Novel Retinal Biomarkers for Hypertension and Cardiovascular Disease

The National Heart, Lung, and Blood Institute (NHLBI), of the National Institutes of Health (NIH), virtually convened a workshop of experts in the fields of hypertension, vascular biology, heart disease, ophthalmology, neurology, imaging, data science, and bioethics on October 13-14, 2022. The goals of the workshop were: (i) to review the current state of science and knowledge about retinal biomarkers obtained from retinal imaging, and their association with hypertension, heart disease, peripheral arterial disease (PAD), and vascular dementia and (ii) to identify research gaps and opportunities which are inhibiting progress in understanding these important connections.

The working group is responsive to NHLBI Strategic Vision Objectives 1-5.

Background
Systemic hypertension is a major risk factor for cardiovascular disease (CVD) and a potential risk factor for vascular dementia, affecting the structure and function of the microcirculation including the retinal vasculature. The presence of high blood pressure- or high blood sugar-induced microvascular alterations can provide prognostic information on the risk of CVD, congestive heart failure, end-organ damage and may be a warning sign for more widespread systemic morbidity. Emerging evidence suggests that there is a close relationship between retinal vessel morphometry and high blood pressure. However, other than some observational association studies, little is known about the mechanisms of retinal vascular changes due to high blood pressure. Retinal biomarkers such as ocular imaging may offer novel pathways for improving the ability to diagnose, evaluate prognosis, and manage hypertension, heart failure, PAD, and vascular dementia. Compared to cardiovascular imaging, ocular imaging is easier to perform, less expensive, non-invasive, and can provide high-resolution images of retinal blood vessels. Moreover, retinal vascular changes can be an indication and risk factor for cardiotoxicity due to drugs or environmental exposure, an important scientific area within NHLBI mission. Finally, obtaining retinal images of a large population is not only readily feasible, but it is also likely that existing data sets, such as ARIC (Atherosclerosis Risk in Communities Study), ACCORD (Action to Control Cardiovascular Risk in Diabetes) and MESA (Multi-ethnic Study of Atherosclerosis), may be utilized for this new area of research. Recent advancements in applying machine learning (ML) and artificial intelligence (AI) techniques to retinal image analysis may provide a significant tool to improve the diagnosis, prognosis, and management of hypertension, CVD, PAD, and vascular dementia in clinical practice.

Discussions
The co-chairs and planning committee identified national and international scientists with expertise in retinal biomarkers and their application in hypertension, heart disease, PAD, and vascular dementia. Experts in bioethics were also invited to discuss the confidentiality and privacy issues when using clinical retinal imaging data sets.

Six plenary sessions were followed by a final group discussion. Six session topics included: (i) Retinal Vascular Abnormalities in Human Diseases; (ii) Retinal Biomarkers for Hypertension; (iii) Privacy and Bioethical Issues in Retinal Biomarkers; (iv) Retinal Biomarkers for Heart Disease; (v) Retinal Biomarkers for Peripheral Vascular Disease; and (vi) Retinal Biomarkers for Vascular Dementia.

Research Gaps, Barriers and Opportunities
The workshop participants identified the following research gaps and barriers, as well as opportunities for future research.

Gaps and Barriers
Four major areas were identified.

