

Recent studies16 have indicated the type of fluid, subretinal fluid or intraretinal fluid could play a role in predicting patient outcomes with anti-VEGF treatments. Such functional patient outcomes are commonly stored in EHR in the form of visual acuity. The types of fluid and the volume of fluid are not generally available in EHR, but could be extracted and quantified with ophthalmic images, namely OCT images. In this case, the true potential of ophthalmology RWD would be unleashed, as images could provide the objective structural outcomes in macular edema patients and these structural outcomes could potentially predict future functional outcomes like visual acuity post anti-VEGF treatment. 

Unlocking the Promise of Imaging Data Linked with Clinical Data

The examples here are but a few that would benefit from the linking and use of real-world imaging data. In reality, many with a stake in ophthalmology stands to benefit. Clinician-scientists could gain a deeper understanding of trends across ophthalmology practices, pharmaceutical companies could be able to segment patient populations further to inform and facilitate effective therapeutic development, individual practices could be able to improve their outcomes, and — most importantly — patients could receive more efficient care. 

Learn more about how Verana Health is curating and linking imaging and clinical data by meeting with us.

Sources for this blog post can be found here.