• Study of the inherent changes in the retina regarding physiologic aging or disease state:
  • The boundary between physiological aging and pathological changes in diseased retina has not yet been defined although it may represent a continuum given the sensitivity of some vascular metrics. 
  • The extent of changes in general retinal vascular responses along with specific mechanism and individual prognosis remains unknown.
  • It is unknown how potential markers co-vary; i.e., Can retinal vascular wall thickness be used as a surrogate for vascular regulation? Can vascular regulation be used as a surrogate for cell loss and basement membrane thickness? Do different retinal metrics provide independent information?
  • The utility of applying retinal measures to prevention needs to be established; i.e., Do microvascular changes during pharmacological, lifestyle, and nutritional interventions relate to clinical outcomes? Could readily visualized and documented changes motivate patients to be more adherent with their medications?
  • It is unknown whether modifiable retinal metrics represent modifiable risk for hypertension, CVD, PAD, or vascular dementia.
  • Associations between arterial wall stiffness and retinal measures are complex and non-linear, and macro-microcirculation crosstalk adds to the complexity.
  • Limitations inherent in vascular dementia and mixed pathology complicate the study of the associations between retinal vascular changes and dementia.
  • The role of rheological factors (i.e., viscosity) on blood flow and how they impact imaging measured such as Optical Coherence Tomography Angiography (OCTA) or flowmetry need to be better understood.
  • The dependence of retinal structural measures such as wall-to-lumen ratio (WLR) and arterio-venule ratio (AVR) from functional measures (i.e., flicker light-induced vasodilation) needs to be better understood.

• Ability to build large data sets:
  • More standardized data sets of well documented cases including acknowledged primary risk markers with ethnicity and gender are needed.
  • Reproducibility and data validation studies are currently lacking.
  • Large-scale longitudinal studies using high-resolution imaging techniques such as OCTA are needed. Most current studies are cross-sectional with small sample sizes and lack racial and ethnic diversity which limits generalizability.
  • The accessibility of imaging technologies may need to be expanded to wider and more diverse populations in order to help reduce health disparities.
  • Issues of data set availabilities and associated privacy issues need to be resolved since retinal scans are biometric (i.e., they are unique identifiers to each patient).

• Need for large collaborative efforts to connect the different disciplines:
  • Collaborative networks are needed to pool matching cohorts for replication studies.
  • More inter-disciplinary collaboration and communication (e.g., between ophthalmologists and cardiologists or neurologists) are needed to better understand the sub-clinical phenotypes to be studied and to improve the diagnostic and therapeutic targeting for individual patients using retinal biomarkers.
  • A central registry is needed where researchers can collaborate to build more powerful tools.

• Need for improvement of current retinal image analysis technology and process of deployment of the AI algorithms (which may need a FDA-approval):
  • Standardizations of OCTA imaging, image analysis, and documentation of techniques in the form of metadata are needed for better interpretation of retinal images.
  • Harmonization between imaging protocols and interoperability between devices are needed.
  • Lack of standardization in arterial stiffness measures and device algorithms limits the ability to combine data across studies; a unifying model of the relation between these measures is needed.
  • For the most advanced forms of imaging, the instrumentation for retinal imaging studies is still mostly custom designed and is not ready for large data sets; there is a need for extensible approaches to building the data sets to incorporate technological advances.
  • Identifying clearer contexts for AI use is needed to focus data scientists on areas with unmet needs.
  • Clinically useful models (AI/ML/Deep learning vs. standard measurements and statistical models) need to be established to identify the retinal measures or combination of measures that best predict events.

Opportunities For Future Research

• Identification and characterization of retinal measurements and biomarkers that can be assessed by AI:
  • Explore the association of retinal microvascular disease (or changes) with hypertension, CVD,  PAD, and vascular dementia pathogenesis as a direct link vs. as a marker for underlying causes.
  • Characterize Retinal Ischemic Perivascular Lesions (RIPLs) as a potential screening tool for CVD in different populations.
  • Test the prognostic value of RIPLs in patients with known CVD.
  • Characterize retinal vessel diameters and flicker light-induced dilation as potential screening tools for CVD in different populations.
  • Design a feasibility trial using retinal vasculometry as a vascular health marker that can be integrated into primary health care for wider population capture.
  • Follow ocular microvascular changes in pharmacological CVD or heart failure trials to determine those most closely associated with outcomes.
  • Explore the possibility of subclinical changes in the retinal microvasculature before they become apparent in conventional images.
  • Evaluate the association of arterial stiffness with novel retinal biomarkers that would allow for a deeper understanding of retinal vessel structure and function (e.g., retinal vessel perfusion and blood flow).
  • Characterize the chronology of retinal microvascular disease relative to hypertension, CVD, PAD, cerebrovascular changes, cognitive decline, and dementia.
  • Identify ocular biomarkers for Alzheimer’s Disease and vascular dementia using advanced retinal imaging techniques such as ultra-wide field retinal images, DARC (Detection of Apoptosing Retinal Cells), and amyloid beta imaging.

• Development of various data sets that can be shared among researchers:
  • Develop longitudinal studies (including populations at an earlier age before onset of disease) to evaluate the temporality of associations between CVD, PAD or dementia and retinal vessel changes.
  • Apply natural language processing tools to interrogate large data sets, e.g., electronic medical records.
  • Create synthetic data sets (e.g., mechanistic model of disease/image formation) that don’t require human subjects.
  • Explore clinical trials comparing standard of care alone to retinal imaging plus standard of care with respect to risk stratification and prediction of CVD-related events and death.

• Development of new AI-based technologies, validation, and deployment of AI use:
  • Develop device-agnostic algorithms to detect RIPLs and to assess retinal vessel diameters and function automatically utilizing existing databases.
  • Explore retinal imaging technology for assessing cardiovascular pathophysiology and integrative systems physiology (e.g., retina-heart and heart-brain interactions).
  • Develop a generic automated imaging system or model that would enable a more homogenous approach for data quality and analysis.
  • Apply AI/ML to retinal Optical Coherence Tomography (OCT) data sets, focusing on functional measurements, to improve sensitivity to small changes.
  • Apply hybrid learning strategies that can combine mechanistic insights with real world data.
  • Improve ocular imaging quality in user-friendly devices that are currently patient and operator-dependent and therefore not readily ported to non-eye care environments.

Publication Plans
The workshop participants plan to prepare a manuscript for publication in a peer-reviewed journal.   
NHLBI Contacts
Young S Oh, PhD, Division of Cardiovascular Sciences (DCVS), NHLBI
Email: yoh@nhlbi.nih.gov

Workshop Invited Participants

Co-Chairs
Stephen A Burns, PhD, Indiana University
Emily Y Chew, MD, National Eye Institute, NIH

Speakers and Moderators

• Alison Abraham, PhD, University of Colorado
• Aaron D Aguirre, MD, PhD, Harvard Medical School
• Mathieu F Bakhoum, MD, PhD, Yale University
• Joshua A Beckman, MD, Vanderbilt University
• Yuen Ping Toco Chui, PhD, New York Eye and Ear Infirmary of Mount Sinai 
• Robert P Finger, MD, PhD, University of Born, Germany
• Alejandro F Frangi, PhD, University of Leeds, UK
• Rebecca F Gottesman, MD, PhD, National Institute of Neurological Disorders and Stroke, NIH
• Maria Grant, MD, University of Alabama at Birmingham
• Henner Hanssen, MD, University of Basel, Switzerland
• Christopher M Kramer, MD, University of Virginia
• Cecilia S Lee, MD, University of Washington
• Michelle Meyer, PhD, University of North Carolina
• Damiano Rizzoni, MD, University of Brescia, Italy
• Alicja R Rudnicka, PhD, St. George’s University of London, UK
• Joel S Schuman, MD, New York University
• Sara B Seidelmann, MD, PhD, Columbia College of Physicians & Surgeons
• Wilson Tang, MD, Cleveland Clinic
• Holly A Taylor, PhD, MPH, Clinical Center, NIH

NIH Workshop Planning Committee

• Bishow Adhikari, PhD, Heart Failure and Arrhythmias Branch, DCVS, NHLBI
• Narasimhan Danthi, PhD, Advanced Technologies and Surgery Branch, DCVS, NHLBI
• Yuling Hong, MD, PhD, Epidemiology Branch, DCVS, NHLBI
• Young Oh, PhD, Vascular Biology and Hypertension Branch, DCVS, NHLBI
• Diane Reid, MD, Vascular Biology and Hypertension Branch, DCVS, NHLBI
• Grace L Shen, PhD, Retinal Diseases Program, National Eye Institute, NIH